KR20210051719A - method for preprocessing Sensor MEASUREMENTS and MACHINE LEARNING METHOD using the same - Google Patents

Abstract

The present invention provides a method for preprocessing sensor measurements and a machine learning method using the same, which can efficiently train a model to be trained. According to several embodiments of the present invention, the method for preprocessing sensor measurements comprises: a step of obtaining measurement time-series data generated by a plurality of adjacent sensors; a step of dividing the time-series data into a plurality of detection time intervals based on non-generation of alarms of all of the plurality of adjacent sensors in the time-series data; a step of merging a first detection time interval and the second detection time interval to form an integrated time interval; a step of using arrangement coordinates and measurements of three key sensors selected in the integrated time intervals and the detection time intervals to determine the problem position; and a step of generating a sequence reflecting results of sequentially connecting the time-series data of the integrated time intervals or the detection time interval with the same problem position and matching the three key sensors.

Description

Method for preprocessing sensor MEASUREMENTS and MACHINE LEARNING METHOD using the same}

The present invention relates to a sensor measurement value preprocessing method and a machine learning method using the same. In more detail, the present invention relates to a method of preprocessing measured value data generated by each of the homogeneous sensors disposed adjacent to each other in an efficient manner for machine learning and a machine learning method using the same.

Conventionally, a sensor used in a process has a high performance and a highly sensitive device for precise detection. However, as the sensitivity of the sensor is high, problems such as an alarm generated even with a too fine detection or easily operated even with noise have occurred. Even in this case, it was impossible to lower the sensitivity of the sensor in order to detect an accident or unexpected problem. Therefore, in the related art, a sensor with high sensitivity is used, but when an unnecessary alarm occurs, there has been a hassle of having to check it again.

Therefore, without lowering the sensitivity of the sensor, whether the alarm generated by the sensor is an alarm caused by noise, or an alarm that is not noise but can be ignored, analyzes the measured value measured by the sensor more accurately. Technology is in demand. Accordingly, an attempt was made to develop an analysis method through machine learning.

In the conventional machine learning method using sensor measurement values, machine learning is performed by learning raw data measured by a sensor as it is or through a preprocessing step of simply removing noise. However, since the raw data measured by the sensor is scattered sporadically or it is not possible to know the data according to any situation, machine learning is performed using these data, and the generated model has numerous errors or errors in the learning stage. The efficiency of learning is low due to unrefined raw data, such as the need to change the algorithm of the model for correction, which incurs time and cost to improve the algorithm.

Accordingly, in order to solve the above-described problems, there is a need for a more efficient and refined data preprocessing method to generate a model capable of accurately analyzing data measured by a sensor.

Unexamined Patent Publication No. 10-2019-0089776 (published on July 31, 2019)

The technical problem to be solved through an embodiment of the present invention is a sensor measurement value preprocessing method in which a model to be learned can be more efficiently learned by preprocessing data measured by a sensor to be suitable for machine learning, and a machine learning method using the same. Is to provide.

Another technical problem to be solved through an embodiment of the present invention is to provide a method for preprocessing measured value data generated by each of the homogeneous sensors disposed adjacent to each other in an efficient manner for machine learning, and a machine learning method using the same. .

Another technical problem to be solved through another embodiment of the present invention is a sensor measurement value that can further improve the performance of machine learning by processing information on events generated by the same or similar causes into standardized data. It is to provide a preprocessing method and a machine learning method using the same.

Another technical problem to be solved through another embodiment of the present invention is a sensor measurement value preprocessing method that can more easily determine a location where a problem occurs by precisely analyzing a sensitively operated homogeneous sensor, and a method using the same. It provides a machine learning method.

The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems that are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above technical problem, a sensor measurement value preprocessing method according to an embodiment of the present invention includes the steps of obtaining measurement value time series data generated by each of a plurality of adjacent sensors, and of all of the plurality of adjacent sensors from the time series data. Dividing the time series data into a plurality of detection time intervals based on no alarm occurrence, selecting at least some of the plurality of adjacent sensors as important sensors for each detection time interval, a first detection time interval, and the first detection time interval When the important sensors of each of the first detection time period and the adjacent second detection time period coincide, the step of forming an integrated time period by merging the first detection time period and the second detection time period, the detection time period and the Determining the location of the problem for each integration time section, but determining the location of the problem using the arrangement coordinates and measurement values of each of the three core sensors selected in the detection time section and the integration time section, and the number of the important sensors And generating a sequence reflecting a result of sequentially connecting time series data of the detection time section or the time series data of the integrated time section in which the problem locations overlap with each other by a reference value or more.

In an embodiment, when the detection time of each of the plurality of adjacent sensors is greater than or equal to a reference detection number, selecting the important sensor.

In one embodiment, the step of configuring the integrated time interval comprises, when the time interval in which the alarm does not occur between the first detection time interval and the second detection time interval is less than an integrated reference value, the first detection time interval and the It may include determining that the second detection time interval is adjacent.

In an embodiment, the determining of the problem location may include selecting the upper three sensors of the measured values from the plurality of adjacent sensors as the core sensors.

In an embodiment, generating the sequence may include processing the time series data from the detection time interval or the integrated time interval having the same problem location into a time interval of the same length.

In the method for learning a problem location determination model according to an embodiment of the present invention, the step of obtaining measurement value time series data generated by each of a plurality of neighboring sensors, based on the occurrence of no alarms of all of the plurality of neighboring sensors from the time series data. Dividing time series data into a plurality of detection time intervals, forming an integrated time interval by merging adjacent first detection time intervals and second detection time intervals of the plurality of detection time intervals, the detection time interval and the integrated time Determining a problem location using the arrangement coordinates and measurement values of each of the three core sensors selected in each section, and time series data of the detection time section in which the number of the important sensors coincide and the problem location overlaps a reference value or more Generating a sequence reflecting a result of sequentially connecting time series data of the integrated time period, and generating a model that detects the problem location based on the measured value of the time series data by learning the sequence. Can include.

1 is a schematic diagram illustrating a problem positioning model learning system according to an embodiment of the present invention.
2 is a flow chart of a method for preprocessing a sensor measurement value according to an embodiment of the present invention.
3 is an exemplary diagram of grouping sensors located within a predetermined radius among a plurality of sensors into adjacent sensors.
4 is a diagram for explaining whether an alarm occurs according to a measured value measured by a sensor.
5 is a diagram for explaining a method of dividing measurement value time series data based on no alarm occurrence.
FIG. 6 is a diagram for explaining in more detail some operations of the sensor measurement value preprocessing method described with reference to FIG. 2.
7 is an exemplary diagram illustrating a method of selecting an important sensor of FIG. 6.
FIG. 8 is a diagram for describing a time section integrated by the operation of FIG. 6.
9 is a diagram for explaining in more detail some operations of the method for preprocessing a sensor measurement value described with reference to FIG. 2.
10 is an exemplary diagram showing measured values measured by a core sensor.
11 is an exemplary diagram showing arrangement coordinates of core sensors.
12 is a diagram for explaining a method of dividing an area in which a core sensor is disposed according to the operation of FIG. 9.
FIG. 13 is a diagram for describing a method of determining a problem location in a divided area in FIG. 12.
14 is an exemplary diagram showing whether an alarm occurs according to time of a core sensor.
15 is a diagram illustrating a method of generating a corresponding code according to whether an alarm is generated by a core sensor.
16 is an exemplary diagram illustrating a sequence generated through a method of preprocessing a sensor measurement value according to an embodiment of the present invention.
17 is a hardware configuration diagram of a sensor measurement value preprocessing apparatus according to another embodiment of the present invention.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings. Advantages and features of the present invention, and a method of achieving them will become apparent with reference to the embodiments described below in detail together with the accompanying drawings. However, the technical idea of the present invention is not limited to the following embodiments, but may be implemented in various different forms, and only the following embodiments complete the technical idea of the present invention, and in the technical field to which the present invention pertains. It is provided to completely inform the scope of the present invention to those of ordinary skill in the art, and the technical idea of the present invention is only defined by the scope of the claims.

In adding reference numerals to elements of each drawing, it should be noted that the same elements are assigned the same numerals as possible, even if they are indicated on different drawings. In addition, in describing the present invention, when it is determined that a detailed description of a related known configuration or function may obscure the subject matter of the present invention, a detailed description thereof will be omitted.

Unless otherwise defined, all terms (including technical and scientific terms) used in the present specification may be used with meanings that can be commonly understood by those of ordinary skill in the art to which the present invention belongs. In addition, terms defined in a commonly used dictionary are not interpreted ideally or excessively unless explicitly defined specifically. The terms used in the present specification are for describing exemplary embodiments and are not intended to limit the present invention. In this specification, the singular form also includes the plural form unless specifically stated in the phrase.

In addition, in describing the constituent elements of the present invention, terms such as first, second, A, B, (a) and (b) may be used. These terms are for distinguishing the constituent element from other constituent elements, and the nature, order, or order of the constituent element is not limited by the term. When a component is described as being "connected", "coupled" or "connected" to another component, the component may be directly connected or connected to that other component, but another component between each component It will be understood that elements may be “connected”, “coupled” or “connected”.

As used in the present invention, "comprises" and/or "comprising" refers to the recited components, steps, actions and/or elements of one or more other elements, steps, actions and/or elements. It does not exclude presence or addition.

Hereinafter, some embodiments of the present invention will be described in detail with reference to the accompanying drawings.

1 is a schematic diagram illustrating a problem positioning model learning system according to an embodiment of the present invention.

Referring to FIG. 1, a problem positioning model learning system according to an embodiment of the present invention includes a plurality of sensors 10, a sensor measurement value preprocessing device 20, a problem positioning model learning device 30, and a problem positioning. Device 40 may be included.

The problem location determination model system may be a system that determines and provides a problem location of a measurement value measured by the sensor 10. Here, the sensor 10 may be composed of a gas detection sensor 10, but is not limited to such a type of sensor 10, and a physical sensor 10 that senses temperature, pressure, sound, gas, etc., or physical A virtual sensor 10 that generates a new value using the sensor 10 may be composed of various sensors 10 that detect a physical quantity or a change thereof. The sensor measurement value v1 may be a value measured by the sensor 10, and the problem location may be a location where a substance sensed by the sensor 10 is generated or leaked.

However, it should be noted that the plurality of sensors 10 are all sensors of the same type. That is, some embodiments of the present invention are understood to include an operation or a related configuration that receives the measurement values of time series sensors generated by a plurality of homogeneous sensors arranged in adjacent locations as raw data, and preprocesses them to be suitable for machine learning. It will be possible.

In one embodiment of the present invention, the problem location determination model system detects the gas leaked from the plurality of sensors 10 and transmits the sensor measurement value v1 to the sensor measurement value preprocessing device 20 through the network, the sensor The measured value preprocessing device 20 may preprocess the sensor measured value v1 and time series data 1 to be used for machine learning.

Specifically, the sensor measurement value preprocessing apparatus 20 may obtain measurement value time series data 1 generated by each of the plurality of adjacent sensors 10. The sensor measurement value preprocessing device 20 divides the time series data 1 into a plurality of detection time intervals based on the non-alarm occurrence of all of the plurality of adjacent sensors 10 in the time series data 1, and detects a plurality of times. An integrated time period may be formed by merging the adjacent first detection time period t1 and the second detection time period t2 of the time period.

The sensor measurement value preprocessing device 20 may determine the problem location using the arrangement coordinates and measurement values of each of the three core sensors selected in the detection time section and the integration time section. Thereafter, the sensor measurement value preprocessing device 20 is a sequence reflecting the result of sequentially connecting the three core sensors and the detection time interval having the same problem location or the time series data 1 of the integrated time interval ( Sequence, 2) can be created.

The problem positioning model learning device 30 receives the sequence 2 from the sensor measurement value preprocessor 20 and learns it, and generates a model that detects the problem location based on the measured value of the time series data 1. I can.

The problem positioning device 40 may acquire a model learned from the problem positioning model learning device 30. When the problem location determining device 40 receives the time series data 1 from the plurality of sensors 10 through a network, the problem location may be determined based on the learned result. For example, when the problem positioning device 40 receives the sensor measurement value v1, and inputs the sensor measurement value v1 into the problem positioning model, the problem location can be predicted using the sensor measurement value v1. have. As another example, when the problem location determination apparatus 40 receives measurement values for gas from the plurality of sensors 10, it is possible to predict the location where the gas has leaked.

In FIG. 1, the sensor measurement value preprocessing device 20, the problem positioning model learning device 30, and the problem positioning device 40 are each represented in different configurations, but the devices are not necessarily classified as described above, and are implemented. Depending on the method, all of the devices may be included in one computing device, and operations and functions of the devices may be distributed and performed in two or more devices.

Hereinafter, a method of preprocessing a sensor measurement value according to another embodiment of the present invention will be described with reference to FIG. 2. The method of preprocessing sensor measurement values according to the present embodiment is performed by a computing device. For example, the computing device may be the preprocessing device 20 of FIG. 1. However, the computing device may be the learning device 30 of FIG. 1. Alternatively, the preprocessing method of the sensor measurement value according to the present embodiment may be performed separately by the preprocessing device 20 and the learning device 30. Hereinafter, in describing each operation belonging to the method according to the present embodiment, when the subject is not specified, the subject may be understood as the computing device.

The method of preprocessing sensor measurement values of the present invention has an advantage in that a model to be trained can be more efficiently learned by preprocessing data measured by a sensor to be suitable for machine learning.

2 is a flow chart of a method for preprocessing sensor measurement values according to an embodiment of the present invention, FIG. 3 is an exemplary diagram of grouping a sensor located within a predetermined radius among a plurality of sensors into adjacent sensors, and FIG. 4 is This is a diagram for explaining whether an alarm occurs according to a measured value.

Referring to FIG. 2, in the method of preprocessing sensor measurement values according to an embodiment of the present invention, a sequence reflecting the result of sequentially connecting time series data is generated after obtaining time series data from a plurality of adjacent sensors and performing several operations. Can be.

Specifically, in step S110, measurement value time series data generated by each of a plurality of adjacent sensors may be obtained. Here, as shown in FIG. 3, the plurality of adjacent sensors may be designated as a plurality of adjacent sensors by grouping sensors located within a preset radius r1 and r2 based on coordinates among the plurality of sensors. The first to fourth sensors s1 to s4 are located within the radius of r1, so they can be designated as group 1 (g1), and the fifth to seventh sensors are located within the radius of r2, so group 2 (g2). ) Can be specified. In addition, the plurality of adjacent sensors may be located within a predetermined radius and may be composed of only sensors that generate an alarm or operate within a predetermined time range.

As illustrated in FIG. 4, the measured value time series data may be data indicating whether an alarm is generated or detected as 0 or 1 using a measured value measured over time. In the measured value time series data, a signal for generating an alarm may be generated when a measured value measured by a plurality of sensors exceeds a preset threshold UL. In other words, the measurement value time series data is when a plurality of sensors acquires a measurement value equal to or greater than the threshold value (UL), data indicating 1 is generated at the corresponding time (t1, t2), and when a measurement value less than the threshold value (UL) is obtained. Data indicating 0 at which no alarm is generated may be generated at that time.

Referring back to FIG. 1, in step S120, time series data may be divided into a plurality of detection time intervals based on the occurrence of no alarms of all of the plurality of adjacent sensors in the time series data. A detailed method of dividing time series data into a plurality of detection time intervals will be described in detail with reference to FIG. 5.

In step S130, at least some of the plurality of adjacent sensors may be selected as the important sensor k for each detection time interval. Here, the important sensor k is a sensor in which an alarm has been generated more than a certain number of times, and may mean a sensor that should be importantly determined in time series data.

In step S140, the first detection time section t1 and the second detection time section t2 among a plurality of detection time sections may be merged to form an integrated time section. In this step, if the critical sensor (k) operated in the detection time section coincides and the section where the alarm does not occur is narrow, it can be viewed as a measurement value generated by the same cause, so the sections are combined and treated as one section for analysis. Can be. Details about this will be described in FIG. 6.

In step S150, a problem location may be determined for each time interval. Here, the problem location may mean a location where the problem occurs. In this step, a core sensor is selected based on the intensity (or absolute value) of the measured value v1, not based on the occurrence time, and the location where the problem occurs may be determined based on the selected core sensor. This will be described with reference to FIG. 9.

In step S160, a sequence may be generated by sequentially connecting time series data between time intervals having the same core sensor and the same problem location.

The sensor measurement value preprocessing method according to the present invention has an effect of generating refined data so that necessary information from the data can be easily processed.

5 is a diagram for explaining a method of dividing measurement value time series data based on no alarm occurrence.

As shown in FIG. 5, the measured value time series data may be composed of a set of data received from a plurality of adjacent sensors, and the plurality of adjacent sensors are group 1 consisting of first to fourth sensors s1 to s4 ( g1). The measured value time series data receives a measured value from each sensor, and the horizontal axis indicates time, and the vertical axis indicates whether an alarm occurs (0, 1) for each sensor.

In this step, in the time series data, the time series data may be divided into a plurality of detection time intervals based on the occurrence of no alarms from all of the plurality of adjacent sensors. No alarm may mean a section in which an alarm is not generated by all of the plurality of adjacent sensors at the same time. In FIG. 5, the alarm-free times of the first to fourth sensors s1 to s4 are 6, 15, 16, 21 to 29, and 33 to 35. Accordingly, in this step, it may be divided into a plurality of detection time intervals based on intervals of 6, 15, 16, 21 to 29, and 33 to 35, which are times when no alarm occurs.

That is, the time series data is a'first detection time interval t1' of times 1 to 5, a'second detection time interval t2' of times 7 to 14, and a'third detection time interval t3 of times 17 to 20'. )'and a'fourth detection time interval' of times 29 to 32.

6 is a view for explaining in more detail some operations of the sensor measurement value preprocessing method described with reference to FIG. 2, and FIG. 7 is an exemplary view showing a method of selecting an important sensor k of FIG. 6, and FIG. 8 is a diagram for describing a time section integrated by the operation of FIG. 6.

6, when the step (S140) of forming an integrated time interval by merging the first detection time interval (t1) and the second detection time interval (t2) is performed, each detection time in step S141 It may be determined whether the important sensors k determined in the section are the same. In step S130, at least some of the plurality of adjacent sensors are selected as important sensors k for each detection time interval, and a specific method will be described with reference to FIG. 7.

Referring to FIG. 7, the important sensor k may be a sensor selected for each detection time interval in a plurality of detection time intervals.

The important sensor k may be selected when the detection time of each of the plurality of adjacent sensors is greater than or equal to the reference detection number. Here, the reference detection number may be a value obtained by multiplying the length of the detection time interval by a preset reference coefficient for each detection time interval.

As shown in the table of FIG. 7, in the case of the first detection time section t1, the section width of the section is 5, and assuming that the reference coefficient k1 is 0.3, the number of reference detections is 1.5, and the second detection time section t2 In the case of, the section width of the section is 8, and the reference number of detection is 2.4. Similarly, in the case of the third detection time section t3, the section width of the section is 4, the reference number of detection is 1.2, and in the case of the fourth detection time section t4, the section width of the section is 4, and the reference detection number is 1.2. .

As shown in the table of FIG. 5, in the first detection time period t1, the first sensor s1 is detected 5 times, the second sensor s2 is detected 4 times, and the third sensor s3 is detected once. , The fourth sensor s4 may be detected once. Sensors that exceed the reference number of detections 1.5 in the first detection time period t1 are the first sensor s1 and the second sensor s2. Accordingly, the first sensor s1 and the second sensor s2 may be selected as important sensors k. The third sensor s3 and the fourth sensor s4 are not selected as important sensors k because the reference number of detection is 1.5 or less. Also in the second detection time period t2 to the fourth detection time period t4, the important sensor k may be selected in the same manner.

Referring to FIG. 6 again, it may be determined in step S143 whether the time interval in which the alarm does not occur is equal to or less than the integrated reference value. In step S143, when the first detection time interval t1 and the second detection time interval t2 are less than the integrated reference value, the first detection time interval t1 and the second detection time interval It may be determined that (t2) is adjacent. The integrated reference value may be a value obtained by multiplying a larger time interval among the first detection time interval t1 and the second detection time interval t2 by a preset integration coefficient k2.

Referring to FIG. 8, since the second detection time interval to the fourth detection time interval t2 to t4 are regions related to four sensors, integration may be considered. First, assuming that the integration coefficient k2 is 0.5, the integration reference value is 4 multiplied by the integration coefficient 0.5 by the second detection time period t2 8, and the second detection time period t2 and the third detection time period Since the size of the no-alarm occurrence between (t3) is 2, the integrated reference value is larger than the size of the no-alarm occurrence, so the first integration time by integrating the second detection time period t2 and the third detection time period t3 The section i1 can be configured.

If it is determined as'Y' in both steps S141 and S143, an integrated time interval may be formed by merging adjacent detection time intervals. However, if it is determined as'N' in any one of steps S141 and S143, adjacent detection time intervals may not be merged. Even if the detection time interval is not merged, if the condition of step S140 is satisfied by comparing the time series data with the adjacent detection time interval, it can be merged. Even if the integrated detection time interval has already been merged, it is compared with the adjacent detection time interval in the time series data. If the condition of step S140 is satisfied, it may be recursively merged again.

The first integration time period (i1) and the fourth detection time period (t4) have an alarm-free size of 8, and the value obtained by multiplying the first integration time period (i1) by the integration factor of 0.5 is 7 The interval i1 and the fourth detection time interval t4 are not integrated.

The sensor measurement value preprocessing method of the present invention has the effect of being able to further purify and refine detailed analysis data by treating data due to the same cause as one by merging detection time intervals.

9 is a view for explaining in more detail some operations of the sensor measurement value preprocessing method described with reference to FIG. 2, FIG. 10 is an exemplary view showing measurement values measured by a core sensor, and FIG. 11 is It is an exemplary diagram showing the arrangement coordinates.

9, in step S150, a problem location is determined for each of the detection time section and the integration time section, but the arrangement coordinates and measurement values of each of the three core sensors (c) selected in each of the detection time section and the integration time section The problem location may be determined using. The core sensor c refers to the sensor having the highest average measured value in each time section, and will be described with reference to FIGS. 10 and 11.

Referring to FIG. 10, measurement value time series data generated from a plurality of adjacent sensors may include measurement values as described in FIG. 4. In this step, an average measured value for a corresponding time interval may be calculated. As shown in FIG. 10, the first sensor (s1) to the third sensor (s3) may be selected as the three core sensors (c) having the highest average measured value, and the average measured value is the first to the first core sensor (c1) to the third sensor. The three core sensors c3 may be 50, 35 and 10, respectively.

Referring to FIG. 11, corresponding location information may be obtained in a step in which all of a plurality of adjacent sensors are grouped, and coordinate values in which both the first core sensor (c1) to the third core sensor (c3) are arranged are analyzed in this step. Can be.

Referring to FIG. 9 again, in step S153, three boundary lines divided into two may be generated around an outer center point of the core sensor c. Thereafter, the measured values are compared based on the divided boundary in step S155, and in step S157, while repeatedly selecting an area on the side with a larger average measured value, the area selected by the most overlapping in step S159 may be determined as the problem location. . Hereinafter, a specific operation of FIG. 9 will be described with reference to FIGS. 12 and 13.

FIG. 12 is a diagram for explaining a method of dividing an area in which the core sensor c is disposed according to the operation of FIG. 9, and FIG. 13 is a diagram for explaining a method of determining a problem location in the divided area in FIG. 12 to be.

Referring to FIG. 12, in an embodiment, an area in which the core sensor c is disposed may be divided into six areas. Thereafter, the measurement values of the core sensor c are compared in each of the six areas, and an area in which a large measurement value is located may be selected. In this step, the selected area is selected as shown in FIG. 13, and the area overlapping the maximum may be determined as a problem location.

For example, when the area where the core sensor c is disposed is divided into 6 areas by the first to third border lines d1 to d3, the upper right side of c1 is called the first area, and it is counterclockwise. It is assumed that the second to sixth regions are set in the direction.

The first boundary line d1 is divided into a first region to a third region and a fourth region to a sixth region. Among the core sensors (c) located in the two divided areas, the core sensor (c) with the largest average measurement value is the second core sensor (c2) in the fourth to sixth areas, and the third core sensor (c2) in the first to third areas. It is the core sensor (c3). When the average measured values of the second core sensor c2 and the third core sensor are compared, the average measured value of the second core sensor c2 is 35, which is greater. Accordingly, when comparing the areas divided by the first boundary line d1, as shown in q1 of FIG. 13, the area where the second core sensor c is located may be selected. Similarly, in the case of the first, second, and sixth regions divided by the second boundary line d2 and the third, fourth, and fifth regions, the region in which the first core sensor c1 is located as shown in q2 of FIG. 13 may be selected. . Similarly, in the case of the first, second, and sixth regions divided by the third boundary line d3 and the third, fourth, and fifth regions, the region in which the first core sensor c1 is located as shown in q3 of FIG. 13 may be selected. . In FIG. 13, the sixth area, which is most frequently selected three times in this manner, may be selected as the problem location.

In another embodiment, the first to third boundary lines d1 to d3 of FIG. 12 may be generated using an outer center of the position of the core sensor c. The first to third boundary lines d1 to d3 are formed by lowering the foot from the triangular outer center circle connecting the arrangement position of the core sensor c to the triangle, and connecting the line heading from the outer circle to the triangle. Can be created. Thereafter, an area with a large measurement value is selected by comparing the measured value of the core sensor c based on the divided boundary for each of the three boundary lines, and the area where the selected area maximally overlaps can be determined as the problem location. have.

14 is an exemplary diagram illustrating whether an alarm occurs according to time of the core sensor c, and FIG. 15 is a diagram illustrating a method of generating a corresponding code according to whether an alarm occurs in the core sensor.

Referring to FIG. 14, whether an alarm (v2) of the first to third core sensors c3 is generated may be extracted from time series data. As shown in FIG. 15, an alphabet may be generated as a new event code according to a combination when an alarm for each sensor occurs. For example, in FIG. 15, when an alarm is generated in the first core sensor c1 and no alarms are generated in the second and third core sensors c3, an A code may be generated. When an alarm is generated in the second core sensor c2 and no alarm is generated in the first and third core sensors c3, a B code may be generated. Eight event codes may be generated based on a combination of a total of three core sensors (c).

16 is an exemplary diagram illustrating a sequence generated through a method of pre-processing a sensor measurement value according to an embodiment of the present invention.

Referring to FIG. 16, a sequence reflecting a result of sequentially connecting time series data of the detection time interval or the integrated time interval in which the number of important sensors is the same and the problem location overlaps by a reference value or more may be generated. Whether the problem location overlaps more than the reference value may be determined based on the coordinate section of the region obtained according to the method described with reference to FIG. 13. For example, if the problem location in the first detection time section and the problem location in the second detection time section overlap a specified area or more, the number of important sensors in the first detection time section and the number of important sensors in the second detection time section are If the same, the time series data of the first detection time interval and the time series data of the second detection time interval are sequentially connected, and a sequence reflecting the sequentially connected time series data may be generated.

If the number of critical sensors is the same and the problem locations overlap, it is likely to indicate the same situation. Therefore, in the case of learning by linking time series data indicating the same situation to each other in consideration of this point, the learning effect may be enhanced.

The first detection time period t1 and the fourth detection time period t4 have the same problem location as the sixth area, and the number of important sensors is also the same, so each of them has the same length of time as shown in FIG. 16 to form a connected sequence. can do. That is, the event code according to the configured sequence will be'ADGDGGGG'.

Unlike FIG. 16, if the first detection time interval t1 and the fourth detection time interval t4 have different time lengths, the time of a time interval having a shorter time length among each detection time interval t1 and t2 Based on the length, time series data may be extracted from each detection time period t1 and t2. For example, assuming that the first detection time period t1 includes 4 time slots, and the second detection time period t2 includes 6 time slots unlike FIG. 16, the first detection time period ( In t1), all data may be extracted as time series data, but in the second detection time period t2, only data of 4 time slots among 6 time slots may be extracted as time series data. In this case, in the second detection time period t2, data of 4 time slots among 6 time slots may be extracted as the time series data.

Even if the first detection time interval t1 and the fourth detection time interval t4 have different lengths of time, it is preferable that the same amount of information is input in the machine learning process as long as they indicate the same situation. With this in mind, when forming a sequence by sequentially connecting each time section having the same core sensor and the same problem area, based on the time section having the minimum length among each time section, It may be understood that the lengths are matched equally.

The methods according to the embodiments of the present invention described so far can be performed by executing a computer program implemented in computer-readable code. The computer program may be transmitted from a first computing device to a second computing device through a network such as the Internet and installed in the second computing device, thereby being used in the second computing device. The first computing device and the second computing device include all of a server device, a physical server belonging to a server pool, and a fixed computing device such as a desktop PC.

The computer program may be stored in a recording medium such as a DVD-ROM or a flash memory device. Hereinafter, with reference to FIG. 15, a hardware configuration of a malicious code detection model learning apparatus 100 according to another embodiment of the present invention will be described.

17 is a hardware configuration diagram of a sensor measurement value preprocessing apparatus 100 according to another embodiment of the present invention.

Referring to FIG. 17, the sensor measurement value preprocessing apparatus 100 according to another embodiment of the present invention includes one or more processors 110, a problem positioning program 151 and a sequence generation program executed by the processor 110. A memory 120 for loading 152, a bus 130, a network interface 140, and a storage 150 for storing the problem location determination program 151 and the sequence generation program 152 may be included. have. Here, only components related to an embodiment of the present invention are shown in FIG. 17. Accordingly, those of ordinary skill in the art to which the present invention pertains can recognize that other general-purpose components may be further included in addition to the components shown in FIG. 17.

The processor 110 controls the overall operation of each component of the sensor measurement value preprocessing device 100. The processor 110 includes a CPU (Central Processing Unit), MPU (Micro Processor Unit), MCU (Micro Controller Unit), GPU (Graphic Processing Unit), or any type of processor well known in the technical field of the present invention. Can be. In addition, the processor 110 may perform an operation on at least one application or program for executing the method according to the embodiments of the present invention. Here, the sensor measurement value preprocessing apparatus 100 shown in FIG. 17 is shown to include one processor 110, but is not limited thereto and may include a plurality of processors.

The memory 120 stores various types of data, commands, and/or information. The memory 120 may load the problem location determination program 151 and the sequence generation program 152 from the storage 150 in order to execute a method/operation according to various embodiments of the present disclosure. When the problem positioning program 151 and the sequence generating program 152 are loaded in the memory 120, the processor 110 is configured to perform one or more instructions 121 constituting the problem positioning program 151 and the sequence generating program 152. , 122) to perform the method/action. The memory 120 may be implemented as a volatile memory such as RAM, but the technical scope of the present invention is not limited thereto.

The bus 130 provides a communication function between components of the sensor measurement value preprocessing device 100. The bus 130 may be implemented as various types of buses such as an address bus, a data bus, and a control bus.

The network interface 140 supports wired/wireless Internet communication of the sensor measurement value preprocessing device 100. In addition, the network interface 140 may support various communication methods other than Internet communication. To this end, the network interface 140 may be configured to include a communication module well known in the art. In some embodiments, the network interface 140 may be omitted.

The storage 150 may non-temporarily store the problem location determination program 151 and the sequence generation program 152. When an application program is executed and manipulated through the sensor measurement value preprocessing device 100, the storage 150 may store various data on the application program executed according to the execution and manipulation. For example, the storage 150 may store information on an application program to be executed, operation information on the application program, information on a user who has requested execution of the application program, and the like.

The storage 150 is a nonvolatile memory such as a ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), flash memory, etc., a hard disk, a removable disk, or well in the technical field to which the present invention belongs. It may be configured to include any known computer-readable recording medium.

The problem positioning program 151 and the sequence generation program 152, when loaded into the memory 120, cause the processor 110 to perform a method/operation according to various embodiments of the present invention. 122). That is, the processor 110 may perform the method/operation according to various embodiments of the present disclosure by executing the one or more instructions 121 and 122.

In one embodiment, the problem location determination program 151 and the sequence generation program 152 are instructions for obtaining measurement value time series data generated by each of a plurality of adjacent sensors, and alarms of all of the plurality of adjacent sensors from the time series data. An instruction for dividing the time series data into a plurality of detection time intervals based on occurrence, an instruction for selecting at least some of the plurality of adjacent sensors as important sensors for each detection time interval, a first detection time interval, and the first detection When the important sensors of each of the second detection time periods adjacent to the time period coincide, an instruction for forming an integrated time period by merging the first detection time period and the second detection time period, the detection time period, and the integration time Determine the location of the problem for each section, but the instruction for determining the location of the problem using the arrangement coordinates and measured values of each of the three core sensors selected in the detection time section and the integration time section, and the three core sensors match And an instruction for generating a sequence reflecting a result of sequentially connecting time series data of the detection time interval or the integrated time interval having the same problem location.

So far, various embodiments of the present invention and effects according to the embodiments have been mentioned with reference to FIGS. 1 to 17. The effects according to the technical idea of the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those skilled in the art from the following description.

Although the embodiments of the present invention have been described with reference to the accompanying drawings, those of ordinary skill in the art to which the present invention pertains can be implemented in other specific forms without changing the technical spirit or essential features. I can understand. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not limiting.

Claims (13)

In the method performed by the computing device,
Obtaining measurement value time series data generated by each of a plurality of adjacent sensors;
Dividing the time series data into a plurality of detection time intervals based on the occurrence of no alarms from all of the plurality of adjacent sensors in the time series data;
Selecting at least some of the plurality of adjacent sensors as important sensors for each detection time interval;
When the first detection time interval and the important sensors of each of the second detection time intervals adjacent to the first detection time interval coincide, the first detection time interval and the second detection time interval are merged to form an integrated time interval. The step of doing;
Determining a problem location for each of the detection time section and the integration time section, and determining the problem location using the arrangement coordinates and measurement values of each of the three core sensors selected in the detection time section and the integration time section. ; And
Generating a sequence reflecting a result of sequentially connecting time series data of the detection time section or the time series data of the integrated time section in which the number of the important sensors is the same and the problem location overlaps by a reference value or more. ,
How to preprocess sensor readings. The method of claim 1,
The plurality of adjacent sensors,
Determined by grouping sensors located within a predetermined radius based on the coordinates of a plurality of sensors,
How to preprocess sensor readings. The method of claim 1,
The step of selecting as the important sensor,
Including the step of selecting the important sensor when the detection time of each of the plurality of adjacent sensors is greater than or equal to the reference detection number,
How to preprocess sensor readings. The method of claim 3,
The number of times of the reference detection,
A value obtained by multiplying the length of the detection time section by a preset reference coefficient for each detection time section,
How to preprocess sensor readings. The method of claim 1,
The step of configuring the integrated time interval,
Comprising the step of determining that the first detection time interval and the second detection time interval are adjacent when the time interval in which the alarm does not occur in the first detection time interval and the second detection time interval is less than an integrated reference value,
How to preprocess sensor readings. The method of claim 5,
The integrated reference value is,
A value obtained by multiplying a predetermined integration coefficient by a larger time interval among the first detection time interval and the second detection time interval,
How to preprocess sensor readings. The method of claim 1,
The step of determining the problem location,
Including the step of selecting the top three sensors of the measured value from the plurality of adjacent sensors as the core sensor,
How to preprocess sensor readings. The method of claim 7,
The step of determining the problem location,
Further comprising the step of dividing the area in which the core sensor is disposed into 6 areas,
How to preprocess sensor readings. The method of claim 8,
The step of determining the problem location,
Comparing the measured values of the core sensors in each of the six areas, selecting an area in which a large measurement value is located, and determining an area where the selected area maximally overlaps as the problem location,
How to preprocess sensor readings. The method of claim 7,
The step of determining the problem location,
Generating three boundary lines for dividing an area in which the core sensor is disposed based on the repair line by lowering the foot to the triangle in a triangular outer center circle connecting the arrangement position of the core sensor;
Comparing the measured values of the core sensors based on the divided boundaries for each of the three boundary lines and selecting a region having a large measured value; And
Further comprising the step of determining a region where the selected region maximally overlaps as the problem location,
How to preprocess sensor readings. The method of claim 1,
Generating the sequence,
Comprising the step of processing the time series data from the detection time section or the integrated time section in which the problem location overlaps by a reference value or more into a time section of the same length,
How to preprocess sensor readings. The method of claim 11,
Generating the sequence,
Comprising the step of processing the time series data based on a minimum time interval among the detection time interval or the integrated time interval in which the problem location overlaps by a reference value or more,
How to preprocess sensor readings. In the method performed by the computing device,
Receiving a sequence in which time series data generated by each of a plurality of adjacent sensors is preprocessed; And
Learning the sequence and generating a model for detecting the problem location based on the measured value of the time series data,
Receiving a sequence in which the time series data has been preprocessed,
Obtaining measurement value time series data generated by each of the plurality of adjacent sensors;
Dividing the time series data into a plurality of detection time intervals based on the occurrence of no alarms from all of the plurality of adjacent sensors in the time series data;
Forming an integrated time interval by merging the adjacent first detection time interval and the second detection time interval of the plurality of detection time intervals;
Determining a problem location using the arrangement coordinates and measurement values of each of the three core sensors selected in the detection time section and the integration time section; And
Generating a sequence reflecting a result of sequentially connecting time series data of the detection time section or the time series data of the integrated time section in which the number of the important sensors is the same and the problem location overlaps by a reference value or more. ,
How to train the problem positioning model.

Images