CN101004799A - Tense generation formula system - Google Patents

Abstract

A production system of tense type is prepared for using databank to store data relating to problem to be solved, using rule set to present normal knowledge rule of problem field, using a reasoning machine to select and to use rules in said rule set as well as to operate databank according to reasoned result, containing tense attribute in both of fact and said rule.

Description

A kind of tense production system

Technical field

The invention belongs to the knowledge representation method and the inference method thereof of artificial intelligence field.

Technical background

Production system is that logician Post has proposed a kind of production notion in nineteen forty-three at first, by Simon and Newell it has been carried out perfectly then in nineteen sixty-five, and is incorporated into KBS Knowledge Based System.Be most popular a kind of knowledge representation method in the expert system at present, generally this system be called rule-based system.A production system comprises: database, rule base, a control module (also claiming rule-interpreter sometimes)." database " is used for depositing the data relevant with the problem of finding the solution." rule base " deposits rule, or is production, and these rules have been represented the general knowledge in the problem domain." control module " selected to use to rule, and the result has produced increasing, delete, changing database.In traditional production system, control module adopts the method for data-driven, and promptly circulation searching in rule base can satisfy the rule of its condition part to look for database.If find such rule, just carry out its action part.In most of the cases, the result of action has changed the fact in the database, thereby makes that other regular conditions are satisfied.Control module continue to be carried out below satisfying at least till any one condition:

(1) problem be resolved (target reaches);

(2) current condition without any rule can be met.

In the process of the reasoning of production system, relate to multiple temporal information: true effective time, the effective time of rule and the consistance of different time information.Such as in intelligent transportation system, may be different in time and change to some extent to the definition of certain unlawful practice, and different punishment measure when whether effectively having (promptly regular state property) also can be arranged because of the variation of time to same unlawful practice.And mostly the production system reasoning of present stage is not have time attribute, and effective time is in fact being given in a spot of reasoning, and rule does not have the constraint of effective time, and this has just limited the application and the popularization of knowledge reasoning.

Time is the immanent objective attribute of nature, and all information all have corresponding tense attribute, and knowledge is no exception.Time attribute is the pith of knowledge, particularly is in the information age that this renewal of knowledge is now maked rapid progress, and whether useful knowledge is closely related with the time, and the reasoning of knowledge also is like this.

The present invention proposes a kind of new tense production system and inference mechanism thereof, provide effective support tense knowledge representation and reasoning.

Summary of the invention

The objective of the invention is to overcome the deficiencies in the prior art, a kind of new tense production system of supporting tense knowledge representation and reasoning is provided.

In order to realize the foregoing invention purpose, the technical scheme of employing is as follows:

A kind of tense production system, comprise database, regular collection and deduce machine, described database is deposited the data relevant with the problem of finding the solution, the rule of the general knowledge in the described regular collection problem of representation field, described deduce machine is selected to use to the rule in the regular collection, and the result operates database by inference, includes the tense attribute in the described rule.

So far, production system of the present invention is made up of set, total data storehouse and the tense deduce machine of tense production rule.By comprise the tense attribute on true and rule, what knowledge with tense attribute in the real world and rule just can be compared shows, and carries out reasoning.

In the technique scheme, described rule comprises regular header, regular former piece and regular consequent;

Described regular header comprises rule numbers, regular effective time, rule name;

All preconditions of described its description rule of regular former piece comprise the tense attribute of description condition; Regular former piece of the present invention is a disjunction expression, and the subitem of disjunction expression is a conjunction expression.Each relational expression wherein all is the precondition of a minimum, just the atom condition.

Described regular consequent is described the action that will carry out after this rule of coupling, and also comprises the tense attribute of describing action in the regular consequent; Regular consequent of the present invention is the conjunction expression of several relational expressions, and each relational expression is described an action or function, and this action is certain atomic action that has defined.

The stored data of database of the present invention have the tense attribute, and it comprises following subdata base:

The system metadata storehouse that provides working rule to represent required basic function;

User-defined word and the table of actual database and the correspondence of field in the storage practical application, the user metadata storehouse of metadata such as definition atomic action;

Factual database or context database;

The rule base that utilizes user metadata, system metadata to produce together;

The volatile data base that produces in the reasoning process.

According to above-mentioned regular collection and database, the inference step of deduce machine of the present invention is as follows:

(1) receives relevant facts and the effective time thereof that the user imports;

(2) from rule base, get a regular R[Vs, Ve], if can not select then failure; Otherwise change step (3);

(3) with regular R[Vs, Ve] regular former piece TF:Pre and the known true coupling in the factual database;

(4) if need then to give regular R[Vs, Ve] conclusion give the TF ' effective time of regular former piece, whether the conclusion of judgment rule R is the new fact, be Already in whether conclusion in the factual database, if not new fact, then change step (5), if new fact, then together with depositing in the factual database as the known fact its effective time, and write down rule and the relevant facts thereof that the match is successful, reexamine whether qualified conclusion of new fact, if meet, then finish this reasoning, if the ineligible conclusion of new fact is then changeed step (2);

(5) from rule base, get another regular R+1, change step (3).

Wherein, the coupling concrete steps of described step (3) are as follows:

(3.1) with the TF:Pre negate, intersect with the effective time of regular R again, obtain [Vs, Ve] ∧ TF:( Pre), deposit among the S set et;

(3.2) there is the fact of common factor to deposit among the S set et with [Vs, Ve] effective time with the known fact in the factual database;

(3.3) select two the expression formula TFa:A that do not have stipulations to cross, TFb:B from Set, if choosing is not come out, then former piece [Vs, Ve] ∧ TF:Pre is false, and turns to step (5);

(3.4) expression formula TFa:A, TFb:B are carried out stipulations, reduction rule is as follows: for predicate TFa:A and TFb:B, if A and B can stipulations when considering effective time, A ∧ B=C, and TFa ∧=NULL, TFa:A then, TFb:B can stipulations, and TFa:A ∧ TFb:B=TFa ∧ TFb:C is arranged; Otherwise these two expression formulas can not stipulations, if one of them expression formula is Pre, then after the stipulations success, with upgrading TF ': Pre stipulations result's effective time;

(3.5) if empty clause appears in stipulations, then former piece TF ': Pre sets up; Turn to step (4); Otherwise turn to (3.3).

In the above-mentioned inference step, step (2) is chosen the method for a regular R, can continue to use the method for existing production system, as can not cancelling method, and exploratory method, graph search method or the like.

The fact in the step (3) coupling adopts is that stipulations (Resolution) mode in the predicate logic is carried out, but has carried out the tense expansion.If be [Vs the effective time of this rule, Ve], the expression formula of rule former piece is TF:P, because must and there be regular effective time the effective time of regular former piece the ability of common factor meaningful, so can ship calculation doing the effective time of the effective time of regular former piece and this rule, be that regular former piece is now: TF ∧ [Vs, Ve]: P.(is regular former piece negate TF ∧ [Vs, Ve]: ( P)), sum up with the expression formula of the known fact again, if sum up empty clause, then regular former piece is set up (can change the effective time of regular former piece along with the carrying out of summing up, and TF ' after end is finished: the TF ' effective time of ( P) is the effective time of final regular former piece); The else rule former piece is false, and selects next bar rule to continue.The basis that can do like this is: if expression formula P is true ([Vs, Ve]: P is true) in the time period [Vs, Ve], then expression formula  P must be that false ([Vs, Ve]: ( P)) is false at time period [Vs, Ve]).

During stipulations, earlier the expression formula of not having effective time and this regular effective time common factor (establishing expression formula effective time is [Es, Ee], promptly satisfies Ve＜Es or Vs〉Ee) is filtered out.In remaining expression formula, sum up.

=NULL, then expression formula A and B can sum up, and establish A ∧ B=C, and be TFc the effective time of then summing up back C, promptly has: TFa:A ∧ TFb:B=TFcC; If TFa ∧ TFb=NULL, then expression formula TFa:A and expression formula TFb:B can not sum up, and should select other expression formulas else and sum up this moment.

If regular former piece is set up, and this moment regular former piece effective time be TF ', then use this rule, regular consequent be true, and the effective time of regular consequent also assignment be TF ', and regular consequent adding knowledge base.

The present invention is by adding the tense attribute in true and rule, to rule is to add the tense attribute in its integral body and its former piece and consequent specifically, and then can with computing machine the incident with tense attribute in the real world be described by production system, and carry out reasoning, overcome existing production to the deficiency in the expression of tense attribute.

Description of drawings

Fig. 1 is the predicate logic stipulations procedure chart that has effective time of the present invention;

Fig. 2 is the reasoning process figure of tense production system of the present invention;

Fig. 3 is database one-piece construction figure of the present invention;

Fig. 4 is rale store method figure.

Embodiment

The present invention is described further below in conjunction with accompanying drawing.

The present invention includes regular collection, total data storehouse, deduce machine.

Rule comprises regular header, regular former piece and regular consequent;

According to above-mentioned statement, the BNF of tense production of the present invention is defined as:

＜tense production system〉∷=＜tense production rule〉}

＜tense production rule〉∷=＜the rule head〉' ∷ ' ' IF '＜regular former piece〉' THEN '＜regular consequent〉[interpretation of rules]

＜rule head〉∷=＜rule numbers〉＜rule name〉＜effective time 〉

＜rule numbers〉∷=＜positive integer 〉

＜rule name〉∷=＜character string 〉

＜regular former piece〉∷=＜the condition union 〉

＜condition union〉∷=＜the condition union〉' ∨ '＜condition common factor〉|＜condition is occured simultaneously〉| NULL

＜condition is occured simultaneously〉∷=＜the condition common factor〉' ∧ '＜atom condition〉|＜atom condition〉| NULL

＜atom condition〉∷=＜TE〉': '＜true title〉＜comparison operator〉＜fiducial value 〉

＜TE〉∷=＜measure word〉＜variable name〉[＜time variable interval 〉] ', '＜variable expression〉|＜IEI〉＜unit 〉

＜time variable interval〉∷=＜left interval number＜IEI or variable〉', '＜IEI or variable〉＜right interval number＜unit 〉

＜IEI or variable〉∷=＜IEI〉|＜variable name 〉

＜variable expression〉∷=＜variable expression〉' ∨ '＜variable Boolean expression〉|＜variable Boolean expression 〉

＜variable Boolean expression〉∷=＜the variable arithmetic expression〉＜comparison operator〉＜IEI〉＜unit 〉

Interval number an of＜left side〉∷=' [' | ' ('

＜right interval number ∷='] ' | ') '

＜IEI〉∷='? ' | ' ∞ ' | ' NOW ' |＜positive integer 〉

＜measure word〉∷='  ' | '  '

＜unit〉∷=' year ' | ' moon ' | ' day ' | ' time ' | ' branch ' | ' second '

＜comparison operator〉∷='==' | '＜=' | ' 〉=' | '＜' | '〉' | '!='

＜true title〉∷=＜character string 〉

＜fiducial value〉∷=＜constant value〉|＜true title 〉

＜constant value〉∷=＜integer〉|＜real number〉|＜character string 〉

＜effective time〉∷=＜time interval 〉

＜time interval〉∷=' ['＜time point〉', '＜time point〉'] '

＜time point〉∷=＜positive integer〉' year '＜positive integer〉' moon '＜positive integer〉' day '＜positive integer〉' time '＜positive integer〉' branch '＜positive integer〉' second '

＜regular consequent〉∷=＜the condition common factor 〉

＜interpretation of rules〉∷=' // '＜character string〉| NULL.

Wherein＜character string,＜integer,＜real number,＜positive integer implication as its name suggests, be character string, integer, real number, positive integer in general sense.'? ' expression uncertain time point in future, ' the maximum in the future time point of ∞ ' expression, ' NOW ' expression tense argument NOW, be the current time, on behalf of union, ' ∧ ' representative ship calculation symbol, ' ∨ ' accord with, '  ' be existential quantifier, '  ' is a generality quantifier; ' the variable arithmetic expression ' be the arithmetic expression of the variable in the general mathematics.The action scope of the variable that the optional position occurs in rule former piece and the regular consequent all is whole regular former piece and regular consequent.As variable x in the regular former piece and the variable x in the regular consequent is a variable.

The total data storehouse comprises the system metadata storehouse that provides working rule to represent required basic function as shown in Figure 3; The user metadata storehouse of the metadata such as correspondence of the table of user-defined word and actual database and field in the storage practical application; The rule base that utilizes user metadata, system metadata to produce together; Factual database or context database; The volatile data base that produces in the reasoning process.

The reasoning process of deduce machine of the present invention is as shown in Figure 2, and is specific as follows:

(1) receives relevant facts and the effective time thereof that the user imports;

(2) from rule base, get a regular R[Vs, Ve], if can not select then failure; Otherwise change step (3);

(3) with regular R[Vs, Ve] regular former piece TF:Pre and the known true coupling in the factual database;

(4) give regular R[Vs, Ve] conclusion give the TF ' effective time of regular former piece, whether the conclusion of judgment rule R is the new fact, be Already in whether conclusion in the factual database, if not new fact, then change step (5), if new fact, then together with depositing in the factual database as the known fact its effective time, and write down rule and the relevant facts thereof that the match is successful, reexamine whether qualified conclusion of new fact, if meet, then finish this reasoning, if the ineligible conclusion of new fact is then changeed step (2);

(5) from rule base, get another regular R+1, change step (3).

Wherein, the stipulations process of step (3) during stipulations, earlier filters out the expression formula of not having effective time and this regular effective time common factor (establishing expression formula effective time is [Es, Ee], promptly satisfies Ve＜Es or Vs〉Ee) as shown in Figure 1.In remaining expression formula, sum up.

=NULL, then expression formula A and B can sum up, and establish A ∧ B=C, and be TFc the effective time of then summing up back C, promptly has: TFa:A ∧ TFb:B=TFcC; If TFa ∧ TFb=NULL, then expression formula TFa:A and expression formula TFb:B can not sum up, and should select other expression formulas else and sum up this moment.

If regular former piece is set up, and this moment regular former piece effective time be TF ', then use this rule, regular consequent be true, and the effective time of regular consequent also assignment be TF ', and regular consequent adding knowledge base.

In the present invention, obtaining mainly by the mode of artificial treatment of knowledge finished by the interface that system provides.

The interpolation of rule and revise processing procedure and be described below: at first user's system metadata of utilizing system to provide produces the user metadata that meets own application demand, and the user utilizes defined metadata to describe relevant rule subsequently.In order to reduce the meaning of one's words conflict of morphology, the input of rule is finished with combination by the selection to user metadata and operational symbol, can deposit in the total data storehouse through the rule that satisfies system restriction after the collision detection.True interpolation and modification also are like this.

System only allows the current effective rule is deleted, but the deletion of rule and traditional deletion are different, the deletion of rule is deletion forever from knowledge base in traditional production system, and in the tense production system, deleted rule is not a permanent delet from knowledge base, but still be stored in the tense knowledge base, the rule effective time the interval the end be modified to the current concrete time, from then on use as historical rule, though these rules are still effective to wherein historical data to no longer generation effect of current effective data in the total data storehouse.Such as the inquiry " how many Zhang San's wages before 5 years is? ", system just needs him historical data and the rule before 5 years to answer.

The storage scheme of rule specifically comprises as shown in Figure 4:

The logical storage scheme: a rule can be considered as a tuple in logic and store, the content of rule is as the value of user metadata.

The physical store scheme: for the first-class structured message of rule, utilize tables of data to store, for rule body non-structured data such as (former piece, consequents), elder generation stores respectively after it is split as former subitem.Rale store is divided into 6 tables like this: and rule list (regular number, rule name, zero-time, concluding time, former piece ID, consequent ID), former piece table (ID, former piece ID, former piece union ID), former piece union table (ID, former piece union ID, atom prerequisite ID), atom prerequisite table (atom prerequisite ID, unit's relation, attribute, value, temporal constraint), consequent table (ID, consequent ID, atomic action ID), atomic action table (atomic action ID, unit's relation, attribute, value, temporal constraint).

The storage scheme of the fact of the present invention also comprises logical storage scheme and physical store scheme:

Logical storage scheme: still be considered as a tuple and store.

The physical store scheme: for the simple fact, with single table storage of appointing, as cash statement (goods ID, Description of Goods, quantity, zero-time, concluding time), this table has been stored the historical record of goods.If run into the complicated fact, as a people's information etc., can divide several tables storages, to satisfy the normal form requirement, eliminate redundantly, but each sublist all has unique ID of this people, is linked to be a table with view again, can directly take out this people's information during reasoning from view.

The implementation detail of reasoning of the present invention is:

Can have dual mode according to concrete application the opportunity of reasoning: tense triggers and customer incident triggers; According to the automatic triggering rule reasoning of relation between the tense of the tense of rule and user data, the latter specifies the object and the scope of reasoning, manual triggers reasoning by the user by system for the former.

Backward reasoning is adopted in reasoning, elder generation is by the purpose of reasoning, whose need to promote as wage, who needs quilt fine etc., from the action schedule of rule base, find the action that comprises this result, the reverse search validity event that goes out which rule satisfies condition and consequent comprises this action then, and these rale store are gone in the temporary table.

Needing to have obtained the rule of coupling, just can mate one by one according to top inference step.For each bar rule, can once take out whole atom former pieces of its former piece conjunction union, avoid repeatedly reading database, cause performance bottleneck, mate one by one then, if it fails to match for a former piece, then cancellation coupling is left, and takes out the whole former former piece that gives of next conjunction union.The word of the former piece of every rule all by particular data table specific fields in meta data definition and the database one to one.If the free variable in former piece the inside, in internal memory, preserve the value of this time variable, and change it constantly, when reasoning is finished, if necessary, this value assignment is given the variable of the same name of consequent.

See the binding of " NOW " value in the reasoning process at last, the tense of tense production system is an one-dimensional among the present invention, promptly has only effective time, and has time argument " NOW ", needs to bind in reasoning process.In the process of reasoning, " NOW " the value binding principle in the system of the present invention is as follows:

1, the temporal constraint of regular former piece based on " NOW " value, all bind historical logic time of the incident that causes this reasoning;

2, Gui Ze effective time, intrafascicular approximately " NOW " value was then bound current time of system;

3, regular consequent intrafascicular approximately " NOW " effective time is worth, and also binds the historical logic time of incident.Can decide according to the time of former piece.

It is as follows that the present invention is applied to the traffic violations event description:

0: 0: 0 on the 1st January of the regulations of making a dash across the red light 1[1995,23: 59: 59 on the 31st Dec in 1999] ∷ IF  x, some ∧ x＜=23, some ∨ x＞=18, some ∧ x＜=6, x＞=0 o'clock: the THEN that makes a dash across the red light (x, ∞): impose a fine 100 yuan of ∧ deduction of points 5.

More than this rule represented during 23: 59: 59 on the 31st 0: 00 second to 1999 on the 1st Dec of January nineteen ninety-five, all people that make a dash across the red light in 0 o'clock to 6 o'clock morning and 6 o'clock to 12 o'clock evening are imposed a fine 100 yuan, and deducted points 5 fens, can also find out that this conclusion began effectively from that time of making a dash across the red light from this rule.Here as can be seen, this kind method for expressing can be expressed union and the various time period with recursive nature of various time periods.

Claims (5)

1, a kind of tense production system, comprise database, regular collection and deduce machine, described database is deposited the data relevant with the problem of finding the solution, the rule of the general knowledge in the described regular collection problem of representation field, described deduce machine is selected to use to the rule in the regular collection, and the result operates database by inference, it is characterized in that including in the described rule tense attribute.

2, tense production system according to claim 1 is characterized in that described rule comprises regular header, regular former piece and regular consequent;

Described regular header comprises rule numbers, regular effective time, rule name;

All preconditions of described its description rule of regular former piece comprise the tense attribute of description condition;

Described regular consequent is described the action that will carry out after this rule of coupling, and also comprises the tense attribute of describing action in the regular consequent.

3, tense production system according to claim 1 and 2 is characterized in that described database comprises:

The system metadata storehouse that provides working rule to represent required basic function;

The user metadata storehouse of the metadata such as correspondence of the table of user-defined word and actual database and field in the storage practical application;

The rule base that utilizes user metadata, system metadata to produce together;

Factual database or context database;

The volatile data base that produces in the reasoning process.

4, tense production system according to claim 1 and 2 is characterized in that the inference step of described deduce machine is as follows:

(1) receives relevant facts and the effective time thereof that the user imports;

(2) from rule base, get a regular R[Vs, Ve], if can not select then failure; Otherwise change step (3);

(3) with regular R[Vs, Ve] regular former piece TF:Pre and the known true coupling in the factual database;

(4) give regular R[Vs, Ve] conclusion give the TF ' effective time of regular former piece, whether the conclusion of judgment rule R is the new fact, be Already in whether conclusion in the factual database, if not new fact, then change step (5), if new fact, then together with depositing in the factual database as the known fact its effective time, and write down rule and the relevant facts thereof that the match is successful, reexamine whether qualified conclusion of new fact, if meet, then finish this reasoning, if the ineligible conclusion of new fact is then changeed step (2);

(5) from rule base, get another regular R+1, change step (3).

5, tense production system according to claim 4 is characterized in that the coupling concrete steps of described step (3) are as follows:

(3.1) with the TF:Pre negate, intersect with the effective time of regular R again, obtain [Vs, Ve] ∧ TF:( Pre), deposit among the S set et;

(3.2) there is the fact of common factor to deposit among the S set et with [Vs, Ve] effective time with the known fact in the factual database;

(3.3) select two the expression formula TFa:A that do not have stipulations to cross, TFb:B from Set, if choosing is not come out, then former piece [Vs, Ve] ∧ TF:Pre is false, and turns to step (5);

(3.4) expression formula TFa:A, TFb:B are carried out stipulations, reduction rule is as follows: for predicate TFa:A and TFb:B, if A and B can stipulations when considering effective time, A ∧ B=C, and TFa ∧=NULL, TFa:A then, TFb:B can stipulations, and TFa:A ∧ TFb:B=TFa ∧ TFb:C is arranged; Otherwise these two expression formulas can not stipulations, if one of them expression formula is Pre, then after the stipulations success, with upgrading TF ': Pre stipulations result's effective time;

(3.5) if empty clause appears in stipulations, then former piece TF ': Pre sets up; Turn to step (4); Otherwise turn to (3.3).

Images