JPH0793161A - Inference device - Google Patents

Abstract

PURPOSE:To quickly obtain a better inference result by shortening the inference time and making the correction result reflect in the next inference. CONSTITUTION:The inference device in which an inference result for input data 1 is registered on an inference result file 5, and a correction content 8 can be obtained by correcting the inference result by a correction mechanism 7 when it is judged as an inapropriate one is constituted by providing an inference pattern extracting mechanism 13 which holds a pattern condition 11 equivalent to the input data 1 and registers the inference result in the inference result file 5 as a pattern inference result 12, a correction pattern sampling mechanism 16 which registers by fetching the correction content by the correction mechanism 7 as a pattern correction content, and a pattern condition inference mechanism 14 which registers by fetching the pattern inference result or pattern correction content as the inference result this time on the inference result file 5 when the pattern condition of the input data this time is equal to that of corresponding past input data in the content.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inference device used in a train schedule creation device and the like, and more particularly to an inference device that effectively uses past inferences and eliminates the waste of modification of inference results.

[0002]

2. Description of the Related Art Conventionally, an inference device as shown in FIG. 5 has been used. This device is provided with an inference section 3 having an inference mechanism 2 for deriving an appropriate solution for input data (for example, timetable information) 1 and when data 1 having a certain content is input, based on past knowledge and experience in advance. An appropriate inference condition 4 is extracted from the inference condition 4 which is a knowledge base including set application rules, and is applied to the inference condition 4 extracted for the input data 1 to execute inference, and the inference result is obtained. It is derived and stored in the inference result file 5. The inference result of the inference result file 5 is sent to the correction unit 6 having a display function and displayed on the display unit.

Here, the operator judges whether or not the inference result on the display unit is appropriate. At this time, an inference result such as leaving the train on the 10th line was obtained. Above, 1 when considering the first train of the next day
If the 1st line number is preferable, the inference result is corrected by using the correction mechanism 7 having a keyboard or other input mechanism, and the correction content for the inference result is stored in the correction content file 8.

Therefore, the inference apparatus as described above executes inference from the given input data and the inference condition 4, and when the obtained inference result is unsatisfactory, a person corrects the inference result, I am trying to get it.

At this time, when the inference condition 4 is slightly changed, the inference condition 4 is changed, and the inference is executed again based on the changed inference condition 4, and the inference result is further obtained. If it does not look right, humans are carrying out the same correction process again.

[0006]

Therefore, in the inference apparatus as described above, when the inference condition 4 is slightly changed, it is necessary to restart the inference from the beginning based on the changed inference condition 4. Therefore, there is a problem that the inference time is about the same as the first inference time.

Further, since the inference unit 3 does not grasp how the human has modified the inference result at all, after the second inference due to the change of the inference condition 4, the same inference result is obtained. There is a problem that a correction process is required, it takes time and labor, and a reasonably good reasoning result cannot be obtained for the time.

The present invention has been made in view of the above circumstances, and in addition to obtaining the inference result in a short time by effectively utilizing the past inference result, the correction content of the inference result is sufficiently reflected in the subsequent inference, It is an object of the present invention to provide an inference device that can obtain better inference results quickly.

[0009]

In order to solve the above problems, the invention according to claim 1 registers an inference result obtained by applying an inference condition to input data in an inference result file, When it is determined that is unsuitable, the inference device that corrects by the correction mechanism and obtains the correction content extracts and holds the required pattern condition from the input data and registers it in the inference result file. An inference pattern extraction mechanism that takes in the inference result and registers it as a pattern inference result, and a correction pattern extraction mechanism that takes in the correction contents by the correction mechanism and registers it as the pattern correction contents when the correction mechanism makes a correction. Input data and pattern conditions corresponding to past input data are compared, and when the contents are equal, the pattern inference result or Is a inference apparatus provided with a pattern condition inference engine to be registered in the inference result file as the inference results of this input data takes in the pattern corrections.

[0010]

Therefore, in the invention corresponding to claim 1, by taking the above-described means, the inference conditions are applied to the sequentially input data and the inference results are obtained and registered in the inference result file. However, at this time, the inference pattern extraction mechanism holds necessary input data from the input data as a pattern condition, while registering the inference result obtained by the inference as a pattern inference result. Further, when a correction is added to the inference result, the correction pattern extraction mechanism registers the correction content as the pattern correction content.

Therefore, in the pattern condition inference mechanism, when the same input data as the past input data is input again, when the current input data and the pattern condition obtained from the past input data are compared, Will match.

Therefore, in this case, the already registered pattern inference result is diverted and output as the inference result, or if the inference result is corrected, the pattern correction content is diverted and the inference result is diverted. Since it is output as, the reasoning result can be output in a short time when similar reasoning is performed in the past, the past correction content can be easily reflected in the next reasoning, and a better reasoning result can be obtained. You can

[0013]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the device of the present invention. In the figure, the same parts as those of the conventional device are designated by the same reference numerals, and detailed description thereof will be omitted. Hereinafter, particularly different parts will be described.

First, in addition to the conventional inference mechanism 2, the inference unit 3 takes in the input data 1 and sets the pattern conditions of the same data as the input data 1 to the pattern condition file 11.
The inference pattern extraction mechanism 13 which stores the inference result in the input data from the inference result file 5 and stores it as the pattern inference result in the pattern inference result file 12 while comparing the input data with the pattern condition. And the pattern condition inference mechanism 14 which takes in the past pattern inference result of the pattern inference result file 12 or the past pattern amendment content of the pattern amendment content file described later and stores it in the inference result file 5 as the inference result of this time. Is provided.

Further, in addition to the conventional correction mechanism 7, the correction unit 6 takes in the correction contents of the correction contents file 8 corrected by a human when the correction mechanism 7 gives an instruction to correct the pattern and corrects the pattern. A correction pattern extraction mechanism 16 that stores the contents in the pattern correction content file 15
Is provided.

Therefore, with the above configuration, when the inference is performed by the conventional inference mechanism 2, the inference pattern extraction mechanism 13 outputs the basic data from the input data 1 given at the same time as the conventional inference mechanism 2. If you capture and store it as a pattern condition in the pattern condition file 11,
After that, when the input data 1 is given, the pattern condition inference mechanism 14 compares the input data with the pattern condition to determine whether the current input data and the pattern condition corresponding to the past input data match. Can be easily determined.

At this time, since the past inference result is stored in the pattern inference result file 12, and when the inference result is corrected, the past correction content is stored in the pattern correction content file 15, the pattern If you preferentially capture the pattern modification content out of the modification content and the pattern inference result as the inference result, and if there is no modification, capture the pattern inference result as the inference result, you can significantly reduce the inference time and modify it. The content can be easily reflected.

Next, description will be given of a specific embodiment in which planned reasoning for placing a train in a garage is incorporated. FIG. 2 is a diagram showing a configuration of an inference device and an inference result correction device when a train is detained in a garage. First, as input data 1, for example, train information 1 that trains are assigned on that day
a and the inspection / work information 1b that records the inspection / work that each train must perform on that day are listed.
On the other hand, as the pattern conditions of the pattern condition file 11, a diamond information pattern condition 11a and an inspection / work information pattern condition 11b are created corresponding to the diamond information 1a and the test / work information 1b.

Next, the inference unit 3 is provided with a conventional inference mechanism 2, an inference pattern extraction mechanism 13 and a pattern condition inference mechanism 14 as in FIG. Further, the detention plan result file 5a is a file that stores the inference result of which train in the garage is detained on which train from what time to when, and the pattern detention inference result file 12a.
Is a file that stores the inference result of each train as it is.

Further, the correction section 6 is provided with a conventional correction mechanism 7 and a correction pattern extraction mechanism 16 as in FIG.
Further, the placement correction content file 8a and the pattern placement correction content file 15a respectively correspond to the correction content file 8 and the pattern correction content file 15 shown in FIG.

Next, the operation of the above apparatus will be described with reference to FIG. (1) Regarding inference for the first input data group and modification of the inference result. Now, as input data 1, timetable information 1 as shown in FIG.
It is assumed that a and the inspection / work information 1b are input to the inference unit 3. As a result, since the inference pattern extraction mechanism 13 is the first input data 1, the pattern information file 1a and the inspection / work information 1b are directly used as the pattern information file condition 11a and the inspection / work information pattern condition 11b. Register with.

That is, as the diamond information pattern condition,
The patterns "train 8000 is assigned to the diamond K1", "train 8001 is assigned to the diamond K2", and "train 8002 is assigned to the diamond K3" are registered. In addition, train 800
0, 8001, 8002 diamonds K1, K2, K3, ...
Each contains information such as the track running time zone.

Further, the inference pattern extracting mechanism 13 "train 8000 does not inspect / work" based on the inspection / work information 1b as the inspection / work information pattern condition.
The inspection / work information pattern conditions of “8001 only perform inspection A” and “train 8002 only perform cleaning work” are registered in the pattern condition file 11.

Here, inference conditions 4 are applied to the diamond information 1a and the inspection / work information 1b to execute inference,
After obtaining the detention plan result as shown in FIG. 3, it is registered in the detention plan result file 5a. On the other hand, since the inference pattern extraction mechanism 13 is the first inference, it takes in the detention plan result of the detention plan result file 5a as it is and registers it in the pattern detention inference result file 12a. Therefore, as is apparent from FIG. 3, the detention plan result and the pattern detention inference result have the same information content. In other words, as the result of the pattern detention inference, "the train 8000 is detained on the 10th line from 18:00 to 24:00," as in the detention plan result.
"Train 8001 stays on line 11 from 19:00 to 24:00", "Train 8002 from 20:00 to 2
The pattern "detain on line number 12 until 4:00" is registered in the pattern detention inference result file 12a.

By the way, when the result of the detention plan is sent to the conventional correction mechanism 7 and is displayed and confirmed, when it is determined that the detention plan result 5a is not bad, that is, the train 8001 is judged from the detention plan result. 1 to be detained
When it is determined that the 1st wire number is unfavorable, the wire number is corrected to "21" which is a preferable wire number. As a result, the detention correction content as shown in FIG. 3 is registered in the detention correction content file 8a.

On the other hand, when the conventional correction mechanism 7 makes a correction, an instruction indicating that the correction has been made is issued to the correction pattern extraction mechanism 1.
Sent to 6. Here, the correction pattern extraction mechanism 16
Detention correction contents file 8 based on the instruction of the correction
The contents of the detention correction of the train 8001 are fetched from a and registered in the pattern detention correction file 15a. FIG. 3 shows the contents of the pattern placement correction. (2) Regarding inference for the second input data group and modification of the inference result.

It is assumed that the diagram information 1a and the inspection / work information 1b as shown in FIG. 4 are input to the inference unit 3 as the second input data 1. Here, the second information 1a, 1
Looking at b and the pattern conditions corresponding to the first information 1a and 1b, the train 8000 is exactly the same, and the train 80
01 is exactly the same. The train 8002 had cleaning work in the first time, but no cleaning work in the second time. Therefore, at this time, the inference pattern extraction mechanism 13 has already performed the same train 80 as the first train.
, 00, 8001, 8002, ..., The first time diamond information pattern condition 11a and inspection / work information pattern condition 11b are registered as they are.

On the other hand, when the pattern condition inference mechanism 14 compares the diamond information 1a, the inspection / work information 1b with the diamond information pattern condition 11a and the inspection / work information pattern condition 11b, the train information 8000 is completely the same. , The obtained inference result is also the same. Therefore, as shown in FIG.
The pattern detention inference result regarding 00 is diverted and registered in the detention plan result file 5a as the detention plan result.

Next, regarding the train 8001, the schedule information 1a, the inspection / work information 1b and the schedule information pattern condition 11a and the test / work information pattern condition 11b are the same, but the pattern detention correction content file for the same train. Since the pattern placement correction contents are registered in 15a, the pattern placement correction contents of the train 8001 are fetched and registered in the placement plan result file 5a as shown in FIG.

When the pattern condition inference mechanism 14 registers the pattern detention correction content in the detention plan result file 5a, the deduction is performed by sending a signal to the inference pattern extraction mechanism 13 that the modification content is the detention plan result. The pattern placement correction contents of the plan result file 5a are not registered in the pattern placement inference result file 12a. That is, only the detention plan result without correction is registered in the pattern detention inference result file 12a.

Further, with respect to the train 8002, since the inspection / work information 1b and the inspection / work information pattern condition 11b are different, the pattern detention inference result or the pattern detention correction content cannot be diverted. The conventional reasoning mechanism 2 is notified to that effect. As a result, the conventional inference mechanism 2 is already in the train 800.
Since the detention plan results of 0,8001 are obtained, based on these detention plan results, the detention plan results are derived so as not to overlap each other, and registered in the detention plan result file 5a.

Therefore, according to the configuration of the above embodiment, since the input data of this time and the already registered pattern condition match for the train 8000, the pattern condition inference mechanism 14 already has the pattern condition. Since the pattern detention inference result registered in the detention inference result file 12a is used as it is as the detention plan result, the inference result can be output in a very short time.

Further, for the train 80001, since the pattern placement correction contents already corrected in the past, it can be used as it is as the placement plan result, and the result corrected by humans can be easily reflected in the next inference result. Better reasoning results are obtained quickly.

Although the garage detention of the train has been described in the above embodiment, it goes without saying that the invention is not limited to this field of use. In addition, the present invention can be modified in various ways without departing from the scope of the invention.

[0035]

As described above, according to the present invention, when the same inference is performed in the past, the pattern inference result or the pattern correction content already registered in the file is diverted, so that useless inference can be performed. It can be omitted and the reasoning time can be shortened. Further, when a human corrects the inference result, the pattern correction content of the correction result is diverted, so that it is not necessary to make the same correction again, and a better inference result desired by the human can be quickly obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of an inference device according to the present invention.

FIG. 2 is a configuration diagram of the device of the present invention applied to planned type inference in which a train is detained in a garage.

FIG. 3 is a diagram for explaining an inference result and inference correction for first input data obtained by using the apparatus of FIG.

FIG. 4 is a diagram illustrating an inference result and inference correction with respect to input data for the second time.

FIG. 5 is a block diagram of a conventional inference device.

[Explanation of symbols]

2 ... Conventional inference mechanism, 3 ... Inference unit, 4 ... Inference condition file, 5 ... Inference result file, 6 ... Correction unit, 7 ... Correction mechanism, 8 ... Correction content file, 11 ... Pattern condition file, 12 ... Pattern inference Result file, 13 ... Inference pattern extraction mechanism, 14 ... Pattern condition inference mechanism, 15 ... Pattern correction content file, 16 ... Pattern inference result file.

Claims (1)

[Claims] 1. An inference device that registers an inference result obtained by applying an inference condition to input data in an inference result file and corrects the inference result by a correction mechanism to obtain correction contents when it is judged that the inference result is inappropriate. In the above, the inference pattern extraction mechanism that extracts and holds the necessary pattern condition from the input data and that stores the inference result registered in the inference result file and registers it as the pattern inference result, and the correction mechanism When the correction is performed, the correction pattern extraction mechanism that takes in the correction contents by the correction mechanism and registers it as the pattern correction contents is compared with the pattern conditions corresponding to the current input data and the past input data, and the contents are compared. When they are equal, the pattern inference result or the pattern correction content is taken in and the inference result of the input data this time is acquired. And a pattern condition inference mechanism registered in the inference result file as