JP2006251947A - Rule acquisition system and method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a rule acquisition system capable of easily adding or deleting previous environmental change element information such as economic trends in creating fuzzy or other rules, sampling previously used rules again, and guaranteeing rules in such a manner as to set the estimation of values close to previous results as a goal and in such a way as to change the periods of the environmental change element information and the results. <P>SOLUTION: A knowledge acquisition support section 108 corrects rules temporarily stored in a temporary rule storage section 104. An inference section 106 estimates anticipated results using the rules in the temporary rule storage section 104 and based on the previous environmental change element information in a data storage section 102. A result determining section 107 determines if the result of anticipation is appropriate compared to results stored in the data storage part 102. If it is appropriate, new rules are stored in the rule storage section 103 and the corrected rules are stored in a rule history storage section 105. If it is not appropriate, the processes of gaining knowledge, making inferences and determining the results are repeated. The future can be anticipated by the application of new environmental change element information to the rules thus created. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technique for acquiring an analysis procedure such as an inference process, experience, or intuition of an expert as a rule.

Conventionally, many techniques for fuzzy reasoning and expert system have been proposed as techniques for acquiring analysis procedures such as inference processes, experiences, and intuitions of experts as rules. In particular, regarding fuzzy inference, which is a method for making computers handle ambiguity, a number of methods have recently been adopted as rules for fuzzy analysis procedures such as expert inference process, experience, or intuition. Examples are disclosed. For example, in Patent Document 1, inference is performed based on input values and teacher values, and errors of each rule and proposition are calculated, and the rules are corrected so as to eliminate the fuzzy rule defects. A fuzzy rule acquisition method is disclosed.
JP-A-6-131188

  Such a conventional fuzzy rule acquisition method is characterized by calculating errors of rules and propositions and automatically tuning rules based on them, but it is possible to create rules that always reduce errors by automatic tuning. For example, it is not possible to capture information such as the inference process, experience, or intuition of experts, such as addition / deletion of variables that can be easily conceived by experts in the field. In addition, the conventional fuzzy rule acquisition method has to prepare a teacher value that is a desirable inference output for the input value, but the method of effectively utilizing the past results for fuzzy rule acquisition is considered. Not. In addition, in order to capture information such as the inference process, experience, or intuition of experts, it is expected that there are many cases where it is desirable to re-import rules used in the past, but this method is not taken into account. There was a problem. The present invention is for solving the above-mentioned problems, and when creating various rules not limited to fuzzy or fuzzy, for example, past environmental change element information such as economic trends can be easily added or deleted. Yes, the rules used in the past can be imported again, and when verifying the rules, verification can be performed with the goal of inferring a value close to the past actual results. It is an object to obtain a rule acquisition system that can perform verification by switching between.

The rule acquisition system according to the present invention is:
An inference unit for inferring a prediction result from input variable information, and a result determination for determining whether a past prediction result inferred from past input variable information by the inference unit is valid as compared with a past result result And the inference unit accepts creation of a rule defining an inference analysis procedure, the inference unit infers a past prediction result from past input variable information, and the result determination unit determines the past prediction result. A knowledge acquisition support unit that creates a rule by repeating whether to determine whether it is appropriate as compared to the past result result, or until a determination result that it is appropriate is obtained, the inference unit includes the A prediction result is estimated from input variable information according to a rule created by a knowledge acquisition support unit.

  According to the present invention, an analysis result (currently a result result) that would have been inferred in the past, such as the demand of goods, the market price of goods, etc., from past environmental change element information such as economic trends, for example. By repeating the creation and verification of rules as a target, expert know-how can be acquired as rules, and new environmental change element information can be applied to the acquired rules to predict the prediction results.

Embodiment 1 FIG.
Embodiment 1 of the present invention will be described below. FIG. 1 is a diagram illustrating a configuration of a rule acquisition system. The rule acquisition system 101 includes a data storage unit 102, a rule storage unit 103, a rule temporary storage unit 104, a rule history storage unit 105, an inference unit 106, a result determination unit 107, a knowledge acquisition support unit 108, an input unit 109, and a display unit 110.

  As shown in FIG. 2, the data storage unit 102 stores time unit information 203, variable information 204, environment change element information 205, results results 207, and prediction results 206. Here, the time unit information 203 defines a time unit of each variable in the variable information 204, and includes, for example, a year, year, half year, quarter, month, week, and day. The variable information 204 defines attributes of variables used for the data of the environment change element information 205, the result result 207, and the prediction result 206, and designates the time unit defined in the above time unit information 203 as the attribute. The environmental change element information 205, the prediction result 206, and the actual result 207 include information on the variables, period start dates, and values defined in the variable information 204 described above, and the data structure is the same. , How data is generated is different.

  The environmental change element information 205 is data of an environmental change element such as an economic trend that is necessary for analysis, and is input by the input unit 109 and is data that becomes an inference input in the inference unit 106, which will be described in detail later. The prediction result 206 is an analysis result of items such as the demand for goods and the market price of the goods, and is data serving as an inference OUTPUT in the inference unit 106, which will be described in detail later. The actual result 207 is data of the past actual result 207 for items such as the product demand and the market price of the product, which are analysis results, and is input by the input unit 109 instead of the value derived by the inference unit 106. As will be described in detail later, the result determination unit 107 compares the result value with the prediction result 206 that is the OUTPUT of the inference unit 106 described above.

  As shown in FIG. 3, the rule storage unit 103 has a plurality of inference engines. One inference engine includes a plurality of rules, and each rule includes one IF-THEN rule information, one weighting information, one or a plurality of input variable information, and one result information. Here, the IF-THEN rule information is a logic for inferring a result from an input. Weighting information is information defining weighting for obtaining one value after weighting a plurality of derived values when estimating values related to the same result variable information from a plurality of IF-THEN rules. It is. The result variable information is information that defines a combination of a variable and a time as a result of the IF-THEN rule. Data stored in the rule storage unit 103 is created by the knowledge acquisition support unit 108.

  The rule temporary storage unit 104 also stores the same data as the rule storage unit 103. However, the rule temporary storage unit 104 stores the rules being verified in the knowledge acquisition support unit 108, and the rule storage unit 103 supports the knowledge acquisition support. The unit 108 stores the rule that has been verified and confirmed.

  As shown in FIG. 4, the rule history storage unit 105 stores IF-THEN rule information, weighting information, input variable information, and result variable information together with information indicating a modification time, an inference engine, a rule, and the like. The rule history storage unit 105 stores, as a history, when a rule stored in the rule storage unit 103 is deleted or changed in accordance with an instruction from the knowledge acquisition support unit 108.

  When the inference unit 106 receives an inference instruction based on a rule in the middle of verification from the knowledge acquisition support unit 108, the inference unit 106 uses the rules stored in the rule temporary storage unit 104 to store the environment change element information 205 stored in the data storage unit 102. From this, the prediction result 206 is estimated and stored in the data storage unit 102. The inference unit 106 displays a screen prompting the user to instruct the display unit 110 and receives an inference instruction from the user via the input unit 109, thereby using the rules stored in the rule storage unit 103. The prediction result 206 can be estimated from the environment change element information 205 stored in the data storage unit 102 and stored in the data storage unit 102.

  When the result determination unit 107 receives an instruction from the knowledge acquisition support unit 108, the result determination unit 107 acquires the prediction result 206 obtained by the inference unit 106 via the knowledge acquisition support unit 108, and stores the prediction result 206 and the data storage unit 102. A comparison with the stored result 207 is displayed on the display unit 110, and it is determined whether or not the prediction result 206 is valid from the user, and is returned to the knowledge acquisition support unit 108.

  The knowledge acquisition support unit 108 displays a screen for prompting rule creation on the display unit 110, receives a command from the user via the input unit 109, creates a rule, and requests the temporary rule storage unit 104 to store the rule. . Further, the knowledge acquisition support unit 108 repeats the rule correction, the inference by the inference unit 106, and the determination by the result determination unit 107 until an appropriate determination result is obtained, and then the rule temporary storage unit 104 stores the data during verification. The new rule that has been stored is stored in the rule storage unit 103.

  As described above, according to the present invention, the inference unit 106 uses the past environmental change element information 205 such as economic trends to analyze the past results of items such as product demand and product market price. With the goal of inferring a value close to the actual result 207, it is possible to make inferences for various periods, and the result determination unit 107 can compare the predicted result 206 with the actual result 207. The knowledge can be easily determined, and the knowledge acquisition support unit 108 can acquire expert know-how as a rule, starting with the inference by the inference unit 106 and the determination by the result determination unit 107.

  Further, the knowledge acquisition support unit 108 easily increases the period of the input variable information and the result variable information to be inferred according to the increment / decrement unit defined in the time unit information 203 stored in the data storage unit 102. Since the phenomenon can be caused, the rules can be verified for various past cases, and the accuracy of the rules can be improved. Further, the knowledge acquisition support unit 108 requests the rule history storage unit 105 to store the past rules, and can reuse the rules stored in the rule history storage unit 105 when creating the rules. Rules can be created efficiently.

  Note that the details of the data stored in the data storage unit 102, rule storage unit 103, rule temporary storage unit 104, and rule history storage unit 105 are described in the processing of the inference unit 106, result determination unit 107, and knowledge acquisition support unit 108. A screen example displayed on the display unit 110 regarding the flow will be described later.

  FIG. 8 is a diagram illustrating the H / W configuration of the rule acquisition system. The rule acquisition system includes an external storage device 801, a CPU (central processing unit) 802, a main memory (main storage device) 803, an input device 804, a display device 805, and a communication device 806 connected by a bus.

  The data storage unit 102, the rule storage unit 103, and the rule history storage unit 105 are configured by an external storage device 801 such as a hard disk, a floppy (registered trademark) disk, an MO, a CD, a DVD, or a magnetic tape. The rule temporary storage unit 104 includes a main memory (main storage device) 803 such as a RAM (Random Access Memory), or an external storage device 801. The inference unit 106, the result determination unit 107, and the knowledge acquisition support unit 108 include an external storage device 801, a main memory (main storage device) 803, and a CPU (central processing unit) 802. A program related to data processing loaded in the (main storage device) 803 is sequentially read into the CPU (central processing unit) 802 for processing.

  The input unit 109 includes an input device 804 such as a mouse and a keyboard, a communication device 806 connected to a LAN or WAN, an external storage device 801, a main memory (main storage device) 803, and a CPU (central processing unit) 802. Configured to accept an input instruction and data from the input device 804 or the communication device 806, and a program related to data input loaded from the external storage device 801 to the main memory (main storage device) 803 is stored in the CPU (central processing unit) 802. It is read and processed sequentially. The display unit 110 includes a display device 805 such as a display.

  FIG. 5 is a data example of the time unit information 203 that is input by the input unit 109 and stored by the data storage unit 102 as described above. The time handled in the present embodiment is a date-based time concept for performing analysis, and time unit data is also based on the date. As shown in FIG. 5, one line is data defining one time unit. Next, the meaning of each column will be described. The time unit ID 501 is an ID for specifying a time unit. The time unit name 502 is a name representing the contents of the time unit. The time unit order 503 represents a priority order for each time unit, and a numerical value corresponding to the range of the time unit is assigned. In FIG. 5, a large numerical value is assigned to a time unit having a wider time range, and a small numerical value is assigned to a time unit having a narrower time range. The time range 504 represents a time range in a corresponding time unit.

  FIG. 6 is a data example of the variable information 204 that is input by the input unit 109 and stored by the data storage unit 102 as described above. The variable information 204 defines the attributes of variables used for the data of the environment change element information 205, the result result 207, and the prediction result 206. As shown in FIG. 6, one line is data defining one variable. Next, the meaning of each column will be described. The variable ID 601 is an ID that identifies a variable. The variable name 602 is a name representing the contents of the variable. A value unit 603 indicates a unit of a value input to the variable. The time unit ID 604 represents the time unit of the variable, and the time unit ID 501 of the time unit data defined in the above time unit information 203 is set.

  FIG. 7 is a data example of the environment change element information 205 input by the input unit 109 and stored by the data storage unit 102 as described above. The environmental change element information 205 is data of an environmental change element such as an economic trend necessary for analysis. As shown in FIG. 7, one row is the actual value of a certain period of certain environment change element information. Next, the meaning of each column will be described. The variable ID 701 is a variable ID that identifies the type of the environment change element information 205, and the variable ID 601 of the variable information data defined in the variable information 204 is set. In the period start date 702, the period start date of the environment change element information 205 is set.

  Here, the variable name, value unit, and time unit ID of certain environmental change element information data can be obtained by specifying variable information data having the same variable ID 601 as the variable ID 701 of the environmental change element information data. . Further, the time unit name, time unit order, and range of certain environment change element information data specify time unit information 203 data having the same time unit ID 501 as the time unit ID 604 of the variable information data specified as described above. Can be obtained. That is, the value 703 is obtained from the period start date 702 of the environment change element information 205 having characteristics such as the variable name 602, the value unit 603, the time unit name 502, the time unit order 503, and the time range 504 specified by the variable ID 701. The value of the period determined by the starting time range 504 is shown.

  For example, in the environment change element information data in the row 704 of FIG. 7, the variable ID 701 is “M1”, the period start date 702 is “2004/01/01”, and the value 703 is “18”. Furthermore, since the variable ID 601 in the row 605 in FIG. 6 is “M1”, it can be seen that the variable information data (row 605) corresponds to the environment change element information data (row 705). That is, the variable name of the environmental change element information data (line 704) is “business trend index (DI)”, the value unit is “%”, and the time unit ID is “MONTH”. Furthermore, since the time unit ID 501 in the row 509 in FIG. 5 is “MONTH”, the time unit information data (row 509) corresponds to the environment change element information data (row 705) and the variable information data (row 605). I understand that That is, the time unit name of the environment change element information data (line 704) is “month”, the time unit order is “3”, and the time range is “January”. That is, row 704 represents that the economic trend index (DI) in January 2004 is 18%.

  The actual result 207 and the prediction result 206 have the same configuration as the environment change element information 205 in FIG. 7, but the contents of the input data are different. The environmental change element information 205 is data of an environmental change element such as an economic trend that is necessary for analysis, and is input by the input unit 109 and is data that becomes an inference input in the inference unit 106, which will be described in detail later. The prediction result 206 is an analysis result of items such as the demand for goods and the market price of the goods, and is data serving as an inference OUTPUT in the inference unit 106, which will be described in detail later. The actual result 207 is data of the past actual result 207 for items such as the product demand and the market price of the product, which are analysis results, and is input by the input unit 109 instead of the value derived by the inference unit 106. As will be described in detail later, the result determination unit 107 compares the result value with the prediction result 206 that is the OUTPUT of the inference unit 106 received through the knowledge acquisition support unit 108.

  Next, the concept of rules in the present embodiment will be described. The present invention relates to a technique for acquiring an analysis procedure such as an inference process, experience, or intuition of an expert as a rule. As shown in FIG. 3, the rule storage unit 103 has a plurality of inference engines. One inference engine is composed of a plurality of rules, and each rule is composed of one IF-THEN rule information, one weighting information, one or a plurality of input variable information, and one or a plurality of result information. Here, the IF-THEN rule information is a logic for inferring a result from an input. The weighting information is used to obtain one value related to the same result variable information after weighting a plurality of derived values when a value related to the same result variable information is estimated from a plurality of IF-THEN rules. This information defines weighting. The input variable information is information that defines a combination of a variable and time to be input to the IF-THEN rule. The result variable information is information that defines a combination of a variable and a time as a result of the IF-THEN rule. In this embodiment, fuzzy inference is applied as an inference method, and the details will be described later. However, in the IF-THEN rule information, the IF part and the THEN part are defined by a fuzzy set. -Estimating a result set for each THEN rule, and further applying weighting to the result set of each IF-THEN rule, then finding the center of gravity of the area of each set related to the same result variable information by the min-max-centroid method To find one value. Note that a method other than fuzzy inference may be applied as an inference method. For example, the IF-THEN rule is a simple calculation formula. At the time of inference execution, a value is estimated by calculation for each IF-THEN rule, and each IF-THEN rule is further calculated. A single value may be obtained by multiplying the value of the result of the -THEN rule by weighting and then obtaining the average value of each value related to the same result variable information. In addition, each IF-THEN rule may be defined such that different result variable information is defined, and weighting may not be performed.

  FIG. 15 is a schematic diagram showing what type of rules are obtained by applying a fuzzy inference method in the present embodiment to information such as an inference process, experience, or intuition of an expert. One inference engine 1501 is defined by rule 1 (1502), rule 2 (1503), and rule 3 (1504). Each rule defines IF-THEN rule information, weighting information, input variable information, and result variable information. For example, the IF-THEN rule information of Rule 1 (1502) is explained by the phrase “IF small THEN less”, and “small” following IF is input variable information “April 2003 Business Trends” The “index (DI)” is “small”, and the “small” following THEN means that the result variable information “2004 product xx sales” is “small”. Apply the value of the corresponding environment change element information 205 to the input variable information of each rule, infer the result set corresponding to the result variable information according to the logic defined in the IF-THEN rule information, and multiply by the weight The result centroid of the area of the result set multiplied by the weight is further obtained for the same result variable information, and the prediction result 206 of the product xx sales inference engine 1501 is obtained. Here, for example, it is assumed that there is environment change element information 205 necessary for estimating “2004 product xx sales”, and a rule that uses this environment change element information 205 as input variable information is added. In addition, it is possible to determine specific numerical values represented by the words “small”, “small”, and “large” in each IF-THEN rule information by repeating the verification, that is, the inference process, experience, or intuition of an expert. Is acquired as a rule using inference.

  FIG. 16 is a schematic diagram of a rule representing FIG. 15 in a format closer to data. Rule 1 (1502), rule 2 (1503), and rule 3 (1504) in FIG. 15 are represented as rule 1 (1602), rule 2 (1603), and rule 3 (1604) in FIG. 16, respectively. The IF-THEN rule is a rule in which an IF part and a THEN part are defined by fuzzy sets. In FIG. 15, the IF-THEN rule of rule 1 (1502) is “IF small THEN less”, the weight is “1.0”, and the input variable information is “April 2003 Business Trend Index (DI)”. The result variable information is “FY2004 Product xx Sales”. How to express such a rule using a fuzzy set will be described.

  In FIG. 16, 1606 is a fuzzy set of the IF unit. The horizontal axis 1616 is “April 2003 Business Trend Index (DI)” defined by the input variable information of FIG. 15, and the vertical axis 1612 is “small” of the IF-THEN rule “IF small THEN less” of FIG. Is a numerical value between 1 and 0. The fuzzy set of the IF section is a space surrounded by a membership function 1614, a horizontal axis 1616, and a vertical axis 1612. The membership function is represented by a straight line connecting the threshold value 1613 and the threshold value 1617.

  Reference numeral 1607 denotes a fuzzy set of the THEN part. Here, the horizontal axis 1624 is “April 2003 product xx sales” defined by the result variable information of FIG. 15, and the vertical axis 1618 is “less” of the IF-THEN rule “IF small THEN less” of FIG. "Is a numerical value between 1 and 0. A fuzzy set of the THEN part is a space surrounded by a membership function 1621, a horizontal axis 1624, and a vertical axis 1618. The membership function is a straight line connecting a threshold value (1619), a threshold value (1620), and a threshold value (1623). It is represented by

  Next, the flow of fuzzy inference will be described with reference to FIG. The value of “April 2003 Business Trend Index (DI)” is acquired from the environmental change element information 205, assigned to the horizontal axis of Rule 1, and the grade value (from the intersection 1615 with the membership function 1614 of the IF section fuzzy set) Value on the vertical axis). Next, a set 1625 is obtained by cutting the fuzzy set of the THEN part with the above grade value. Further, 1625 is obtained by reducing the vertical axis to 100% by weighting “1.0”. In the example of rule 1, since the weight is 100%, the vertical axis does not change. However, if the weight is 0.9, the weight is reduced to 90%, and if the weight is 0.8, the scale is reduced to 80%. Similarly, 1626 is obtained from rule 2 and 1629 is obtained from rule 3. The center of gravity of the area of these three sets is the prediction result 206 of “2004 product xx sales”. Such a method is called the MIN-MAX centroid method in fuzzy reasoning. FIG. 28 explains the fuzzy inference process and inference calculation by the MIN-MAX centroid method centroid method. In the rule of FIG. 28, the intersection of the value of the variable in the IF part and the membership function at the time of prediction is obtained. Assume that three rules are intersected and adopted as shown in FIG. In this example, the variable of the IF part is DI, and the DI value of the predicted value is 20%. When the intersection point is hit by the membership function of the THEN part in rule 1, an area portion representing the degree of adoption of rule 1 is selected. Since this rule has a weight of 1.0, it indicates a larger shape. Similarly, the rule 2 and the rule 3 adopted also extract areas representing the degree of adoption. Since the weight of rule 2 is 0.6 and that of rule 3 is 1.0, an area corresponding to these is selected. Each of the next selected area portions is discretized at a certain interval. In the example of FIG. 28, it is divided into 20 million units. The formula for obtaining the center of gravity can be obtained by the formula “centroid = Σ (xi × yi) / Σ (xi)” when the center of gravity method is employed and the x-axis and y-axis are used. According to the actual calculation formula of FIG. 28, the inferred value in this case is 233.4 billion and is defined as the center of gravity.

  FIG. 9 is an example of rule data created by the knowledge acquisition support unit 108 and stored by the rule storage unit 103 as described above. As shown in FIG. 9, one line is data defining IF-THEN rule information, weighting information, input variable information, or result variable information that is an element of the rule. Next, the meaning of each column will be described. The inference engine ID 901 is an ID that identifies the inference engine. As shown in FIG. 3, a rule is created for each inference engine, but an inference engine ID 901 identifies this inference engine. The rule number 902 is an ID that identifies the rule. As shown in FIG. 3, IF-THEN rule information, weighting information, input variable information, and result variable information are defined for each rule. The rule type 903 specifies the classification of rules such as the IF-THEN rule information, the weighting information, the input variable information, and the result variable information described above.

  The change type 904 indicates whether the corresponding information is addition / deletion / change. The variable ID 905 specifies a corresponding variable when the rule type 903 is input variable information or result variable information. The variable ID specified here is the variable ID 601 defined in the variable information 204 described with reference to FIG. Further, by tracing the time unit information 203 in FIG. 5 from the variable information 204 in FIG. 6, the time range 504 possessed by the variable IDs 905 and 601 can also be specified.

  The period start date 906 specifies the period start date of the corresponding variable when the rule type 903 is input variable information or result variable information. Here, if the information of the time range 504 specified as described above and the information of the period start date 906 are combined, the specific period of the specified input variable information or result variable information can be specified. Furthermore, when the rule type 903 is input variable information, the variable ID 905 and the period start date 906 in FIG. 9 and the variable ID 701 and the period start date 702 in the environment change element information 205 in FIG. Value 703 can be obtained. When the rule type 903 is result variable information, the value of the variable ID 905 is set to the variable ID 701 and the period start date 906 is set to the period start date 703 with respect to the prediction result 206 having the same data configuration as FIG. The value obtained as the inference result can be set as the value 703. Furthermore, the prediction result 206 created as described above can be compared with the result 207 having the same data configuration as that shown in FIG. 7, and the variable 701 can be compared with the value 703 that matches the period start date 702. .

  The weight 907 specifies a numerical value from 0.0 to 1.0 when the rule type 903 is weighting information. The IF part concept 908 specifies the IF part concept of the rule concept described in FIG. This wording does not affect inference, but serves to explain the IF-THEN rule. The IF unit threshold value X1 (909) designates a value on the horizontal axis of the first threshold value of the IF unit fuzzy set described in FIG. The IF unit threshold Y1 (910) specifies the value of the vertical axis of the first threshold of the IF unit fuzzy set described in FIG. The IF part threshold value X2 (911) and the IF part threshold value Y2 (912) designate the values on the horizontal axis and the vertical axis of the second threshold value of the fuzzy set described in FIG. Similarly, it is sufficient that at least three or more sets of IF threshold values can be stored. The THEN concept 913 designates the THEN concept of the rule concept described in FIG. The THEN part threshold value X1 (914) designates a value on the horizontal axis of the first threshold value of the THEN part fuzzy set described in FIG. The THEN part threshold Y1 (915) designates the value of the vertical axis of the first threshold of the THEN part fuzzy set described in FIG. Similarly, it is only necessary that at least three sets of THEN threshold values can be stored.

  When a method other than fuzzy inference is applied as an inference method, data necessary for the inference method to be applied may be stored in the right column of the IF unit concept (908). In the example of FIG. 9, one input variable information is applied to one IF-THEN rule information. However, a plurality of input variable information is applied to one IF-THEN rule information. In this case, for a row in which the rule type 903 is result variable information, a data string for specifying the result variable information in one rule is added, and the rule type 903 is IF- For the line that is the THEN rule information, a data string for specifying the aforementioned result variable information may be added to each of a plurality of variables used in the inference logic. In addition, each IF-THEN rule may be defined such that different result variable information is defined, and weighting may not be performed. In this case, the weight information is not necessary as the rule type 903, and the column of the weight 907 is not necessary.

  Note that the rule temporary data stored in the rule temporary storage unit 104 also has the same configuration as the rule data described with reference to FIG.

  FIG. 10 is an example of rule history data created by the knowledge acquisition support unit 108 and stored by the rule history storage unit 105 as described above. As shown in FIG. 10, one line is one history of rules. Next, the meaning of each column will be described. Each column of the rule history data in FIG. 10 is obtained by adding a correction time 1001 to the previous column of the rule data in FIG. 9 when compared with each column in the rule data in FIG. The correction time 1001 is a date and time at a level at which the rule history of each row can be ordered according to the correction order in the knowledge acquisition support unit 108. Each column after the inference engine ID 901 is the same as each column after the inference engine ID 901 of the rule data described with reference to FIG.

  FIG. 11 is a flowchart for explaining the flow of processing of the inference unit 106 described above. First, designation of either the rule storage unit 103 or the rule temporary storage unit 104 is accepted from the input unit 109 or the knowledge acquisition support unit 108. (Step S1101) Here, in the case of inference execution for prediction, designation is received from the user via the input unit 109, and in the case of inference execution for knowledge acquisition, it is received from the knowledge acquisition support unit 108. Next, designation of an inference engine is accepted from the user from the input unit 109 or the knowledge acquisition support unit 108. (Step S1102) Here, in the case of inference execution for prediction, designation is received from the user via the input unit 109, and in the case of inference execution for knowledge acquisition, it is received from the knowledge acquisition support unit 108. Next, a rule relating to the inference engine specified in step S1102 is acquired from either the rule storage unit 103 or the rule temporary storage unit 104 specified in step S1101. (Step S1103)

  Next, in step S1104 to step S1107, the period is changed. Note that the period change specifically means that the period start date 906 described in FIG. 9 is temporarily changed during the execution of the inference. First, with respect to the input variable information and result variable information of all rules acquired in step S1103, each time unit order is acquired by tracing the variable information 204 and the time unit information 203. More specifically, the variable ID 905 is specified for the input variable information and the result variable information of all the rules acquired in step S1103, the time unit ID 604 is specified from the variable information 204 data having the same variable ID 601, and the same time unit ID 501. The time unit order 503 and the time range 504 are acquired from the time unit information data having (Step S1104) Next, the time range 504 having the largest time unit order acquired in step S1104 is set as a time increase / decrease unit in this inference engine. Here, as described above with reference to FIG. 5, since a large numerical value is assigned to a time unit having a wider time range and a small numerical value is assigned to a time unit having a narrower time range, the time unit order is the largest. The time range is the widest time range. (Step S1105) The change of the period of the input variable information and the result variable information is accepted from the input unit 109. Note that the period change can be increased or decreased by a plurality of units based on the time increase / decrease unit obtained in step S1105. Multiple input variable information and result variable information are defined in one inference engine, but the period of all input variable information and result variable information is matched to the input variable information or result variable information variable with the widest time range. You can change the start date. (Step S1106) The period start date of the input variable information and the period start date of the result variable information are increased or decreased by the increase / decrease number received in step S1106 based on the increase / decrease unit acquired in step S1105. For example, if the increase / decrease unit acquired in step S1105 is one year and the increase / decrease number is +2, all period start dates are increased by two years. (Step S1107) An example of changing the period will be described later with reference to FIGS.

  Next, from the environment change element information 205 stored in the data storage unit 102, the variable and the period start date of the rule input variable information acquired in step S1103 (however, the period start date has been corrected in step S1107). For example, the value corresponding to the corrected period start date) is acquired. (Step S1108) Next, inference is executed. (Step S1109) Although the inference is executed as described with reference to FIGS. 15 and 16, an inference method other than fuzzy inference may be applied as described above. In any case, according to the applied reasoning logic, the value of the environment change element information 205 corresponding to the input variable information of the IF-THEN rule is assigned, and the value related to the result variable information of the IF-THEN rule is obtained. Next, the value obtained in step S1109 is stored in the data storage unit 102 with the variable of the result variable information of the rule acquired in step S1103 and the period start date (provided that the period start date is corrected in step S1105). , The period start date after correction) and the prediction result 206. FIG. 21 shows a screen example of the inference unit.

  FIG. 12 is a flowchart for explaining the flow of processing of the result determination unit 107 described above. First, the prediction result 206 obtained by the inference unit 106 and stored by the data storage unit 102 and the result result 207 stored in advance by the data storage unit 102 are shown to the user in a format that can be easily compared by the display unit 110. (Step S1201) Next, it is accepted from the user via the input unit 109 whether or not the prediction result 206 is more appropriate than the result result 207. (Step S1202) In addition, the example of a screen of a result determination part is shown in FIG. As shown in the example of FIG. 22, the difference between the prediction result 206 and the actual result 207 may be displayed on the result determination unit screen.

  FIG. 13 is a flowchart for explaining the processing flow of the knowledge acquisition support unit 108 described above. First, the user accepts whether to create a new inference engine or specify an existing inference engine and correct it from the user via the input unit 109. (Step S1301) When a new creation is accepted in Step S1301, an inference engine ID 901 is determined. (S1302) When the modification of the existing inference engine is accepted in step S1301, the rule relating to the inference engine accepted in step S1301 is acquired from the rule storage unit 103 and stored in the rule temporary storage unit 104. (Step S1303) Next, a past rule related to the inference engine accepted in Step S1301 is acquired from the rule history storage unit. (Step S1304)

  Next, modify the rules. (Step S1305) The process flow of the rule correction method will be described later with reference to FIG. Next, inference by the inference unit 106 is performed. (Step S1306) The process flow of the inference unit 106 is as shown in FIG. Next, the result determination unit 107 determines whether or not the prediction result 206 by the inference unit 106 acquired via the knowledge acquisition support unit 108 is valid. (Step S1307) The process flow of the result determination unit 107 is as shown in FIG. If the determination is “invalid”, the process returns to step S1305, the rule is corrected again, and the verification operation is repeated. If the determination is “valid”, the rule to be changed this time among the data stored in the rule storage unit 103 is stored in the rule history storage unit. (Step S1308) Next, among the rules stored in the temporary rule storage unit 104, the rule to be changed this time is stored in the rule storage unit 103.

  FIG. 14 is a flowchart for explaining the flow of processing for correcting the rule described in step S1305. First, the rule data regarding the corresponding inference engine acquired from the rule temporary storage unit 104 and the rule history regarding the corresponding inference engine acquired from the rule history storage unit 105 are displayed to the user by the display unit 110. (Step S1401) Next, the rule type 903 to be corrected is received from the user via the input unit 109. (Step S1402) Here, the rule type 903 is IF-THEN rule information, weighting information, input variable information, or result variable information.

  If the rule type 903 received in step S1402 is IF-THEN rule information, the display unit 110 shows the IF-THEN rule information and the correction history to the user. (Step S1403) Next, the input unit 109 accepts correction of IF-THEN rule information from the user and stores it in the rule temporary storage unit. (Step S1404)

  If the rule type 903 received in step S1402 is weighting information, the display unit 110 shows the weighting information and the correction history to the user. (Step S1405) Next, the correction of the weighting information is accepted from the user by the input unit 109 and stored in the rule temporary storage unit. (Step S1406) If the rule type 903 accepted in step S1402 is input variable information, the display unit 110 indicates the input variable information and the correction history to the user. (Step S1407) Next, the input unit 109 accepts correction of input variable information from the user (Step S1410), and further accepts correction of IF-THEN rule information related to the relevant variable (S1403), and the changed input variable information. At the same time, it is stored in the rule temporary storage unit. (Step S1404) If the rule type 903 received in step S1402 is the result variable information, the display unit 110 shows the result variable information and the correction history to the user. (Step S1409) Next, correction of the result variable information is accepted from the user by the input unit 109 and stored in the rule temporary storage unit. (Step S1410)

  In step S1404, step S1406, step S1408, and step S1410, by accepting designation of the past correction history from the user, the latest rule can be replaced with the past rule and reused. As shown in FIG. 10, the past correction history has a configuration in which only the correction time is added to the rule data described in FIG. 9, and the rule history data designated by the user in FIG. The modification time may be removed, and the change type 904 may be added or changed and stored in the rule temporary storage unit. Further, regarding the reuse of the input variable information at step S1408 and the result variable information at step S1410, the IF-THEN rule information and the weighting information that are performed simultaneously with the corresponding variable correction are also reused at the same time.

  FIG. 23 shows a screen example of step S1401, FIG. 24 shows a screen example of step S1403, FIG. 25 shows a screen example of step S1405, and FIG. 26 shows a screen example of step S1407. A screen example of step S1409 is shown in FIG.

  FIGS. 17 and 18 are examples of changing the period from step S1104 to step S1107 described in the processing flow of the inference unit in FIG. For example, in FIG. 17, an example will be described in which the period is changed so as to increase by one year. In FIG. 17, a rule 1701 is stored in the rule storage unit 103, and variable information 1702 and time unit information 1703 are stored in the data storage unit 102. Note that changing the period specifically means that the period start date of the rule 1701 is temporarily changed during the execution of the inference. First, with respect to the input variable information and result variable information of the rule 1701, each time unit order is acquired by tracing the variable information 204 and the time unit information 203. More specifically, each variable ID is specified for data in which the rule type 903 of the rule 1701 is input variable information or result variable information. Next, in the variable information 1702, variable information data having the same variable ID as the variable ID specified in the rule 1701 is specified, and further a time unit ID is specified. Next, in the time unit information 1703, time unit information data having the same time unit ID as the time unit ID specified in the variable information 1702 is specified, and further, the time unit order and the time range are specified. According to such a procedure, the time unit order of the variable N1 in the rule 1701 is 6, the time range is 1 year, the time unit order of the variable M1 is 1, the time range is 1 day, and the time of the variable M2 The unit order is 3, the time range is January, the time unit order of the variable M3 is 5, and the time range is June. (Step S1104) Next, since the time unit order acquired in Step S1104 is the largest, 6, the corresponding year of the time range is set as the time increase / decrease unit in this inference engine. (Step S1105) The input unit 109 accepts a change in the period of +1 input variable information and result variable information, with 1 year as a time increase / decrease unit. (Step S1106)

  Increase / decrease the period start date of all input variable information and the period start date of all result variable information by the increase / decrease number “+1” received in step S1106 based on the increase / decrease unit “1 year” acquired in step S1105. The rule 1704 shows the result obtained, that is, the result increased by one year. The period start date “2005/4/1” of variable N1 has been changed to “2006/4/1” one year later, the time unit name of the relevant variable is the year, and the time range of the relevant variable is one year Therefore, it can be seen that the period related to the variable N1 is changed from FY2005 to FY2006. The period start date “2004/4/1” of variable M2 has been changed to “2005/4/1” one year later, the time unit name of the variable is month, and the time range of the variable is January Therefore, it can be seen that the period related to the variable M2 is changed from April 2004 to April 2005. The period start date “2004/10/1” of variable M3 has been changed to “2005/10/1” one year later, the time unit name of the relevant variable is half-year, and the time range of the relevant variable is June Therefore, it can be seen that the period related to the variable M3 is changed from the second half of 2004 to the second half of 2005. (Step S1107)

  For example, in FIG. 18, an example will be described in which the period is changed so as to decrease in June. In FIG. 18, a rule 1801 is stored in the rule storage unit 103, and variable information 1802 and time unit information 1803 are stored in the data storage unit 102. According to the same procedure as described in FIG. 17, the time unit order of the variable N11 is 5, the time range is June, the time unit order of the variable M11 is 3, the time range is January, and the variable The time unit order of M12 is 2, the time range is 7 days, the time unit order of variable M13 is 1, and the time range is 1 day. (Step S1104) Next, since the time unit order acquired in Step S1104 is the largest, 5 is set as the time increase / decrease unit in this inference engine, since June is the corresponding time range. (Step S1105) The input unit 109 accepts a change in the period of the input variable information and the result variable information of −3, with 1 year as a time increase / decrease unit. (Step S1106)

  Increase / decrease the period start date of all input variable information and the period start date of all result variable information by the increase / decrease number “−3” received in step S1106 based on the increase / decrease unit “June” acquired in step S1105. The rule 1804 shows the result obtained, that is, the result of reduction in 18 months. The period start date “2005/4/1” of variable N11 has been changed to “2003/10/1” 18 months ago, the time unit name of the relevant variable is half-year, and the time range of the relevant variable is June Therefore, it can be seen that the period related to the variable N11 is changed from the first half of 2005 to the second half of 2003. The period start date “2005/3/1” of variable M11 has been changed to “2003/9/1” 18 months ago, the time unit name of the variable is month, and the time range of the variable is January Therefore, it can be seen that the period related to the variable M1 has been changed from March 2005 to September 2003. The period start date “2004/4/3” of variable M12 has been changed to “2002/10/3” 18 months ago, the time unit name of the variable is week, and the time range of the variable is 7 days Therefore, it can be seen that the period related to the variable M12 is changed from one week starting on April 3, 2004 to one week starting on October 3, 2002. The period start date “2004/4/1” of variable M13 has been changed to “2002/10/1” 18 months ago, the time unit name of the variable is day, and the time range of the variable is 1 day. Therefore, it can be seen that the period related to the variable M13 has been changed from April 1, 2004 to October 1, 2002. (Step S1107)

  FIG. 19 and FIG. 20 are schematic diagrams for explaining an operation example of such a rule acquisition system. In the 2004 knowledge acquisition (1901), the knowledge acquisition support unit 108 first creates a 2003 year rule (1902), and the inference unit 106 uses the 2003 year rule and the 2002 environmental change element information (1903) from 2003. The year prediction result (1905) is estimated, and the result determination unit 107 determines whether or not it is appropriate as compared with the 2003 result (1906) until it is determined to be appropriate. Repeat this. If it is determined to be valid, the knowledge acquisition support unit 108 changes the period of the 2003 rule described above to create a 2004 rule (1906). (1908) Since the reasoning unit 106 can easily switch between the fiscal years, not only the fiscal year 2003 rules but also the fiscal year 2002 rules, the fiscal year 2001 rules, and the like are verified, and the fiscal year 2004 rules 1909 are more reliable. It can also be.

  In the 2004 operation (1911), the inference unit 106 estimates the 2004 prediction result (1904) from the 200 rule (1909) and the 2003 environmental change element information (1912) created as described above. At the end of the fiscal year 2004, the environmental change element information (2012) for the fiscal year 2004 and the actual result (2006) for the fiscal year 2004 are stored in the data storage unit 102 by the input unit 109.

  Furthermore, knowledge acquisition (2001) is performed at the end of FY2004. In the 2005 knowledge acquisition (2001), the knowledge acquisition support unit 108 first modifies the 2004 rule (1909) (2002), and the inference unit 106 uses the 2004 rule and the 2003 environment change element information (1912). ) To estimate the 2004 prediction result (2005), and the result determination unit 107 determines whether or not it is appropriate as compared with the newly stored 2004 result (2006) (2007). This is repeated until it is determined to be valid. If it is determined to be appropriate, the knowledge acquisition support unit 108 feedback-changes the 2004 rule described above to create a 2005 rule (2009). (2008) In the 2005 operation (2011), the inference unit 106 estimates the 2005 prediction result (2014) from the 2005 rule (2009) and the 2004 environmental change factor information (2012) created as described above. To do.

  Note that, in the processing flow of the inference unit 106 of the present embodiment, unlike the steps S1104 and S1105 in FIG. In the method shown in FIG. 11, for example, from one year to the next year, or from one month to the next month, the period handled by the rule can be switched using a meaningful period as a unit. In this case, for example, the degree of freedom can be increased by switching from one year starting on April 1 to one year starting on April 15. In the processing flow of the result determination unit 107 of this embodiment, unlike the processing flow shown in FIG. 12, the difference between the prediction result 206 obtained by the inference unit and the result result 207 is calculated, and the difference is determined in advance. If it is smaller than the threshold value, it may be determined that the prediction result 206 is appropriate as compared with the actual result 207, and in other cases, it may be determined that the prediction result 206 is not appropriate. In this case, the result determination can be performed automatically.

  Further, the rule acquisition system shown in FIG. 1 may be configured only by the data storage unit 102, the rule storage unit 103, the inference unit 106, the input unit 109, and the display unit 110. In this case, the processing flow of the inference unit 106 is substantially the same as the processing flow of FIG. 11, the rule storage unit is automatically selected in step S1101, and in step 1102, the inference engine designation is received from the input unit. The only difference is that the rule relating to the inference engine is acquired from the rule storage unit. In such a configuration, the period change and inference can be performed using the rules already acquired as described above. Alternatively, in the rule acquisition support system shown in FIG. 1, the input unit 109 and the display unit 110 are connected to the data storage unit 102, the inference unit 106, the result determination unit 107, and the knowledge acquisition via a public line or a network such as a LAN. You may make it connect with the support part 108. FIG. Alternatively, in the rule acquisition support system shown in FIG. 1, a user management unit is provided, and a user who can use the data storage unit 102, the inference unit 106, the result determination unit 107, and the knowledge acquisition support unit 108, and the inference unit 106 only are used. You may be able to manage the users who can.

  As described above, according to the present embodiment, according to the present invention, for example, an analysis result (e.g., demand for a product, market price of a product, etc.) that has been inferred in the past from past environmental change element information such as an economic trend ( (Currently, it is the actual result), by creating and verifying the rules repeatedly, acquiring expert know-how as rules, and applying new environmental change element information to the acquired rules, A prediction result can be estimated. It is possible to perform inference by assigning separately stored environment change element information to input variable information. Using the variable information and the time unit information, the concept of the period in the environment change element information, the actual result, the prediction result, the rule input variable information, and the rule output variable information can be handled in a unified manner. Using variable information and time unit information, it is possible to associate rule input variable information with environment change element information, rule output variable information, and prediction results. In the rule, by setting the variable with the widest time range as the unit of time increase / decrease, the period of all rules related to one inference engine is changed from one year to the next year, for example, from one month to the next month as shown in FIG. Or, as shown in FIG. 18, it is possible to switch from one half period to the next half period in units of meaningful periods. By making it possible to switch the period on a daily basis, the degree of freedom can be increased, for example, by switching from one year starting on April 1 to one year starting on April 15. Result determination can also be performed automatically. Past rules can be reused when creating rules. The display unit and the input unit are arranged on the client, and the reasoning unit, the result determination unit, and the knowledge acquisition support unit are arranged on the server, so that the client can acquire the rules in a simple environment. Because it is possible to create rules that take into account results accumulated over time, such as when forecasting the number of product demands from various values representing economic trends, matters that cannot be reliably predicted Suitable for winning the rules.

Embodiment 2. FIG.
In this embodiment, an example of a screen displayed by the display unit 110 will be described with reference to FIGS. The inference screen 2101 in FIG. 21 is a screen example for realizing the processing of the inference unit 106 described with reference to FIG. The inference engine 2102 displays the inference engine ID 901 illustrated in FIG. If an inference engine name corresponding to the inference engine ID 901 is separately defined, the inference engine name may be displayed.

  The environment change element information 2103 is a table display of the environment change element information 205 assigned to the input variable of the inference engine indicated by the rectangle 2102. The variable ID 701 of the environment change element information 205 is converted into a variable name 602 of variable information data having the same variable ID 701 and displayed. Similarly, the period start date 702 of the environment change element information 205 is represented by a time unit ID 604 of variable information data having the same variable ID, a time unit name 502 of time unit information data having the same time unit ID 501, and a time range 504. A word suitable for representing the time range 504 starting from the period start date 702 is displayed in the period 2106. Similarly, the value 703 of the environment change element information 205 is displayed as a value 2107 by adding a value unit 603 of variable information data having the same variable ID 701. The prediction result 2104 is derived as a result variable of the inference engine indicated by the rectangle 2102, and the prediction result 206 stored as the prediction result 206 in the data storage unit 102 is displayed in a table format. Note that the value of each column of the prediction result 206 is converted in the same manner as in the case of the environment change element information 205 described above, and is displayed in the variable name 2108, the period 2109, and the value 2110. Note that the value 2110 is blank before inference execution.

  When the selection of “rule storage unit or rule temporary storage unit designation” 2111 in FIG. 21 is accepted by the input unit 109, either the rule storage unit 103 or the rule temporary storage unit 104 can be designated as the reference destination of the rule. When the selection of “inference engine designation” 2112 is accepted by the input unit 109, the inference engine 2112 can be designated from the rule storage unit 103 or the rule temporary storage unit 104 designated by the “rule storage unit or rule temporary storage unit designation” 2111. If the inference engine is changed here, the display of the inference engine 2102, the environment change element information 2103, and the prediction result 2104 is changed.

  When the selection of “period change” 2113 in FIG. 21 is accepted by the input unit 109, the period 2106 or the period 2109 can be changed for each data of the environment change element information 2103 or the prediction result 2104. If the period is changed here, the display of the period 2106 or the period 2109 is changed. When the selection of “save rule or temporarily save rule” 2114 is accepted by the input unit 109, the rule changed by the “change period” 2113 can be stored in the rule storage unit 103 or the rule temporary storage unit 104. When selection of “inference” 2115 is accepted by the input unit 109, inference is executed. When the selection of “end” 2116 is received by the input unit 109, this screen can be ended.

  A result determination screen 2201 in FIG. 22 is an example of a screen for realizing the processing of the result determination unit 107 described with reference to FIG. In the comparison result 2202, the prediction result 206 derived by the inference unit 106 and the result result 2004 having the same variable ID 701 and period start date 702 as the prediction result 206 are displayed side by side. Further, the difference 2205 may be displayed. When selection of “valid” 2206 is accepted by the input unit 2206, the determination result is determined to be “valid”. When selection of “invalid” 2207 is accepted by the input unit 2206, the determination result is determined to be “invalid”. The inference engine 2102 and the end 2116 are as described above.

  A knowledge acquisition support screen 2301 in FIG. 23 is an example of a screen for realizing the processing of the knowledge acquisition support unit 108 described with reference to FIGS. 13 and 14, and in particular, an example of a screen for realizing steps S1401 and 1402. It is. The rule change history summary 2302 displays a summary of the rule data related to the inference engine 2102 acquired from the rule temporary storage unit 104 and the rule history related to the inference engine 2102 acquired from the rule history storage unit 105.

  The rule change history summary 2302 takes the classification (rule type) of rules such as IF-THEN rule information, weighting information, input variable information, and result variable information on the vertical axis, and creates rules and inferences for knowledge acquisition. The horizontal axis represents the time axis for repeated execution, and the change of the rule for bringing the prediction result closer to the actual result is displayed as an icon as exemplified in the rule 2304 in the area indicated by the horizontal axis of the vertical axis. To do. Each icon or the background of the icon may be displayed in a color-sensitive manner depending on the rule type, the type of change, the modification time, and other information. In the present embodiment, the icon is displayed on the display device in a state that can be visually discriminated by a picture or characters, and can receive a selection from the user by an input device such as a mouse or a keyboard. Character strings, pictures, buttons, menus, and combinations thereof. If the vertical axis and the horizontal axis cannot display all the data, respectively, the scroll bar can be displayed and scrolled to display all the data.

  For the icons indicating the rules arranged in the rule change history summary 2302, only information serving as points stored in the rule storage unit 103 and the rule history storage unit 105 is acquired and displayed. In the example of the rule 2304, an outline of the rule is displayed by displaying the rule number 902 and the change type 904. Further, before and after the icon indicating the rule, for example, an arrow is displayed as shown by an arrow 2303 and an arrow 2305 so that the background of the rule correction can be clearly understood. When the input unit 109 accepts selection of an icon indicating a rule (for example, a single click with a mouse), the classification of each rule (rule type) is IF-THEN rule information, weighting information, input variable information, and result variable information. 24, the IF-THEN rule information correction screen 2401 illustrated in FIG. 24, the weighting information correction screen 2501 illustrated in FIG. 25, the input variable information correction screen 2601 illustrated in FIG. The screen changes to an input variable information correction screen 2701 exemplified in 27, and each information can be corrected. When the input unit 109 accepts selection of a different type of icon indicating a rule (eg, moving the mouse cursor over the icon), the rule data acquired from the rule storage unit 103, the rule temporary storage unit, or the rule history storage unit 105 Details can be temporarily displayed on the dialog screen in FIG. Thus, in the rule change history summary 2302, for example, after changing the weight of the rule 3, the input variable information N 1 is added, and the addition of the input variable information N 1, the IF-THEN rules of the rules 8 and 9 are added, You can change the weighting of rule 8, change the IF-THEN rule of rule 2, and change the weighting of rule 1 at a glance, and if you select the icon for each rule, how You can see if it has been corrected and use it to support knowledge acquisition.

  On the left side of the rule change history summary 2302, the history 2309 is located at a position corresponding to each of IF-THEN rule information, weighting information, input variable information, and result variable information arranged on the vertical axis of the rule change history summary 2302. And a correction 2310 icon are arranged. When selection of any one of the history 2309 icons in FIG. 23 is accepted by the input unit 109, the corresponding information depends on whether the corresponding information is IF-THEN rule information, weighting information, input variable information, or result variable information, respectively. 24, the IF-THEN rule information correction screen 2401 illustrated in FIG. 24, the weighting information correction screen 2501 illustrated in FIG. 25, the input variable information correction screen 2601 illustrated in FIG. 26, or the input variable information correction screen illustrated in FIG. Transition to 2701, each information can be corrected. From the IF-THEN rule information correction screen 2401, the weighting information correction screen 2501, the input variable information correction screen 2601, and the input variable information correction screen 2701, it is possible to transition to this screen (knowledge acquisition support screen 2301).

  When the selection of “history start” 2311 in FIG. 23 is accepted by the input unit 109, accumulation of rule history in the main memory 803 is started. When the selection of “history update” 2312 in FIG. 23 is accepted by the input unit 109, the rule history accumulated in the main memory 803 is added and stored in the rule history storage unit 105. When the selection of “history reset” 2317 in FIG. 23 is accepted by the input unit 109, the rule data related to the inference engine 2102 in the rule history storage unit 105 is deleted.

  When the selection of “new creation” 2313 in FIG. When the selection of “inference execution and result determination” 2314 is accepted by the input unit 109, the screen transitions to the inference screen 2101. Since the inference engine ID 901 of the inference engine 2102 that is currently stored in the rule temporary storage unit 104 and being edited is passed to the inference screen 2101, the inference screen 2101 displays the information related to the specified inference engine ID 901 as the inference engine. 2102, environment change element information 2103, and prediction result 2104. Further, when the selection of “inference” 2115 is accepted by the input unit 109 on the inference screen 2101, inference is executed, and then the result determination screen 2201 is displayed, and the result determination screen 2201 displays the comparison result 2202. Further, when the selection of “valid” 2206 is accepted by the input unit 109 on the result determination screen 2201, it is determined to be valid, and the rule storage unit 103 and the rule temporary storage unit 104 are updated. Alternatively, when the selection of “invalid” 2207 is accepted by the input unit 109 on the result determination screen 2201, it is determined that the input is not appropriate, and the screen transitions to the knowledge acquisition support screen 2301.

  When the selection of “temporary rule storage” 2315 in FIG. 23 is accepted by the input unit 109, the latest rule being corrected is stored in the rule temporary storage unit 104. When the selection of “save rule” 2316 is accepted by the input unit 109, the latest rule being corrected is stored in the rule storage unit 103. Note that the inference engine 2102 and end 2116 of the knowledge acquisition support screen 2301 are as described above.

  An IF-THEN rule information correction screen 2401 in FIG. 24 is an example of a screen for realizing the processing from step S1403 to step S1404 of the knowledge acquisition support unit 108 described with reference to FIG. The IF-THEN rule information 2402 displays all IF-THEN rules related to the inference engine 2102. The rule number 2403 is the rule number 902 described in FIG. 9, and the data 2404 may be displayed as it is in the content 2404, or may be displayed as a graph as shown in FIG. . When the input unit 109 accepts the selection of the text of the content 2404 (“normal” in rule 1 and “equivalent to the previous month”), the text describing the rule can be changed. When the input unit 109 accepts selection of the threshold value (graph coordinates) of the content 2404, the threshold value can be changed. When the input unit 109 accepts a change in the shape of the membership function of the content 2404 by an operation such as dragging with the mouse, the threshold value can be changed. When the input unit 109 accepts the selection of the input variable of the content 2404 (in rule 1, “domestic corporate price index for 2004”), the input variable can be changed. When the input unit 109 accepts selection of a result variable of the content 2404 (“Rule 1 demand forecast for 2005”), the result variable can be changed. Also, with support from the user via the input unit 109, a plurality of rules using the same environment change element information as input variable information can be added or deleted all at once. Also, with the support from the user via the input unit 109, the same IF-THEN rule can be set for a plurality of rules using the same environment change element information as input variable information, or continuous period start dates can be set at regular intervals. You can modify the settings. In the history 2405, the history of IF-THEN rules changed in the past is displayed in time series. When the selection of the past history by the input unit 109 is accepted, the corresponding IF-THEN rule can be reused as the latest rule.

  When selection of “screen switching” 2406 is accepted by the input unit 109, the screen can transition to any of the knowledge acquisition support screen 2301, the weighting information correction screen 2501, the input variable information correction screen 2601, or the result variable information correction screen 2701. The inference engine 2102, “inference result and result determination” 2314, rule temporary storage 2315, rule storage 2316, and end 2116 are as described above.

  A weighting information correction screen 2501 in FIG. 25 is a screen example for realizing the processing from step S1405 to step S1406 of the knowledge acquisition support unit 108 described with reference to FIG. The weight information 2502 displays all the weights related to the inference engine 2102. The rule number 2403 is the rule number 902 described with reference to FIG. 9, the content 2504 simply displays the data described with reference to FIG. 9, and only the word describing the rule is displayed. The weight 2504 has a weight 907. indicate. In the history 2505, the weighting history changed in the past is displayed in time series. When the selection of the past history by the input unit 109 is accepted, the corresponding weight can be reused as the latest rule. When the input unit 109 accepts selection of the weighting 2504 (for example, a single click with the mouse), the weighting can be changed. When the input unit 109 accepts a selection of a different type of weighting 2504 (for example, moving the mouse cursor over the icon of the weighting 2504), a past history with a changed weighting is displayed in time series.

  When the selection of “screen switching” 2506 is accepted by the input unit 109, transition is made to any of the knowledge acquisition support screen 2301, IF-THEN rule information correction screen 2401, input variable information correction screen 2601, or result variable information correction screen 2701. it can. The inference engine 2102, “inference result and result determination” 2314, rule temporary storage 2315, rule storage 2316, and end 2116 are as described above.

  An input variable information correction screen 2601 in FIG. 26 is an example of a screen for realizing the processing from step S1407 to step S1408 of the knowledge acquisition support unit 108 described with reference to FIG. The input variable information 2602 displays data related to input variable information of all rules used in the inference engine 2102. The input variable information is specified by a variable ID 905 and a period start date 906, and a value 703 of environment change element information having a variable ID 701 and a period start date 702 having the same value is assigned. The input variable information 2602 is collectively displayed as an icon of one input variable information 2603 when the input variable information having the same variable ID is designated by a plurality of rules. On this icon, for example, a variable name 602 of variable information having a variable ID 601 having the same value as the variable ID 905 of the input variable information is displayed. (The economic trend index (DI) 2604 is an example of a variable name.) When the input unit 109 accepts selection of one input variable information 2603 in FIG. 26, the display unit 111 uses the input variable information. Detailed information such as the rule, variable ID, and period start date is displayed on the dialog screen in FIG. Note that the variable ID of the input variable information may be converted into a variable name and the period start date may be converted into a period 2405 and displayed in the same manner as described in the environment change element information 2105 in FIG. The environment change element information 2603 displays an icon 2606 representing information not used in the input variable information among the environment change element information stored in the data storage unit 102. Alternatively, the environment variable information used in the input variable information may be displayed after being processed so that it can be visually recognized that it is used. For each icon 2606, a variable name 2607 and a graph 2608 of environment change element information are displayed. Note that the variable ID of the environment change element information is converted into a variable name and displayed in the same manner as described for the environment change element information in FIG. Also, the graph 2608 displays the value of the environment change element information stored in the data storage unit 102 as a graph. That is, a change in value and a unit of value as time elapses are visually displayed together. Note that the graph 2608 may be temporarily displayed when one variable name 2607 is selected by the input unit 109. Further, between the input variable information 2602 and the environment change element information 2603, an addition 2609 is always displayed as an icon indicating addition of input variable information, and a deletion 2610 is always displayed as an icon indicating deletion of input variable information. Then, the change 2605 is always displayed as an icon indicating the change of the input variable information. In the history 2611, the history of variables deleted in the past and the history of variables added in the past are displayed in time series.

  When the selection of one environment change element information 2606 and addition 2609 in FIG. 26 is accepted by the input unit 109, information related to the period of the environment change element information 2606 is accepted, and a transition is made to the IF-THEN rule information correction screen. You can accept additional rules. If the addition of the rule is executed, the next time the input variable information correction screen 2601 is displayed, the input variable information 2602 displays the added input variable information. From the environment variable information 2603, the same variable ID is displayed. The environment change element information is deleted or reprocessed so that it is visually recognized that it is used for the input variable information. When selection of one input variable information 2603 and deletion 2610 is accepted by the input unit 109, the screen shifts to an IF-THEN rule information correction screen, and deletion of a plurality of related rules is accepted. If the related rule is deleted, the input variable information 2602 does not display the deleted input variable information when the input variable information correction screen 2601 is displayed next time. When selection of one input variable information 2603 and change 2605 is accepted by the input unit 109, the screen transitions to the IF-THEN rule information correction screen, and changes of a plurality of related rules are accepted.

  When selection of “screen switching” 2612 is accepted by the input unit 109, the knowledge acquisition support screen 2301 of FIG. 23, the IF-THEN rule information correction screen 2401 of FIG. 24, the weighting information correction screen 2501 of FIG. 25, or the result of FIG. Transition to any of the variable information correction screen 2701 is possible. The inference engine 2102, “inference result and result determination” 2314, rule temporary storage 2315, rule storage 2316, and end 2116 are as described above.

  The result variable information correction screen 2701 in FIG. 27 is an example of a screen for realizing the processing from step S1409 to step S1410 of the knowledge acquisition support unit 108 described with reference to FIG. The result variable information 2702 in FIG. 27 displays data related to result variable information related to all rules used in the inference engine 2102. In the result variable information 2702, all variable information 2703 stored in the data storage unit 102 is displayed. Among them, what is used in the rule as result variable information is modified so as to be visually conspicuous, for example, enclosed in a thick frame, and displayed at the top of the list. Note that the variable ID of the result variable information is converted into the variable name 2704 and displayed in the same manner as described for the environment change element information in FIG.

  When the selection of one result variable information 2704 in FIG. 27 is accepted by the input unit 109, detailed information such as a rule in which the result variable information is used and a variable ID is displayed. In addition, when selection of one result variable information 2704 or variable information 2705 is accepted by the input unit 109, the corresponding variable can be added to or deleted from the result variable information. In this case, addition or deletion is performed. The information regarding the period of the result variable information to be received is received, the screen is changed to the IF-THEN rule information correction screen, and the addition or deletion of the corresponding rule is received. If a rule is added or deleted, the next time the result variable information correction screen 2701 is displayed, a screen reflecting the corrected result variable information and variable information is displayed. In the history 2706, the history of variables changed in the past is displayed in time series.

  When the selection of “screen switching” 2707 is accepted by the input unit 109, it is possible to transition to any of the knowledge acquisition support screen 2301, IF-THEN rule information correction screen 2401, weighting information correction screen 2501, or input variable information correction screen 2601. . The inference engine 2102, “inference result and result determination” 2314, rule temporary storage 2315, rule storage 2316, and end 2116 are as described above.

  As described above, in this embodiment, the knowledge acquisition support screen allows the user to visually recognize the rule correction history and the rule classification with respect to the passage of time. The knowledge acquisition support screen allows the user to immediately transition to a correction screen corresponding to the classification of the rule while checking the rule correction process. Since only the rule history for the specified period can be stored, only the rule when the correct answer is approached can be stored, and the useless rule history can be reduced. On the input variable information modification screen, variables can be added and deleted while visually confirming the change in value of environment change element information with respect to time. In particular, inference using fuzzy, a plurality of rules related to the same variable are often defined, but a plurality of rules using the same variable can be added, deleted, and modified in a lump. In this way, in this embodiment, by using each screen comprehensively, based on the actual data with correct answers, simulation such as demand forecast that was necessary at the past time can be performed easily and easily. Add or delete environment change element information necessary to bring it closer to the correct value, and input the IF-THEN rules and weights using the environment change element information graphically into the membership function format. Will continue. You can enter your own scales and variables using the membership function threshold, and the amount and weight of the membership function can be changed freely. Since the input rule is determined immediately, it can be captured as if it were a game, and the history management of thought and error is acquired only when the correct answer is approached, so that useless rule formation can be reduced. In each screen, a rule correction history is displayed by a method according to the rule classification.

It is a block diagram of the rule acquisition system which shows Embodiment 1 of this invention. It is a figure explaining the outline | summary of the data storage part in Embodiment 1 of this invention. It is a figure explaining the outline | summary of the rule storage part in Embodiment 1 of this invention. It is a figure explaining the outline | summary of the rule log | history storage part in Embodiment 1 of this invention. It is a figure of the example of data of the time unit information in Embodiment 1 of this invention. It is a figure of the data example of the variable information in Embodiment 1 of this invention. It is a figure of the example of data of the environmental change element information in Embodiment 1 of this invention. It is a hardware block diagram of the rule acquisition system which shows Embodiment 1 of this invention. It is a figure of the example of data of the rule in Embodiment 1 of this invention. It is a figure of the example of data of the rule history in Embodiment 1 of this invention. It is a figure which shows the processing flow of the inference part in Embodiment 1 of this invention. It is a figure which shows the processing flow of the result determination part in Embodiment 1 of this invention. It is a figure which shows the processing flow of the knowledge acquisition assistance part in Embodiment 1 of this invention. It is a figure which shows the processing flow of the knowledge acquisition assistance part in Embodiment 1 of this invention. It is a schematic diagram explaining the concept of the rule in Embodiment 1 of this invention. It is a schematic diagram explaining the concept of the rule in Embodiment 1 of this invention. It is a schematic diagram which shows the example of the period change in the reasoning part of Embodiment 1 of this invention. It is a schematic diagram which shows the example of the period change in the reasoning part of Embodiment 1 of this invention. It is a schematic diagram explaining operation | movement of the rule acquisition system in Embodiment 1 of this invention. It is a schematic diagram explaining operation | movement of the rule acquisition system in Embodiment 1 of this invention. It is a figure of the example of the inference screen in Embodiment 2 of this invention. It is a figure of the example of the result determination screen in Embodiment 2 of this invention. It is a figure of the example of the knowledge acquisition assistance screen in Embodiment 2 of this invention. It is a figure of the example of the IF-THEN rule information correction screen in Embodiment 2 of this invention. It is a figure of the example of the weighting information correction screen in Embodiment 2 of this invention. It is a figure of the example of the input variable information correction screen in Embodiment 2 of this invention. It is a figure of the example of the result variable information correction screen in Embodiment 2 of this invention. It is a figure of the example of the gravity center calculation in Embodiment 1 of this invention.

Explanation of symbols

  101 rule acquisition system, 102 data storage unit, 103 rule storage unit, 104 rule temporary storage unit, 105 rule history storage unit, 106 reasoning unit, 107 result determination unit, 108 knowledge acquisition support unit, 109 input unit, 110 display unit, 203 time unit information, 204 variable information, 205 environment change element information, 206 prediction result, 207 result result.

Claims (14)

An inference section for inferring a prediction result from input variable information;
A result determination unit that determines whether or not a past prediction result estimated from past input variable information is valid by comparing with a past result result by the inference unit;
The inference unit accepts creation of a rule that defines an inference analysis procedure, the inference unit estimates a past prediction result from past input variable information, and the result determination unit converts the past prediction result into a past performance A knowledge acquisition support unit that creates a rule by repeating whether to determine whether it is appropriate compared to the result until a determination result that is appropriate is obtained;
The inference unit infers a prediction result from input variable information according to a rule created by the knowledge acquisition support unit. Environmental change element information that is a collection of actual data and has a combination of variable, time, and value;
A rule that is a combination of variables and time and is an analysis procedure for inferring prediction results from input variable information that becomes input,
An inference unit that assigns a value of environment change element information having the same variable and time to the input variable information and infers a prediction result according to an analysis procedure of the rule having the input variable information. The rule acquisition system according to claim 1. The environment change element information includes a variable ID, a period start date, and a value. The actual result includes a variable ID, a period start date, and a value. The prediction result includes a variable ID, a period start date, And the rule is IF-THEN rule information, input variable information that is an input of IF-THEN rule and includes variable ID and period start date, and a result of IF-THEN rule that is variable ID and period start It has result variable information including date, the variable ID is defined by variable information having a variable ID and a time unit ID, and the time unit ID is defined by time unit ID and a time unit information having a time range. It is what
The inference unit, the result determination unit, and the knowledge acquisition support unit are the environment change element information, the actual result, the prediction result, the input variable information of the rule, and the result variable information of the rule. For each of the above, the variable ID of each data is obtained, the time unit ID of variable information having the same value of variable ID is obtained, and the time range of time unit information having the same value of time unit ID is obtained, The rule acquisition system according to claim 2, wherein a period is recognized based on a period start date of the data and the time range. The environment change element information includes a variable ID, a period start date, and a value. The actual result includes a variable ID, a period start date, and a value. The prediction result includes a variable ID, a period start date, And the rule is IF-THEN rule information, input variable information that is an input of IF-THEN rule and includes variable ID and period start date, and a result of IF-THEN rule that is variable ID and period start It has result variable information including date, the variable ID is defined by variable information having a variable ID and a time unit ID, and the time unit ID is defined by time unit ID and a time unit information having a time range. It is what
The inference unit obtains the variable ID and period start date of the input variable information of each rule, obtains environment change element information having the same value of variable ID and period start date, and sets the value of the environment change element information. Assigning a value to the input variable information of each rule, executing inference, and using the value derived by inference as a prediction result together with the variable ID of the result variable information of each rule and the period start date The rule acquisition system according to claim 2.   If the difference between the past prediction result and the past actual result is smaller than a predetermined threshold, the result determination unit determines that the prediction result is appropriate as compared with the actual result. The rule acquisition system according to claim 1, wherein:   The rule acquisition according to claim 1, wherein the knowledge acquisition support unit sequentially accumulates and stores the rules before correction as past rules when the rules are corrected, and reuses the past rules when creating the rules. system. Connected to the reasoning unit, the result determination unit, and the knowledge acquisition support unit via a network such as a public line or a LAN, receives a rule from a user and transmits the rule to the knowledge acquisition support unit, and receives a result determination from the user. An input unit that transmits to the result determination unit, receives an inference execution request from the user, and transmits to the inference unit;
A display unit for displaying a rule transmitted from the knowledge acquisition support unit, displaying a determination result transmitted from the result determination unit, and displaying a prediction result transmitted from the inference unit. The rule acquisition system according to claim 1. Accepts the creation of a rule, which is an analysis procedure for inferring the prediction result, and stores and stores the rules before correction as past rules together with the correction time information at the time of rule correction, and reuses the previous rules at the time of rule creation With the Knowledge Acquisition Support Department,
The time information is set as one axis, the classification of the rule is set as the other axis, and in the area indicated by the vertical axis, the rule sequentially stored in the knowledge acquisition support unit is represented by a picture or a character or these A rule acquisition system, comprising: a display unit configured to display icons as combinations. An input unit that accepts selection of an icon representing a rule displayed by the display unit from a user;
9. The rule acquisition system according to claim 8, wherein when the display unit accepts selection of an icon in the input unit, the display unit transitions to a screen for correcting a rule indicated by the icon. The first area for displaying all the input variable information related to one inference or the variables of the input variable information side by side, and the values of the environment change element information related to the variables not used for all or the input variable information over time A display unit for displaying a second region for displaying the graphs shown side by side;
The selection of environment change element information displayed by the display unit and an instruction to add the environment change element information to the input variable information are received from the user, and the input variable information displayed by the display unit or the variable of the input variable information The rule acquisition system according to claim 2, further comprising: an input unit that receives selection and an instruction to delete the input variable information or the input variable information having the variable.   In the inference unit, the result determination unit, and the knowledge acquisition support unit, environmental change element information that is a combination of variables, time, and values and that has accumulated performance data is various values that represent economic trends, The rule acquisition system according to claim 1, wherein the result and the prediction result are the number of demands for the product. An inference step for inferring a prediction result from input variable information;
A result determination step for determining whether or not the past prediction result estimated from the past input variable information is appropriate as compared with the past result result by the inference step;
In the inference step, accepting the creation of a rule defining an inference analysis procedure, inferring past prediction results from past input variable information in the inference step, and past prediction results in the past results in the result judging step A knowledge acquisition support step of creating a rule by repeating the determination of whether or not the result is appropriate as compared with the result until a determination result that is appropriate is obtained,
The inference step infers a prediction result from input variable information according to the rule created in the knowledge acquisition support step. The rule is IF-THEN rule information, input variable information that is an input of the IF-THEN rule and includes a variable ID and a period start date, and result variable information that is a result of the IF-THEN rule and includes a variable ID and a period start date. The variable ID is defined by variable information having a variable ID and a time unit ID, and the time unit ID is a time unit representing an order of the time unit ID, the time range, and the width of the time range. Defined by time unit information having an order,
The inference step obtains variable IDs of input variable information and result variable information of all rules related to one inference, obtains a time unit ID of variable information having a variable ID of the same value as the variable ID, and A time unit order of time unit information having the same time unit ID as the unit ID is obtained, time unit information having the widest time range is obtained from the value of the time unit order, and the time range is obtained from the time unit information. After obtaining the time increase / decrease unit, accept the number of positive or negative time increase / decrease units, and increase / decrease the period start date of the input variable information and the result variable information. The rule acquisition method according to claim 12, wherein the period is changed. The rule is IF-THEN rule information, input variable information that is an input of the IF-THEN rule and includes a variable ID and a period start date, and result variable information that is a result of the IF-THEN rule and includes a variable ID and a period start date. The variable ID is defined by variable information having a variable ID and a time unit ID, and the time unit ID is defined by a time unit ID and a time unit information having a time range. ,
The inference step accepts the number of positive or negative time increase / decrease units with 1 day as the time increase / decrease unit, and increases or decreases the period start date of the input variable information and the result variable information to make one inference. The rule acquisition method according to claim 12, wherein the period of all the rules is changed.

Images