JPH06259255A - Action plan generating device - Google Patents

Abstract

PURPOSE:To make the device easy-to-use and capable of applying it to problems over a much wider range by automatically developing an inputted target to the state description level of an object while considering limit conditions for the function/structure of the object to be applied. CONSTITUTION:A model base 1 stores deep knowledge expressing the limit conditions for the function/structure or the like of the object related to action plan problems as models, and a model expressing means 2 supports the construction of models by defining the expressed forms of models stored in the model base 1. On the other hand, a model satisfying means 3 infers the target states of respective objects required for satisfying the target while considering the limitation of the function/structure for each object corresponding to the target applied to the object. An operation deciding means 4 infers and leads out the operation corresponding to each object by comparing the target state with the initial state of each object before starting the inference. Then, a presupposing condition guiding means 5 generates presupposing conditions to be confirmed before the operation while considering the limit condition to each operation. These operations are performed by an execution control CPU 6 or the like.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention uses a model called deep knowledge to generate an action plan generation device for generating an action plan such as an operation sequence for operation control of a plant or an operation sequence for system fault recovery. Regarding

[0002]

2. Description of the Related Art For example, in a chemical reaction plant or a power generation plant, so-called action plan generation such as generation of an operation plan for operation control and generation of a switching operation plan of a power transmission system is important. Is recognized as a serious problem.

The basic technique for generating such an action plan is described in Reference 1 (RE Fikes, NJNilsson, “STRIPS
: A new approach to the application of theorem pr
ov-ing to problem solving ”, Artificial Intellig
ence 2, pp.251-288, 1971.).

STRIPS is a program designed to solve problems encountered by robots. The case where this program is applied to the world of building blocks will be briefly explained as an example.
In this program, first, the initial state and the target state are set as follows, for example. Initial state on (C, table), on (B, C), on (A, B), cle
ar (A) Target state on (C, table), on (A, C), on (B, table),
clear (A), clear (B)

This is because the building block C is placed on the table, the building block B is placed on the building block C, and the building block A is building block B.
, And nothing on Block A,
In the initial state, building blocks C are placed on the table, building blocks A are placed on the building blocks C, building blocks B are placed on the table, and nothing is placed on the building blocks A and B. It means that the target state is not. The actual action is performed by the robot, and the possible movements of the robot are defined as an operator (hereinafter synonymous with "operation") as follows. Operation pick up (X) Prerequisite: clear (X), on (X, Y) Additional effect: hold (X), clear (Y) Deletion effect: clear (X), on (X, Y) Operation stacK (X , Y) Precondition: hold (X), clear (Y) Additional list: on (X, Y) Deletion list: hold (X), clear (Y)

In STRIPS, first, the initial state is compared with the target state, and the difference is detected. The difference is an item that is not satisfied in the initial state in the target state (or the prerequisite of the operator). First, try to apply an operator suitable for the difference. If there is an item that is part of the difference in the additional effect of the operator, it is determined that the operator is appropriate. In the process of applying the operator, the precondition may not be satisfied. In such a case, generate a sub-problem, transform the given state into the operator's domain, and generate the goal of applying it to the operator who does not satisfy the precondition but wants to apply it. Try to solve this subproblem in the same way as when solving the first problem.

Such a search method is called means-end analysis. By adopting this search method, an operator application sequence that reduces the difference between the initial state given as a problem and the target state is generated as a plan.

Generally, in the action plan problem, the main focus is to search which operator should be applied in what order (or at what timing). Can be said to be given by STRIPS. However, such a conventional method has the following problems.

First, the method of giving a target in STRIPS is based on a state description like the example of building blocks. That is, when a target is given, the state of the building blocks will be immediately known automatically. However, when a goal is specified, the problem of instantly knowing the target state, such as building blocks, is a rather rare case, and it is usually necessary to infer what the target state should be in order to achieve the goal. There is. That is, the “goal”
In general, "state" generally refers to different information.
In such a problem, it is necessary to expand the goal description given as an input into a state description that can satisfy the goal, taking into consideration the constraint conditions inherent in the object. The prior art does not provide any method for solving such a problem.

Next, in STRIPS, the operator is
It is premised that it is given in advance together with the prerequisites and effects necessary for its execution. Therefore, as described above, the main problem is the search problem of how to quickly find the application sequence of the operator. However, in a very large-scale problem, the number of operation targets becomes enormous, and it is enormous to extract all possible operations, preconditions, and effects for each operation target in advance. Costly,
It becomes a factor that hinders the ease of system construction. In such a problem, a mechanism that can automatically derive the operation and its preconditions and effects that can be automatically performed from the function / structure of each operation target is required, but the conventional technology does not provide any method for that. Not not.

[0011]

As described above, in the conventional action plan generation method, not only the description level of the goal is specified, but also preconditions and effects for executing each operator (operation) are given in advance. Therefore, there is a problem that the applicable range is extremely narrow.

Therefore, according to the present invention, it is possible to generate an action plan without having to define in advance the degree of freedom in describing goals and the operators that can be executed for each target and the preconditions relating to the operators, thus expanding the scope of application. It is intended to provide an action plan generation device that can perform.

[0013]

In order to achieve the above object, an action plan generation apparatus according to the present invention models each object related to an action plan problem and defines a terminal for each modeled object. However, the connection relationship between objects and the hierarchical relationship of abstraction / reification are defined as the connection relationship between terminals, and the process flow between objects is defined by the process parameter added to the connection relationship between terminals. Model expressing means for defining the functional / structural constraint conditions of the object as local information of each target, and the functional / structural constraint conditions for each target given by the model expressing means to the target given to the target A model satisfying means for inferring a goal state of each target necessary to satisfy the above goal while considering, and a goal state determined by this model satisfying means and each An operation deciding means for inferring an operation for each object necessary for satisfying a goal by comparing the initial state of the elephant before starting the inference, and the model expressing means for each operation decided by the operation deciding means. And a precondition derivation means for generating a precondition that needs to be confirmed from the functional / structural viewpoint prior to the operation in consideration of the target functional / structural constraint conditions given by the above.

[0014]

[Operation] Even when the specification that the object finally wants to satisfy is given as a goal of a form not directly related to the state description of the object, the functional and structural constraints of the object given by the model expressing means are taken into consideration. Then, the model satisfying means automatically expands the input target to the target state description level. Therefore, it can be applied to a wider range of problems. That is,
The action plan of the target can be generated only by giving the goal of a more general form, rather than the goal of the concrete form developed at the state description level.

Further, the precondition derivation means automatically generates the preconditions for each target operation in consideration of the functional and structural constraints of the target given by the model representation means. Therefore, it is not necessary to define the preconditions for the operation of each target in advance, and only the functional / structural constraints of each target need to be defined as a model, which facilitates acquisition of knowledge (model). Become.

[0016]

Embodiments will be described below with reference to the drawings. FIG. 1 shows a block configuration diagram of an action plan generating apparatus according to an embodiment of the present invention.

This action plan generation apparatus stores a model base 1 in which deep knowledge expressing constraint conditions such as functions and structures of objects related to the action plan problem is stored as a model, and a model stored in the model base 1. Model expression means 2 for defining the expression format of the
And a model satisfying means 3 for performing a satisfying process of the constraint condition defined in the model with respect to the input target, and an operation determining means for deriving an operation for the target by comparing the contents before and after the model satisfying process. 4 and precondition derivation means 5 for determining the preconditions necessary for the operation from the model. Further, a CPU 6 and an input / output device 7 that control execution of the model expressing means 2, the model satisfying means 3, the operation determining means 4, and the precondition deriving means 5 are also provided. Here, the function of each element constituting the action plan generation device will be described by taking the automatic silo device shown in FIG. 2 as an example of a processing target.

In FIG. 2, in order to facilitate the understanding of the modeling concept by the model representation means 2 described later, the terminal definition (black circle) of each element and hierarchical modeling (in this example, subordinate to the conveyor unit 20). A hierarchy is defined, and this lower hierarchy is composed of two systems connected in parallel as shown on the right side of the drawing. First, the configuration of this automatic silo device will be briefly described.

In this automatic silo device, the raw materials put into the silo 11 and the silo 12 are weighed by the hopper 13 and the hopper 14, respectively, and then mixed by the mixer 15.
Release of the raw material from each silo 11, 12 and each hopper 13, 14 is controlled by valves 16, 17, 18, 19. The raw materials mixed in the mixer 15 are output by the conveyor unit 20.

The conveyor unit 20 is a conceptual device, and is actually composed of two systems of conveyors and silos as shown on the right side of the drawing. That is, the mixer 1
The raw material taken out from No. 5 is sent to the conveyor 22 and the conveyor 23 via the distribution pipe 21. Each conveyor 2
2, 23 are connected to a silo 24 and a silo 25.

The final output of this automatic silo device is obtained via the merging pipe 26. At this time, the proper use of the system of the conveyor 22 and the system of the conveyor 23 is
It is controlled by valves 27, 28 mounted at the entrance of each conveyor. Now, as a problem, the final output of this automatic silo device is "production of 210 [kg / hr] mixed raw material".
Is given as a goal. First, the model representation means 2 models the automatic silo device shown in FIG. The features of this modeling method are as follows. (1) Device configuration view

In order to define modeled objects locally, a terminal is defined for each object, and a connection relationship between objects is expressed as a connection relationship between terminals. (2) Process view

The process flow between modeling targets is expressed as the flow between terminals. Here, one terminal and one process are used to facilitate handling.

Hierarchical modeling can be realized by this modeling method. Therefore, a model corresponding to one physical device and a model corresponding to an abstract device obtained by hierarchically organizing several devices can be formed. The former will be called a "physical target model" and the latter will be called an "abstract target model". The description of each model is realized as frame type knowledge or object instance having the following attributes. Physical target model (1) name Specify the target name. (2) export

Designate a terminal that can be referenced from the outside. The terminal name must be unique to the target local. These terminals are used to define the object-to-object connection relationship and process flow. (3) Specify the inheritance destination of the is-a model attribute definition. (4) a-part-of It is used to define the hierarchical structure of the target and expresses that the modeling target is a partial element of the target specified by this attribute. (5) goal

Define process parameters in which goals for an object are defined. Since the process parameter name and the terminal name are based on the principle of one terminal and one process, they can be handled the same in the target local (the description can be made compact). Therefore, this attribute is the same as specifying the terminal name where the target should be defined. (6) forward

The effect of the state is defined for each possible state of the object. Specifically, the effect is specified by the relational expression regarding the process parameter name for transmitting the effect, that is, the terminal name. (7) backward

Preconditions of the state are defined for each state that the target can have. Specifically, the precondition is specified by a process parameter name to which the process amount as the precondition is transmitted, that is, a relational expression regarding the terminal name. In addition, of these prerequisites, specify those important ones that need to be confirmed when performing the operation on the target. (8) states Define the definition of possible state names of the target and the relational expressions of process parameters and internal parameters unique to each state. (9) local Defines the relational expression for the internal parameter. (10) value The values in the initial state and the target state of all parameters defined in the model are stored. (11) status The target initial state and target state are stored. Abstract target model

An object is conceptually abstracted from a collection of physical objects. This allows a hierarchical representation of the model. The role of the abstract object model is only to define connection relationships between objects. (1) name Specify the abstract target name. (2) export

Designate a terminal that can be referenced from the outside. The terminal name must be unique to the abstract target local. These terminals are used to define connection relationships between abstract objects and process flows. (3) Specify the inheritance destination of the is-a model attribute definition. (4) a-part-of It is used to define the hierarchical structure of the abstract target and expresses that the modeling target is a partial element of the target specified by this attribute. (5) import

A relationship opposite to export is defined, and a lower layer target terminal is designated when referring to a lower layer target from a higher layer target (that is, a target designated by name). (6) connection Specify the connection relationship between terminals in the lower hierarchy. This defines the device configuration. (7) correspond

It specifies which terminal of the lower hierarchy object corresponds to the terminal of the upper hierarchy object (that is, the object specified by name) (that is, the terminal defined by export). This clarifies the relationship between the upper layer target and the lower layer target. (8) component Specify the lower layer target.

This model representation format will be described with reference to the example of FIG. 2. In FIG. 2, the circle marks (t1 ...
t37) represents each target terminal, the solid arrows between the terminals represent the target connection relationship and process flow (in the example of Fig. 2, the flow rate process of the raw material), and the dashed line between terminals is the upper hierarchy. Represents the correspondence between the and lower layers. Therefore, when the entire automatic silo device, which is the highest layer, is modeled using the identification name and the terminal name of each device shown in FIG. 2, it becomes as shown in FIG.

Similarly, the conveyor unit 20 is modeled as shown in FIG. 4, the merging pipe 26 is modeled as shown in FIG. 5, and the silo 24 is modeled as shown in FIG. The conveyor 22 is modeled as shown in FIG. 7, the conveyor valve 27 is modeled as shown in FIG. 8, the diversion pipe 21 is modeled as shown in FIG. 9, and the mixer 15 is modeled. As shown in FIG. 10, the hopper 13 is modeled as shown in FIG.

As described above, the model expressing means 2 incorporated in the present embodiment expresses the target hierarchically and declaratively in a concise manner. In this case, a standard model is defined in advance for each typical device such as a valve or silo, and the system provides the standard model, so that the model construction cost can be reduced.

Next, what kind of processing is performed by other elements using this model will be described. First, the operation of the model satisfying means 3 will be described with reference to FIG.

In step 31, a goal is received. This goal need not be a state description of the object, ie the automatic silo system, but may be of a higher dimension. As an example, assume that the following goals are given. Goal: Increase the production amount by the automatic silo device to 210 [kg / hr]. This target is concretely expressed and input in the following form based on the description by the model expressing means 2. Goal: goal (convey-unit, t19,210) At step 32, this goal is recognized as unprocessed and control proceeds to step 33.

As is clear from the model of the conveyor unit 20 described above, the conveyor unit 20 is an abstract object model, and subordinate objects are defined. Therefore, in step 33, the target given to t19 is expanded to the following target by referring to the correspond attribute in the model of the conveyor unit 20. Goal: goal (merge-conduit2, t37,210)

After this expansion processing, steps 31 and 32 are executed again.
Do. As is clear from the model of the merging pipe 26 described above, since the subordinate object is not defined in this merging pipe 26, the process proceeds from step 33 to step 35.

In step 35, the expanded target is input, and the state satisfying all the constraint conditions defined in the merging pipe 26 is determined. In the confluence pipe 26, the state is pa
Since there is only one single state defined, ssive, passi
The constraint expressions for ve states are collected from the target model.
Then, it is checked whether or not the input goal can be satisfied.
Below, all constraint equations in the passive state regarding the merging pipe 26 are shown. t37 = t27 + t35 (from forward attribute) t27 = t37-t35 (from backward attribute) t35 = t37-t27 (from backward attribute) distribute2 (t37, t27, t35) (from states attribute)

Here, the fourth constraint expression requires attention. This is represented by the special name distribute2. This name is called a target expansion primitive, and it is expanded into a concrete set of constraint expressions before the processing of step 35. I.e. distribute2
Expands to the following set of constraints according to priority. Priority 1: t27 = t37, t35 = 0 Priority 2: t35 = t37, t27 = 0 Priority 3: t27 = t37 / 2, t35 = t37 / 2

With respect to this expansion rule, candidates for a constraint expression set to be expanded may be defined in advance in accordance with a case, and various expansion rules other than those shown in this example can be defined. Further, when returning to the process of step 35 from the process of step 38 which will be described later, it is possible to examine all the selection candidates by expanding the constraint formula of the next priority and repeating the process. When this expansion process ends, the constraint expression of t37 = t27 + t35 t27 = t37-t35 t35 = t37-t27 t27 = t37, t35 = 0 (the result of the expansion of priority 1) is obtained first. Currently, the target is t37 = 21
Since 0 is given, the following solution is obtained as a solution that satisfies this constraint set. t37 = 210 t27 = 210 t35 = 0

The local constraint satisfaction process for each object can be realized by a simple algorithm. For example, when all the parameter values on the right side of one constraint expression are known, an algorithm by constraint propagation that propagates the calculated value to the parameter value on the left side, or by solving simultaneous equations and transforming the constraint expression, Algorithms that determine individual parameter values are known, and it is easy to select an appropriate algorithm introduced in Ref. 2 ("Japan Society for Software Science and Technology Summer Tutorial (Constraint Programming)", 1989.). Can be realized. Confluence pipe 2
In step 6, since the state of passive satisfies the constraint equation, the process proceeds from step 36 to step 37.

In step 37, the backward flow of the merging pipe 26 is performed.
A new target value is calculated from the above constraint satisfaction result by referring to the attribute, and then export, a-part-of, conne
Decide the propagation destination by referring to the ction, import and correspond attributes. First, a new target value is calculated as follows. Target value 1: t27 = 210 Target value 2: t35 = 0

Next, as to the propagation destination, expo of the merging pipe 26
Referring to the rt attribute, since both t27 and t35 are defined as external reference terminals, the merging pipe 26 to t2 can be connected to the conveyor unit 20 defined by the a-part-of attribute.
Propagate 7 and t35.

In the conveyor unit 20, the merging pipe 26
Both t27 and t35 are defined in the connection attribute, and the connection destination of each terminal is "t27 of merge-conduit2.
Is t-26 of c-silo ”,“ t35 of merge-conduit2
Is a d-silo t34 ”, and the target and terminal of the connection destination are defined in the import attribute.
At this point, the propagation destination of each target value is determined as follows. Target 1 propagation destination: silo 24 t26 terminal Target 2 propagation destination: silo 25 t34 terminal By combining the above results, a new target is determined as follows, and processing from step 37 to step 31 Is repeated again. Goal 1: goal (c-silo, t26,210) Goal 2: goal (d-silo, t34,0)

Thereafter, similar processing is repeated to perform the constraint satisfaction processing for all the objects shown in FIG. Finally, when the process of step 37 is performed in each of silo 11 and silo 12, each model becomes
Since the backward attribute is not defined, a new target is not generated, there is no unprocessed target when the process proceeds from step 31 to step 32, and the model satisfying process ends. If the constraint satisfaction fails in step 35, the process proceeds from step 36 to step 38 to return to another selection candidate in step 35 and continue the processing.

As described above, the model satisfying means 3 develops the target based on the functional / structural constraint expression of the target with respect to the input target not described in the state level, and satisfies the target. , That is, automatically generate a goal state in a general action plan generation problem. Next, the operation determining means 4
The operation of will be described.

Basically, it is realized by comparing the states before and after the processing by the model satisfying means 2. In the description of the model representation means 2 described above, the value attribute and the status attribute shown in the physical object model description example show an example of the states of the model satisfaction result before and after the model satisfaction process is performed. The following operations are determined. activate c-convey, ie off (initial state) → on
(Target state) open c-convey-vlv That is, close (initial state) → o
pen (target state) activate mixer ie off (initial state) → on
(Target state) open a-hopper-vlv That is, close (initial state) → o
pen (target state) activate a-hopper ie off (initial state) → on
(Target state)

Open a-silo-vlv That is, close (initial state) → open (target state) In this example, the operation is easily derived by comparing the target state and the initial state, but the comparison is made by some knowledge. It can be extended to problems where the results must be analyzed. Next, the operation of the precondition derivation means 5 will be described. The following relationships are established when an operation on a certain object can be safely executed from the viewpoint of the function and structure of the object.

(1) When a process flow exists along the target connection, the precondition of the target located on the downstream side (defined by the backward attribute expressed by the model expressing means 2) is on the upstream side. Guaranteed by the state effect located (defined by the forward attribute represented by the model representation means 2). (2) The state effect of the object located upstream does not adversely affect the precondition of the object located downstream.

The precondition derivation means 5 generates the following local preconditions and global preconditions in consideration of the above relationship. Each precondition is briefly described below. (1) Local preconditions Preconditions that are determined from the local perspective of the operation target only that the prerequisites of the operation target have to be satisfied prior to the operation. (a) Target state KEEP prerequisites

It defines that the preconditions sufficient to keep the transition destination state by the operation are satisfied. It defines the condition in which the prerequisite is propagated to the adjacent target on the upstream side and the effect is obtained. (b) Initial condition KEEP preconditions

It is specified that the preconditions for keeping the pre-operation state are satisfied until just before the execution of the operation. It defines the condition in which the prerequisite is propagated to the adjacent target on the upstream side and the effect is obtained. (2) Global prerequisites

Preconditions determined from a global point of view in consideration of targets affected by the operation target, that is, the preconditions of the neighborhood of the operation target are satisfied prior to the operation. (a) Target condition Effect exertion condition

It is stipulated that the effect of the operation will not spread and the adjacent object will not be adversely affected. It specifies that the effect of the target state propagates to the adjacent target on the downstream side, and that it does not have a precondition that contradicts this. (b) Conditions for extinguishing the initial state effect

It is specified that even if the effect of the pre-operation state disappears by the operation, the adjacent object is not adversely affected. It stipulates that the denial of the effect of the initial state is propagated to the adjacent target on the downstream side, and that it does not have a precondition that conflicts with this.

The above-mentioned preconditions are used to determine the operation determining means 4
FIG. 13 shows a processing flow of the precondition derivation means 5 which is derived for each operation generated by. In addition, here
In the case shown in FIG. 2, the starting operation of the conveyor 22 will be taken as an example, and the derivation process of the preconditions for this operation will be described. In step 41, the operation of starting the conveyor 22 (changing the state from off to on) is input. Operation operation (c-convey, off, on) First, the target state KEEP precondition generation process (step
42 to step 46) will be explained.

In step 42, by referring to the backward attribute in the model of the conveyor 22 generated by the model expressing means 2, precond (t23 = t24) is extracted as the precondition description in the target state on. This expresses that the value of t23 calculated by t23 = t24 propagates as a precondition. By referring to the value attribute in the model of the conveyor 22, it can be seen that the value in the target state of t23 is 210, and finally the following precondition for the target state on is obtained. Here, the symbol "precond" is described by the model representation means 2 as a keyword indicating a precondition of the operation. Assumption: t23 = 210

Since the precondition is found, the process proceeds from step 43 to step 44 and propagates it to the adjacent object. The mechanism of this propagation is the model satisfying means 3 shown in FIG.
The processing is exactly the same as the processing in steps 34 and 37. Therefore, although not described, the following preconditions are eventually propagated to the valve 27 of the conveyor 22. Assumption: t22 = 210 Destination: c-convey-vlv

Next, at step 45, the state of the valve 27 of the conveyor 22 that satisfies the propagated precondition is determined. This state determination mechanism is also shown in FIG.
The process is exactly the same as the process in step 35 of the model satisfying means 3 shown in FIG. Therefore, although the explanation is omitted, as a result, the following is determined as the state of the valve 27 of the conveyor 22. Target: c-convey-vlv State: open Therefore, in step 46, the following target state KEE
P preconditions are generated. Target state KEEP precondition: c-convey-vlv = open Next, generation process of initial state KEEP precondition (step
Steps 47 to 51) will be described. This process is exactly the same as the process flow of the target state KEEP precondition.

In step 47, the backward attribute in the model of the conveyor 22 is referred to, but the initial state is off.
Preconditions on are not defined (keyword prec
The condition specified in ond is not described in the backward attribute). Therefore, this process is ended by steps 47 to 48, and in this case, no precondition is generated. Next, the generation process of the target state effect exertion condition (step 52 to step 56) will be described.

By referring to the forward attribute in the model of the conveyor 22 in step 52, the target state is
As the state effect description for on, t24 = t23 is extracted. This expresses that the value of t24 calculated by t24 = t23 propagates as a state effect. By referring to the value attribute in the model of the conveyor 22, it can be seen that the value in the target state of t24 is 210, and finally the following state effect in the target state on is obtained. State effect: t24 = 210

Since the state effect is found, the process proceeds from step 53 to step 54 and propagates it to the adjacent object. The mechanism of this propagation also corresponds to the model satisfying means 3 shown in FIG.
This is the same as the processing in steps 34 and 37 of. Therefore, although the description is omitted, as a result, silo 2
The following state effects are propagated to 4. State effect: t25 = 210 Destination: c-silo

Next, in step 55, the state of the silo 24 that does not satisfy the propagated state effect is determined. This state determination mechanism is also similar to the process in step 35 of the model satisfying means 3 shown in FIG. That is, in step 35, the state in which the constraint is satisfied is selected, whereas in step 55, the state in which the constraint is failed is selected. Although the description is omitted because it is the same processing, as a result, it is output as follows that such a state of the silo 24 does not exist. Target: c-silo State: NIL Therefore, in step 56, the following target state effect exerting conditions are generated. Target state effect exertion condition: c-silo ≠ NIL

However, since it is meaningless to generate such a condition, no condition is generated in step 56 if there is no state in which the state effect adversely affects as in this example. May be. Finally,
The initial state effect extinguishing condition generation processing (step 57 to step 61) will be described. This processing is similar to the processing flow of the target state effect exerting condition.

In step 57, t24 = t23 is fetched as the state effect description in the initial state off by referring to the forward attribute in the model of the conveyor 22. This is t24 calculated by t24 = t23
The value of is propagated as a state effect. By referring to the value attribute in the model of the conveyor 22, it can be seen that the value of t24 in the initial state is 0, and finally the following state in the initial state off is obtained. State effect: t24 = 0

Since the state effect is found, the process proceeds from step 58 to step 59 and propagates it to the adjacent object. The mechanism of this propagation also corresponds to the model satisfying means 3 shown in FIG.
This is the same as the processing in steps 34 and 37 of. However, the only difference is that the negation of the state effect is propagated. This denial can be easily realized by predefining the conversion rule of the constraint expression described in the model as shown below. Transformation Rule 1: Equality is transformed into inequality. P1 = f (Xi) → P1 ≠ f (Xi) or P1 = f (Xi) → P1> f (Xi) or P1 <f (Xi
) Transformation rule 2: The inequality reverses direction. P1> f (Xi) → P1 = <f (Xi) P1 <f (Xi) → P1 => f (Xi)

Conversion rules other than the above can be easily realized. According to such a conversion rule, the state effect t24 = 0 extracted in step 57 is the negation of the state effect t24 ≠
Converted to 0 and finally propagated to the silo 24 as follows. State effect: t25 ≠ 0 Propagation destination: c-silo

Next, in step 60, the state of the silo 24 that does not satisfy the negation of the propagated state effect is determined. This state determination mechanism is also shown in FIG.
The process is the same as the process in step 35 of the model satisfying means 3 shown in FIG. That is, in step 35, the state in which the constraint is satisfied is selected, whereas in step 60, the state in which the constraint is failed is selected. Although the description is omitted because it is the same processing, as a result, it is output as follows that such a state of the silo 24 does not exist. Target: c-silo State: NIL Therefore, in step 61, the following initial state effect elimination conditions are generated. Initial state effect extinction condition: c-silo ≠ NIL

However, since it is meaningless to generate such a condition, if there is no state in which the state effect adversely affects as in this example, no condition is generated in step 61. May be.

When all the conditions have been generated in this way, the generated conditions are AND-combined in step 62 to generate the following as a final precondition regarding the starting operation of the conveyor 22. IF: c-convey-vlv = open AND C-silo ≠ NIL THEN: activate c-convey Therefore, the action plan of the conveyor 22 is generated here.

The case explained above is based on the precondition deriving means 5.
It is extremely simplified and simplified to explain the basic processing mechanism of. Therefore, a case of generating particularly complicated preconditions is not described.
However, it goes without saying that if the target model is described in detail by the model expressing means 2, complicated preconditions of a real level can be generated.

Further, in the above-mentioned example, a comparatively simple case has been described in order to briefly explain the gist of the invention, but a more complicated case can be applied as follows, for example. That is, not only the flow rate but also other processes such as pressure and temperature can be handled. This can be realized by adding a terminal for each process to each target and adding a relational expression between processes as a constraint expression to the model. For example, inference using a pressure loss equation regarding pressure, flow rate, and resistance can be realized.

Further, regarding the valve, the description has been made only for "fully open" and "fully closed", but it is possible to handle the intermediate opening. That is, since the resistance of the pressure loss equation described above is given as a function of the valve opening degree, and the pressure loss equation is also given by Bernoulli's law, all can be formulated as constraint equations for the model.

Further, in the operation determining means, the case where the operation is inferred only by the difference between the target state determined by the model satisfying process and the initial state before the model satisfying process has been taken up. The present invention can be applied to problems that must be solved.

[0077]

As described above in detail, according to the action plan generating apparatus of the present invention, the specification of the target desired by the object is a goal of a general form not directly related to the state description of the object. , It is easy to use because the model satisfying means automatically expands the input target to the state description level of the target in consideration of the functional and structural constraints of the target given by the model expressing means. Moreover, it can be applied to a wider range of problems.

Further, by providing the precondition deriving means, the precondition for each operation of the target is automatically generated in consideration of the functional and structural constraints of the target given by the model expressing means. You can Therefore, it is not necessary to define the preconditions for the operation of each target in advance, and only the functional and structural constraints of each target need to be defined as a model, which facilitates acquisition of knowledge (model). There is also an advantage.

[Brief description of drawings]

FIG. 1 is a block configuration diagram of an action plan generation device according to an embodiment of the present invention.

FIG. 2 is a schematic configuration diagram of an automatic silo device as an example of a processing target of the device.

FIG. 3 is a diagram showing a representation example in which the entire automatic silo device is modeled by a model representation means.

FIG. 4 is a diagram showing a representation example in which a coin-bearing unit of the same automatic silo device is modeled by a model representation means.

FIG. 5 is a diagram showing a representation example in which a confluent pipe of the automatic silo device is modeled by a model representation means.

FIG. 6 is a diagram showing a representation example in which one silo of the same automatic silo device is modeled by a model representation means.

FIG. 7 is a diagram showing a representation example in which one conveyor of the automatic silo device is modeled by a model representation means.

FIG. 8 is a diagram showing an expression example in which one conveyor valve of the automatic silo device is modeled by a model expression means.

FIG. 9 is a diagram showing a representation example in which the flow dividing pipe of the automatic silo device is modeled by a model representation means.

FIG. 10 is a diagram showing a representation example in which a mixer of the automatic silo device is modeled by a model representation means.

FIG. 11 is a diagram showing a representation example in which one hopper of the same automatic silo device is modeled by a model representation means.

FIG. 12 is a flow chart for explaining the processing by the model satisfying means in the action plan generating device.

FIG. 13 is a flowchart for explaining the processing by the precondition derivation means in the action plan generation device.

[Explanation of symbols]

1 ... Model Base 2 ... Model Representation Means 3 ... Model Satisfaction Means 4 ... Operation Determining Means 5 ... Prerequisite Deriving Means 6 ... CPU 7 ... Input / Output Device

Claims (1)

[Claims] 1. A model of each object related to an action plan problem, a terminal is defined for each modeled object, and a connection relationship between objects and a hierarchical relationship of abstraction / incarnation are defined as a connection relationship between terminals. And a model expression means for defining the process flow between the objects by the process parameters added to the connection relationship between terminals, and further defining the functional / structural constraints of each object as local information of each object, A model that infers the target state of each target that is necessary to satisfy the above target while considering the functional / structural constraints of each target given by the above model expressing means to the target given to the target By comparing the satisfying means and the goal state determined by this model satisfying means with the initial state of each object before the start of inference, each object required to satisfy the goal is compared. The operation determining means for inferring the operation to be performed, and the function / structural constraint conditions of the target given by the model expressing means for each operation determined by the operation determining means are taken into consideration. From the point of view, an action plan generation device comprising: a precondition derivation unit that generates a precondition that needs to be confirmed.