JP2822673B2 - Real-time monitoring expert system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a real-time monitoring type expert system for a system having an internal state, and can be applied to, for example, trouble repair of mechanical equipment or diagnosis of a patient in a hospital.

[0002]

2. Description of the Related Art Conventionally, in a monitoring-type expert system, an inference unit drives a procedure for examining an external environment at predetermined time intervals, performs parameter analysis of obtained fault information, and estimates a state of a target system. . In addition, in the inference method for performing state management, the target system, that is, the electronic exchange, the power receiving and distributing equipment, the medical condition of an emergency patient in a hospital, and the like are regarded as a finite state machine in which the state transition condition does not include a time constraint, so that the event The state transition of the target system is described in such a manner that a state that is premised to cause the error is deleted from the knowledge base, and the resulting state is declared to the knowledge base.

For this reason, the knowledge base holds only the system state at the latest point in time, and does not hold the state change history. Also, once a state was created, it stayed in the knowledge base unless an event to update the state occurred.

[0004] When explicitly describing a time constraint associated with a state transition, a time logic for strictly managing the duration of the state and the time of occurrence of the state transition and its application system have been proposed. In such a time logic application system, a large storage capacity is required because time information is added to the states and state transitions of all the constituent elements of the target system.

[0005] For this reason, the use form in which the state estimation is performed while searching the history information off-line is predominant, and is not suitable for real-time processing which requires completion of the state estimation within a fixed time.

[0006]

In the system configuration of the prior art, a real-time dispatcher function is realized by the inference unit.

However, since the inference unit is originally described in a high-level language for knowledge description, it is not suitable for low-level processing such as asynchronous input / output, and processing efficiency as a dispatcher is low. For this reason, the ability of the target system to follow the state change is weak, and when the state change frequently occurs, the state of the target system cannot be monitored. Also,
The inference unit sometimes operated even when there was no change in the state of the target system, and the load on the CPU was high.

Further, in many of the inference methods in the prior art described above, time relations are not explicitly handled, and when the system reset occurs N times or more in M minutes, the state becomes "...". Within a period of N minutes from now on "cannot be properly expressed. Even a system that explicitly deals with time relations requires a large storage capacity because time information is added to the states and state transitions of all components of the target system.

In the above-described inference method in the prior art, events such as "command input by an operator from an external terminal that does not monitor messages" and "offline processing such as manual work" cannot be detected as messages. For this reason, once some states are declared in the knowledge base, they remain in the knowledge base until they are manually deleted, causing a difference from the state of the target system.

The present invention has been made to solve the above-mentioned problems of the prior art, and provides a system configuration for realizing a high-speed follow-up function in monitoring the state of a target system, including time constraints. Calculation means for handling the state transition of the target system and storage management means for continuing to monitor the state of the target system while keeping the knowledge base storage usage constant considering the unobservable events generated by offline processing etc. The purpose is to provide.

[0011]

According to the present invention, in a real-time monitoring type expert system for inferring in real time the state of a target system composed of a plurality of devices, the device notifies the change of the device state. A message processing unit for format analysis of an event to be performed, a command processing unit for format analysis of an event for controlling the device, and an internal representation of a knowledge base from the event format-analyzed by the message processing unit and the command processing unit. An event monitoring / generating unit to be generated; a unit for inferring the device state and the situation based on the internal expression generated by the event monitoring / generating unit; and deleting the event exceeding a time range handled by the knowledge base. Event deletion rule means to perform, and when the time range handled by the knowledge base is exceeded An apparatus state recovery rule unit for returning to the apparatus state before the state change, a state processing unit having a state deletion rule unit for deleting the state beyond a time range handled by the knowledge base, and the state processing unit. A measure search unit for searching for a measure procedure for avoiding a failure from the inferred device state and the situation, and an output unit for outputting the measure procedure searched by the measure search unit; A command according to the measure procedure output from the real-time controller configured to control the device via the command processing unit and the event monitor generation unit and to be notified to the measure search unit. It is a monitoring expert system.

Furthermore, the status processing unit, when the time t ₁ to the situation G is no longer satisfied exists in the time range Tmax, said from the knowledge base, from the time t ₀ when the status G is satisfied all events that caused the individual device status defining the status G up to the time t _1, the deleted by the event delete rule means, said status G is configured to be deleted in the status delete rule means Is preferred. This can prevent a remarkable increase in the amount of information stored in the knowledge base.

[0013] Further, the status processing unit may be configured so that the event generating the device state exceeds a time range handled by the knowledge base, and transitions from another device state to the device state within the time range. It is preferable that the device state is returned to the device state before the state transition when no event has occurred. As a result, when an event that is out of the time range is deleted from the history of event occurrence time, the device state caused by the unobserved event is restored to the previous state, so that the knowledge base does not match the actual state. Can be prevented from increasing the mismatch between the knowledge base and the actual state in continuous monitoring for a long time.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS First, modeling of a device state of a target system as a finite state machine will be described.

The real-time monitoring operation of the target system is a dialog between the system and a monitor via a console. The message from the target system output to the console is "event notifying the change of the device status of the target system", the command from the monitor input to the console is "event to change the device status of the target system", the target system Is considered as "the intersection of the device states of all the devices constituting the target system", the target system can be modeled as a finite state machine that performs a state transition by these events.

In order to model the target system, for each device constituting the target system, all events that change the device state are defined as state transitions of the finite state machine.

The concept used in modeling the target system is defined as follows.

The "status" is composed of a device status, a status, and a system status. -"Device status" refers to the operating status of each device constituting the target system. "Situation" refers to a composite state defined as an intersection of the states of a plurality of devices. "System status" refers to the intersection of the device statuses of all devices constituting the target system.

An "event" is composed of an operation, a message, and a measure. "Operation" refers to a basic event that cannot be resolved anymore, such as a command from a supervisor input to the console. "Message" refers to an event that is autonomously output from the target system to the console and notifies a change in the device state of the target system. -"Measure" refers to a sequence of operations and messages that is significant in the operation management of the target system.

Next, the "state" and the "event"
A method for inferring the device state based on the above will be described.

First, an inference rule for obtaining a temporal sequence relationship among device status, status, and system status from the history of event occurrence times will be specifically described. For an event e that causes a state transition of the device state, a predicate occur (e, t) to which an occurrence time t is added is managed in a knowledge base. It is o that event e occurs at time t2.
ccur (e, t2). Event e is state p
Af is an event that causes 1 to transition to state p2.
It is expressed by a predicate of fact (e, p1, p2). oc
cur (e, t) declares to the knowledge base every time event e occurs at time t.

As for the device state, a predicate from which the time information is not added is managed in a knowledge base. The device state p is expressed by a predicate active (p), and the device state p is called active. The time information is not attached to the device state, and only whether or not it is currently active is declared to the knowledge base.

The inference of the change in the device state is a forward inference triggered by an event composed of an operation, a message, and a measure, and is defined by the device state transition rule of Expression 1.

Equation 1 Device state transition rule

[1] [[active (p ₁ ) Λoccur]
(E, t) Λ affect ( e, p 1, p 2) Λ n
ot (active (p ₂ ))] → [remove (a
active (p ₁ )) Λ assert (active
(P ₂ ))]], where t is the actual time when this rule fires, remove (p) is an operation to remove p from the knowledge base, and assert (p) is an operation to declare p in the knowledge base.

Next, a situation inference method will be described.

When the device state p ₁ holds in the time interval I ₁ and the device state P ₂ holds in the time interval I ₂ , the device state p ₁
Status G defined from the device state _{p 2} and when the interval _{(I 1}
{I ₂ ). The status is updated by recalculating the status when the device status changes. Tru is an algorithm for such state management.
the Maintainance System (Doy
le, J .; , '' Atruth maintenanc
e system '', Artificial Int
alliance 12, 1979, 231-27
2) is known.

As to the situation G, the information of the start time Ts and the end time Te is represented by start (G, Ts) and end.
It is managed in a knowledge base by a predicate (G, Te).
In the real-time monitoring of the target system, since the history of the fault status is significant, the start and end of the status defined from the active device status are declared to the knowledge base by the following procedure.
In such a method, the individual device states at any time t are stored in the knowledge base in the stored ｛occur
(E, t ') \ start (G, t'') \ end
(G, t ′ ″)}.

If the situation G is the device state p ₁ , p ₂ ,. . . , P
_n is defined as the intersection of _n , TMS (True Maintance) that creates and deletes the situation G
Examples of the (System) rule are shown in Expressions 2 and 3, respectively. The rule of Equation 2 is that p ₁ , p ₂ ,. . . , _Pn are activated at time t. The rule of Equation 3 is that if the situation G that occurred at the past time t ₀ is active, the device states p ₁ , p ₂ ,. . . ,
Indicates that condition G ends at time t when any of _pn becomes inactive.

Equation 2 Situation Start Rule from Device Status

(Equation 2)

Equation 3 Situation termination rule from device status

(Equation 3) Here, t ₀ is an arbitrary past time, and t is an actual time when this rule fires.

Next, a description will be given of a means for automatically eliminating a past apparatus state.

When the above-described state inference is applied to a long-term system state monitoring operation, states that can only be removed manually remain in the knowledge base, and the available storage capacity of the knowledge base is proportional to time. And decrease. In order to keep this constant, the width of time inference handled in the knowledge base is set to T
Max is determined, and knowledge about the time range from the current time t to (t-Tmax) is handled, and memories relating to past system states and events outside this range are automatically excluded from the knowledge base. The deletion procedure will be described below.

The rule of Equation 4 is that the occurrence time t of event e
₀ indicates that when the time range handled by the knowledge base is exceeded, it is deleted from the knowledge base. The rule of Equation 5 is that the device state p generated at the past time t ₀ is active, the time t ₀ exceeds the time range handled by the knowledge base, and the event e occurs at the time t ₀ , and when the time t ₀ after the cause transition from the apparatus state p ₁ to the device state p ₂ event e is not generated, returning the device state to p _1. The rule of Equation 6 indicates that the situation G that has ended before the time range handled by the knowledge base is deleted.

The time t ₀ situation G, which was started in is, in a state that does not end for unobservable events, even in the case it exceeds the time range to be handled in the knowledge base, following the logical unwanted situation G It can be deleted by the procedure of.
First, since the apparatus state is restored by the apparatus state recovery rule of Equation 5, the start condition of the situation G on the left side of Equation 2 is not satisfied, and the end of the situation G is declared by the rule of Equation 3. Here, the situation G is deleted from the knowledge base by applying the rule of Expression 6.

Equation 4 Rule for Deleting Events Exceeding Time Range Handled by Knowledge Base

[Number 4] _{[occur (e, t 0)} Λ ((t-t 0)>
Tmax) → remove (occur (e,
t ₀ ))]

Equation 5 Device Status Recovery Rule

[Number 5] _{[occur (e, t 0)} Λ ((t-t 0)>
Tmax) Λactive (p ₂ ) Λeffect (e, p ₁ , p ₂ ) Λnot (ac
five (p ₁ )) Λnot (existst
₁ [effect (e, p ₁ , p ₂ )} occur
(E, t ₁ ) Λ (t ₁ > t ₀ )]) → remove (a
active (p ₂ )) Λ assert (active
(P ₁ ))]

Expression 6: Rule for deleting a situation beyond the time range handled by the knowledge base

[Start (G, t ₀ ) Λend (G, t
₁ ) Λ ((t−t ₁ )> Tmax) → remove (s
start (G, t ₀ )) Λ remove (end
(G, t ₁ ))], where t ₀ and t ₁ are arbitrary past times, and t is an actual time when this rule fires.

[0038]

FIG. 1 shows an embodiment of an operation management support expert system for an electronic exchange based on the configuration of the expert system of the present invention.

In FIG. 1, 1 is an exchange, 2 is a message input / output device for transmitting and receiving data between the exchange and the expert system, 3 is a message processing unit for packetizing messages from the exchange, and 4 is a message processing unit and device. An event monitor generation unit that monitors the timing of event transmission and reception between the state inference unit and the command processing unit and creates an internal expression for inference, and 5 also has an internal expression of an event declared in the knowledge base by the event monitor generation unit. A situation processing unit for inferring a device state or a situation caused by an event; 6 a measure search unit for searching for a measure procedure for avoiding a failure in the situation generated by the situation processing unit;
7 is a device state inference unit composed of a situation processing unit and a measure search unit, 8 is a screen output device that receives display information from the event monitor generation unit, the situation processing unit, and the measure search unit and outputs a screen, and 9 is an operator Is a command processing unit for packetizing a command input by the operator, 10 is a command input device for the operator to input a command, and 11 is an operator.

The expert system is located between the exchange 3 and the operator 10 and monitors a message from the exchange 3 and a command from the operator 10.

The command processing unit 9 and the message processing unit 3 are modules for format-analyzing text data and packetizing the results, and include a syntax analysis unit and a lexical analysis unit.

FIG. 2 shows a configuration diagram of the message processing unit.
In FIG. 2, reference numeral 21 denotes a syntax analysis unit, 22 denotes a lexical analysis unit, and 23 denotes a message processing unit. The message processing unit 23 receives a message from the exchange via the message input / output device, packetizes the message, and sends it to the event monitoring and generation unit.

FIG. 3 shows a configuration diagram of the command processing unit. In FIG. 3, reference numeral 31 denotes a syntax analysis unit, 32 denotes a lexical analysis unit, and 33 denotes a processing unit. The command processing unit 9 receives a command from the operator 10 via the command input device, packetizes the command, and sends the packet to the event monitoring generation unit 4.

The event monitoring / generating unit 4 monitors packets sent from the message processing unit 3 and the command processing unit 9, and upon receiving the packets, performs the processing in the following order. (1) Convert a packet into an internal representation as a "message" or "operation" and register it in the knowledge base.
At this point, an internal representation occur (e, t) for event e observed at time t is generated and time updated in the knowledge base. (2) Analyze the meaning of the event parameters and send the result to the screen output device

The status processing unit of the device state inference unit performs processing in the following order. (1) Calculate a change in device state caused by an event according to a state transition rule. (2) A change in the situation caused by the update of the device state is determined by the Truth Maintenance System.
Recalculate with m. (3) By updating the time in the knowledge base in the event monitoring and generation unit, an event or a situation outside the time range handled by the knowledge base is searched and deleted from the knowledge base.

The action search section 6 of the apparatus status inference section 7 searches for an action procedure to be applied using the active apparatus status and status calculated by the status processing section 5 as search keys, and displays the search results on a screen output device. Send to The procedure of the switch operation management is stored in the knowledge base as the conclusion part of the rule that uses the equipment status and status as a condition part.

[0047]

According to the real-time monitoring type expert system of the present invention, a distributed processing configuration of the monitoring function of the event monitoring and generation unit and the state estimation function of the device state inference unit is adopted. Only the information on the occurrence time needs to be notified to the device state estimating unit, and the load is light.
A high-speed follow-up function can be realized even for a target system whose status changes frequently.

Further, the device state inference unit of the present invention handles the state transition of the target system including the time constraint, so that the state consisting of the device state / situation / system state is
By adopting a knowledge expression based on events consisting of operations, messages, and measures, and providing an inference method that generates time constraints between other state descriptions from only history information that records the event occurrence time, Expert systems can handle knowledge that includes time constraints that are difficult to handle.

Further, when the target system is logically modeled as a finite state machine, the problem of the accuracy of the model itself for simulating the target system and the problem of the heuristics of operation management are clearly separated. By doing, the knowledge about the target system and the operation management, wasteful state by general rules,
Detecting and deleting redundant and inconsistent states facilitates debugging of the knowledge base.

Further, the apparatus state inference unit of the present invention recovers the apparatus state caused by an unobserved event to the previous state when deleting an event out of the time range from the history of event occurrence items. By doing so, the inconsistency between the knowledge base and the actual state is eliminated, so that it is possible to prevent an increase in the inconsistency between the knowledge base and the actual state in continuous monitoring for a long time.

[Brief description of the drawings]

FIG. 1 is an embodiment of an “operation management support expert system for an electronic exchange” based on a method of configuring an expert system according to the present invention.

FIG. 2 is a configuration diagram of a message processing unit of the operation management support expert system of the electronic exchange shown in FIG. 1;

FIG. 3 is a configuration diagram of a command processing unit of the operation management support expert system of the electronic exchange shown in FIG. 1;

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 switch 2 message input / output device 3 message processing unit 4 event monitoring / generation unit 5 status processing unit 6 measure search unit 7 device state inference unit 8 screen output device 9 command processing unit 10 command input device 11 operator

Continuation of front page (58) Field surveyed (Int.Cl. ⁶ , DB name) G06F 9/44

Claims (3)

(57) [Claims] 1. A real-time monitoring expert system for inferring a state of a target system composed of a plurality of devices in real time, a message processing unit for format-analyzing an event in which the device notifies a change in the device state. A command processing unit that format-analyzes an event for controlling the device; an event monitoring generation unit that generates an internal representation of a knowledge base from the event that is format-analyzed by the message processing unit and the command processing unit; A means for inferring the device state or situation based on the internal expression generated by a monitoring generation unit; an event deletion rule means for deleting the event exceeding a time range handled by the knowledge base; and the knowledge base. Return to the device state before the state change when the time range A status processing unit having a status deletion rule unit for deleting the status exceeding the time range handled by the knowledge base; and a device status and a status inferred by the status processing unit. It has a measure search unit for searching for measure procedures for avoiding obstacles, and an output unit for outputting the measure procedure searched by the measure search unit, according to the measure procedure output from the output unit. A real-time monitoring type expert system, wherein the command is controlled so as to control the device via the command processing unit and the event monitoring generation unit, and to be notified to the measure searching unit. . Wherein said status processing unit, when the time t ₁ to the situation G is no longer satisfied exists in the time range Tmax, said from the knowledge base, from the time t ₀ when the status G is satisfied all events that caused the individual device status defining the status G up to the time t _1, the deleted by the event delete rule means, said status G is configured to be deleted in the status delete rule means 2. The real-time monitoring type expert system according to claim 1, wherein: 3. The status processing unit according to claim 1, wherein the event generating the device state exceeds a time range handled by the knowledge base, and transitions from another device state to the device state within the time range. If no event has occurred,
The real-time monitoring type expert system according to claim 1, wherein the device state is configured to return to the device state before the state transition.