WO2014188475A1

WO2014188475A1 - Information processing system

Info

Publication number: WO2014188475A1
Application number: PCT/JP2013/063885
Authority: WO
Inventors: 篤志宮本; 純一宮越; 幸二福田
Original assignee: 株式会社日立製作所
Priority date: 2013-05-20
Filing date: 2013-05-20
Publication date: 2014-11-27

Abstract

With a conventional simulation method there is a problem in that in order to analyze simulations with respect to multiple scenarios a user must execute a simulation for each scenario, and compare the respective logs to predict relationships, which is extremely inefficient. In order to solve this problem, this information processing system has a user interface that receives the selections of a first and a second case, as subjects, and a storage that stores time-series data and information about the relationships between the sources of the time-series data. Events corresponding to the first and the second cases are extracted from the time series data, and on the basis of the extracted events and the relationship information, stochastic process graphs for the first and the second cases are generated, and scenarios for which there is a match or a similarity between the generated stochastic process graphs are extracted.

Description

Information processing system

The present invention relates to an information processing system, and more particularly, to a technique for simulating a plurality of scenarios.

Social simulation is very important in providing appropriate decision support in various scenes. In this type of simulation, generally, a technique that handles the interaction behavior of a micro model such as an agent base is often used. While unexpected scenarios that cannot be expressed with a macro model can be created, in order to obtain effective results (useful laws), it is necessary to analyze a large amount of simulation log data from multiple scenarios.

In order to convince the user in the analysis work, not only the simulation results are shown, but the relationship that the user wants to observe is related to "what caused the event?" It is important to show. However, this work was only verified by the user on a log basis, and the log was visualized and displayed on the screen, and the analysis was performed while directly confirming the operation and predicting it. In particular, analysis of the cause investigation of multiple scenarios and complex problems with multiple factors is a big issue, and as a result, it is easy to have specialized knowledge and give users new discoveries and awareness including convincing and unexpected. It wasn't.

As a simulation causal analysis device, there is a technique described in Patent Document 1. Patent Document 1 discloses a simulation apparatus that can execute a causal relationship analysis of simulation results in accordance with the purpose of the simulation while limiting the log size. An apparatus disclosed in Patent Literature 1 includes an event table holding unit that holds in advance an event table that associates an event that occurs for each evaluation object with a variable that changes when the event occurs, and logs according to an evaluation item. A log output setting unit that holds an event acquisition rule that prescribes an event and a variable to be output in advance, and generates a specific event table for performing log output necessary for the analysis of the causal relationship based on the event acquisition rule and the evaluation item; Scenario execution means for executing a simulation and log recording means for storing data defined in a specific event table as a log output when the simulation is executed. It is described that a log acquisition function and a causal relationship analysis function can be realized in accordance with the purpose of simulation by utilizing an event table and event acquisition rules stored in advance.

Japanese Patent Laid-Open No. 2007-220078

However, the conventional techniques have the following problems.

The apparatus of Patent Document 1 enables analysis of the causal relationship of simulation results, but basically targets analysis of simulation for one scenario, and cannot be used for analysis of simulation for a plurality of scenarios. When analyzing simulations for multiple scenarios, it is necessary to apply a series of flows repeatedly. However, since there is no means for associating simulation logs for multiple scenarios and using them for analysis, the user executes simulations for each scenario and compares the logs by comparing each log as in the conventional general simulation method. There is a problem that it is necessary to predict and is very inefficient.

The present invention has been made to solve the above-described problems, and aims to improve the analysis efficiency of simulation for a plurality of scenarios.

An information processing system according to the present invention includes a user interface that accepts selection of first and second target cases, and a storage that stores time-series data and information on the relationship between the sources of the time-series data. Then, an event corresponding to each of the first and second cases is extracted from the time series data, and the stochastic process for each of the first and second cases is extracted based on the extracted event and the relationship information. The above-mentioned problem is solved by generating a graph and extracting a scenario that matches or resembles between the generated stochastic process graphs.

According to the present invention, it is possible to improve the analysis efficiency of simulation for a plurality of scenarios.

It is a block diagram of the information processing system which is an Example of this invention. It is a flowchart of operation | movement of the graph process of an information processing system. It is a figure which shows the example of time series data. It is a figure which shows the example of time series data. It is a figure which shows the example of relationship data. It is a figure which shows the example of relationship data. It is a figure which shows the example of sensor arrangement | positioning on the road used as the origin of the example of relationship data. It is a flowchart of a case generation process. It is a figure which shows the specific example of event extraction. It is a figure which shows the specific example of event extraction. It is a figure which shows the example of example graph data. It is a figure which shows the specific example of a case production | generation process. It is a figure which shows the example of example graph data. It is a flowchart of a scenario generation process. It is a figure which shows the example of example graph data. It is a figure which shows the example of the graph description of event graph data. It is a figure which shows the example of a similar vertex list | wrist. It is a figure which shows the example of the graph description of scenario graph data. It is a figure which shows the example of scenario graph data. It is a figure which shows the structure of the information processing system which is an Example of this invention, and a user terminal. It is a schematic diagram which shows the concept that the vertex is integrated about the inputted example graph data and a weighted probability process graph is created. It is a functional block diagram of the part which inputs / outputs information between a scenario graph analyzer and a scenario graph analyzer in a user I / F and a user. It is a schematic diagram of the process for demonstrating operation | movement of a 1st scenario graph process part. It is a figure which shows the example of the table of the correspondence of 1st scenario graph (stochastic process graph) data and 2nd scenario graph (stochastic process graph) data. It is a figure which shows an example of the process which calculates | requires a response | compatibility of 2nd scenario graph (stochastic process graph) data with respect to the scenario candidate designated by the user in 1st scenario graph (stochastic process graph) data. FIG. 6 is a flowchart for explaining the operation of a scenario graph analyzer 124. It is an example of the process which calculates | requires the response | compatibility of the scenario candidate in 2nd scenario graph (stochastic process graph) data with respect to the scenario candidate designated by the user. It is a schematic diagram for demonstrating operation | movement of a 2nd scenario graph process part. It is the figure which showed the example of the chain bankruptcy simulation. It is the figure which showed the example of the chain bankruptcy simulation. It is the figure which showed the example of the chain bankruptcy simulation. It is the figure which showed the example of the chain bankruptcy simulation.

Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, functionally identical elements are denoted by the same numbers.

An embodiment of the information processing system of the present invention will be described below. FIG. 1 shows a block diagram of an information processing system 101 of this embodiment.

The information processing system 101 includes a graph processing device 110, a database 125, and a case database 126. The graph processing apparatus 110 includes an object selector 120, a case generator 122, a scenario graph generator 123, and a scenario graph analyzer 124. The graph processing apparatus 110 receives the time series data 107 and the relationship data 108 and outputs the scenario graph data 105 and the scenario analysis result 102. The scenario analysis result 102 is output to a user interface (I / F) 127 for presentation to the user.

In the present embodiment, as an example, a graph process for a traffic jam will be described as a processing target. Here, an example of the time-series data 107 is shown in FIGS. 3A and 3B. The time series data 107 is stored in the database 125. FIG. 3A shows time-series data 300 of the number of passing vehicles by time, and FIG. 3B shows time-series data 301 of average vehicle speed corresponding to the data 300. The time series data 107 is acquired for each sensor arranged on the road, and the number of vehicles passing through the road and the average vehicle speed are acquired. These “number of passing vehicles” and “average vehicle speed” are called feature information.

Examples of relationship data 108 are shown in FIGS. 4A and 4B. FIG. 4C shows the arrangement of sensors with respect to the road that is the basis of the relationship data 108 shown in FIGS. 4A and 4B.

The relationship data 108 is data representing the relationship between the sources of the time-series data, and in this embodiment is data representing the relationship between the sensors. The relationship data 108 is stored in the database 125. When the sensors 1 to 7 arranged in a distributed manner on the road including the intersection 402 shown in FIG. 4C are expressed as a graph by connecting each sensor as an apex and connecting adjacent sensors with edges, a graph 400 in FIG. 4A is obtained. Further, the relationship data 401 between the sensors in FIG. 4B is the data 400 of the graph 400 in FIG. 4A. In the relationship data 401 between sensors, “1” indicates that there is a relationship between sensors, and “0” indicates that there is no relationship. In the present embodiment, the relationship represents a connection by roads. For example, the sensor 1 arranged at the intersection 402 of the road 1 with the road 2 and the sensor 2 arranged at the road 2 have the road 1 and the road 2 at the intersection. There is a relationship because they are connected and the sensors are connected adjacent to each other on the graph 400, and the sensor 2 and the sensor 3 are not connected adjacent to each other on the graph 400, so there is no relationship. In this example, a traffic jam is used as an example, and the relationship is a spatial arrangement relationship of sensor groups. However, as another example, a company bankruptcy is used as an example, and the relationship is a business relationship between companies. You can also.

FIG. 11 shows the hardware configuration of the information processing system 101 and the user terminal 1101. The information processing system 101 and the user terminal 1101 are wirelessly connected via the network 1102 and the wireless base station 1103. In this embodiment, an example of wireless connection is shown, but it can also be implemented by wired connection. A sensor group (

sensors

1, 2,...) 1104 arranged on the road according to the present embodiment is connected to the information processing system 101 via the network 1102. As a result, the information processing system 101 can collect information from the sensor group 1104 that is a source of time-series data, create time-series data 107, and store it in the database 125. Further, the time series data 107 stored in the database 125 can be not only data measured by a sensor or the like, but also a simulation result by a computer such as an agent-based simulation. The information processing system 101 includes a simulator 128 for performing computer simulation such as agent-based simulation. The simulator 128 stores the simulation result in the case DB 126. Further, the simulator 128 may be operated from the user I / F 127.

The information processing system 101 includes a central processing unit (CPU) 1105, a main memory 1106, a storage 1107, and a network interface (I / F) 1108. The object selector 120, the case generator 122, the scenario graph generator 123, and the scenario graph analyzer 124 of the information processing system 101 are stored as programs in the storage 1107 and executed by the CPU 1105 and the main memory 1106.

User terminal 1101 is a mobile terminal such as a tablet terminal, for example, and implements user I / F 127. The user terminal 1101 includes a CPU 1109, a main memory 1110, a storage 1111, a touch panel 1112, and a network I / F 1113. The user terminal 1101 receives input from the user such as selection of a target case on the touch panel 1112, and displays the scenario analysis result 102 and the like on the liquid crystal screen of the touch panel 1112.

Next, graph processing by the information processing system 101 of this embodiment will be described. FIG. 2 shows a flowchart of the graph processing operation of the information processing system 101.

First, in the target selection step 201, the user selects a first target case and a second target case. In the present example, the user selects “congestion” as the second target candidate among the target candidates presented from the target selector 120 via the touch panel 1112 of the user terminal 1101, and sets “congestion on the road” as the second target. Assume that each case has been selected as a target case. The object selector 120 that has received that “congestion” and “road congestion” are selected as the objects, the definition “congestion” and the definition of “road congestion” are defined as the object definition 103 in the case generator 122. Send. In this embodiment, the definition of “traffic jam” is an average vehicle speed of 20 km / h or less. The definition of “road congestion” is that the number of passing vehicles per unit time is 10 or more. The target definition 103 may be stored in the storage 1107 in advance, or may be input by the user from the touch panel 1112. Further, the object selector 120 further designates the case of yesterday and the day before yesterday as the object definition 103, and transmits it to the case generator 122. The period of the cases of yesterday and yesterday may be input by the user from the touch panel 1112 or may be set in advance.

The case generator 122 that has received the target definition 103 acquires the time series data 107 and the relationship data 108 corresponding to the received target definition 103 from the database 125. In FIG. 2, this process corresponds to Step 202 and Step 203.

The case generation process 204 will be described. The case generation process 204 is executed by the case generator 122. The case generator 122 receives the target definition 103, the time series data 107, and the relationship data 108 as input, and outputs the case graph data 106 as output.

The detailed flowchart of the case generation process 204 is shown in FIG. In the loop process between the processes 501-1 to 501-2, the case generator 122 scans the time series data 107, extracts an event (congestion or congestion in this embodiment) (process 502), and extracts the extracted event. Vertex information corresponding to is generated (process 503). The case generation process 204 is performed for each of the first target case and the second target case. Hereinafter, an operation regarding “traffic jam” will be described as a representative.

FIGS. 6A, 6B, and 6C show examples of processing 502 and processing 503. FIG. The case generator 122 extracts an event that matches the target definition 103 in the time series data 301 of the average vehicle speed. Here, events 1 to 3 surrounded by ellipses 601 to 603 in FIG. 6A, which are events that meet an average speed of 20 km / h or less, are extracted. The extracted event is generated as vertex information, and is represented as a vertex of the graph 610 shown in FIG. 6B in the graph notation, and becomes an entry of the case graph data 620 shown in FIG. 6C when expressed in the data structure. Information on each vertex is associated with time series data information such as sensor identification information (ID) and sensor data acquisition time as vertex data, as in each entry of the case graph data 620, and is stored in the main memory 1106. Or stored in the storage 1107 (process 504).

Subsequently, the case generator 122 executes loop processing (processing 505-1 to 505-2) for any two vertices from the generated vertex information. Here, for the sake of explanation, it is assumed that vertices selected as two arbitrary vertices are a vertex A and a vertex B, respectively. The case generator 122 evaluates the time interval for acquiring sensor data between vertices in the process 506 from the time data held in the information of the two vertices. If the time interval (the time between the acquisition time of the sensor data of the vertex B and the acquisition time of the sensor data of the vertex A) is greater than 0 and smaller than the predetermined threshold value, the case generator 122 performs processing 507. The relationship between the two vertices is acquired from the relationship data 401. The case generator 122 evaluates the acquired relationship in the process 508, and when the relationship is “1”, that is, there is a relationship, in the process 509, between the two vertices (here, between the vertex A and the vertex B) ) Output edge information. The outputted edge information is stored as “1” in the vertex B column of the edge data of the vertex A information. On the other hand, when the condition is not satisfied (N) in any of the evaluations of the processing 507 and the processing 509, the case generator 122 does not output the edge information. Therefore, the edge information remains “0”. As described above, the case generator 122 generates edge information.

7A and 7B show processing examples of processing 505-1 to processing 505-2. Here, it is assumed that the threshold is 2, and vertex 1 (611) and vertex 2 (612) in FIG. 6C are taken as an example. Vertex 1 (611) and vertex 2 (612) have a time interval (time of vertex 2 (= 4) −time of vertex 1 (= 3) = 1) from case graph data 620, and processing 506 The evaluation formula (0 <1 <2) holds. Further, there is a relationship from the relationship data 401 between the sensors, and the evaluation formula (relationship = 1) of processing 508 is established. Accordingly, the information on the edge 1 (701) of the graph notation 700 in FIG. 7A is output between the vertex 1 and the vertex 2 as the edge data 711 in FIG. 7B. Similarly, the case generator 122 outputs information on the side 2 (702) between the vertex 1 and the vertex 3, and does not output information on the side between the vertex 2 and the vertex 3. Here, the side is a directed side having a direction with respect to a direction increasing in the time direction. When the generated edge information is shown in data notation, it becomes like each entry of the case graph data 710 in FIG. 7B.

Through the above processing, the case generator 122 receives the target definition 103, the time series data 107, and the relationship data 108, and outputs the case graph data 106.

Next, the scenario graph generation process 205 will be described. The scenario graph generation process is a process for generating a scenario graph (stochastic process graph) by integrating a plurality of entries of case graph data. The scenario graph generation processing 205 is performed for each of the first target case and the second target case. Hereinafter, the operation regarding “traffic jam” will be described as a representative.

Scenario graph generation processing 205 is executed by the scenario graph generator 123. The scenario graph generator 123 receives the example graph data 106 and outputs the scenario graph data 105.

FIG. 8 shows a flowchart of the scenario graph generation process 205. First, in process 801, the scenario graph generator 123 acquires case graph data. Thereafter, the scenario graph generator 123 executes a loop process (802-1 to 802-2) for selecting an arbitrary vertex from the case graph data (the selected vertex is a vertex A).

The scenario graph generator 123 evaluates whether or not the vertex A has already been selected as a similar vertex in the process 803. If not selected, the scenario graph generator 123 proceeds to the process 804. Subsequently, the scenario graph generator 123 executes a loop process (805-1 to 805-2) for selecting a vertex other than the vertex A from the acquired case graph data (with the selected vertex as the vertex B). In process 806, the scenario graph generator 123 calculates the similarity between the vertex A and the vertex B. The similarity is, for example, an absolute value of a difference of data such as an average speed held by each vertex as an index. In the process 807, the scenario graph generator 123 compares the calculated similarity with a predetermined threshold value and evaluates. If the evaluation formula (similarity <threshold) is satisfied, the scenario graph generator 123 proceeds to process 808. Vertex B is registered in the similar vertex list as a similar vertex of vertex A.

Specific examples of processing from processing 801 to 805-2 are shown in FIGS. 9A, 9B and 9C. As an example, the case graph data acquired by the scenario graph generator 123 in the process 801 is the case graph data 1 (900) and the case graph data 2 (920) shown in FIG. 9A. Each case graph data is specified by the target definition 103, and the case graph data 1 corresponds to yesterday's data, and the case graph data 2 corresponds to the data yesterday. The graph notations of event graph data 1 and event graph data 2 are shown in graph 910 and graph 930 in FIG. 9B, respectively.

Scenario graph generator 123 first selects vertex 1 of case graph data 1 as vertex A. Vertex A has never been selected as a similar vertex. In the loop process 805, the scenario graph generator 123 selects the vertex 2 of the case graph data 1 as the vertex B. Then, the scenario graph generator 123 calculates the similarity from the average vehicle speed of the vertex A and the vertex B (average vehicle speed of the vertex A (= 15) −average vehicle speed of the vertex B (= 10) = 5) and compares it with the threshold value. To do. Here, if the threshold is 1, the evaluation formula (5 <1) of the processing 807 is not satisfied, and the vertex B is not added to the similar vertex list. The scenario graph generator 123 returns to the processing 805-1, selects the vertex 3 of the case graph data 1 as the vertex B, and performs the same processing. Again, the evaluation formula is not satisfied, so vertex B is not added to the similar vertex list. Subsequently, similarly, the scenario graph generator 123 processes the vertex 1 of the case graph data 2. Since the vertex 1 of the case graph data 2 satisfies the evaluation formula of the process 807, it is added to the similar vertex list. By repeating this, the similar vertex list 940 shown in FIG. 9C is generated. Here, the vertex 1 of the case graph data 2 is not included in the column of the vertex A of the similar vertex list 940, but this is because the vertex 1 of the case graph data 2 is a similar vertex because the case graph data 1 was processed first. Because it was selected as.

Next, processing after step 810 will be described. In process 810, the scenario graph generator 123 defines an edge identifier and initializes it with “2”. Thereafter, the scenario graph generator 123 executes a loop process (811-1 to 811-2) for selecting a vertex from the similar vertex list for the vertex A (with the selected vertex as a similar vertex).

In the process 812, the scenario graph generator 123 acquires edge data of similar vertices, and sets the edge data initialized to “2”, with the edge data set to “1”, that is, the edges corresponding to the related vertices. Change (process 813). Then, in process 814, the scenario graph generator 123 stores the vertex data of the similar vertex and the changed edge data in the vertex A information. The scenario graph generator 123 adds 1 to the edge identifier at the end of the loop. The reason why 1 is added to the edge identifier is to avoid having duplicate values in the edge identifier, and as a modification, it can be realized by a number other than 1 or a duplicate list. By repeating the above until there are no vertices in the similar vertex list, it is possible to integrate vertices similar to the vertex A into the vertex A and distinguish the integrated side and the pre-integration side by the side identifier. . By repeating these, a plurality of entries of the case graph data can be integrated into one graph data. Here, the integrated graph data is referred to as scenario graph data.

Specific examples of processing from processing 810 to processing 811-2 are shown in FIGS. 10A and 10B. In FIG. 10, for simplicity, vertex 1 of case graph data 1 is described as vertex 11, and vertex 1 of case graph data 2 is described as vertex 21. When the vertex A is the vertex 11, the corresponding similar vertex in the similar vertex list 940 is the vertex 21. Accordingly, in process 812, the scenario graph generator obtains edge data of the vertex 21. In the edge data of the vertex 21, since the edge corresponding to the vertex 22 is “1”, the scenario graph generator 123 rewrites the edge data to the edge identifier “2” in process 813. In step 814, the scenario graph generator 123 adds the data held by the vertex 21 to the vertex 11. By repeating the above, the scenario graph data 1010 shown in FIG. 10B is generated. At this time, the graph notation of the scenario graph data 1010 can be drawn as a graph 1000 in FIG. 10A, and it can be seen that the entries of a plurality of case graph data are integrated into one. Further, in the scenario graph data 1010, the edge before integration can be distinguished from the edge identification information data 1011.

Through the above processing, the scenario graph generation processing 205 inputs the case graph data 106 and outputs the scenario graph data 105. The scenario graph data 105 is output for each of the first target case and the second target case. The scenario graph data 105 includes a stochastic process graph in which the transition probabilities are weighted because the transition probabilities increase due to the integration of the vertices when each of the integrated similar vertices has an edge to the similar vertex to be integrated separately. Become. Therefore, from the scenario graph data 105, for example, it is possible to extract a scenario that is most likely to occur, and conversely, it is possible to extract a scenario that has a very low possibility of occurrence although the possibility is not zero.

FIG. 12 is a schematic diagram showing a concept that a vertex is integrated with respect to input case graph data and a weighted probability process graph is created. Between the two case graphs 1201 on the left side of FIG. 12, the two pairs (1202, 1203) of the vertices are merged because they have similar characteristics, and weights are given to other edges as shown on the right side of FIG. A stochastic process graph 1204 with large edges is created.

Next, the scenario graph analysis process 206 will be described in detail. The scenario graph analysis process 206 is performed by the scenario graph analyzer 124 and the user I / F 127.

The scenario graph analyzer 124 is a means for obtaining the scenario analysis result 102 using the scenario graph (stochastic process graph) data 105 generated by the scenario graph generator 123 as an input. Specifically, the scenario analysis result 102 is, for example, from vertices, edges, or a series of vertices and edges between the first scenario graph data and the second scenario graph data whose target cases are different from each other. This is the correspondence of the case. A request from the user is input to the scenario graph analyzer 124 via the user I / F 127.

The scenario graph analyzer 124 extracts a scenario that corresponds between two scenario graphs having different target cases, presents the extracted scenario to the user via the user I / F 127, and the event that the user wants to evaluate Enables analysis of case causality. In addition, by changing a target case and repeating a series of operations interactively, it derives the feature information of the event to be evaluated and the main component of the case and the relationship between the feature information, and supports deeper causal relationship elucidation. It can be carried out.

FIG. 13 is a functional block diagram of a part for inputting / outputting information between the scenario graph analyzer 124 and the scenario graph analyzer 124 in the user I / F 127 and the user.

The scenario graph analyzer 124 includes a first scenario graph processing unit 1302 that performs processing on the first scenario graph (stochastic process graph) data 1301 based on the first target case, and the second target case. A second scenario graph processing unit 1304 that performs processing on the second scenario graph (stochastic process graph) data 1303 based thereon. Here, the combination of the first scenario graph data 1301 and the second scenario graph data 1303 includes, for example, the scenario graph data for the above-described case of “traffic jam” and the case for “road congestion”. Scenario graph data that is a combination of the scenario graph data, and the scenario graph data that targets “more than the predetermined number of bankruptcies” described later, and scenario graph data that targets the case where “the sum of the profits and losses of each company exceeds the predetermined loss amount” There are combinations.

The first scenario graph processing unit 1302 includes a scenario extraction unit 1305, an important situation extraction unit 1306, and a scenario selection unit 1307. The second scenario graph processing unit 1304 includes a corresponding scenario selection unit 1308, a scenario extraction unit 1309, and an important situation extraction unit 1310.

The part of the user I / F 127 that inputs / outputs information between the scenario graph analyzer 124 and the user will be described below. The user I / F 127 includes a first display unit 1311 that displays each output from the first scenario graph processing unit 1302 and a second display unit 1312 that displays each output from the second scenario graph processing unit 1304. In order to realize interactive analysis with the user 1313, a user input receiving unit 1314 that receives an input from the user 1313 is provided. The first display unit 1311, the second display unit 1312, and the user input reception unit 1314 are realized by the touch panel 1112 of the user terminal 1101.

The first scenario graph processing unit 1302 will be described below. The scenario extraction unit 1305 extracts a scenario group 1315 from the input first scenario graph (stochastic process graph) data 1301. The scenario graph data is a graph representing a state transition stochastic process in which weights are added to edges, and a new scenario can be generated by the scenario extracting unit 1305 extracting a path of a probable stochastic process. Therefore, the scenario group 1315 extracted here includes scenarios other than the cases used by the scenario graph generator 123 to generate scenario graph data.

The important aspect extraction unit 1306 extracts the important aspects of the first scenario graph data from the input first scenario graph (stochastic process graph) data 1301, and outputs the extracted important aspects information 1316. Here, the important aspect of the scenario graph data is an aspect (vertex) where the degree of the vertex and the value of PageRank are high, such as the hub vertex in the scale free graph. For example, the important aspect extraction unit 1306 extracts an important aspect by extracting vertices having an order equal to or higher than a predetermined order from vertices in the input scenario graph data. Further, for example, the important situation extraction unit 1306 extracts an important situation by extracting vertices having a PageRank value equal to or larger than a predetermined value from the vertices in the input scenario graph data. The predetermined order and value may be set in advance, or may be set by the user inputting from the user I / F 127. The important aspect appears as an important point where the case branches in the scenario graph data.

The scenario selection unit 1307 is designated by a user given via the input first scenario graph (stochastic process graph) data 1301, important phase information 1316 from the important phase extraction unit 1306, and the user I / F 127. The scenario candidate 1317 is input, and one or more scenario candidates related to the scenario candidate 1317 designated by the user are selected.

FIG. 14 is a schematic diagram of processing for explaining the operation of the first scenario graph processing unit 1302. A process 1401 for extracting a scenario group is a process realized by the scenario extraction unit 1305. Further, the process 1402 for extracting an important situation is a process realized by the important situation extraction unit 1306. Further, the process 1403 for selecting one or more scenarios based on the important situation is a process realized by the scenario selection unit 1307.

As shown in FIG. 14, a process 1401 for extracting a scenario group from the input first scenario graph (probability process graph) data 1301 is performed, and a scenario group 1315 surrounded by a dotted line is extracted. The scenario extraction unit 1305 extracts the scenario group 1315 in consideration of the probability transition of the scenario graph data. For example, a scenario that can occur with a predetermined probability or higher is extracted based on the transition probability. As a result, a scenario that may occur more than a predetermined probability is presented to the user via the user I / F 127. Conversely, by extracting a scenario that occurs with a probability equal to or lower than a predetermined probability, a scenario that may occur with a predetermined probability or less can be presented to the user via the user I / F 127. The predetermined probability may be set in advance or may be input from the user via the user I / F 127. In FIG. 14, as an example, four scenarios surrounded by dotted lines are extracted.

Next, a process 1402 for extracting important aspects is performed, and important aspects are extracted. An important aspect is, for example, a vertex having a high vertex order or PageRank value. In FIG. 14, three important aspects surrounded by an ellipse are extracted.

Then, when the scenario candidate 1317 designated by the user indicated by the alternate long and short dash line is input (arrow 1404), a process 1403 for selecting one or more scenarios based on the important situation is performed. The scenario selection unit 1307 searches for related case candidates based on important aspects included in the scenario candidates 1317 designated by the user. In FIG. 17, in addition to the scenario candidate 1317 designated by the user, one scenario candidate is newly selected, and a scenario candidate group 1405 indicated by a two-dot chain line together with the scenario candidate 1317 designated by the user is shown. ing.

Next, the second scenario graph processing unit 1304 will be described. The corresponding scenario selection unit 1308 outputs a second scenario graph (stochastic process) based on the second target case for the scenario candidate 1317 specified by the user in the input first scenario graph (probability process graph) data 1301. Graph) The correspondence of scenario candidates in the data 1303 is obtained. The correspondence of the scenario candidates between the scenario graph data is obtained, for example, by searching for the correspondence between the vertices of the scenario candidates designated by the user. Vertices can be associated, for example, by associating the vertices of the scenario graph data with the information on the vertices of the case graph data output from the case generator 122 using a table or the like. For example, the correspondence between the first scenario graph data 1301 and the second scenario graph data 1303 can be obtained by searching using the vertex identification information (ID) of the case graph data as a key.

The scenario extraction unit 1309 extracts a scenario group from the input second scenario graph (stochastic process graph) data 1303. The scenario graph data is a graph representing a stochastic process of state transition in which weights are added to edges, and a new scenario can be generated by the scenario extracting unit 1309 extracting a path of a probabilistic process. Therefore, the scenario group extracted here also includes scenarios other than the examples used by the scenario graph generator 123 to generate scenario graph data.

The important situation extraction unit 1310 extracts the important aspects of the second scenario graph (stochastic process graph) from the input second scenario graph (stochastic process graph) data 1303, and outputs the information of the extracted important aspects. . Here, the important aspect of the scenario graph data is an aspect (vertex) where the degree of the vertex and the value of PageRank are high, such as the hub vertex in the scale free graph. The important situation extraction unit 1310 extracts an important situation by, for example, extracting vertices having an order equal to or higher than a predetermined order from vertices in the input scenario graph data. For example, the important situation extraction unit 1310 extracts an important situation by extracting vertices having a PageRank value equal to or larger than a predetermined value from the vertices in the input scenario graph data. The predetermined order and value may be set in advance, or may be set by the user inputting from the user I / F 127. The important situation appears as an important point where the case branches in the stochastic process graph.

FIG. 15 shows an example of the correspondence table between the first scenario graph (stochastic process graph) data 1301 and the second scenario graph (stochastic process graph) data 1303 (table 1501, table 1502). The feature information 1 and feature information 2 of the case graph data 1503 output from the case generator 122 correspond to, for example, “average vehicle speed” in FIG. 3B and “number of vehicles on the road” in FIG. 3A, respectively. A table 1501 is a correspondence table between the case graph 1503 and the first scenario graph data 1301. A table 1502 is a correspondence table between the case graph 1503 and the second scenario graph data 1303. Note that these pieces of information do not necessarily have to be expressed in a data structure using a table, and may be expressed in, for example, a data structure such as a list, DB, or queue, or other data.

FIG. 16 shows an example of a process for obtaining the correspondence of the second scenario graph (probability process graph) data 1303 to the scenario candidate designated by the user in the first scenario graph (probability process graph) data 1301. FIG. In order to extract a matching scenario candidate from the second scenario graph data 1303 with respect to a scenario candidate specified by the user in the first scenario graph data 1301, the corresponding scenario selection unit 1308 has a corresponding vertex. And search for edges. However, as shown in FIG. 15, the way in which the vertices are integrated differs depending on the feature information, so that there is a case where the corresponding route does not exist as shown in FIG. 16. In this case, even if the corresponding vertex (vertex A, B, C, D in FIG. 16) can be obtained, the corresponding scenario candidate cannot be obtained.

Here, strict scenario matching is not important for the result, and it may be sufficient to extract similar scenarios. Therefore, the corresponding scenario selection unit 1308 further determines the correspondence of the second scenario graph data 1303 to the scenario candidate similar to the scenario candidate specified by the user in the first scenario graph data 1301. At this time, one or more scenario candidates selected by the scenario selection unit 1307 and related to the scenario candidate specified by the user are similar to the scenario candidates specified by the user in the first scenario graph data 1301. Use as a candidate.

FIG. 17 shows a flowchart for explaining the operation of the scenario graph analyzer 124. First, in step S1701, the scenario extraction unit 1305 extracts the scenario group 1315 and outputs the extracted scenario group 1315 to the user via the user I / F 127 with respect to the first scenario graph data 1301. Next, the important situation extraction unit 1306 extracts the important situation, and outputs the extracted important situation information 1316 to the user via the user I / F 127 (step S1702).

Next, in step S1703, the user selects a scenario candidate 1317 from the scenario group 1315 indicated by the user I / F 127, and inputs the selection result from the user I / F 127. In step S1704, the scenario selection unit 1307 causes the scenario candidate 1317 designated by the user to be given via the first scenario graph data 1301, the important phase information 1316 from the important phase extraction unit 1306, and the user I / F 127. As an input, one or more scenario candidates related to the scenario candidate 1317 designated by the user are selected.

In step S1705, the corresponding scenario selection unit 1308 obtains the correspondence of the scenario candidate in the second scenario graph data 1303 to the scenario candidate 1317 designated by the user, and the corresponding scenario candidate is obtained through the user I / F 127. To present. By comparing the scenario correspondence between scenario graph data with different feature information, the cause of the case or event in one feature information can be caused by another feature information. You can analyze what happened.

In step S1706, the scenario extraction unit 1309 of the second scenario graph processing unit 1304 extracts the scenario group in the second scenario graph data 1303 and presents it to the user via the user I / F 127. In step S1707, the important situation extraction unit 1310 extracts each important situation in the second scenario graph data 1303 and presents it to the user via the user I / F 127. Note that whether or not to provide step S1706 and step S1707 is arbitrary.

FIG. 18 shows an example of processing for obtaining the correspondence of the scenario candidate in the second scenario graph (stochastic process graph) data 1303 to the scenario candidate 1317 designated by the user. In the example shown in FIG. 18, from the scenario group extracted by the scenario extraction unit 1305 from the first scenario graph data 1301 in step S1701, the scenarios of vertices A, B, C, and D indicated by the user in dashed lines in step S1703. Is designated as a scenario candidate, but based on the information that the important phase extraction unit 1306 has extracted the vertex A and the vertex C as the important phase, the scenario selection unit 1307 is indicated by a dotted line in step 1704 as the related scenario candidate Select scenarios with vertices A, E, C, and D. In step S1705, the corresponding scenario selection unit 1308 selects not only the scenario candidate specified by the user but also the scenario that matches the scenario selected by the scenario selection unit 1307 from the second scenario graph data 1303. The corresponding scenario candidate indicated by the broken line is selected. Thereby, similar scenarios as well as scenarios that match the scenario candidate 1317 designated by the user can be extracted from the second scenario graph data 1303. By extracting these scenarios, for example, the user can evaluate the probability of the scenario based on the existence of a scenario corresponding to different feature information, or investigate the causal relationship between feature information. It becomes possible to improve the analysis efficiency of the simulation for the scenario.

FIG. 19 is a schematic diagram for explaining the operation of the second scenario graph processing unit 1304. In step S1705, a corresponding scenario candidate is shown. In step S1705, a scenario group surrounded by a dotted line is extracted from the second scenario graph data 1303. An important aspect surrounded by an ellipse is extracted from the scenario group extracted in step S1706. Extracted.

As an explanation of a more specific example, FIGS. 20A to 20D show an example of a chain bankruptcy simulation. FIG. 20A shows an example 2001 of first scenario graph (stochastic process graph) data created by extracting the events having the number of bankruptcies of 100 or more with the characteristic information as “number of bankruptcies”. Assume that the number of companies in the simulation model is 5000.

In FIG. 20A, two types of trends are observed: a direction 2002 in which chain bankruptcy converges in the number of bankruptcies surrounded by a dotted ellipse, and a direction 2003 in which chain bankruptcy progresses surrounded by a dashed ellipse. Here, it is assumed that the user wants to elucidate the cause of chain bankruptcy by multidimensional analysis with changed feature information. A scenario surrounded by an alternate long and short dash line is a chain bankruptcy scenario candidate 2004 designated by the user.

FIG. 20B shows an important situation 2005 enclosed by an ellipse and a scenario candidate 2006 enclosed by a two-dot chain line, which are obtained in step S1704. The scenario candidate 2006 includes a scenario candidate 2004 and a scenario candidate similar to the scenario candidate 2004 obtained using the information on the important aspects 2005 and the scenario candidate 2004 as input.

In FIG. 20C, the second scenario graph (probability) is created by extracting the events where the total profit / loss is −10.0 billion yen or more, with the characteristic information as “total profit / loss of 5000 companies in the model (total profit / loss)”. Process graph) Data example 2007 is shown.

FIG. 20D shows a scenario candidate 2008. The scenario candidate 2008 is a scenario candidate that is obtained from a match or similarity with the chained bankruptcy scenario candidate 2004 designated by the user, that is, a scenario candidate corresponding to the scenario candidate 2006. Accordingly, the scenario candidate 2008 is a scenario candidate in a direction in which chain bankruptcy proceeds in the same manner as the scenario candidate 2004 and the scenario candidate 2006. A vertex group 2009 is a vertex group corresponding to the important situation 2005. The important situation 2010 is an important situation extracted from the second scenario graph data example 2007 by the important situation extraction unit 1310 in step S1707.

The first scenario graph (stochastic process graph) data created by extracting events with the number of bankruptcies of 100 or more, with the feature information as “number of bankruptcies” shown in FIG. In addition, the second scenario graph (stochastic process graph) data created by extracting the events where the total profit and loss is -10.0 billion yen, with the characteristic information as “total profit and loss of 5000 companies in the model (total profit and loss)” By comparing the above, it can be seen that the cause of a specific chain bankruptcy is related to the fact that the total profit or loss is -10 billion yen or more in the critical situation 2010 before the bankruptcy of 100 companies or more.

As described above, if a correspondence relationship is observed between the scenario graph data of each feature information, for example, the position information of the important phase or the characteristic pattern structure, the feature information is deeply related to the cause of the case. You can see that If no such correspondence is found, it can be concluded that the feature information is not an important factor for the cause. By the multidimensional analysis in which the feature information is changed, the principal component analysis between the scenario graph data can be interactively executed by the user. In particular, in the case of issues such as social simulation, it is difficult for the device to uniquely determine what is happening, and the user can interactively recognize and proceed with the analysis while confirming the detailed cause. Analysis and elucidation are possible.

The above-described configurations, functions, processing units, processing means, etc. may be realized by hardware by designing a part or all of them, for example, with an integrated circuit. The present invention can also be realized by software program codes that implement the functions of the embodiments. In this case, a storage medium in which the program code is recorded is provided to the computer, and the computer (or CPU or MPU) reads the program code stored in the storage medium. In this case, the program code itself read from the storage medium realizes the functions of the above-described embodiments, and the program code itself and the storage medium storing the program code constitute the present invention. Examples of storage media for supplying such program codes include flexible disks, CD-ROMs, DVD-ROMs, hard disks, SSDs (Solid State Drives), optical disks, magneto-optical disks, CD-Rs, magnetic tapes, A non-volatile memory card, ROM, or the like is used. Moreover, the program code for realizing the functions described in the present embodiment can be implemented by a wide range of programs or script languages such as assembler, C / C ++, perl, Shell, PHP, Java (registered trademark), and the like.

Further, by distributing the program code of the software that realizes the functions of the embodiment via a network, it is stored in a storage means such as a hard disk or memory of a computer or a storage medium such as a CD-RW or CD-R. In use, the computer (or CPU or MPU) may read and execute the program code stored in the storage means or the storage medium.

101: Information processing system, 110: Graph processing device, 120: Object selector, 122: Case generator, 123: Scenario graph generator, 124: Scenario graph analyzer, 125: Database, 126: Case database, 1101: User Terminal 1102: Network 1103: Wireless base station 1104: Sensor group 1105: Central processing unit (CPU) 1106: Main memory 1107: Storage 1108: Network interface (I / F) 1109: CPU 1110 : Main memory, 1111: Storage, 1112: Touch panel, 1113: Network I / F.

Claims

A user interface that accepts selection of the first and second cases of interest;
Storage for storing time series data and information on the relationship between the sources of the time series data,
Extracting events corresponding to each of the first and second cases from the time series data,
Generating a stochastic process graph for each of the first and second cases based on the extracted events and the relationship information;
An information processing system for extracting a scenario that matches or resembles between generated probabilistic process graphs.
The information processing system according to claim 1,
The user interface further accepts designation of a scenario in the stochastic process graph for the first case from a user;
When extracting a scenario similar between the generated stochastic process graphs, the user specified from the stochastic process graph for the first case based on an important aspect of the stochastic process graph for the first case. A scenario related to a scenario is extracted, and a scenario that matches the related scenario is extracted from the stochastic process graph for the second case, thereby extracting similar scenarios between the generated stochastic process graphs. Information processing system.
The information processing system according to claim 2,
The information processing system, wherein the important aspect is extracted based on the degree of each vertex of the stochastic process graph for the first case.
The information processing system according to claim 1,
The information processing system, wherein the user interface includes a touch panel.
The information processing system according to claim 4,
The information processing system, wherein the touch panel is provided in a mobile terminal.
The information processing system according to claim 1,
An information processing system, wherein a result of computer simulation is stored in the storage as the time series data.