JPH0756740A - Monitoring/diagnostic inference device - Google Patents

Abstract

PURPOSE:To inform a user of a result by executing the analysis and diagnostic inference of data expressing a state to be monitored and continuously executing the processing for a long period without losing a real time property. CONSTITUTION:Input data are stored in an element buffer in a shared memory 6 for storing the input data time sequentially as history data without preparing a data base. In the case of storing input data exceeding the upper limit of storage capacity of the element buffer, a noise removing device 2 finds out a representative value for the element buffer and stores the found value in another element buffer so that a newer representative value is stored on an upper position in addition to noise removing processing for removing the uncertainty of data. Consequently a data retrieving time can be shortened. Since the shared memory 6 has said structure, the monitoring/diagnostic inference device capable of executing the processing of data as opened data in real time while storing the data is obtained.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a monitoring / diagnosis inference apparatus for recognizing and predicting the state of a monitoring / diagnosis target using a computer, and uses AI data in real time from an input device. The present invention relates to a method of observing the state of a monitored object such as a device by various inference methods, diagnosing a failure or the like as necessary, processing information at high speed, and drawing a conclusion.

[0002]

2. Description of the Related Art FIG. 9 is a block diagram showing an example of a conventional monitoring / diagnosis inference apparatus disclosed in Japanese Patent Laid-Open No. 4-236938. In FIG. 9, 11 is a patient basic data storage unit that stores basic patient information, 12 is an examination data history storage unit that stores the examination contents of the patient, and examination results,
Reference numeral 10 denotes a condition history data storage unit that stores a history of how to deal with a patient's condition abnormality. Reference numeral 15 is a patient test data registration unit that registers each test data in the test data history storage unit, and 13 is a condition data registration / change test unit that registers / changes the contents of the corresponding processing in the condition history data storage unit 10. Reference numeral 14 is a patient examination unit that searches the patient basic data storage unit 11.

Reference numeral 16 is used when the condition is abnormal by using the AI method based on the input test data and the data from the test data history storage unit 12, the condition history data storage unit 10 and the patient basic data storage unit 12. It is an AI condition determination unit that identifies and determines correspondence. Reference numeral 23 denotes the input inspection data, the inspection data history storage unit 12, the condition history data storage unit 10,
It is a patient information editing unit that edits data from the patient basic data storage unit 11. 27 is an output device (display 25, printer 2) for the data edited by the patient information editing unit 23.
6) Display control section for displaying, 19 is AI condition judging section 16
It is an inspection instruction unit for instructing a necessary inspection execution according to the result output by. Reference numeral 20 denotes an input / output control unit that sends an instruction to each inspection device 22 according to the inspection instruction and sends the data sent by each inspection device 22 to the AI condition determination unit 16. Reference numeral 24 is an input identifying unit for instructing the patient information editing unit 23 and the AI condition determining unit 16.

In the conventional method, a plurality of data storage units such as a condition history storage unit and a patient basic data storage unit are provided for each data having different uses. Therefore, when a certain history is required, it is necessary to first access a storage area such as a database in which the data is stored. Then, the data storage areas are physically separated, and the real-time status of the monitoring target is not stored therein. The current state of the monitoring target was stored based on a certain judgment, and the history was not stored for each data from each sensor. The data has a unique structure in the storage area divided for each purpose, and is stored as one element of the storage area divided for each purpose.

In case-based reasoning, cases are stored and monotonically increasing, and gradually increasing. As a result, many case searches are time-consuming and compared with rule-based inference, it has been considered that it is not suitable for real-time use, but an effective method to improve the efficiency of case searches. There is a method called indexing. In this method, the values that appear in the case group as a value of the attribute item of the case are held as a list, and the case (group) corresponding to each value is indexed. The above attribute items are considered to be elements for determining the causal relationship. For example, if the result of case A occurs,
When the cause of the case A is a ₁ , a ₂ , a ₃ , ..., These a ₁ , a ₂ , a ₃ , ... Are attribute items. When searching for a case close to the current situation, the candidates are narrowed down based on this index in advance, and the similarity is calculated to obtain a case to be obtained. However, even with this method, the efficiency deteriorates rapidly as the number of cases increases.

[0006]

[Problems to be Solved by the Invention] Using historical data,
In the case of recognizing, judging, and predicting the current state, conventionally, when having a database that stores a history for a long period of time for each purpose of use of data and searching for data,
It takes time because it is necessary to retrieve the necessary data from the huge amount of data. Further, even when data is processed, there is a problem in that when there are many objects to be processed, a single process becomes large and lacks real-time responsiveness.

The present invention has been made in order to solve the above problems, and avoids having a storage area such as a database divided for each purpose of use of data, and allows data to be opened and accessed at any time. The purpose is to save the data in the form, speed up the retrieval process, and remove the uncertainties in the data in real time.

In the conventional method, when the number of cases increases,
The time required for execution also increases rapidly, and as a result, it becomes unsatisfactory when there is a restriction on execution time. In this invention,
The purpose of the present invention is to realize a system that guarantees real-time execution in the long term by mitigating this increase in execution time, that is, deterioration in execution efficiency.

[0009]

[Means for Solving the Problems] Monitoring according to the first invention
The shared memory in the diagnostic inference device is divided into a plurality of element buffers, and the results obtained by the input device and the processing are stored in the element buffer in chronological order. Further, the data stored in the element buffer can be passed to another element buffer.

A monitoring / diagnostic reasoning device according to the second invention is
Further, the present invention is characterized in that the element buffer storing the data is provided with a noise removing device for analyzing in real time and removing noise that causes uncertainty.

A monitoring / diagnostic reasoning device according to a third invention is
When the case is stored, the feature amount indicated by a numerical value is calculated from each value of the item group considered to be the cause (or symptom) of the causal relationship. One file is allocated for the calculated value and the case is stored therein. The simplest way is to map the value to the file as a filename. In addition, at the time of search, the corresponding feature amount is calculated from the current situation corresponding to the cause by the same method, the candidate group of the value of the feature amount to be searched is determined, and the corresponding file group is determined. Search for similar cases.

In the monitoring / diagnostic reasoning apparatus according to the fourth aspect of the invention, the upper limit number of cases that can be stored in the file as described in the third aspect of the invention is determined. Reaching this limit,
Further, when storing cases, the maximum total number is protected by selecting and deleting the case that is evaluated as the least effective from the file. As a method of deletion, for example, the current case is stored as the latest case, and the result of similarity search for the latest case is determined to have the highest similarity. Alternatively, the oldest one, the one having the same value, or a method combining these can be appropriately designated.

[0013]

In the monitoring / diagnostic reasoning device according to the first aspect of the present invention, the shared memory is divided into a plurality of element buffers, so that the history is not stored in the database for each purpose of use of data as in the conventional example. Therefore, the time required for the data search processing is reduced, and the real-time property is improved.

The monitoring / diagnosis inference apparatus according to the second aspect of the present invention can improve the accuracy of data retrieval by removing the uncertainty of the data by the noise removing apparatus.

The monitoring / diagnostic reasoning device according to the third aspect of the present invention appropriately classifies the case groups in advance and accumulates the feature values corresponding to the causes of the cases as keys so that only the corresponding files can be searched at the time of retrieval. And As a result, the number of cases to be searched is limited, and it is possible to suppress a decrease in efficiency in case search.

The monitoring / diagnostic inference device according to the fourth aspect of the invention sets an upper limit on the number of cases in each file, and every time one more case is added beyond that, the effectiveness value is the lowest. Cases that are evaluated to be selected are deleted by a predetermined method. As a result, it is possible to increase or maintain the effectiveness of the information that the entire case has. Furthermore, it is guaranteed that the execution efficiency in the case search will not drop below a certain value.

[0017]

【Example】

Example 1. FIG. 1 is a block diagram showing a monitoring / diagnostic inference apparatus. FIG. 2 is a block diagram showing a shared memory. In FIG. 1, reference numeral 1 denotes an input device, which receives data such as a sensor or a file from the outside and transfers it to the inference unit 4. A data management unit 3 manages the input / output of data in the inference unit 4, and the data format (string, number, etc.), type (in
(t, float, etc.) and the time when the data is written in the shared memory 6 are attached. The shared memory 6 is composed of a plurality of element buffers 9 as shown in FIG. Reference numeral 5 is a schedule section. Reference numeral 8 is a knowledge source that describes the contents of processing, and is executed according to the status of the shared memory 6 and the scheduling result of the scheduling unit 5. Reference numeral 2 is a noise eliminator, the details of the function of which will be described in a third embodiment below. 7 is an output device,
This is for notifying the user of the state of the shared memory 6, that is, the state of the monitoring target and the inference result via the data management unit 3.

The schedule section 5 will be described. The scheduling unit 5 schedules the processing written in the knowledge source 8 on the data in each element buffer 9 on the shared memory 6. At this time, depending on the knowledge source to be executed, the rules written in the knowledge source may conflict with the rules, or there may be a plurality of knowledge sources to be executed and the knowledge sources may compete with each other. In that case, in order to eliminate the processing conflict, the scheduling unit 5 performs the scheduling so that the processing with the higher priority is executed.

For example, the following method can be considered as a scheduling method. First, in order to eliminate the conflict between knowledge sources, the scheduling unit 5 determines whether the urgency is high based on the value of the data in the shared memory (whether the urgency is high or low depends on the system. In the case of a system, the one with higher mortality rate is judged to be more urgent). The other is a method in which, in order to eliminate the conflict between the rules in the knowledge source and the rules, the scheduling unit 5 gives priority to the rule having the higher reliability based on the certainty that each rule has conventionally.

Next, the shared memory 6 will be described. Shared memory 6 replaces the prior art history database. The element buffer 9 obtained by subdividing the shared memory 6 has a data name, a data format,
It is possible to store data of the same type having the data value and the time in the same data format up to a predetermined upper limit number. The upper limit number referred to here may be limited by time or may be limited by the number as long as the data can be stored in time series. Data is stored up to the upper limit number, and when the upper limit number is reached, a representative value is calculated based on the history data stored in the element buffer, and the calculated representative value is written in another element buffer 9. Here, the representative value is also written in another element buffer in time series. In this way, past data is converted into a representative value and stored in time series without being limited by the upper limit number of storage in the element buffer.

An example of calculating a representative value and storing the obtained representative value in the element buffer will be described with reference to FIGS. 2 and 3. The element buffer 1 of FIG. 2 stores data from the input sensor, but stores input data input from the current time to one hour before. A representative value is obtained for the data that has been stored for one hour or more, the obtained representative value is stored in the element buffer 2, and the representative value is erased from the element buffer 1. Also, the maximum number of element buffers 1 that can be stored is 10. Element buffer 1 has element buffer 2
A representative value calculated based on the data stored in is stored. The timing at which the representative value is stored is calculated based on the data stored in the element buffer 1 every one hour after the storage process is started, and the calculated representative value is stored in the element buffer 2. However, the maximum number that can be stored in the element buffer 2 is 10.

As shown in FIG. 3A, the start time at which the sensor inputs data to the element buffer is 9:00. Input data D1, D2, D3, ..., D13 are input to the element buffer 1 from the start time 9:00 to the current time 10:00. The maximum storage number of the element buffer 1 is 1.
Since it is 0, the data of D4 to D13 is stored at the current time of 10:00. The current time is 10: 0
Since one hour has passed from the start time when it reaches 0, a representative value (root mean square) is calculated based on the data stored in the element buffer 1, and the calculated representative value is stored in the element buffer 2. .

Next, the input data D14, D15, ..., D21 from the input sensor are stored in the element buffer 1 by the current time 11:00. When the current time reaches 11:00, one hour has passed since the time when the representative value was calculated last time, so the representative value is calculated again based on the data stored in the element buffer 1, and the calculated representative value Is stored in the element buffer 2. Next, the input data D22, D23, ... From the input sensor until the current time becomes 12:00.
When D36 is stored, the maximum number of storages in the element buffer 1 is 10. Therefore, the numbers D27 to D36 are stored. At the time when the current time is 12:00, one hour has passed since the time when the representative value was obtained last time, so the representative value is calculated again and the calculated representative value is stored in the element buffer 2. The above process is repeated to obtain a representative value every one hour, and the obtained representative value is stored in the element buffer 2.

Since up to 10 data can be stored in the element buffer 2, when the number of stored data exceeds 10,
The representative value is calculated again based on the representative value stored in the element buffer 2, and the calculated representative value is stored in the element buffer 3. In this way, the data input from the generated input sensor is stored in the element buffer 1 in time series. Since the representative value is calculated for the data that cannot be stored beyond the upper limit number and the calculated representative value is stored in the element buffer 2, it is possible to store data close to the history of past data. In other words, if you want to refer to the data that occurred within 1 hour before the current time, access the element buffer 1, and refer to the data that occurred 1 hour before the current time (in this case, there is a tendency). If desired, access the element buffer 2.

The procedure of the inference processing using the monitoring / diagnostic inference apparatus configured as described above will be described in the second embodiment below.

Example 2. In this embodiment,
The procedure of the inference processing will be described with reference to the flowchart of FIG. 4 using the monitoring / diagnostic inference apparatus configured as described above.

First, an element buffer is defined on the shared memory 6 (S1). The contents to be defined are the number of element buffers, the storage restriction condition of each element buffer, and the like. Next, the input data from the sensor device is received from the input device 1 (S2). The input data received in S2 is written in the shared memory 6 via the data management unit 3 (S3).
The data written in the shared memory 6 is exchanged with the knowledge source 8 via the data management unit 3 (S4).
At this time, when the data is competingly processed by a plurality of knowledge sources or the data is competingly processed by a plurality of rules in the knowledge source (S5), the method described in the first embodiment is used. Thus, the scheduling section 5 resolves the conflict (S6).

The data in the element buffer 9 is processed according to the processing content designated in the knowledge source 8 (S7). The contents of the process considered here may include writing the data in another element buffer, reading data from another element buffer, and performing comparison operation, arithmetic operation, fuzzy operation, etc. with the data. Finally, the deduced result and the underlying data are taken out from the shared memory 6 via the data management unit 3 and output to the output device 7 such as a file, a display device, or a printer (S8). Input device 1
When a newer input is made (S9), the processing from S2 to S8 is repeated. Inference results can be obtained in this way.

For example, an inference processing procedure will be described by taking a steel plant system as an example. FIG. 5 is a diagram showing the contents of the element buffer in this embodiment. The target temperature, the actual temperature, and the water temperature in FIG. 5 are data input from the sensor, and the learning coefficient of the element buffer 4 is the target temperature, the actual temperature,
Based on the water temperature, the value calculated according to the calculation formula designated by the knowledge source for obtaining the learning coefficient is stored. The calculation formula for obtaining the learning coefficient is as follows. (Actual temperature-Water temperature) / Actual temperature The target temperature is the target temperature in the iron cooling system from 9:00 to 1 minute, and the actual temperature is from 9:00 to 1 minute in the iron cooling system. Actual temperature, water temperature is the temperature of the water in the iron cooling system from 9:00 to 1 minute. The learning coefficient is ideally 1.0, and the value decreases as the deviation from the target temperature increases.

When the target temperature, the actual temperature, and the water temperature are input from the input device, these data are written in the shared memory together with the data generation time via the data management unit 3. After writing to the shared memory, the learning coefficient is obtained from the target temperature, the actual temperature, and the water temperature according to the calculation formula instructed by the knowledge source for obtaining the learning coefficient. When the learning coefficient is calculated according to the equation for calculating the learning coefficient using the target temperature, the actual temperature, and the water temperature generated at each data generation time, the data value as shown in the element buffer 4 of FIG. 5 is calculated. . The obtained data value is the element buffer 4 of the shared memory.
The value of the learning coefficient, the target temperature, the actual temperature, the water temperature, the data generation time, etc. are displayed to the user of this system at the same time as the writing. Since the target temperature, the actual temperature, and the water temperature are input from the input device 1 every one minute, the learning coefficient is calculated each time and the output device 7
More displayed to system users.

At this time, in addition to the knowledge source for obtaining the learning coefficient, it is assumed that there is another knowledge source for obtaining the stratified classification of the actual temperature based on the actual temperature. When the actual temperature is input from the input device 1, the knowledge source for obtaining the learning coefficient and the knowledge source for obtaining the stratified classification of the actual temperature may compete with each other. In this case, the scheduling unit 5 gives priority to a highly urgent knowledge source. For example, if it is determined that the knowledge source of the learning coefficient is more urgent, the processing instructed by the knowledge source for the learning coefficient is executed, and the processing content instructed by the knowledge source for the stratification is learned. It will be processed after determining the coefficients. Based on the learning coefficient thus obtained, the operator of the system can manually adjust the temperature of the part that the automatic cooling system cannot reach.

By configuring the monitoring / diagnostic reasoning device as described above, the shared memory 6 can input / output data in real time, unlike a conventional database. At the same time, when searching for data, the data name (identification name) of the element buffer 9 and the depth (storage position) from the data on the surface of the element can be specified. The depth in this case is considered to be the depth because the element buffer 9 has the time when the data was written or the time when the data was generated as information. For example, when it is desired to refer to the value of the learning coefficient of 9:03, the value at the time of the learning coefficient of 9:03 can be referred to by specifying the data name as the learning coefficient and the depth as 4.

Example 3. In this embodiment,
A noise removal process using the noise removal device of the monitoring / diagnostic reasoning device configured as described above will be described.

When the number of data stored in a certain element buffer 9 reaches the limit number of the element buffer 9, the representative value of the history is successively passed to another element buffer 9 in a carry expression (for detailed processing, Was described in Example 1 with reference to FIG. 3). At the same time, in the first embodiment, the root mean square described with reference to FIG. 3 was used to calculate the representative value. However, this also results in noise removal processing such as filtering, which reduces the uncertainty of the data. You can By appropriately reducing the upper limit number of each element buffer 9, the number of times the representative value is written to the other element buffer 9 increases, so the number of noise removal processes also increases.
Inference can be performed using data with low uncertainty. FIG. 6 shows a representative value, compares the value of the element buffer in which the latest data is stored with the value of the element buffer in which the past data history is stored, and performs a trend analysis. It shows the case of outputting.

The processing procedure of the noise eliminator 2 shown in FIG. 6 will be described. The noise eliminator 2 checks whether or not the number of element buffers for writing processed data has reached the storage limit and compares the trends at a certain timing. An example of this processing will be described below using a flowchart. It is confirmed whether the number of stored element buffers has reached the limit (S1).
1). If it has, the same as in the first embodiment, the root mean square is performed based on the history data stored in the element buffer to obtain a representative value (S12). The obtained representative value is also stored in another element buffer as in the first embodiment (S1).
3). If not reached, the slope value (trend) of the element buffer in which the latest data is written is compared with the slope value (trend) of the element buffer in which the past representative value is stored (S14). Output device 7
It is displayed to the user from (printer or display device) (S15).

As described above, in this embodiment, when the past representative value and the number of stored element buffers exceed the limit,
Data noise is removed by using the root mean square. This is also effective in improving the accuracy of the results obtained by performing the trend analysis. Further, the noise of the data can be removed in real time by performing the calculation of the representative value using the noise removing device.

Example 4. In the third embodiment, the calculation of the representative value is combined with the noise removing apparatus 2 to remove the noise of the data by the root mean square to improve the accuracy of the representative value, that is,
I realized the improvement of the trend comparison result, but another method,
For example, instead of the root mean square, the standard deviation may be used to calculate the representative value, that is, the data noise may be removed.

Example 5. In this embodiment, a memory of an accident to be monitored such as a plant is stored in a file as an example,
The feature is that the file name of the file is determined based on the feature amount of the data that causes the case.

When the memory of an accident to be monitored such as a plant is stored in a file as an example, in the part corresponding to the cause (or sign) of the causal relationship of the accident, the sign preceding the accident, that is, the value indicated by a specific sensor or Memorize fluctuations. In the portion corresponding to the result, the state of the failure that has occurred thereafter, the coping method at that time, the result, the severity, and the like are stored. Fault memories need to be stored for several to several decades, and in order to match with the current state, similar searches are performed at regular intervals to see if there are any similar signs of past faults. I need to find it.

Therefore, the characteristic amount is calculated based on the level of the value of the specific sensor, the variation amount in a predetermined time, and the like, and the corresponding case file name is determined and stored. If the number of cases stored in this file has reached the maximum value, the data showing the highest similarity to the case to be stored this time, if it exceeds the predetermined similarity value determined by the system in advance,
Deem this case as a duplicate case and delete it. Alternatively, the cases are checked in the oldest order, and if there is a case where the case is lower than a predetermined severity, the case is deleted as information to be forgotten.

FIG. 7 is a flow chart showing the procedure of case storage in this embodiment. The procedure of storing a case will be described below according to the flowchart of FIG.
First, the feature amount is calculated based on the above-described sensor value level and the variation amount in a predetermined time, and the feature amount is used as a file name (S21). The file name here is a feature amount, and is therefore represented by a numerical value. It is confirmed whether the determined file name already exists and the number of cases stored in the file has reached the upper limit (S22). In the case of the condition of S22, the case less effective than the file is deleted (S23).

To judge whether the validity is low or high, as described above, the severity levels are checked in ascending order of the storage dates of the cases stored in the file, and the severity levels are determined to be lower than the severity levels set in advance by the system. If there is one, delete this case.
The severity here is the severity of the accident, and can be calculated by a certain calculation formula based on the case data stored in the file. As another determination method, a case having the case data closest to the case data to be stored this time is searched from the file, the similarity is calculated, and the calculated similarity exceeds the predetermined similarity. In that case, the retrieved case is deleted from the file. The degree of similarity here is assumed to be obtained by a certain calculation formula based on the data stored in the file as the case data.

After deleting the case, the case is written to the file (S24). When writing a case, the data used as the basis for calculating the feature amount is the same in the file, but the values of other data are different, so an index is provided to distinguish these data. The method of determining the index is arbitrary depending on the system, but for example, if the date of occurrence of the case data (including the time) is unique in the file. Although the case data to which the index is added is written to the file, it is possible that all the same data has already been written to the file, and having the same data in duplicate deteriorates the usage efficiency of the file storage capacity. Therefore, the maximum number of duplicated data in the system is determined in advance, and when the same data is written exceeding the upper limit, the oldest data is deleted (this is called index maintenance) to improve the file usage efficiency ( S2
5).

Next, the processing procedure of case retrieval will be described with reference to the flowchart of FIG. For example, when judging the current status of a plant and investigating whether an accident has occurred or whether there are any signs of an accident, first enter the current status as data, and then enter the data in the entered data. The feature amount is calculated from certain specific data, and the file name is determined (S31). When the file name is determined and the file to be searched is determined, an index is selected to narrow down the search target (S32). The most effective selection can be made by selecting the index based on the input data. This choice depends on how the system found the index when storing the case data in a file. As described above, if the index is determined according to the date of occurrence of the case data, there is no relation between the data indicating the current situation and the index, and it becomes difficult to select the index.

For example, assume that there are five items of input data from A to E. It is assumed that the five items are classified according to the range of values. For example, the item A is 10
If the value of the item of B is between 30 and 50, it is classified as Category 1 if the range is from 20 to 20. Item A
Assuming that the index of this data is 12, for example, using the division determined by the value of item B and the value of item B, when the index to be searched is narrowed down, the index having at least 10 digits is 1 and the 1s digit is 2. If the index is the search target, it is guaranteed that similar data is stored, and the search efficiency can be improved.

The data stored in the selected index is sequentially referred to, and a case similarity search is performed (S33).
Check if there are cases with the same data (S3
4). If the same case exists, diagnostic inference processing is performed based on the obtained case data (S36). The diagnostic method is to execute the processing content written in the knowledge source 8. The processing content written in the knowledge source 8 can be arbitrarily set by the system.

When the same case does not exist, the case having the case data most similar to the data to be searched is searched. Items that can be linearly approximated by similar case data are approximately interpolated, and the data is processed so as to match the data to be searched (S35). Based on the processed data, the diagnostic reasoning process similar to S36 is performed.

As described above, in this embodiment, the event data is stored by using the feature amount corresponding to the cause of the causal relation between the events as the key of the file, and the upper limit is given to the number of the stored event data and the upper limit is set. When it is reached, the validity is confirmed by the prescribed method. By deleting the case data determined to be low in effectiveness from the file, the case data can be searched quickly and efficiently. In addition, it is possible to perform diagnostic inference processing in real time using the retrieved result.

[0049]

According to the present invention, the following effects can be obtained.

According to the first aspect of the present invention, the data is written in the element buffer on the shared memory in time series while vertically shifting from upper to lower. For the search, the element buffer identification name and the absolute position from the upper level of the element buffer are specified. Therefore, since complicated processing is not used for data retrieval, there is an effect that data retrieval can be performed at high speed. Also, because the element buffer that stores the latest data of history data and the past history data are stored in different element buffers, when comparing the trend of recent data and the tendency of data that is distant in time, each data Processing such as trend analysis can be performed only by searching and analyzing only the element buffer that stores the. In addition, since the data stored as history is stored in appropriately divided element buffers and can be handled openly, when analyzing the data, the data stored in each element buffer is the analysis target, The number of data used for calculation is small and the processing speed can be increased.

According to the second aspect of the present invention, since the noise removing device is used to remove the noise that causes the uncertainty of the data, the accuracy of the value obtained as the analysis result can be improved when the trend analysis is performed. It has the effect of increasing the price.

According to the third aspect of the invention, since the file name for storing the case is obtained from the data corresponding to the cause of the causal relationship of the case to be stored, no complicated processing procedure is required when storing the case data in the file. become. Further, when searching for similar case data, the file name can be determined based on the case data, so that the number of cases to be searched can be limited, and the similar case data can be easily searched.

According to the fourth aspect of the invention, when the case data is stored in the file, the case data having a low validity is deleted from the file, so that the similar data can be searched at high speed. Further, since the deterioration of the validity of the data retrieved as the similar data can be prevented, the accuracy of the obtained results will be improved if the diagnostic inference investigation is performed based on this data.

[Brief description of drawings]

FIG. 1 is a block diagram showing a monitoring / diagnosis inference apparatus according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a monitoring / diagnosis inference apparatus according to an embodiment of the present invention, particularly a shared memory.

FIG. 3 is a graph showing a representative value obtained in the first embodiment of the present invention.
It is a figure which shows the timing which stores the calculated representative value in another element buffer.

FIG. 4 is a flowchart showing a procedure of inference processing according to the second embodiment of the present invention.

FIG. 5: Element buffers 1 to 2 in Embodiment 2 of the present invention
It is a figure which shows the data content of No. 4.

FIG. 6 is a flow chart diagram showing a procedure of processing by a noise removal device in embodiment 3 of the present invention.

FIG. 7 is a flowchart showing a procedure of storing a case according to the fifth embodiment of the present invention.

FIG. 8 is a flow chart diagram showing a procedure of a case search according to the fifth embodiment of the present invention.

FIG. 9 is a block diagram showing a conventional monitoring / diagnosis inference apparatus using historical data.

[Explanation of symbols]

 1 Input Device 2 Noise Removal Device 3 Data Management Unit 4 Inference Unit 5 Scheduling Unit 6 Shared Memory 7 Output Device 8 Knowledge Source 9 Element Buffer 10 Condition History Data Storage Unit 11 Patient Basic Data Storage Unit 12 Examination Data History Storage Unit 13 Condition Data Registration / change inspection unit 14 Patient inspection unit 15 Patient inspection data registration unit 16 AI condition determination unit 17 Knowledge base unit 18 Data storage unit 19 Inspection instruction unit 20 Input / output control unit 21 Data analysis device 22 Inspection equipment 23 Patient information editing unit 24 Input identification unit 25 Display 26 Printer 27 Display control unit 28 Keyboard

Claims (4)

[Claims] 1. An input device for taking in data to be monitored / diagnosed, a data management unit for managing input / output of data,
A shared memory in which the data used for prediction and recognition of the state of the monitoring target managed by the data management unit is written and can be stored as a history, a knowledge source that executes inference, and the order of processing by the knowledge source is determined. In a monitoring / diagnostic inference device that includes a schedule unit that outputs a result derived by inference / processing and that recognizes and predicts the state of a monitoring / diagnostic target, the shared memory is divided into a plurality of element buffers, A monitoring / diagnostic inference apparatus characterized by setting an upper limit number of data that can be stored in each element buffer and storing the history of data in each element buffer by temporally dividing it. 2. The monitoring / diagnostic inference apparatus further includes a noise removing apparatus that analyzes in real time an element buffer in which data is stored, and executes a process of removing noise that causes uncertainty in real time. The monitoring / diagnostic inference apparatus according to claim 1, further comprising: 3. The monitoring / diagnostic inference device selects a cause and an actual result that are causal relationships between events from data input from an input device, obtains a feature amount corresponding to the cause of the causal relation, and obtains the feature amount. A monitoring / diagnostic inference apparatus, which is characterized by generating a file that stores a result using a feature amount as a key. 4. The monitoring / diagnostic inference device puts an upper limit on the number of achievements stored in the generated file, and when the upper limit is reached, deletes the achievement selected as the least effective by a predetermined method. Monitoring according to claim 3, characterized in that
Diagnostic reasoning device.