JPH0778681B2 - Man-machine interface of plant - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to a man-machine interface of a plant, and more particularly to an interface capable of input / output in natural language.

[Conventional technology]

Conventionally, in the operation and maintenance of a plant such as a nuclear power plant, complicated instructions to a process computer rely on a menu selection or a switch on a keyboard. Therefore, it has been desired to develop a man-machine interface that simplifies command operations in order to reduce the burden on the operator and prevent human error in the event of abnormal operations involving a large number of operations.
As one of the countermeasures, conventionally, for example, a method has been developed in which a keyword is recognized from the natural language of an operator and the meaning content of the input sentence is recognized based on the keyword (Japan Atomic Energy Society “ 1986 meeting "E8).

[Problems to be Solved by the Invention]

According to the above conventional technique, it is possible to input in a natural language,
The burden of command input during plant operation is reduced. But,
Since this is a method centered on keyword recognition, the expressions of input sentences that can be accepted and the types of input commands are limited.

In order to improve man-machine performance for plant operation / maintenance, it is necessary to have a method that can input various instructions for operation / maintenance using natural expressions that are normally used by operators / maintenance staff. Is. An object of the present invention is to provide an interface capable of coping with various command inputs by various expressions from operators and maintenance personnel.

[Means for Solving the Problems]

The present invention recognizes a plurality of words included in an input natural language sentence and analyzes the meaning to determine the meaning of the word, and the identification for identifying the type of instruction to be executed in the plant based on the natural language sentence judged by the analyzing means. Means, information acquisition means for acquiring information items necessary for the operation content to be executed according to the type of execution instruction, and processing based on the type of execution instruction identified by the identification means and information items acquired by the information acquisition means And an instruction generation unit that generates an instruction.

Further, in the present invention, the character pattern used for identifying the content of the execution instruction is prepared by a word, its meaning, and one of the meaning categories, or a combination thereof. Furthermore, the information acquisition means in the present invention acquires the information item that defines the operation content from one or more of the following: an input sentence reference, dialogue history storage content reference, plant related knowledge reference, and user inquiry. Is configured to.

[Action]

According to the present invention, a plurality of words included in an input natural language sentence are identified, the type of instruction to be executed in the plant is identified by the identifying means based on the natural language sentence in which the meaning of the word is determined by the analyzing means, and the identified execution is performed. The information acquisition means acquires the information items necessary for the operation content to be executed depending on the type of instruction. Then, the processing instruction is generated based on the type of the execution instruction identified by the identification means and the information item acquired by the information acquisition means. Since the instruction type is identified and necessary information items are obtained in two independent steps, it is possible to identify many existing instruction types using a small number of sentence patterns, and the identified instruction The process of acquiring the necessary information items is limited to the above. As a result, since the sentence pattern required for analyzing the natural language sentence and the data for obtaining the information are separately prepared, it is possible to simplify each of them.

〔Example〕

Hereinafter, a plant man-machine interface according to an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the apparatus according to the embodiment. In the figure, 11 is a keyboard and 3 is an input sentence analysis device. This device is composed of a morpheme / syntactic analysis unit 31, an instruction content identification unit 32, a necessary information acquisition unit 33, and a function / argument association unit 34. Further, reference is made to a storage device 35 of data prepared in advance such as a dictionary / grammar, instruction identification rules, and case frames, and a dialogue history stack 36.

Reference numeral 4 is a response creating device, in which a regular test consulting program for searching the knowledge base 41 is prepared. Reference numeral 5 is a response sentence generation device, and 6 is a CRT.

First, a brief description will be given of the consulting program for the regular test in the response creating device 4.

FIG. 2 is a diagram showing knowledge in a frame format, which is a part of the knowledge base used in the consulting program. As shown in the figure, TEST is displayed in the frame showing the test name.
-PURPOSE (test purpose), TEST-FREQUENCY (test frequency)
Slots such as are provided, and the corresponding knowledge is given as the value of the value (variation) function. This knowledge search can be performed using the function
TEST-FREQ may be executed by giving the test name as an argument value. As a result, the process of extracting the value from the test frequency slot of the frame of the corresponding test name is executed.

FIG. 3 is a diagram showing knowledge in the common sense algorithm (CSA) format, which is the remaining part of the knowledge base.
As shown in the figure, in the CSA format knowledge, an operation procedure or a change in the device state caused by the operation is represented by an event and a link connecting the events. The event includes an "Action" event that represents an operation or confirmation, a "State" event that represents the state of a device, and the like. In addition, the links include "Sequential" links that represent the context of actions, and "One-" that represents the causal relationship between "Action" and "State".
There is a "Shot Causal" link, an equivalence relation between "States", or a "State Coupling" link that represents a causal relation.
Also, a link is a "Gate" that represents the conditions under which it holds.
Can have

Regarding this CSA knowledge, it is sufficient to give a function and its argument value and execute it, as in the case of frame format knowledge. For example, when searching for a test procedure, the test name may be given as an argument and the function TEST-PROCEDURE may be executed. When this function is executed, the CSA knowledge corresponding to the test given as an argument is processed such that the "Action" events linked by the "Sequential" link are sequentially extracted from the first "Action" event. It

As described above, for example, the inquiry about the regular test can be performed by executing the function of the consulting program by specifying the argument value. Incidentally, in general, instructions to various application programs prepared on a normal process computer can also be given by similarly designating corresponding functions and argument values.

Next, the processing in the input sentence analysis device for converting the input sentence from the operator / maintenance staff into the function and argument value of the consulting program of the response creating section will be described.

FIG. 4 is a flowchart showing an outline of processing in the input sentence analysis device. The input sentence analysis device analyzes the input sentence input from the keyboard 11 to process instructions to the process computer,
That is, the function and argument values are converted and output. Or, let us consider the case where the input sentence is “gensiroscrumnogenin.” (101).

In the input sentence analysis device, first, the morphological / syntactic analysis (102) of the input sentence is performed with reference to the part-of-speech category of the word stored in the dictionary 202 and the grammar 201. As a result, the input sentence is divided into words, and the meaning category and meaning are added to each word. Here, as a morpheme / syntactic analysis method, a commonly used method, for example, a predictive bottom up method is used. Here, the grammar is a rule regarding the structure of a sentence, for example, S (sentence) → NP (noun phrase) S (sentence) represents a rule that S on the left side can be composed of NP and S on the right side. Is. Further, the dictionary is provided with four pieces of information for each headword: a part-of-speech category, a display word obtained by converting the headword into kana-kanji, a meaning category, and a meaning. However, only two parts, the part-of-speech category and the display word, are given for particles.

Note that the words included in the input sentence can be similarly recognized by dictionary lookup by the longest matching method that is usually used in word processors and the like.

Next, a process (103) for identifying the instruction content is executed. In this processing, the result of syntax analysis is taken in and the instruction content is identified using the instruction identification rule (203). The instruction is prepared corresponding to the processing function in the regular test consulting program, such as EVENT-CAUSE (search for the event causing the specified event).

The instruction identification rule gives each instruction a standard sentence pattern for identifying the instruction.
In the instruction identification, the matching between the input sentence and each sentence pattern is sequentially tried, and the instruction corresponding to the matched sentence pattern is the input instruction. In the example shown in the figure, the sentence pattern “Cause% + Post” matches the input sentence,
The identification result that the input command is EVENT-CAUSE is obtained. Here, a symbol is prepared to give a sentence pattern. "%" Is a symbol representing a word used for matching only when the input sentence has a particle. As a result, it is possible to cope with the omission of particles, which is often seen in the dialogue between operators and maintenance staff. "+" Is a symbol that represents an abbreviation word and is used to correspond to a word stop. Further, "?" Is a symbol representing an arbitrary number of words, which makes it easy to set a sentence pattern. Furthermore, "<" represents matching with the meaning category of the word.

Here, the flow of processing in the instruction content identification unit will be described. FIG. 5 is a flowchart showing the processing. In the instruction identifying unit, first, the sentence pattern number k is set to 1 as shown in the figure. For the sentence pattern, i representing the number of the word in the sentence pattern is set to the total number I of words in the sentence pattern. Further, j indicating the number of words in the input sentence is set to the total number J of words in the input sentence (step 401).

Next, the i-th word P _{i of} the k-th sentence pattern and the j of the input sentence
Bring the th word W _j (step 402). This P _i and W _j
The matching process described below is performed.
Since i is set to I and j is set to J at the beginning,
The words are compared from the end of the input sentence and the sentence pattern toward the beginning of the sentence.

The matching process differs depending on the shape of the word P _i of the sentence pattern, that is, the presence or absence of the above-mentioned symbol (step 403). First, the process when no symbol is attached will be described. In this case, P _i and W _j are compared (step 404), and if they are the same, i and j are subtracted (step 405). That is, the word to be compared is the previous word. If the result j is 0 (step 406), it means that all the words of the sentence pattern match the words of the input sentence. Therefore, the command name corresponding to the sentence pattern k is regarded as the input command content. Output (step 410). On the other hand, when i is 1 or more (step 406), the judgment for j is performed (step 407). As a result, if j is 1 or more, the process from step 402 is repeated. On the other hand, if j is 0, the matching between the sentence pattern k and the input sentence has failed, so preparation is made for matching with the next sentence pattern (k + 1) (step 408). Next, if the sentence pattern still remains (step 409), the process from step 402 is executed for the new sentence pattern. On the other hand, when the sentence pattern is exhausted, a message indicating that the instruction cannot be identified is output and the processing is terminated (step 412).

To P _i, if you are marked with the symbol "<" compares the semantic categories of P _i and W _j (step 413). Depending on the result, the same processing as described above is performed.

When the symbol "%" is attached to P _j , it is determined whether the part-of-speech category of W _j is a particle (step 414).
As a result, if it is a particle, compare P _i and W _j (step 41
5) According to the result, perform the same process as above. On the other hand, if W _j is not a particle, j is decremented by 1, and if j is 1 or more (step 417), the process returns to step 402. on the other hand,
If j is 0, the words of the input sentence have been lost, so that the matching between the sentence pattern k and the input sentence has failed, and the process for the next sentence pattern, that is, the process from step 408 is executed.

When P _i has a symbol “+”, P _i is compared with W _j (step 418), and if they are the same, the process of step 405 and the like is executed. On the other hand, if different, i is decremented by 1, and if i is 1 or more, the process returns to step 402. i is 0
In the case of, it means that all the words of the sentence pattern k have been included in the input sentence (except those which can be omitted), so that the process proceeds to step 410 and the instruction identification is completed.

If P _i is the symbol “?”, Then the word P _i-1 before the “?”
Compare W _j (step 421). As a result, if the word is matched, 1 is subtracted from i, and the matching process for the next word, from step 405, is executed. On the other hand, if they are not the same, j is decreased by 1, and if j is 1 or more (step
424), from step 402, matching is performed using the word immediately preceding the input sentence. If j is 0, the matching with the sentence pattern k has failed, so the next sentence pattern k +
The matching process with 1, that is, from step 408 is executed.

Next, a process for acquiring information necessary for executing an instruction (fourth
Perform 104) in the figure.

Some instructions require some information to execute. As a method for controlling this information acquisition, necessary information items and their acquisition methods were prepared as frame format data. This is called a case frame. The case frame is defined for each instruction, and each slot of the instruction name frame is provided with an item of information necessary for executing the instruction. In the example of the figure, it is shown that two pieces of information, "test name" and "event number", are necessary for the instruction EVENT-CAUSE. The data and the value necessary to obtain the information are given to the root and the value of each slot. VA
The LUE function is given the category name of the meaning of the word to be searched from the input sentence, and the IF-NEEDED function is given the name of the function to be activated when the input sentence does not contain the necessary word. Here, the function STACK is a function that refers to the dialogue history stack, and the function ASK is a function that requests the questioner to input additional information. The function GET-EVENT-NO is a function that obtains the event number from the input sentence. This function represents the event name, ACT (act), COMP (device name), ATT
An event number is obtained by searching the input sentence for a word having RIB (device attribute) and STATE (device state) as semantic categories, and comparing the result with the event name of the knowledge expressed in CSA format.

The flow of this processing will be described with reference to FIG. FIG. 6 is a flowchart showing the processing. In this process as shown in the figure,
First, a case frame having an instruction name as a frame name is taken out (step 501). Then set i to 1 (step 5
02). After this initial processing, the value of the VALUE flag of the i-th slot is read and set as the value of Vi (step
503). If the value of Vi is not nil (step 504),
Search the input sentence for a word having the semantic category Vi (step 505). If the word can be found (step 506), the obtained word is set as the value of the information item of the i-th slot (step 507), and if the slot still remains (step 51).
0), i is incremented by 1 (step 511), and the process for the next slot is started. On the other hand, if the word cannot be found in step 506, or if the value of Vi is nil in step 504, the value of the IF-NEEDED function is read and set as the value of Ni (step 508). Then, the Ni functions are sequentially executed. As a result, if a value is obtained, it is set as the value of the information item of the i-th slot (step 50
9), move on to the next slot.

The outline of the process will be described with reference to the example of FIG. Instruction EVENT
-In the case frame corresponding to CAUSE, there are two slots, "test name" and "event number". Regarding "test name", first, a word having the meaning category TEST-NAME given as the value of the VALUE argument is searched for in the input sentence. At present, there are no words with this semantic category in the input sentence. Therefore, the value of the IF-NEEDED function is referred to next, and the function STACK given first is executed. As a result, the "test name" of the dialogue history stack
Part is referred to. In the example in the figure, MSIV 10% closed test (MSIV-
TEST) is being answered, and it is assumed that MSIV-TEST is included as a value in the “Test name” stack. Therefore, by executing the function STACK, the MSIV-T
The EST is obtained and is set as the value of "Test name".

As for "event number", there is no value in the VALUE flag (nil). Therefore, the value of the IF-NEEDED function is referred to and the function GET-EVENT-NO is executed. This function first searches the input sentence for words with semantic categories such as ACT, COMP. As a result, the input sentence has a semantic category COMP
It is discovered that there is "reactor" as a word with "," and "scrum" as a word with "STATE." This result is compared with the event name in CSA format shown in Fig. 3, and COMP, STATE
You get 204 as the number of the event that has "reactor" as the value of and "scrum". In this example, we considered the case where both "reactor" and "scrum" were entered.
If only "Scrum" is entered, "Scrum"
The same process is performed by referring to the plant-related knowledge that the is in the “reactor” state.

Finally, the process for associating with the function / argument (105)
Carry out. In this processing, the values of the command and the information item obtained in the previous processing are associated with the completion of the program and the argument value, and output to the corresponding program. In the example shown in FIG. 4, since the function name and argument value in the response creating unit are the same as the command and information item values in the input sentence analysis unit, the obtained command and information item value are listed as they are. And output it. On the other hand, when the input sentence analysis unit and the response creation unit have different command names and the like, they are associated with each other by using the conversion table.

According to the processing command from the input sentence processing device, the response creating unit carries out the processing for knowledge retrieval. The result is sent to the response sentence generation device as needed. In the response sentence generator,
Generate an output statement based on the result and output it on the CRT.
The outline of the processing in this output statement generation device will be described by taking the result of the consulting program as an example.

FIG. 7 is a diagram showing an example of processing in the response sentence generation device. For example, the event name is expressed as shown in the figure in the knowledge base used in the consulting program. So the "Action" event is
It is expressed by a word indicating an act (ACT), a device name (Comp), an attribute (ATTRIB), and a state (STATE), and a remark (301). A rule (302) for generating an output sentence is prepared in the output sentence generating device, and an output sentence is generated using this. In the example in the figure, if the event is "Actio
The output sentence "Confirm plant stability [before test]" is generated using the rule that should be used when n "and the word indicating the action includes" confirm ". Here, the word order is changed and particles are inserted.

As another example, the content of the output sentence generation process when a question is asked about the possibility of an erroneous operation is shown. At this time, if event A is obtained as the correct operation and event B is obtained as the incorrect operation (304), rule 3
05 is applied, and the output sentence "Event B when event A should occur" (306) is generated.

FIG. 8 is a diagram showing the operation of the interface according to the above-mentioned embodiment. In response to the input request from the computer, "HPCS
Pump Shiken no Mokutekiha. When you input "" from the keyboard, first, as a result of the morphological / syntactic analysis, the sentence written in pieces is displayed on the CRT. Next, consulting program knowledge base search function (search for search purpose)
Is executed and the result is shown on the CRT. Next, enter "HINDO SHISEN." And the response will be presented in the same way.
Here, since there is no test name in the input sentence, the test name required for command execution is obtained by referring to the dialogue history stack.

Next, from the keyboard, select "MSIV
Hioji. Is entered, and the purpose of the MSIV closed test is presented. Then, "Sequence Yuukiha." Is input and a response is presented. The above is the one in which a response is obtained by searching for knowledge in the frame format.

Next, "Sequence Talent" is entered,
The answer is the search of knowledge in CSA format in the response creation section,
It is obtained by generating a response sentence based on the result.

As described above, the apparatus according to the present embodiment can handle input sentences in various expressions including omission of particle and word stop. Also, when giving a sentence pattern, not only the display word of the word but also its meaning and meaning category can be used, so the number of patterns to be prepared should be reduced compared to the case of giving only the display word of the word. You can Furthermore, by using the symbols, it is possible to easily express the omission of particle and the word stop as a sentence pattern.

Also, by referring to the dialogue history, the omitted information of the input sentence can be supplemented, so it is not necessary to input the same word every time,
The input can be simplified. Furthermore, by using the plant-related knowledge, it is possible to reduce the amount of information that needs to be input. Further, a more natural dialogue can be realized by supplementing the information by the inquiry to the user.

Therefore, by using this device, the operator / maintenance staff can input commands more naturally and easily. Furthermore, this device can present the response from the computer in the form of a natural language sentence. As a result, man-machine performance for operation and maintenance is improved. Moreover, since the response sentence generation method can be given in the form of a rule, the maintenance of the response sentence generation unit is easy.

Although a keyboard and a CRT are used as the input / output devices in the device according to the present embodiment, if a commercially available voice input / output device is used, a natural language voice can be processed by the same processing method as this device. Dialogue is possible. Also, a liquid crystal display device or the like can be used as the display device.

The man-machine interface described above can be applied not only to plant operation and maintenance but also to general interactive systems such as consulting systems and design support systems.

〔The invention's effect〕

As described above, according to the present invention, the identification of the instruction type and the acquisition of the necessary information items are performed in two independent steps, so that the instruction types that exist in large numbers are identified using a small number of sentence patterns. In addition, it is possible to obtain necessary information items only for the identified command. As a result, character patterns required for analyzing natural language sentences and data for information acquisition can be prepared separately, which simplifies each and reduces the burden of operator and maintenance personnel inputting commands and recognizing output from computers. Can reduce the possibility of human error.

[Brief description of drawings]

FIG. 1 is a block diagram of a man-machine interface of a plant which is a preferred embodiment of the present invention, FIGS. 2 and 3 are explanatory views showing the contents of the knowledge base of FIG. 1, and FIG. 1 is an explanatory view showing the processing in the input sentence analysis device in FIG. 1, FIG. 5 is an explanatory view showing the processing in the instruction content identification unit in FIG. 1, and FIG. 6 is the processing in the necessary information acquisition unit in FIG. Explanatory drawing showing
FIG. 8 is an explanatory diagram showing the processing in the response sentence generation device of FIG. 1, and FIG. 8 is the operation of the interface of the embodiment of FIG. 1, especially CR.
It is explanatory drawing which shows the example of a screen of T. 3 ... Input sentence analysis device, 4 ... Response generation device, 5 ... Response sentence generation device, 6 ... CRT, 11 ... Keyboard, 31 ...
Morphological / syntactic analysis unit, 32 ... Instruction content identification unit, 33 ... Necessary information acquisition unit, 34 ... Function / argument association unit, 35 ...
... storage device, 36 ... dialogue history stack, 41 ... knowledge base, 101 ... input sentence, 102-105 ... processing step, 201
~ 204 …… use data, 301,304 …… input, 302,305 ……
Rules for sentence generation, 303, 306 ... Output sentence, 401 to 424,5
01-511 …… Step of processing.

Front Page Continuation (72) Inventor Tetsuo Ito 1168 Moriyama-cho, Hitachi-shi, Ibaraki Energy Research Laboratory, Hiritsu Manufacturing Co., Ltd. (56) Reference JP-A-62-49436 (JP, A) JP-A-1-229321 ( JP, A) JP 63-219073 (JP, A) JP 63-317878 (JP, A)

Claims (4)

[Claims] 1. An analyzing unit for recognizing a plurality of words included in an input natural language sentence to determine the meaning of the word, and a type of instruction to be executed in a plant based on the natural language sentence determined by the analyzing unit. Identification means for identifying, information acquisition means for acquiring information items necessary for the operation content to be executed according to the type of the execution instruction, type of execution instruction identified by the identification means, and information acquired by the information acquisition means A man-machine interface for a plant, comprising: a command generation unit that generates a processing command based on an information item. 2. An analysis unit for recognizing a plurality of words included in an input natural language sentence to determine the meaning of the word, and a word indicating the type of the execution command for fetching the natural language sentence determined by the analysis unit. Identification means for identifying the type of execution command to be executed in the plant by comparison with the sentence pattern, information items required to define the operation content to be executed by the type of execution command, and acquisition of this information item An information acquisition unit for determining a method, and a command generation unit for generating a processing command based on the type of the execution command identified by the identification unit and the information item acquired by the information acquisition unit. Man-machine interface. 3. The sentence pattern according to claim 2, wherein the sentence pattern used for identifying the content of the execution instruction is prepared by one of a word, its meaning, and a category of meaning, or a combination thereof. Man-machine interface of the plant. 4. The information acquisition means according to claim 1 or 2, wherein the information acquisition unit refers to an input sentence for an information item that defines an operation content,
Refer to memory contents of dialogue history, refer to plant-related knowledge,
And a man-machine interface of the plant, characterized in that it is obtained from one or more of the queries to the user.