JPH0774997B2 - Qualitative reasoning system - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inference method of a qualitative inference system that infers a temporal change of a dynamic system by using knowledge of an object consisting of a conditional part and a consequent part and a physical law, and particularly, By referring to the physical quantity at the previous time when expressing deep knowledge, it is possible to make expressions and inferences that could not be expressed until now.

[0002]

2. Description of the Related Art A qualitative inference system, which is a system for inferring temporal changes, generally conducts simulations,
It is used to predict and explain the dynamics of a dynamic system over time.

For example, a qualitative reasoning system Qu, which is one of the qualitative reasoning systems that is a system for reasoning temporal changes.
pras (Qualitative reasoning based on physical laws, qualitative reasoning, 8th
In Chapter, Kyoritsu Shuppan), a state at a certain time is created using rules equivalent to fundamental knowledge about objects and laws of physics. A rule corresponding to the physical law that a light bulb shines when electricity flows can be expressed as shown in FIG.

This rule is an IF-THEN format rule that indicates that the expression written in relations becomes valid when the condition written in conditions fires. The rule is that the temperature of the light bulb (light_bulb) (light_generatin
g_temperature @ Light_bulb), it means that the change in the light emission (light_generation @ Light_bulb) of the light bulb (enclosed by ddt) is proportional to the change in the temperature. This consequent expression is considered as a constraint and is stored in the constraint pool where the constraints are gathered to see if it is consistent. The constraint pool is used to evaluate the conditional part of other rules. The constraint pool corresponds to the working memory of a normal rule-based inference system, but when evaluating conditions, it is evaluated using a constraint solver that solves expressions (or inequalities) instead of simple pattern matching.

The evaluation order of rules is the same as that of forward inference. To evaluate the rule in order to obtain the state at the current time (the state is defined by the rule that satisfies the physical quantity value and the condition) is called the in-state analysis.

Once all rules have been evaluated, the system finds a variable in the constraint pool whose derivative is non-zero and the value that variable should take at the next time. This inference is called marginal analysis. Qualitative reasoning is
The behavior of the target system is inferred by repeating the in-state analysis to obtain the state at a certain time and predicting the value of the next time of the changing physical quantity by the limit analysis.

In a conventional system for inferring temporal change, for example, Qupras, the reference of physical quantity in the condition part and the consequent part is limited to the physical quantity at the time of inference. For example, if it becomes necessary to convert the voltage from digital to analog, (1) if the digital level does not change at 0, the analog voltage is 1.0 volt or less, and (2) if the digital level is 1 If it doesn't change, the analog voltage is more than 1.4V, and (3) if the digital level is changing, the analog voltage is between 1.0V and 1.4V. It is possible to use the law, but to describe this law, it is necessary to refer to the value of the physical quantity at the previous time. In the third rule, the condition that the digital level is changing is
It can be described that the digital level at the current time and the digital level at the previous time are different. In the first rule, the condition that the digital level does not change at 0 can be described as if the digital level at the current time and the digital level at the previous time are 0. However, in the conventional systems, such a rule could not be expressed because there is no function to refer to the physical quantity at the previous time.

[0008]

In a qualitative inference system that infers temporal changes, the condition part and consequent part of the rule cannot refer to the physical quantity at the previous time, so the physical quantity at the previous time must be referenced. There was a problem that rules that could not be expressed could not be described.

An object of the present invention is to enable a description of a rule that needs to refer to a physical quantity at a previous time by making it possible to refer to a physical quantity at a previous time in a conditional part and a consequent part of the rule. It is to be.

[0010]

[Means for Solving the Problem] In a qualitative inference system that infers temporal changes, the conditional part and consequent part of a rule are
Allows you to write a prev function that takes a physical quantity as an argument to refer to the physical quantity at the previous time. The prev function refers to the time before the physical quantity. The means to implement the prev function is to store a dictionary that collects the values of physical quantities at each time as a history pool, and if the prev function is used, retrieve the dictionary of the current previous time from the history pool and apply it. Retrieve the value of physical quantity.

As an example using the prev function, if the digital level, which is the first rule of converting the voltage from digital to analog, does not change at 0, the analog voltage is 1.0 volt. It is as shown in Fig. 8 when it is described using functions. Here, in order to indicate that the digital level does not change, it is described that the current time and the previous time are the same.

As in this example, by using the prev function that refers to the physical quantity at the previous time, it is possible to describe a rule that needs to refer to the value at the previous time.

[0013]

When the time change is inferred, the physical quantity values for each time are collected as a dictionary, the time and the dictionary are paired and stored in the history pool, and the physical quantity at the past time is referred to later. Then retrieve the dictionary for that time from the history pool,
The physical quantity at the previous time is taken out by providing a mechanism for taking out the corresponding physical quantity from the dictionary.

[0014]

FIG. 2 is a block diagram of an inference apparatus which is an embodiment of the present invention. An outline of the configuration and operation of the entire device will be described.

The device comprises a control device (CPU) 200, a main memory 206, a magnetic disk 208, a CRT 202, and a keyboard 204. Knowledge base 104 and initial data 10
0 is stored in the magnetic disk device 208. Further, the program of the inference unit 108 is also the magnetic disk device 20.
8 is stored. The keyboard 204 is used for program instructions. The CRT 202 is used for displaying results and the like.

The programs of the inference unit, the contents of the knowledge base and the initial data are loaded into the main memory 206 as required. This operation is performed from the keyboard 204. The keyboard 20 is also used to specify a command for calling the program of the inference unit into the main memory 206 and an execution start command.
It starts from 4. CRT20
It is displayed in 2.

The present invention will be specifically described below.

First, the outline of the embodiment will be described. FIG. 1 shows the entire program and data structure. The inference unit 108 stores the content of the initial data 100 in the constraint pool 124 via the constraint solver 116, and evaluates the conditional part of the rule in the knowledge base using the constraint pool and the constraint solver. If all the rules are evaluated, the value of physical quantity is calculated and stored in the dictionary. The dictionary stores the values of physical quantities in the form of pairs of physical quantities and values. Then, the identifier of the time at that time and the dictionary are paired and stored in the history pool 120. The inferred result is 112
Sought.

The inference unit 108 stores the initial data 100 in the constraint pool 124 using the constraint solver 116 of the inference unit before starting the inference.

First, the inference of the in-state analysis will be described with reference to FIG. First, in the in-state analysis, we check whether all rules have been evaluated (30
2). If it is finished, the values of all physical quantities are calculated, pairs of physical quantities and their values are created, and the collection of them is used as a dictionary. Then, the identifier of the current time and the dictionary are paired and stored in the history pool (312). The history pool is
The dictionary can be retrieved from the time identifier.
Also, by specifying a physical quantity, its value can be retrieved from the dictionary. When the dictionary is stored in the history pool, the limit analysis shown in FIG. 4 is performed to predict the next time (314). If all the rules have not been evaluated, one rule that has not been evaluated is taken out from the knowledge base (304). Then, the constraint solver 116 is used to evaluate the conditional part of the rule (30
6). The evaluation method will be described with reference to FIG. It is checked from the evaluation result whether the condition part of the rule is satisfied (308). If it is not satisfied, the process proceeds to step 302 again. If so, the consequent expression is added to the constraint pool using the constraint solver (310). The detailed method of 310 will be described with reference to FIG.

The process of limit analysis 314 will be described with reference to FIG. First, the history pool is used to check whether the state obtained by the previous in-state analysis has already been obtained (40
2). If it is the same as the already obtained state, the process is ended. If not, it is checked whether there is a physical quantity changing in the state obtained by the in-state analysis (406). If there is no changing physical quantity, the process ends.
If there is a changing physical quantity, the value to be taken at the next time of the physical quantity is obtained (408). And 300 in FIG.
By performing the processing of (4), the in-state analysis at the next time is performed based on the value (410). As shown in FIG. 3 and FIG. 4, the state at a certain time is obtained by the in-state analysis, and the value at the next time of the changing physical quantity is obtained by the limit analysis based on the obtained state. Furthermore, based on it, the in-state analysis is executed. In this way, the processing of the in-state analysis and the limit analysis are performed recursively. The methods of in-state analysis and limit analysis shown in FIGS. 3 and 4 are qualitative inference systems.
Qupras (Qualitative reasoning based on laws of physics, qualitative reasoning, 8th
Chapter, Kyoritsu Publishing) is equivalent to.

Next, the condition evaluation method will be described with reference to FIG. This is different from the processing of the conventional qualitative inference system because the prev function can be handled in the conditional expression. First, it is checked whether or not the prev function is included in the conditional expression (502). If not, the conditional expression is evaluated using the constraint solver (508) as was done in conventional systems. If the prev function is included in the conditional expression, from the current time identifier,
The identifier of the previous time is obtained, and the dictionary of the corresponding time is retrieved from the history pool 120 using the identifier (50
4). Then, the value of the physical quantity that is the argument of the prev function is extracted from the dictionary, and the value of the prev function is set to that value (506). The conditional expression that includes the prev function is 50
The value obtained in 6 is evaluated using a constraint solver.

Next, the method of processing the consequent part of the rule will be described with reference to FIG. First, it is checked whether the expression of the consequent part includes the prev function (602). If not included, the expression is added to the constraint solver of the constraint solver and evaluated (608). If included, the identifier of the previous time is obtained from the current time identifier, as in the process of FIG.
Using the identifier, the dictionary of the corresponding time is retrieved from the history pool 120 (604). Then, the value of the physical quantity that is the argument of the prev function is extracted from the dictionary, and pr
The value of the ev function is set to that value (606). The expression of the consequent part including the prev function is evaluated by adding the expression to the constraint solver using the prev function as the value obtained in 606 (608).

By using the prev function, as a rule for converting a voltage from digital to analog as described above, if the digital level does not change at 0, the analog voltage is 1.0 volt. It becomes like FIG.
As another example, when the digital level is 1 and does not change and the digital level is 1, the rule for representing that the analog voltage is 1.4 V or higher is as shown in FIG. Also, if the digital level is different from the previous state, that is, if a level transition occurs,
The rule indicating that the analog voltage is 1.0 volt or more and 1.4 volt or less is as shown in FIG.

Even in a rule that cannot be described without referring to the previous time as in these examples, it is possible to describe such a rule by using the prev function that refers to the physical quantity of the previous time. .

[0026]

According to the present invention, in a qualitative inference system that infers a temporal change, it becomes possible to refer to a physical quantity at a previous time in a conditional part and a consequent part of a rule, and to calculate a physical quantity at a previous time. It becomes possible to describe rules that need to be referenced. As a result, qualitative reasoning can be applied to a wider range of physical phenomena.

[Brief description of drawings]

FIG. 1 is a block diagram of an inference system showing an embodiment of the present invention.

FIG. 2 is a configuration diagram of the present invention.

FIG. 3 is a flowchart showing a process of in-state analysis.

FIG. 4 is a flowchart showing a limit analysis process.

FIG. 5 is a flowchart showing a conditional expression evaluation process.

FIG. 6 is a flowchart showing a result expression evaluation process.

FIG. 7 is a rule showing a physical law that a light bulb shines when electricity flows.

FIG. 8 shows a rule when the digital level is 0 and does not change.

FIG. 9 shows a rule when the digital level is 1 and does not change.

FIG. 10 is a rule when a level transition occurs.

 ─────────────────────────────────────────────────── ─── Continued Front Page (56) References ICOT TECHNICAL REP ORT TR-738 (1992-2)

Claims (1)

[Claims] 1. A conditional part and a consequent part are inequalities or mathematical expressions.
Knowledge base that stores rules that contain expressions and constraints that evaluate at least the conditional part of the rules above
The solver, the constraint pool that stores the consequent expression, and the rules stored in the knowledge base are extracted and
Approximately solver evaluates the above condition part, and above results part
If the evaluation of the conditional part is satisfied,
In the state that stores in the constraint pool and performs the in-state analysis processing
The analysis processing means and the evaluation of the rules stored in the above knowledge base are completed.
If so, find a variable that changes over time and
The value that predicts the value at the next time of the number and predicts the value of that variable
, And start the processing of the above in-state analysis at the next time
Limit analysis processing means to apply
Qualitative inference system that qualitatively infers a dynamic system, and at least one of the conditional part and consequent part of the above rules
The time before the time-varying variable in one of the expressions
Write a prev function that takes as an argument the value of the physical quantity of
A history pool that stores the value of the physical quantity at the time is provided, and the in- state analysis unit is configured to reduce the number of condition part consequent parts of the rule.
If at least one of the expressions contains the above prev function,
Check if the above prev function is included.
Get the value of the physical quantity at the previous time from the history pool
And configured to perform the above-mentioned in-state analysis processing.
Characteristic qualitative reasoning system.