JP2005339529A - Expert system creation device and executing method for the same device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a versatile expert system creation device and its executing method capable of lessening a burden on a knowledge engineer and producing an appropriate expert system without requiring so much time and labor. <P>SOLUTION: The expert system creation device comprises a case acquiring means 10 forming a case base by acquiring a case composed of an attribute and an attribute value, a rule generation means 20 generating a determining tree from the case base, and a consulting means 30 obtaining an attribute value of conclusion by referring to the generated determining tree. The case acquiring means 10 has a built-in case acquiring interface 10a for acquiring cases from an external system, and automatically stores the cases from a memory of a control unit arranged in the external system via the case acquiring interface 10a, and forms the case base, and updates the case base. The rule generation means 20 accumulates the cases, and automatically updates the determining tree. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention derives a statistical tendency that results from the situation from the accumulation of the observable situation and the accompanying results. In particular, the situation and the target of the result are the situation judgment and action realized by a person (expert). In some cases, the present invention relates to an expert system generation apparatus and its implementation method for guiding knowledge about human behavior as a statistical tendency, and ruled and utilizing the derived tendency. These inventions are suitable for using expert knowledge in industry, mining, agriculture, forestry, fishery, transportation, medical, financial, commerce, and all other industries.

  Conventionally, an expert system is known in which causal relations are derived from all events (examples) in social activities, ruled out, and results found under certain conditions. In this expert system, various events (cases) in social activities are memorized as a case base, and the computer makes an inference based on the case base, and substitutes for experts and the like. The main operation of the expert system is to search for a rule that matches the inference object from the case base, and interpret and execute the corresponding rule.

  In the inductive learning method for knowledge acquisition, a case is expressed by a set of “condition attribute” and “conclusion attribute”, and a rule that estimates the “conclusion attribute value” from the “condition attribute value” observed based on the case is determined by a decision tree. Or in the form of a Prolog equation.

For example, a method for generating a rule as a decision tree outputs a propositional logic rule that depends on the environment in the form of a decision tree (see, for example, Patent Document 1).
JP 09-330224 A

  Such an expert system that uses a decision tree as an output format can output a conclusion attribute value from an attribute value according to a rule prepared in advance in response to a preset question item (attribute). it can. In other words, it is possible to provide a guideline “in this situation, do this”. For this reason, systems that generate this expert system in industries, mining, agriculture, forestry, fisheries, transportation, medical, financial, commercial, and all other industries (eg case learning systems, data mining systems, experience learning systems) , Statistical analysis, etc.), it is possible to absorb and use the knowledge of experts and the like.

  However, in the above-mentioned conventional expert system, in order to derive the causal relationship from the case and make a rule and find the result under a certain condition, a skilled knowledge engineer selects what factor as an attribute and what attribute. It is necessary to judge and decide whether to select as a conclusion attribute. That is, a knowledge engineer must extract expertise from experts, organize and select those expertise. For this reason, knowledge engineers need to have long interviews with experts. In addition, after the interview is over, the problem must be organized from the records, and facts and rules, questions and answers, causal links, assumptions and conclusions, probability functions, etc. must be confirmed. This requires well-trained and skilled knowledge engineers.

  For this reason, in order to use an expert system, enormous effort, time, and expense will be required. Further, in order to change the rule once fixed so as to be more appropriate, it is necessary to select a new attribute, input an observation attribute value, and the like, which again requires enormous effort, time and cost. Furthermore, no matter how good the knowledge engineer is, there are human limitations and it is difficult to do any further work. In addition, an estimation rule must be set for each application target of the expert system, and special work is required to perform partial segmentation of the rule or combination with other rules.

  The present invention has been made in view of the above-described conventional situation, has a low burden on knowledge engineers, can create rules without much time and effort for use, and is versatile. It is a problem to be solved to provide an expert system generation apparatus and an implementation method thereof.

The expert system generation device of the present invention is an expert system generation device for constructing an expert system,
Case acquisition means for accumulating a situation observed from an external system to be constructed of the expert system and a result associated therewith,
Rule generation means for automatically learning the accumulated situation and result, and for deriving as a knowledge a statistical tendency resulting from the situation;
A consulting means for automatically judging the result for a certain observed situation from the derived statistical tendency,
Update the situation and the results by operating the case acquisition means, the rule generation means, and the consulting means independently or in cooperation with each other, and automatically update them to more appropriate knowledge. It is characterized by.

  In the expert system generation apparatus of the present invention, first, the situation that can be observed from the external system that is the construction target of the expert system and the accompanying results are accumulated by the case acquisition means. Then, the situation and the result accumulated by the rule generation means are automatically learned, and a statistical tendency for producing the result is derived as knowledge from the situation. Here, the external system means not only an artificially manufactured apparatus such as a manufacturing apparatus or a processing apparatus, but also all systems to which the expert system is applied, such as natural changes and human behavior. The knowledge of statistical trends derived in this way is not unchanging, and the situation and results are updated by operating the case acquisition means, rule generation means and consulting means independently or in cooperation with each other. Automatically updated to a more reasonable knowledge. For this reason, even if knowledge is automatically judged by consulting means at first, even if knowledge is lacking in validity, the situation and results are subsequently updated, and knowledge is automatically updated. It can change to. For this reason, in order to obtain highly relevant knowledge from the beginning, the burden of hiring skilled knowledge engineers, accurately extracting expert knowledge from experts, organizing and selecting those expertise is reduced. The It also reduces the burden of organizing problems from expert interviews and their records to accurately confirm facts and rules, questions and answers, causal links, assumptions and conclusions, and probability functions. This makes it possible to create an expert system without being a well-trained and skilled knowledge engineer. In addition, while the burden on the conventional knowledge engineer is continuous, this expert system generation apparatus greatly reduces the continuous burden. Furthermore, labor, time, and cost for constructing the expert system are reduced.

  Therefore, the expert system generation device of the present invention has a low burden on knowledge engineers, can create an appropriate expert system without much time and effort, and is excellent in versatility.

  In the expert system generation apparatus of the present invention, the case acquisition means acquires and accumulates the status and results relating to the external system that is the construction target of the expert system as data, and the rule generation means is acquired and stored by the case acquisition means. A statistical trend is automatically extracted from the data in the form of a rule, and the consulting means refers to the rule generated by the rule generating means to automatically determine the result for the situation from the current data of the external system, and The case acquisition means, the rule generation means and the consulting means can be operated independently, and the rule generation means automatically cooperates with the case acquisition means and the consulting means and automatically collects the rules along with the accumulation of the data. Updated to a more appropriate rule. Can.

  In this case, first, data relating to the external system to be constructed by the expert system is acquired by the case acquisition means, and the causal relationship between the data is automatically extracted in the form of rules by the rule generation means. Here, the data relating to the external system means a numerical value or a symbol value representing the state of the external system. The rules generated in this way are not immutable, and are updated to a more suitable rule by automatically updating the rules together with the accumulation of data acquired thereafter. For this reason, even if the rule generated initially lacks validity, the rule can gradually change to a highly valid rule with the subsequent acquisition of data. For this reason, there is little burden on knowledge engineers, it is possible to create an appropriate expert system without much time and effort, and it is excellent in versatility.

  In such an expert system generation device, data acquired by the case acquisition unit can be used as a case composed of attributes and attribute values, and a conclusion derived by the consulting unit can be obtained as a conclusion attribute value.

  The rules generated by the rule generation means in the expert system generation apparatus of the present invention can be generated in the form of a decision tree, a Prolog expression, a logical expression, a function in a statistical analysis method, or the like.

  When the rule generation means generates a rule in the form of a decision tree, it is preferable that the decision tree can be automatically generated regardless of the number of cases. When actually trying to acquire a case from the industrial field, a part of the case may be unknown. Also, there are cases where it is not possible to accurately determine in advance what cases are necessary and what cases are unnecessary. Therefore, if a decision tree can be automatically generated regardless of the number of cases, such as cases with unknown condition attribute values or cases with unknown conclusion attribute values, a decision tree is generated for the time being. In addition, it is possible to update to a more appropriate decision tree later by the decision tree automatic updating means. For this reason, the burden on knowledge engineers can be further reduced, and the applicable range can be further expanded. Regardless of the number of cases, the specific method for automatically generating decision trees is to skip unknown values when calculating the amount of information so as not to affect the probability. And so on.

  Moreover, it is preferable that a rule production | generation means can set a conclusion attribute value to a fixed value with respect to a predetermined example. In this case, if the attribute value of the conclusion attribute is clearly wrong, it can be corrected to an appropriate value. Further, when a conclusion is uniquely determined under an arbitrary condition, it can be set as an appropriate attribute value of the conclusion attribute. For this reason, the estimation rule obtained by the expert system generation device of the present invention can be tailor-made more adapted to the current situation.

  Further, the rule generating means, when prior knowledge about the distribution of attribute values of the acquired cases is obtained, for the unknown conclusion attribute value derived as a result of tracing the decision tree generated once, It is preferable to add a case including an attribute value. If this is the case, an appropriate decision tree can be generated even if a small number of conclusion attribute values are unknown by generating missing cases based on condition attributes that appear on the decision tree before reaching an unknown conclusion attribute value. Can be generated.

  Further, it is preferable that the consulting means automatically obtains the condition attribute value from the outside and derives the conclusion attribute value from the obtained decision tree. In this way, consulting can be automated. As such consulting means, an external system in the form of an execution program is automatically started, and an output value of the external system is converted into a condition attribute value and acquired. If this is the case, for example, if there is a program for accessing the outside via a network on the computer on which the expert system generating apparatus is operating, the outside of another device coupled by the network via the system You can control the program and get examples.

  Furthermore, the consulting means preferably applies the conclusion attribute value to the external system. If it is like this, it will become possible to control an external system automatically based on the result of consulting. Examples of such consulting means include automatically starting an external system in the form of an execution program, converting a conclusion attribute value obtained by the consulting means into an input value of the external system, and the like.

  Further, it is preferable that the rule generation means evaluates the validity of the generated decision tree and updates the decision tree with a higher validity based on the result. Examples of such rule generation means include that exceptions of cases can be made, exceptions of condition attribute items, and pruning of branch branches can be made. If case exceptions or condition attribute items can be made exceptions, a more realistic decision tree can be generated. In addition, if the branch branch can be pruned, the decision tree is simplified and the analysis is facilitated when the case becomes more complex than necessary and analysis of the case becomes difficult. Here, it is preferable that pruning of branch branches can be performed between any condition attribute items and between condition attributes and conclusion attributes. In this way, only a part of the decision tree can be changed, simplified, or deleted, and the decision tree can be edited more realistically. In addition, when a plurality of conclusion attribute values are derived for one conditional case as a result of tracing the decision tree, it is possible to regenerate the decision tree with the corresponding case as an exception. . In this case, even in the case where the conclusion attribute values conflict with each other in the current case, it is possible to generate a decision tree.

  The case acquisition means is preferably configured to be able to automatically acquire a case. If this is the case, it is possible to automatically update the decision tree, and it is possible to greatly reduce the labor of the worker required to acquire the case. Examples of the case acquisition means include automatically starting an external system in the form of an execution program and converting the output value of the external system into a condition attribute value. If this is the case, for example, if there is a program for accessing the outside via a network on the computer on which the expert system generating apparatus is operating, the outside of another device coupled by the network via the system You can control the program and get examples.

  When rule generation means generates rules in the form of decision trees, information obtained by experts as experience is accumulated as examples, and multiple knowledge structures that include the same knowledge elements and knowledge elements that constitute knowledge are separated. Knowledge that has a multi-view knowledge structure that is a hierarchical knowledge structure that is a superordinate concept or subordinate concept of knowledge, or a special knowledge structure of the multiple knowledge structure, and is a collection of knowledge from many different viewpoints It is also preferable to extract as a decision tree.

  In this way, knowledge with multiple knowledge structures, hierarchical knowledge structures, and multi-view knowledge structures can be extracted as decision trees, making it easy to grasp knowledge elements from various aspects in multiple layers Thus, a more accurate decision tree can be generated.

  In this case, the rule generation means is preferably configured as follows in order to enable generation of a decision tree according to the purpose. That is, (1) For each attribute item of the case, it is possible to arbitrarily select whether it is a condition attribute or a conclusion attribute. (2) Generate only decision trees corresponding to specific conclusion attributes.

  Further, the rule generation means can have a plurality of conclusion attribute items for the case base in which cases are accumulated, and can include a user interface for comparing decision trees corresponding to the respective conclusion attributes. In this way, it is easy to determine what is an important conclusion attribute among a plurality of conclusion attribute items.

  Further, the rule generation means may be capable of expanding and concatenating another decision tree with the conclusion attribute value of the generated decision tree. If this is the case, it is possible to further develop the causal relationship with respect to the causal relationship, and it is possible to develop the analysis of the causal relationship.

  In addition, the rule generation means is capable of performing detailed concatenation of the generated decision tree condition attribute item with another decision tree having all attribute values of the condition attribute item as conclusion attribute values. You can also. In this way, a more detailed decision tree can be generated for an arbitrary condition attribute value, so that a more detailed causal relationship can be obtained for the causal relationship.

  Furthermore, when the conclusion attribute value is a numerical type, the rule generation means can approximate the conclusion attribute value with a regression line and obtain an approximate residual as the width of the conclusion attribute. If this is the case, it is possible to divide the numerical value of the conclusion attribute statistically as a numerical range and configure a rule as a conclusion of the decision tree, and thus it is possible to estimate the range of the conclusion value that has not been observed as an example.

  The implementation method of the expert system generation device of the present invention can be implemented using the expert system generation device of the present invention. That is, the implementation method of the expert system generation device of the present invention is characterized in that expert knowledge is ruled by using the expert system generation device according to any one of claims 1 to 25.

  Furthermore, in the implementation method of this expert system production | generation apparatus, the expert's action can also be automated by the ruled expert's knowledge. In this way, it is possible to absorb and use the knowledge of experts in industries, mining, agriculture, forestry, fisheries, transportation, medical, financial, commercial and other industries. For example, it is possible to support a person's work or educate a person with the knowledge of a ruled expert.

  An expert can be an expert on an industrial process. If this is the case, use the knowledge of specialists in industrial processes as an asset and use it for automatic operation of industrial processes, support of industrial process drivers, education of industrial process drivers, etc. Can do. The industrial processes here include petrochemical plants such as refining plants using rectification towers, robots, steel production sites, ceramics production sites, power plants, medical sites, and other industrial equipment and machinery. Is done. A refinery plant using a rectification column is based on the temperature of the rectification column, the sound generated during boiling, the state of bubbles observed from the observation window provided in the rectification column, etc. It has been operated based on many years of experience and intuition, and if the method of implementing the expert system generation apparatus of the present invention is used in such a refinery plant, control equivalent to that of a skilled expert is possible.

  When the industrial process is a refining plant using a rectifying tower, the acquisition of the case from the refining plant is at least one of a temperature sensor, a pressure sensor, a flow sensor, a sound sensor, and an image sensor provided in the rectifying tower. Can be made through If this is the case, the temperature, pressure, flow rate, sound generated at the time of boiling, the sound generated at the time of boiling, which are important attributes and attribute values for controlling the purification plant using the rectifying column, It is possible to automatically capture the state of generated bubbles as an example. In addition, as a case acquisition method in the case acquisition means, a device for acquiring data already instrumented and the case acquisition means can be directly connected. In this case, it is possible to effectively use past data recorded in a data acquisition device installed in an industrial process.

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, an embodiment embodying an expert system generation device and an implementation method of the invention will be described with reference to the drawings.

  As shown in FIG. 1, the expert system generation apparatus according to the embodiment includes case acquisition means 10, rule generation means 20, and consulting means 30. The case acquisition means 10 incorporates a case acquisition interface 10a for acquiring a case from an external system as a case acquisition target. Data related to various cases sent from various sensors provided in the external system is automatically stored in the data file as cases consisting of the attributes and attribute values of each data via the case acquisition interface 10a to form a case base. It is supposed to be. The case acquisition means 10 can also input data directly from the keyboard. Furthermore, the case acquisition means 10 directly accesses the storage unit of the control device via the case acquisition interface 10a, and stores the recorded various operation data externally. It is possible to automatically acquire a case by controlling the sequencer of the system or to read a case from a text file in a specified format.

  The rule generation unit 20 includes a generation preparation block 21, a tree generation block 22, and a tree check block 23. The generation preparation block 21 checks whether or not a decision tree can be generated from a case (for example, when there is no conclusion attribute (hereinafter referred to as “target element”) in the case or when there is no case). Or a boundary value in a numeric conclusion attribute value (hereinafter referred to as “target element value”) is determined using a regression line.

Further, as shown in FIG. 2, the tree generation block 22 includes a part related to the block function related to ID3 and a part related to the block function related to the branch generation of the decision tree, and is based on the data created by the generation preparation block 21. Then, arithmetic processing for decision tree generation is performed. The block function related to ID3 is composed of a numeric type binarized block function and an information amount calculation block function. The numeric binary block function binarizes binary numeric or date type element values and divides the binarized element values into two clusters. The information amount calculation blocking function performs calculation related to the information amount such as information amount and information entropy of each element.

Further, in the tree check block 23, as shown in FIG. 3, a check regarding the decision tree generated by the tree generation block 22, such as presence / absence of duplicate target element values, is performed.

Moreover, the consulting means 30 is comprised from the consult preparation block 31, the consult block 32, and the output block 33, as shown in FIG.

  The consult preparation block 31 is a block for acquiring information necessary for consulting (data to be traced to a decision tree, information on detailed connection and expansion connection, information on a fixed target case described later, and the like). The consult block 32 is a block for performing consulting by tracing consult data to a decision tree. Further, the output block 33 is a block for outputting the result of consulting by the consulting block 32. The output of the target element obtained by consulting, its value, and the element information of the conditional branch after the trace fails, And output the trace result.

<Flowchart>
The expert system generation apparatus configured as described above generates a decision tree in the following procedure, and further consults based on the decision tree.

(Decision tree generation)
The decision tree is generated according to the flowchart shown in FIG. That is, as step S000, the exception flag is checked based on the case acquired from the external system by the case acquisition means 10, and the case excluding the exception case is registered in the memory. Next, in step S100, whether or not a case exists, and all cases are exception cases (where an exception case is an exceptionally determined target regardless of the value of the target element value based on the generated rule) Whether or not a decision tree can be generated is checked. As a result, when it is determined that the decision tree cannot be generated, the processing is ended there. On the other hand, if it is determined that the decision tree can be generated, the range of the numerical target is acquired in step S200 when the target element is a numerical target. That is, a single regression analysis is performed on the condition attribute value (hereinafter referred to as “element value”) and the target element value in each case to obtain a regression line. Then, using the obtained regression line, the ratio of the difference between each target element value and the regression value is obtained, and the boundary value is obtained from the ratio.

In step S300, a decision tree is generated. FIG. 6 shows step S300 in more detail. That is, first, in step S301, it is checked whether or not a decision tree already exists. If there is a decision tree, all nodes in the decision tree are deleted in step S302. If they do not exist, the theme name of the target element, the target Are registered in the tree database. In step S304, the head node is registered in the tree database, and in step S305, calculation based on ID3 is performed.

FIG. 7 is a flowchart showing the calculation by ID3 in step S305 in more detail. First, in step S3051, preprocessing for calculating the information amount is performed. Specifically, as shown in FIG. 8, it is determined whether the case data is a logical type (that is, a numeric type or a date type) or not a logical type. If it is not a logical type, binarization is performed in step S30511. Do. FIG. 9 is a flowchart showing step S30511 in more detail. That is, first, each element value is counted as step S305111, and an expected value is calculated as step S305112. If the expected value is less than or equal to the predetermined value, Fisher's test is performed (step S305113), and if the expected value exceeds the predetermined value, a chi-square test is performed (step S305114). As a result, if binarization is possible, the boundary value is acquired (step S305115), and the above processing is repeated from step S30511 for another element value. If binarization could not be performed, the above processing is repeated from step S30511 for another element value without acquiring a boundary value. Thus, after repeating the process only (number of cases-1) times, a cluster is created from the boundary value obtained in S305116. Then, as shown in FIG. 8, in step S30512, the number of element values and the number of target element values are counted for each element value in the case. In this case, when an unknown value value exists in the element value value, the unknown value value is not counted. When the value of the element value is a numeric type or a date type, the value is counted for each value.

  Then, as shown in FIG. 7, in step S3052, an information element is obtained from each element, and this is looped by the number of elements, thereby obtaining an overall set entropy and obtaining an information amount. Furthermore, as step S3053, the acquisition of the element with the largest amount of information is repeated for the number of elements. In step S3054, an element and an element value are registered. Furthermore, if the flag for branch pruning is on in step S3055 and the number of nodes from the beginning matches the specified number of pruning nodes, the target element and the target element value are registered. . That is, in step S3055, as shown in FIG. 10, first, in step S30551, it is determined whether or not the value is the same as a predetermined target element (hereinafter referred to as “fixed target element”). Determines whether a value exists in step S30552. If there is a value, in step S30553, the operation of acquiring the target value of the matching case is repeated by the number of cases. In step S30551, if the value is the same as the value of the fixed target element, it is determined whether or not the value exists. If there is a value, in step S30553, the operation of acquiring the target value of the matching case is repeated by the number of cases. In this way, the fixed target value is acquired and step S3055 is ended. On the other hand, if no value exists in step S30552, step S3055 is terminated without performing steps S30553 and 30554.

  On the other hand, when the flag for branch pruning in FIG. 7 is in an on state and the number of nodes from the top does not satisfy any of the specified pruning node numbers, As shown in FIG. 7, a target arrival determination is performed in step S3056. That is, in this step S3056, as shown in FIG. 11, when the element value entropy is zero and the information amount of all the elements is zero or the number of element values is 1 or less, the target in the previous branch is set. To reach. Further, when the element value entropy is zero and the information amount of all the elements is zero or the number of element values is other than one, the target value is reached.

  On the other hand, if the entropy of the element value is not zero, in step S30561, a target arrival determination with information amount is made. That is, in step S30561, as shown in FIG. 12, when the element has the largest amount of information and the information amount of the element is less than or equal to zero, the target is reached at the previous branch. For elements other than those with the largest amount of information, it is determined whether the information amount of the element is greater than zero. If the information amount of the element is greater than zero, the next conditional element is searched, and so on. If not, reach the target.

As a result of the target arrival determination in step S3056 shown in FIG. 7, when the target has not been reached, in step S3057, a case is reconstructed for the remaining elements for each element value of the previous element, and further in step S3058. , Calculation based on ID3 is performed.

  After the decision tree is created in step S300 shown in FIG. 5, the target value is checked in step S400. That is, as shown in FIG. 13, first, in step S401, the target value number of the node is acquired. If there are two or more target values, a process of multiple target values is performed as step S402. FIG. 14 is a flowchart showing step S402 in more detail. That is, (1) a plurality of target values are displayed as they are, (2) a value is invalid and a target value node is deleted, (3) generation of a decision tree is canceled, or (4) target values are forced Select whether to decide.

  When the number of target values is 0 (that is, the target value is unknown), processing without target value is performed as step S403. That is, as shown in FIG. 15, (1) the target value is ignored as invalid, (2) decision tree generation is stopped, or (3) an element appears on the tree before reaching an unknown target value. Or (4) whether to forcibly assign the target value by the value selection requester.

Finally, as step S500 shown in FIG. 5, only the bits to be generated for the exception processing are set, and the generation of the decision tree ends.

(consulting)
Based on the decision tree generated as described above, consulting is performed according to the flowchart shown in FIG. In other words, in step S600, the case file to be consulted, the consulting environment setting file (element value is automatically acquired from the outside, manually acquired, the remaining elements that have not been traced are output, the style is set again. A file in which selection information to be generated is registered) and a history file (a file in which elements already traced in the process of tracing a tree are recorded) are read. Next, as step S700, the decision tree is traced. In step S700, as shown in FIG. 17, when the tree node is an element, the element value is traced as step S701, and when the tree node is the target element, the target element value is traced as S702.

  The element value tracing in step S701 is performed according to the flowchart shown in FIG. That is, first in step S7011, it is determined whether or not value can be acquired from the history file called in in step S600. If value acquisition is possible, a log is taken and the tree is traced in step S7012. On the other hand, if the value cannot be acquired, it is determined whether or not there is a detailed connection flag. If there is a detailed connection flag, a tree related to the detailed connection is traced in step S7013. On the other hand, if there is no detailed connection flag, it is determined whether or not acquisition is performed by an external program. If so, value is acquired from the external program, a log is taken as step S7012, and the tree is traced. Also, when not acquiring value from an external program, (1) trace the target element value after acquiring the value by taking the standard deviation, (2) trace the tree by inputting the value by manual input, (3) Select the remaining elements and output them to a file, or (4) Reconstruct the tree again.

  In addition, as shown in FIG. 19, the target element value trace in S <b> 702 compares the fix case with the consult process (condition attribute passed in the middle of tracing the tree) as S <b> 7021, and further determines the presence or absence of expansion connection, If there is no expansion connection, an external program is started based on the obtained target element value in step S7022. If there is an expansion connection, a tree based on the connection is traced in step S7023. This completes the consulting.

Hereinafter, a case where a decision tree is generated using this expert system generation apparatus and consulting is performed will be described more specifically.

<Case acquisition process S1>
First, as the case acquisition step S1, a theme to be performed by the expert system is determined, and an element (condition attribute) and an element value (condition attribute value) to be used for the theme are registered. Further, a target element (conclusion attribute) and a target element value (conclusion attribute value) are registered. The registration can be automatically performed from various sensors provided in the external system via the case acquisition interface 10a shown in FIG. It is also possible to input data directly from the keyboard. If the case base is in a text file format, it can be called up. Furthermore, it is also possible to directly access the storage unit of the control device via the case acquisition interface 10a and automatically acquire various recorded operation data as cases. The case registered in this way is displayed on the computer screen as a case editing panel as shown in FIG. 20, and is filed in the computer. In addition, registered elements and target elements can be freely added or deleted. Furthermore, “examples” composed of combinations of elements, element values, target elements, and target element values can be freely added or deleted. Further, the element and the target element can be freely exchanged. Furthermore, when it is determined that the target element value is clearly wrong from the conditions, or when the target element value is determined under any condition, a fixed value may be registered as the fixed target element value. Yes (FIG. 21). Also, any case can be selected as an exception process by selecting a case to be an exception in the case editing panel and clicking “Make exception” on the edit menu.

<Decision Tree Generation Step S2>
Next, on the case editing panel displayed on the computer screen, by selecting decision tree generation from the function menu, a decision tree corresponding to all target elements is generated, and the currently selected decision tree Select only when to select. By selecting display of the decision tree from the function menu, a decision tree display panel shown in FIG. 22 is displayed. Furthermore, the decision tree is pruned by selecting an arbitrary node in the decision tree display panel with the mouse and clicking pruning in the function menu (FIG. 23). Here, pruning the decision tree is the same operation as making the corresponding case an exception. Thereafter, by moving to the case editing panel, cases that become unnecessary when the decision tree is reconstructed are automatically handled as exceptions (FIG. 24). Note that pruning can be performed not only between elements but also between an element and a target element.

  Also, when analyzing the generated decision tree, a decision tree based on other attributes can be generated by making the top node an exception in the style editing panel. For example, FIG. 25 is a top node portion of a decision tree obtained by baking a test piece of incinerated ash by sludge treatment, determining a melting grade from its gloss and dissolution degree, and collecting various data in sludge treatment along with this. is there. Here, with the exception of the element “incinerator” at the top node, a decision tree that does not depend on the incinerator can be derived. As a result, as shown in FIG. 26, the element “MgO” comes to the top node, and it can be seen that the content of magnesium oxide is an important factor in addition to the factor of the element of the incineration plant.

Further, in the generation of a decision tree, if the majority of the decision tree is occupied by the same element, there is a high possibility that the element is noise that specializes the rule. In such a case, it becomes easier to find an important element that has been missed by excluding the elements that have become identical as exceptions. For example, by excluding the “date” element in the decision tree shown in FIG. 27, the “SiO ₂ ” element appears as shown in FIG. Since silicon oxide has a function of enhancing the melting of the substance, it is presumed that the silicon oxide has a great influence on the value of the melting grade. This inference can be easily made by observing a decision tree with the exception of the “SiO ₂ ” element. In this way, it is possible to efficiently analyze the decision tree by repeating inference and rule induction.

  In addition, when performing expansion concatenation that joins another decision tree to the target element value of the generated decision tree, the target element value selection requester is displayed from the edit menu, and the target element value list can be expanded. Select the target element value to expand. When the expansion connection is thus completed, the expansion connection mark is displayed in front of the target element value name in the style editing panel as shown in FIG. In this way, further decision trees can be developed and generated for any target element value, and further causal relations can be developed for causal relations, and analysis of causal relations can be developed.

  In addition, when performing detailed concatenation that combines another decision tree with all element values as target element values to the element value of the generated decision tree, the attribute item selection requester is displayed using the edit menu. Click to select the element to be detailed from the element. When the detailed connection is completed in this way, a detailed connection mark is displayed in front of the connected element name in the style editing panel as shown in FIG. In this way, since a more detailed decision tree can be generated for any element value, it becomes possible to obtain a more detailed causal relationship for the causal relationship.

<Consulting process S3>
After the decision tree is generated in the decision tree generation step S2, “style-consulting” in the upper menu of the screen is selected in the case editing panel (FIG. 31). As a result, as shown in FIG. 32, a target element selection dialog is displayed, and the target element can be selected. When a target element is selected, consulting according to a decision tree for each target element is started. The consulting means 3 can automatically start an external system in the form of an execution program, convert an output value of the external system into an element value, and acquire it. The consulting means can automatically start an external system in the form of an execution program, convert the target element value obtained by the consulting means to an input value of the external system, and deliver it. Then, as shown in FIG. 33, responses corresponding to the element type such as logical type, numerical type, and date type are selected or entered, and when all the responses have been entered, the value of the target element is concluded as a conclusion. Is derived. Here, when the element is of a logical type, the input is merely a selection. Even in numerical type elements such as age, in the case of a person whose age is unknown, it can be treated as “logic unknown” as “age unknown”. Thus, even if there is a shortage of data, it is possible to generate a decision tree and consult.

(Operation of recovery solvent purification plant using rectification column)
Using the expert system generator of the embodiment, the recovery solvent purification plant using a rectification column was operated.

  This recovery solvent plant is for separating and recovering tetrahydrofuran (THF) and methyl ethyl ketone (MEK) from the recovery solvent. As shown in FIG. 34, the storage tank stores the recovery solvent in which THF and MEK are mixed. 11 and two rectifying towers 12 and 13. Pipes 11 a and 11 b are connected to the storage tank 11, and the other ends of the pipes 11 a and 11 b are connected to the ninth stage of the rectification columns 11 and 12. Heat exchangers 12a and 13a are provided at the lower ends of the rectifying columns 12 and 13, and the lower ends of the rectifying columns 12 and 13 can be heated by supplying steam from the outside. Further, the bottom ends of the rectifying columns 12 and 13 are connected to can outlets 12b and 13b, respectively, so that the bottoms collected at the lower ends of the rectifying columns 12 and 13 can be taken out. Further, distillation pipes 12 c and 13 c are connected to the upper ends of the rectifying columns 12 and 13, and the distillation pipes 12 c and 13 c are connected to one end side of the capacitors 14 and 15. The other ends of the capacitors 14 and 15 are connected to the catcher tanks 16b and 17b via the recovery pipes 16 and 17, respectively. Further, reflux pipes 16a and 17a branch from the middle of the collection pipes 16 and 17 and are connected to the can pipes 12b and 13b so that a part of the bottom can be refluxed to the rectification columns 12 and 13. .

  As information transmission means for grasping the operation status of the recovered solvent refining plant and obtaining information on the cases, thermometers TR1 to 11, flow meters V1 to 8 and pressure gauges P1 and P2 (not shown) are located at the positions shown in Table 1. Is installed.

  The operation of the recovered solvent purification plant was performed by three skilled operators.

(Case acquisition)
The value of each of the thermometers TR1 to 11, the flow meters V1 to 8 and the pressure gauges P1 and P2, the temperature and the date / time are automatically transferred from the recording unit of the control device (not shown) to the case acquisition means 1 via the automatic acquisition interface 1a. I took it in. Further, the weather, operator, presence / absence of instrument check, etc. were directly input to the case acquisition means 1 from the keyboard. In this way, the case base divided into Tables 2 to 4 was created by the case acquisition means 1.

(Decision tree generation)
Furthermore, a decision tree is generated by the rule generation means 3 based on the case base.

(consulting)
Furthermore, consulting was conducted based on the decision tree generated in this way.

  By applying this expert system generator to the recovery solvent purification plant, the case base is automatically formed by the case acquisition means 1 based on the information from the thermometers TR1 to 11, the flow meters V1 to 8, and the pressure gauges 1 and 2. The decision generation unit 2 can automatically generate a decision tree from this case base. And the consulting means 3 can obtain | require the target element value of a target element automatically with reference to this decision tree. For this reason, the decision tree initially generated by the rule generation means 3 is constantly updated automatically with the accumulation of the case base, and is updated to a more suitable decision tree for case learning. It becomes. Even when the cause-and-effect relationship changes for some reason as time passes, an appropriate decision tree is generated following the change. For this reason, the target element value obtained by the consulting means 3 can also be set to a more appropriate value.

  In the above-described embodiment, the information from the thermometers TR1 to 11, the flow meters V1 to 8 and the pressure gauges 1 and 2 is taken from the storage unit of the control device, but the information is obtained in real time via the interface 1a. It can also be sent to the case acquisition means.

  The present invention can be used in all industries by utilizing the knowledge of experts in industries, mining, agriculture, forestry, fisheries, transportation, medical, financial, commercial, and all other industries.

It is a block diagram of the expert system production | generation apparatus of embodiment. It is explanatory drawing of the function of a consultation tree production | generation block. It is explanatory drawing of the function of a tree check block. It is a block diagram of a consulting means. It is a flowchart regarding decision tree generation. It is a detailed flowchart of step S300. It is a detailed flowchart of step S305. It is a detailed flowchart of step S3051. It is a detailed flowchart of step S30511. It is a detailed flowchart of step S3055. It is a detailed flowchart of step S3056. It is a detailed flowchart of step S30561. It is a detailed flowchart of step S400. It is a detailed flowchart of step S402. It is a detailed flowchart of step S403. It is a flowchart regarding consulting. It is a detailed flowchart of step S700. It is a detailed flowchart of step S701. It is a detailed flowchart of step S702. It is a figure which shows a case edit panel. It is a display screen for registering a fixed value as a fixed target element value. It is a display screen which shows the produced | generated decision tree. It is a display screen which shows the decision tree after the end of pruning. It is a display screen which shows the example automatically exception-processed by pruning. It is a display screen which shows the decision tree obtained about the operation | work which bakes the test piece of the incinerated ash by sludge processing. It is a display screen which shows before and after performing exception processing of an element. It is a part of a display screen showing a decision tree before an exception of a date element. It is a part of a display screen showing a decision tree after the date element is an exception. It is a display screen which shows the style edit panel after completion | finish of expansion | deployment connection. It is a display screen which shows the style edit panel after detailed connection is completed. It is a display screen which shows the case where consulting is selected in a case edit panel. It is a display screen which shows the case where a target element is selected in a case edit panel. It is a figure which shows the input screen in the case of performing consulting. It is a schematic diagram of a recovery solvent purification plant.

Explanation of symbols

1 ... Case acquisition means 2 ... Rule generation means 3 ... Consulting means

Claims (35)

An expert system generator for building an expert system,
Case acquisition means for accumulating a situation observed from an external system to be constructed of the expert system and a result associated therewith,
Rule generation means for automatically learning the accumulated situation and result, and for deriving as a knowledge a statistical tendency resulting from the situation;
A consulting means for automatically judging the result for a certain observed situation from the derived statistical tendency,
Update the situation and the results by operating the case acquisition means, the rule generation means, and the consulting means independently or in cooperation with each other, and automatically update them to more appropriate knowledge. An expert system generator characterized by Case acquisition means acquires and accumulates the status and results related to the external system for which the expert system is built as data,
The rule generation means automatically extracts a statistical tendency in the form of rules from the data acquired and accumulated by the case acquisition means,
The consulting means automatically determines the result for the situation from the current data of the external system with reference to the rule generated by the rule generating means,
The case acquisition means, the rule generation means and the consulting means can be operated independently,
The rule generation unit automatically updates the rule together with the data accumulation in cooperation with the case acquisition unit and the consulting unit, and updates the rule to a more suitable rule. The expert system generator according to 1. The data acquired by the case acquisition means is a case consisting of attributes and attribute values.
3. The expert system generation apparatus according to claim 2, wherein the conclusion derived in the consulting means is a conclusion attribute value.   4. The expert system generation apparatus according to claim 2, wherein the rule generation means generates a rule in the form of a decision tree.   5. The expert system generation apparatus according to claim 4, wherein the rule generation means is capable of automatically generating a decision tree regardless of the number of cases.   6. The expert system generation apparatus according to claim 5, wherein the rule generation means is capable of generating a decision tree even in cases where unknown condition attribute values exist.   7. The expert system generation apparatus according to claim 5, wherein the rule generation means can generate a decision tree even in a case where an unknown conclusion attribute value exists.   8. The expert system generation apparatus according to claim 4, wherein the rule generation means can set a conclusion attribute value to a fixed value for a predetermined case.   In the case where prior knowledge about the distribution of attribute values of the acquired cases is obtained, the rule generation means performs the decision attribute value for an unknown conclusion attribute value derived as a result of tracing the decision tree generated once. 9. The expert system generation device according to claim 5, wherein a case including the above is added.   10. The expert system generation apparatus according to claim 4, wherein the consulting means automatically acquires a condition attribute value from the outside and derives a conclusion attribute value from an already obtained decision tree.   11. The expert system generation apparatus according to claim 10, wherein the consulting means automatically starts an external system in the form of an execution program, converts the output value of the external system into a condition attribute value, and acquires the condition attribute value.   12. The expert system generation apparatus according to claim 10 or 11, wherein the consulting means applies the conclusion attribute value to the external system.   The consulting means can automatically start an external system in the form of an execution program, convert a conclusion attribute value obtained by the consulting means into an input value of the external system, and deliver it. 12. The expert system generation device according to 12.   14. The rule generation means evaluates the validity of the generated decision tree, and updates the decision tree with a higher validity based on the result. Expert system generator.   15. The expert system generation device according to claim 14, wherein the rule generation means can make an exception of a case.   16. The expert system generation apparatus according to claim 14, wherein the rule generation means is capable of making exceptions to condition attribute items.   The expert system generation device according to claim 14, wherein the rule generation means is capable of pruning a branch branch.   18. The expert system generation according to claim 17, wherein the pruning of the branch branch can be performed between any condition attribute item and between any condition attribute and any conclusion attribute. apparatus.   The rule generation means regenerates the decision tree with the corresponding case as an exception when a plurality of conclusion attribute values are derived for one condition case as a result of following the decision tree. The expert system production | generation apparatus of any one of thru | or 18.   20. The expert system generation apparatus according to claim 3, wherein the case acquisition means can automatically acquire a case.   Information obtained by experts as examples is accumulated as examples, multiple knowledge structures that include the same knowledge elements, and hierarchical knowledge structures in which the knowledge elements that make up the knowledge are superordinate concepts or subordinate concepts of different knowledge The knowledge structure is a special knowledge structure of the multiple knowledge structure, and it is possible to extract knowledge having a multi-view knowledge structure, which is a collection of knowledge from many different viewpoints, as a decision tree. The expert system generation device according to any one of claims 4 to 20.   Rule generation means can generate decision trees that express various causal relationships inherent in cases or groups of mutually related decision trees, and analyze the causal relationships using the generated decision trees or decision tree groups The expert system generation device according to claim 21, wherein   The expert system generation apparatus according to claim 22, wherein the rule generation means can arbitrarily select a condition attribute or a conclusion attribute for each attribute item of the case.   The expert system generation device according to claim 22 or 23, wherein the rule generation means is capable of generating only a decision tree corresponding to a specific conclusion attribute.   The rule generation means can have a plurality of conclusion attribute items for a case base in which cases are accumulated, and has a user interface for comparing decision trees corresponding to each conclusion attribute. The expert system generation device according to any one of claims 22 to 24.   The expert system according to any one of claims 22 to 25, wherein the rule generation means is capable of expanding and concatenating another decision tree with a conclusion attribute value of the generated decision tree. Generator.   The rule generation means is characterized in that detailed concatenation is possible for a condition attribute item of the generated decision tree, in which another decision tree having all attribute values of the condition attribute item as conclusion attribute values is combined. 27. The expert system generation device according to any one of claims 22 to 26.   The rule generation means, when the conclusion attribute value is a numerical type, approximates the conclusion attribute value with a regression line and obtains the approximate residual as the width of the conclusion attribute. The expert system generator according to claim 1.   29. A method for implementing an expert system generation device, characterized in that expert knowledge is ruled using the expert system generation device according to any one of claims 1 to 28.   30. The method of implementing an expert system generation apparatus according to claim 29, wherein the expert's action is automated based on the ruled expert's knowledge.   31. The method of implementing an expert system generation apparatus according to claim 30, wherein a person's work is supported by the knowledge of a ruled expert.   32. The method for implementing an expert system generation apparatus according to claim 30 or 31, wherein a person is educated based on ruled expert knowledge.   33. The method of implementing an expert system generation apparatus according to any one of claims 30 to 32, wherein the expert is an expert on an industrial process.   34. The method for implementing an expert system generation apparatus according to claim 33, wherein the industrial process is a refining plant using a rectification column.   35. The expert according to claim 34, wherein the acquisition of the case from the refinery plant is performed through at least one of a temperature sensor, a pressure sensor, a flow rate sensor, a sound sensor, and an image sensor provided in the rectifying column. Implementation method of system generation device.