WO2017150263A1

WO2017150263A1 - Abnormality detection device, abnormality detection system, and method thereof

Info

Publication number: WO2017150263A1
Application number: PCT/JP2017/006258
Authority: WO
Inventors: 進吾足立; 信補高橋; 基朗小熊; 剛武本
Original assignee: 株式会社日立製作所
Priority date: 2016-02-29
Filing date: 2017-02-21
Publication date: 2017-09-08
Also published as: JP2017156818A; JP6605357B2

Abstract

In the present invention, even a process abnormality that gives a moderate change to measured values is detected. An abnormality detection device has a prediction and assessment unit, an effect data selection unit, an assessment integration unit, and an output part. In the prediction and assessment unit: the prediction value of a first measurement value of a sensor in a monitored process at the point in time an assessment is made is predicted, using each of a plurality of prediction and assessment methods, on the basis of measurement values from the sensor over a data range corresponding to the plurality of prediction and assessment methods; the process to be monitored is assessed to be abnormal on the basis of (1) the difference between the prediction value and the first measurement value at the point in time the assessment is made, and/or (2) the first measurement value outside the range between the upper and lower limits of the prediction values; and the abnormality assessment results for when an abnormality was assessed to have occurred as obtained with the plurality of predictive assessment methods are outputted. In the effect data selection unit, a range that includes, as effect data, a second measurement value affected by the abnormality included in the abnormality assessment results is selected as the effect data range. In the assessment integration unit, the reliability of those, among the abnormality assessment results, that are abnormality assessment results based on the effect data included in the effect data range is lowered, and the abnormality assessment results of the plurality of prediction and assessment methods are integrated. The output part outputs the integrated abnormality assessment results.

Description

Anomaly detection apparatus, anomaly detection system and method

The present invention relates to an anomaly detection device, an anomaly detection system, and a method thereof.

Non-Patent Document 1 calculates the predicted value of the next time from the observed value of the past time by applying support vector regression to the measured value time series of the flow rate and pressure of the water distribution process. Disclosed is a method for detecting an abnormality in a water distribution process, such as the occurrence of water leakage, by performing property detection (Novelty detection).

When performing abnormality detection of the water distribution process such as the occurrence of water leakage by the method of Non-Patent Document 1, it is impossible to detect abnormality such as the occurrence of small-scale water leakage. This method uses a measurement value affected by the occurrence of water leakage to calculate a predicted value at the next time after the occurrence of water leakage. For this reason, in the case of small-scale water leakage, the predicted value is close to the measured value after the occurrence of water leakage, and the difference between the predicted value and the measured value is reduced. Therefore, if the water leak is large enough that the difference between the predicted value and the measured value is sufficiently large immediately after the occurrence of water leakage, the predicted value can be detected before being affected by the occurrence of water leakage, but small-scale water leakage cannot be detected.

Here, the scale of the water leakage is a scale obtained by evaluating the change that the water leakage gives to the measured values of the sensor such as the flow rate and the pressure by a relative ratio to the fluctuation of the measured values in the normal time when no water leakage occurs. Even if the water flow leaks from the water pipe with a large absolute value of the flow rate, if the water leak occurs in a distribution area with a large amount of water distribution, the water leak will be small.

Therefore, it is desirable that even small-scale water leakage can be detected from measured values such as flow rate and pressure. In other words, it is desirable to be able to detect even a process abnormality that gives a gradual change in the measured value.

The disclosed abnormality detection device uses each of the plurality of prediction determination methods, based on the measurement value of the sensor of the monitored process in the data range according to the plurality of prediction determination methods, the first sensor of the determination time Predict the predicted value of the measured value, (1) the difference between the predicted value and the first measured value at the determination time, and (2) the first measured value outside the range between the upper limit value and the lower limit value of the predicted value Based on at least one of the above, a prediction determination unit that determines an abnormality of the monitoring target process and outputs an abnormality determination result of determining an abnormality of a plurality of prediction determination methods, a second that an abnormality included in the abnormality determination result affects The influence data selection unit that selects the range including the measured value of the measurement data as the influence data range as the influence data range, among the abnormality determination results, the reliability of the abnormality determination result based on the influence data included in the influence data range is lowered, and a plurality of Predictive judgment Having the abnormality determination integrating unit that integrates the determination result, and an output unit for outputting the integrated anomaly determination result of the expression.

According to the disclosed abnormality detection device, even a process abnormality that gives a gradual change in the measured value can be detected.

It is a block diagram of an abnormality detection system. It is a hardware structural example of an abnormality detection apparatus. It is a structural example of the water pipe network which an abnormality detection apparatus makes the monitoring object. It is a figure which shows the prediction and abnormality determination by a different prediction determination system. It is a processing flowchart of a judgment integration part. It is a display example of the detection result of the occurrence of water leakage by the abnormality detection device. It is a block diagram of the abnormality detection system of Example 2.

Hereinafter, examples will be described with reference to the drawings.

FIG. 1 is a configuration diagram of an abnormality detection system 100 that monitors a water distribution process (water pipe network) based on sensor measurement values such as flow rate and pressure to detect an abnormality such as occurrence of water leakage. The abnormality detection system 100 includes an abnormality detection device 101, a sensor 191 that obtains measurement values such as flow rate and pressure, a measurement value collection device 102, and an alarm display device 103.

The abnormality detection apparatus 101 includes an influence data selection unit 111, a determination integration unit 112, a prediction determination unit 131, a prediction determination unit 132, a measurement value collection unit 151, and an output unit 152, and a measurement value storage unit 121. Each storage unit includes an auxiliary data storage unit 122 and a prediction determination method data storage unit 123.

The influence data selection unit 111 inputs an abnormality determination result from the prediction determination units 131 and 132 (hereinafter, the prediction determination unit 131 is representative when the

prediction determination units

131 and 132 need not be individually described), and the abnormality If it is determined that the determination result is abnormal, a data range of measurement values affected by the abnormality is selected, and the selected data range is output to the determination integration unit 112 as an influence data range (described later). Details of the processing of the influence data selection unit 111 will be described later.

The determination integration unit 112 inputs the abnormality determination result from the prediction determination unit 131, inputs the influence data range from the influence data selection unit 111, and lowers the reliability (described later) of the abnormality determination result based on the input influence data range. The abnormality determination result is integrated, and the integrated abnormality determination result is output to the output unit 152. Details of the processing of the determination integration unit 112 will be described later.

The prediction determination unit 131 reads measurement values from the measurement value storage unit 121, reads auxiliary data from the auxiliary data storage unit 122 as necessary, and reads prediction determination method data from the prediction determination method data storage unit 123. The prediction determination unit 131 uses a predetermined prediction determination method to calculate a sensor predicted value (a value that the sensor will output as a measurement value at the prediction time) from the sensor measurement value in the data range defined by the prediction determination method data. The difference between the predicted value and the measured value of the predicted time (details will be described later, but the predicted value has the upper and lower limit values of the predicted value itself and the predicted value. Including)), the abnormality of the water pipe network to be monitored is determined, the abnormality determination result is obtained, and the abnormality determination result is output to the influence data selection unit 111 and the determination integration unit 112. The prediction determination unit 132 also outputs an abnormality determination result based on a prediction determination method different from the prediction determination unit 131.

Also, the prediction determination unit 131 adjusts the prediction determination method parameters (prediction determination method data) based on each input data including the measurement values, and sets the adjusted prediction determination method parameters. Stored in the prediction determination method data storage unit 123.

The prediction determination unit 131 learns the prediction determination method, and executes prediction processing of the predicted value of the sensor, and processing for determining abnormality of the monitored water pipe network from the difference between the predicted value and the measured value.

The prediction determination unit 131 uses a known technique such as a method using support vector regression or a neural network described in Non-Patent Document 1 for prediction and abnormality determination. Details of the processing of the prediction determination unit 131 will be described later.

The abnormality detection apparatus 101 may include a plurality of

prediction determination units

131 and 132 in order to perform abnormality determination using different prediction determination methods. Or the prediction determination part 131 may implement | achieve a different prediction determination system by adjusting the parameter of a prediction determination system to a different value by making the logic of a prediction determination system the same.

The measurement value storage unit 121 stores the measurement value of the sensor installed in the monitored water pipe network from the measurement value collection unit 151. The prediction determination unit 131 reads the measurement value from the measurement value storage unit 121.

The auxiliary data storage unit 122 stores auxiliary data used for the processing of the prediction determination unit 131. The prediction determination unit 131 reads auxiliary data from the auxiliary data storage unit 122. The auxiliary data storage unit 122 is a water pipe network to be monitored, such as a calendar, weather, social event, etc. for specifying the season, month, day of the week, etc., as auxiliary data for prediction determination by the prediction determination unit 131 Information that influences the state of is stored in advance. The abnormality detection device 101 may collect these pieces of information from other devices outside the abnormality detection system 100 and store them in the auxiliary data storage unit 122.

The prediction determination method data storage unit 123 stores information such as parameters for determining each prediction determination method for a plurality of prediction determination methods used by the prediction determination unit 131. The prediction determination unit 131 reads information on each prediction determination method from the prediction determination method data storage unit 123. In addition, the prediction determination method data storage unit 123 receives and stores parameters of each prediction determination method adjusted by the prediction determination unit 131 through learning.

The information such as the parameters stored in the prediction determination method data storage unit 123 includes, for each prediction determination method, data relating to the prediction determination method ID, the adjusted parameter of the prediction determination method, and the measurement value used for the adjustment of this parameter. Range (described later).

The measurement value collection unit 151 receives the measurement values of the sensor 191 such as the pressure and flow rate installed in the monitored water pipe network from the measurement value collection device 102 and stores them in the measurement value storage unit 121.

The output unit 152 inputs the abnormality determination result integrated from the determination integration unit 112, and presents the abnormality determination result to the operator of the abnormality detection system 100. Further, the output unit 152 outputs an abnormality determination result to the alarm display device 103.

For example, the output unit 152 displays the abnormality determination result on the display for the operator. Alternatively, a smart device such as a smartphone or tablet held by the operator is used as the alarm display device 103, and an abnormality determination result is transmitted in response to a request from such a smart device. When the abnormality determination result satisfies a preset condition, for example, when it is determined that water leakage of a predetermined size or more has occurred, the output unit 152 may notify the alarm display device 103 by a push type such as an email or an alarm. Good.

The measurement value collection device 102 collects measurement values from the sensor 191 that measures the state of the monitored water pipe network, and transmits the collected measurement values of the sensor 191 to the abnormality detection device 101.

The alarm display device 103 receives the abnormality determination result from the output unit 152 and displays the received abnormality determination result. Further, the alarm display device 103 may receive a notification of the abnormality determination result from the output unit 152 and display the abnormality determination result.

In addition, the abnormality detection apparatus 101 may be arrange | positioned in the remote place from the water pipe network of the monitoring object where the sensor 191 and the measured value collection apparatus 102 are arrange | positioned.

FIG. 2 is a hardware configuration example of the abnormality detection apparatus 101. The abnormality detection apparatus 101 connects a CPU 201, a memory 202, a media input / output unit 203, an input unit 205, a communication control unit 204 connected to a network, a display unit 206 such as a display, and a peripheral device IF unit 207 via a bus 210. It is a so-called computer.

Therefore, the CPU 201 executes the processing of each processing unit, and the memory 202 stores the program of each processing unit and the data of each storage unit.

Note that the abnormality detection apparatus 101, the measurement value collection apparatus 102, and the alarm display apparatus 103 may be implemented as different programs on the same computer.

FIG. 3 is a configuration example of a water pipe network to be monitored by the abnormality detection apparatus 101. The water pipe network is composed of a plurality of DMAs (Distributed Metered Areas). The DMA is an area of the water pipe network, and there are a small number of pipes (inflow / outflow pipes) into and out of the adjacent pipe network (inflow / outlet pipes), and in many cases there is only one. It is measured.

FIG. 3 is a configuration example of a water pipe network supplied from the distribution reservoir 301 and includes

DMAs

340 and 341. This water pipe network includes pipes such as the distribution reservoir 301 and the distribution pipe 351, and is composed of sensors such as a flow sensor 310 (rectangle in the figure) and a pressure sensor 320 (o in the figure), and incidental equipment such as a valve 361. Has been.

The DMA 340 has one inflow / outflow pipe in which a flow rate sensor 311 is installed, and a pipe valve 361 connected to an adjacent area (DMA 341) is closed. Further, pressure sensors 320 to 322 are installed in the DMA 340. The DMA 341 has an inflow / outflow pipe in which a flow rate sensor 312 and a flow rate sensor 313 are installed.

In the example of the water pipe network of FIG. 3, the sensors 191 for collecting the measurement values by the measurement value collection device 102 are flow rate sensors 310-313 and pressure sensors 320-324.

FIG. 4 is a diagram showing prediction and abnormality determination by two prediction determination methods different from the measurement value of the flow rate sensor 311. Processing of the prediction determination unit 131 and the influence data selection unit 111 will be described with reference to FIG.

The prediction determination unit 131 is activated at a collection period of measurement values of the sensor 311, for example, at a period of 5 minutes. As a data input process, the prediction determination unit 131 reads the measurement value of the sensor 311 from the measurement value storage unit 121, reads auxiliary data from the auxiliary data storage unit 122, and receives each prediction determination method data from the prediction determination method data storage unit 123. read out. The prediction determination unit 131 obtains a prediction value corresponding to the latest measurement value of the sensor 311 using each prediction determination method as a prediction process. As a determination process, the prediction determination unit 131 determines whether or not an abnormality has occurred in the monitored water pipe network based on the difference between the predicted value using each prediction determination method and the latest measured value. The prediction determination unit 131 outputs an abnormality determination result using each prediction determination method to the influence data selection unit 111 and the determination integration unit 112 as a result output process.

4, the prediction determination results 491 and 492 are based on the prediction determination method 1 and the prediction determination method 2 output by the prediction determination unit 131. In each prediction determination graph, the horizontal axis represents time, and the vertical axis represents the measurement value (flow rate) of the flow sensor 311. Circles 401 to 404 in the figure are measurement values of the flow sensor 311 at each time for each measurement value collection cycle. Note that the measurement values that the prediction determination unit 131 determines to be normal in each prediction determination method are white circles, and the measurement values that are determined to be abnormal are black circles.

The time series prediction values 411 and 412 are point prediction results by each prediction determination method, and the time series prediction upper limit values 421 and 422 and the prediction

lower limit values

431 and 432 are section prediction results by each prediction determination method. The point prediction result is the time-series data of the prediction value at each prediction time (predicted time), and the interval prediction result has a prediction upper limit value determined by the prediction accuracy because the prediction accuracy of the prediction value differs depending on each prediction determination method. It is time series data of a prediction lower limit. The difference between the prediction upper limit value and the prediction lower limit value is called an interval.

Prediction determination method 1 performs prediction and determination based on measured values in a time range (time zone) up to the latest time, as in the method described in Non-Patent Document 1. When calculating the prediction value 411, the prediction upper limit value 421, and the prediction lower limit value 431, the prediction determination unit 131 uses the measurement value of the time range 470 for prediction of the determination time (same as the above-described prediction time) 480, and the determination time 481 The measurement value in the time range 471 is used for the prediction. For example, the predicted value corresponding to the measured value at the determination time is predicted using the measured value in the time range up to 5 minutes before the determination time (the latest measurement value collection time).

Prediction determination method 2 performs prediction and determination based on measurement values in a time range before a predetermined time from the determination time. In other words, there is a predetermined time between the time when the last measured value in the time range is obtained and the determination time. When calculating the prediction value 412, the prediction upper limit value 422, and the prediction lower limit value 432, the prediction determination unit 131 calculates the measurement value of the time range 475 for prediction of the determination time 480 and measures the time range 476 for prediction of the determination time 481. Use each value. For example, the measurement value at the determination time is predicted using the measurement value 75 minutes before the determination time.

In both prediction determination method 1 and prediction determination method 2, the predicted value at the determination time is a value predicted to be obtained as a measurement value at the determination time, and the upper limit of the prediction accuracy (vibration width) of the predicted value is It is a prediction upper limit, and a lower limit is a prediction lower limit.

As described above, a plurality of prediction determination methods are used by setting different times as the predetermined time of the prediction determination method based on the measurement values in the time range before the determination time. The prediction determination method 1 is a method in which the predetermined time is 5 minutes, and the prediction determination method 2 is a method in which the predetermined time is, for example, 75 minutes. Further, by changing the time range (changing the number of measurement values used for prediction), prediction determination methods with different prediction accuracy can be obtained. The predetermined time used by the prediction determination method is used as a parameter, and the time range or the number of measured values is stored as a data range in the prediction determination method data storage unit 123 in association with the ID of the prediction determination method.

The prediction determination unit 131 uses the prediction determination method 1 and the prediction determination method 2 when the difference between the measured value and the predicted value at the determination time is equal to or greater than a predetermined threshold, and when the measured value deviates from the section prediction (upper limit value). If it exceeds or falls below the lower limit), it is determined as abnormal.

Referring to the example shown in FIG. 4, the case where the measured value deviates from the section prediction will be described. FIG. 4 shows data (measured values and predicted values) in which water leakage occurred in the DMA 340 immediately before the measurement time of the measured value 401. The prediction determination unit 131 using the prediction determination method 1 determines that the measurement values 401 to 404 are normal because the measured values 401 to 404 are between the prediction upper limit value 421 and the prediction lower limit value 431 (section). On the other hand, the prediction determination unit 131 using the prediction determination method 2 determines that the measurement value 402-404 exceeds the prediction upper limit value 422, and thus is abnormal.

In this example, the prediction determination method for the measurement value of the flow sensor 311 is described, but a value obtained by applying various processes such as filtering such as moving average and normalization may be used. Moreover, it is good also as performing prediction and determination which considered the correlation with the measured value of another sensor.

Further, when the prediction determination unit 131 determines that an abnormality has occurred, the prediction determination unit 131 also estimates the abnormality attribute including the abnormality occurrence time, type, location, occurrence location, and the like. Even when it is difficult to estimate each attribute included in the abnormal attribute, the prediction determination unit 131 estimates at least one attribute such as an occurrence time. In the prediction determination method 2, the measured value of the flow sensor 311 that measures the amount of flow into the DMA 340 continuously exceeds the predicted value from the time 480, indicating the characteristic of water leakage. For this reason, the prediction determination unit 131 determines that the abnormality occurrence time is around the determination time 480, the abnormality type is water leakage, the abnormality location is DMA 340, and the abnormality occurrence location is near the flow sensor 311. The prediction determination unit 131 uses a technique such as pattern matching of the increasing tendency of the difference between the predicted value and the measured value as described above in order to estimate the abnormal attribute.

The influence data selection unit 111 inputs the abnormality determination result and the abnormality attribute from the prediction determination unit 131. If the abnormality determination result is abnormal, the influence data selection unit selects the influence data range affected by the abnormality, and selects the selected influence data range. The result is output to the determination integration unit 112.

The selection process of the influence data selection unit 111 selects the influence data range affected by this abnormality based on the abnormality attribute input from the prediction determination unit 131. The influence data selection unit 111 uses the position of the abnormality type and the location where the abnormality occurs and the measured value of the sensor as the influence data affected by the abnormality in the time zone after the occurrence time of the abnormality, and the range is set as the influence data range. Select. In other words, the influence data range depends on the regional range and time range (time range) in which the abnormality affects, so the influence data is in the regional range in which the abnormality affects the region, and the abnormality is in time. It is a measured value of the time zone (time range) that has an influence on the environment.

In the example of FIG. 4, in the prediction determination result 492 of the prediction determination unit 131 using the prediction determination method 2, the occurrence of water leakage is estimated by the DMA 340 after the determination time 480. In the influence data selection unit 111, after the determination time 480 (time zone), the sensor having a hydraulic connection relationship with the DMA 340 (in the regional range), that is, the flow rate sensor 310-311 and the pressure sensor 320- The measurement value of 322 is selected as the influence data range in which the abnormality affects.

For example, if the abnormality is estimated to be a trouble that a DMA boundary closing valve such as the valve 361 is opened, which is called DMA breach, the influence data selection unit 111 is hydraulically connected to one of the DMAs 340 and 341. All relevant sensors are selected as influence data ranges as regional ranges.

In addition, even if there is a hydraulic connection relationship, the influence data selection unit 111 excludes a sensor that is not affected by the abnormality determined by the prediction determination unit 131 by controlling the water pipe network from the influence data range. . For example, if there is a pump facility in the distribution reservoir 301 and this pump facility is controlled so that the pressure sensor 320 is set to a predetermined control target amount, the influence data selection unit 111 is based on this control information. Thus, the pressure sensor 320 determines that it is not affected by the abnormality determined by the prediction determination unit 131.

FIG. 5 is a process flowchart of the determination integration unit 112. When the determination integration unit 112 starts processing (S501), the abnormality determination result is input from the prediction determination unit 131, and the influence data range is input from the influence data selection unit 111 (S502).

The determination integration unit 112 calculates an initial value of reliability for the abnormality determination result by each prediction determination method input from the prediction determination unit 131 (S503). The reliability is to quantify the reliability of the abnormality determination result by the prediction determination method. For example, the reciprocal of the size of the section (difference between the prediction upper limit value and the prediction lower limit value) in prediction time determination is used. Can do. That is, since the prediction accuracy is poor if the section is large, the reliability is low. The initial value of reliability may be a fixed value for each prediction determination method, stored in advance in the prediction determination method data storage unit 123, and the determination integration unit 112 may read the initial value of reliability.

The determination integration unit 112 determines whether there is a result determined to be abnormal among the abnormality determination results by each prediction determination method input from the prediction determination unit 131 (S504). The judgment integration unit 112 proceeds to S505 if there is an abnormality, and proceeds to S506 if there is no abnormality.

The determination integration unit 112 determines whether there is an abnormality determination result using the influence data included in the influence data range of the abnormality determination extracted in S504 among the abnormality determination results by each prediction determination method input from the prediction determination unit 131. And the process of lowering the reliability of the abnormality determination result using the influence data is performed. In the process of reducing the reliability, for example, the reliability of the abnormality determination result using the influence data is set to 0. The degree of reliability may be calculated more precisely according to the number of measurement values as influence data used by the prediction determination unit 131 for prediction determination. Further, the determination as to whether or not the influence data is used includes not only the measurement value used for the prediction but also the measurement value used for adjusting the parameter of the prediction determination method.

The determination integration unit 112 integrates the abnormality determination result by each prediction determination method input from the prediction determination unit 131 based on the reliability (S506). As the integration method, for example, the abnormality determination result with the highest reliability is adopted. Alternatively, an integration method may be used, such as taking a majority vote of abnormality determination results having a reliability level equal to or higher than a predetermined value, and averaging the abnormality determination results after weighting with reliability.

The determination integration unit 112 outputs the integrated abnormality determination result to the output unit 152 (S507), and ends the process (S508).

In the example of FIG. 4, the prediction determination unit 131 for the measurement value 404 at the determination time 481 determines that the prediction determination result 491 using the prediction determination method 1 is normal and the prediction determination result 492 using the prediction determination method 2 is abnormal. . Since the prediction determination method 1 is determined using the most recent measurement value compared to the prediction determination method 2, the initial value of the reliability is higher in the prediction determination method 1 than in the prediction determination method 2. Is expensive.

However, the prediction determination unit 131 uses a measurement value (influence data) on which the abnormality determined by the prediction determination method 2 temporally affects the determination at the determination time 481 by the prediction determination method 1. For this reason, the determination integration unit 112 sets the reliability of the prediction determination method 1 to 0 (lowers the reliability), and adopts the result of the prediction determination method 2 with the highest reliability as a result.

FIG. 6 is a display example of the detection result of the occurrence of water leakage by the abnormality detection device 101. A method in which the output unit 152 presents the abnormality determination result to the operator will be described with reference to FIG.

The abnormality detection result window 601 displayed on the display unit 206 such as a display by the output unit 152 includes a prediction determination display panel 602 and a determination integrated display panel 603. The output unit 152 displays the abnormality determination result input from the determination integration unit 112 on each panel of the abnormality detection result window 601.

The output unit 152 displays the abnormality determination result on the prediction determination display panel 602. In FIG. 6, information corresponding to the prediction determination result 492 of FIG. 4 that is the basis for the abnormality determination is displayed. In particular, an influence data range that is determined to be abnormal and the abnormality affects temporally is highlighted as an abnormal influence time range 621. The operator operates the sensor selection box 611 and the prediction determination method selection box 612 to change the sensor and prediction determination method displayed on the prediction determination display panel 602 by the output unit 152.

The output unit 152 displays the determination result of each prediction determination method and information used for integration on the determination integrated display panel 603. In the table of the judgment integrated display panel 603, as information used for the integration, the ID of the prediction judgment method, its reliability, the judgment result of normal or abnormal, the presence or absence of use of the influence data, the abnormal part, the occurrence time of the abnormal, and The type is displayed.

With the above configuration, the abnormality detection device 101 can detect even the occurrence of small-scale water leakage based on the measured values of the sensors such as the flow rate and pressure.

In addition, an abnormality detection device that does not include a prediction determination unit that performs prediction determination by a plurality of prediction determination methods and an integrated determination unit that integrates determination results, detects the occurrence of small-scale water leakage by increasing the sensitivity of abnormality determination. As a result, misreports frequently occur that determine that a change in the measured value of the sensor within the normal range is abnormal. For this reason, only the occurrence of large-scale water leakage can be detected practically.

According to the abnormality detection device of the present embodiment, it is possible to detect even the occurrence of an abnormality such as a small-scale water leak that gives a gradual change in measured values such as flow rate and pressure.

More specifically, the abnormality detection device according to the present embodiment evaluates the reliability of the prediction determination method based on whether or not the influence data affected by the abnormality is used, and then integrates the determination results to generate a false alarm. It is possible to detect the occurrence of small-scale water leakage by keeping the ratio of low.

In addition, by detecting small-scale water leakage, the abnormality detection device of the present embodiment can detect events that increase gradually and become large-scale water leakage earlier.

FIG. 7 is a configuration diagram of the abnormality detection system 100 of the present embodiment. In addition to the configuration of the first embodiment, the abnormality detection apparatus 101 according to the present embodiment includes a method selection unit 701 that selects a prediction determination method used by the prediction determination unit 131 based on the abnormality determination result of the determination integration unit 112. The description will focus on the differences from the first embodiment.

The prediction determination unit 131 is selected by the method selection unit 701 among the prediction determination methods stored in the prediction determination method data storage unit 123 (strictly, the prediction determination method determined by the prediction determination method data). Prediction and determination are performed by a prediction determination method. In the present embodiment, the prediction determination method stored in the prediction determination method data storage unit 123 is referred to as a prediction determination method candidate.

The method selection unit 701 inputs the abnormality determination result and the influence data range from the determination integration unit 112. The method selection unit 701 reads the prediction determination method candidate list from the prediction determination method data storage unit 123, refers to the input abnormality determination result and the influence data range, and uses the prediction determination unit to use the prediction determination unit from the read candidate list A method is selected, and the selected prediction determination method is output to the prediction determination unit 131.

As time elapses after an abnormality such as occurrence of water leakage is detected, the influence data range of the abnormality expands, and the number of prediction judgment methods using the abnormality influence data increases. In the example of FIG. 4, the prediction determination unit 131 using the prediction determination method 2 does not use the influence data at the determination time 481, but the prediction determination unit 131 uses the prediction determination method 2 when more time elapses. Impact data must be used.

Specifically, the method selection unit 701 measures the data range from which the influence data (measurement value) affected by the abnormality is excluded from the prediction determination method candidates when the determination integration unit 112 determines that the abnormality is present. Select the prediction judgment method that uses the value.

For example, the method selection unit 701 selects a prediction determination method that uses a measurement value in the past time range further than the prediction determination method 2. Alternatively, the method selection unit 701 does not use the measurement value of the DMA 340 sensor, and various auxiliary data such as day of the week and weather stored in the auxiliary data storage unit 122, and the water distribution flow rate of the DMA highly correlated with the DMA 340 For example, the prediction determination method for predicting the flow rate of the flow rate sensor 311 is selected.

The criterion for abnormality determination that serves as a trigger for the method selection unit 701 to select the prediction determination method may be lower in certainty (the probability of the abnormality determination result) than the criterion for abnormality determination presented to the operator. That is, the prediction determination unit 131 provides a two-stage time interval (data range) to obtain a two-stage certainty and determines an abnormality, and the method selection unit 701 predicts using an abnormality determination result with a low certainty as a trigger. A determination method may be selected. In this case, the determination integration unit 112 outputs an abnormality determination result with a high degree of certainty to the output unit 152. When the prediction determination unit 131 provides a time interval of three or more stages to obtain the certainty level and determines an abnormality, the method selection unit 701 selects the prediction determination method when the certainty level is less than a predetermined threshold, and the certainty level is predetermined. The abnormality determination result may be output to the output unit 152 when the threshold value is equal to or greater than the threshold value.

As described above, the abnormality detection apparatus 101 according to the present embodiment selects the prediction determination method in which the method selection unit does not use the abnormality influence data, so that the degree of certainty of the abnormality determination result can be increased even when the abnormality continues for a long time. Maintained abnormality detection can be continued.

Moreover, the abnormality detection apparatus 101 of the present embodiment can suppress the processing load of the abnormality detection apparatus 101 by selectively using a prediction determination method that is effective for determining abnormality.

In addition to the second embodiment, the present embodiment is an abnormality detection apparatus 101 and an abnormality detection system 100 in which the method selection unit 701 instructs collection of measurement values.

The anomaly detection system 100 includes a sensor 191 that can collect measurement values according to an explicit instruction from the measurement value collection device 102, which corresponds to both periodic operation and demand-based operation, although the measurement value collection cycle in normal times is long. Including. As a specific example of a sensor, for example, it is a charge meter of a large-volume water consumer, and the water usage is measured and recorded in a cycle of 30 minutes, but the recorded measurement value is transmitted in a cycle of one day. It is a set smart meter.

The anomaly detection apparatus 101 uses the measurement value of the sensor corresponding to the measurement value collection instruction (demand) in the prediction determination method data storage unit 123 as a candidate for the prediction determination method for performing highly reliable prediction determination. As stored.

In response to the input of the abnormality determination result from the determination integration unit 112, the method selection unit 701 selects the above-described candidate prediction determination method (data), outputs the selected prediction determination method to the prediction determination unit 131, and the selected prediction determination method. The measurement value collection unit 151 is instructed to collect the measurement values required by. In response to the instruction from the method selection unit 701, the measurement value collection unit 151 instructs the measurement value collection device 102 to collect necessary measurement values.

For example, in the example of FIG. 4, if there is a consumer in the DMA 340 whose use of water is similar to the estimated water leakage, and the consumer's toll meter is a smart meter that corresponds to an instruction to collect measurement values, The method selection unit 701 instructs the measurement value collection device 102 to collect the water usage measured by the smart meter, and selects the prediction determination method that uses the water usage measured by the smart meter as the measurement value. In addition, use of the water similar to a water leak can be discriminate | determined by whether the diameter of the water supply pipe to a consumer is the diameter through which the flow volume equivalent to the estimated water leak flows.

According to the anomaly detection apparatus 101 of the present embodiment, for example, by collecting the measured values from the smart meter for the water leakage and the water usage of the consumer, it is possible to suppress false reports of the occurrence of water leakage.

In the abnormality detection device of the present embodiment, the prediction determination unit 131 outputs an abnormality determination result at a period according to each feature of the plurality of prediction determination methods.

For example, the prediction determination method used by the prediction determination unit 131 includes a prediction determination method for performing prediction and determination based on the nighttime flow rate. The nighttime flow rate is the minimum value of the flow rate observed in a specific time zone at midnight. Prediction / determination based on the nighttime flow rate is highly reliable, but it can be determined only once a day (the determination cycle is one day), and it takes time from the occurrence of water leakage to its detection. The abnormality determination result is also output once a day.

Therefore, the determination integration unit 112 determines that the determination period is long and the reliability of the abnormality determination result when it is determined to be abnormal by the short determination period prediction determination method corresponding to the measurement value collection period as described in the first embodiment. Therefore, the abnormality determination result is integrated so as to select the determination of abnormality without adopting the determination result of the prediction determination method having a high value, that is, the prediction determination method based on the nighttime flow rate.

According to the abnormality detection device 101 of the present embodiment, it is possible to detect the occurrence of water leakage with high reliability based on the nighttime flow rate and to detect the occurrence of water leakage in a short time from the occurrence of water leakage.

In the above embodiment, the abnormality detection device for monitoring the water pipe network has been described. However, the monitoring target is not limited to the water pipe network, and the abnormality detection device can be used for various processes in which abnormality determination based on prediction is effective. It can be applied to anomaly detection. The anomaly detection device sets the measurement items of the sensor (type of sensor) and auxiliary data to be used appropriately so that the water supply process, the water purification process, and the gas supply that supplies resources through the pipe network and pipeline Processes and chemical plant operation processes can be monitored. For example, the auxiliary data may be an operation plan or a control target value in a specific line of the plant, and various prediction judgment methods using these data may be used.

According to the described embodiment, it is possible to detect the occurrence of small-scale water leakage from measured values such as flow rate and pressure.

100: Anomaly detection system, 101: Anomaly detection device, 102: Measurement value collection device, 103: Alarm display device, 111: Influence data selection unit, 112: Determination integration unit, 121: Measurement value storage unit, 122: Auxiliary data storage Unit, 123: prediction determination method data storage unit, 131: prediction determination unit, 151: measurement value collection unit, 152: output unit, 191: sensor, 701: method selection unit.

Claims

A predicted value of the first measured value of the sensor at the determination time based on the measured value of the sensor of the monitoring target process in the data range corresponding to the multiple predicted determination methods using each of the plurality of prediction determination methods. (1) The difference between the predicted value and the first measured value at the determination time, and (2) the first measured value outside the range between the upper limit value and the lower limit value of the predicted value Based on at least one of the above, a prediction determination unit that determines an abnormality of the monitoring target process and outputs an abnormality determination result of determining the abnormality of the plurality of prediction determination methods,
An influence data selection unit that selects, as the influence data range, a range including the second measurement value affected by the abnormality included in the abnormality determination result as the influence data;
Among the abnormality determination results, a determination integration unit that lowers the reliability of the abnormality determination results based on the influence data included in the influence data range and integrates the abnormality determination results of the plurality of prediction determination methods, and An abnormality detection apparatus comprising an output unit that outputs the integrated abnormality determination result.
The abnormality detection device according to claim 1,
The abnormality detection device further comprising: a method selection unit that selects the prediction determination method used by the prediction determination unit from a plurality of the prediction determination methods based on the abnormality determination result integrated by the determination integration unit. .
The abnormality detection device according to claim 2,
The abnormality detection device, wherein the method selection unit selects the prediction determination method in which the measurement value excluding the influence data is the data range.
The abnormality detection device according to claim 3,
When the prediction determination unit determines that the monitored process is abnormal, it estimates an abnormality attribute including at least one of the occurrence time, type, and occurrence location of the abnormality,
The influence data selection unit uses the second measurement value of the sensor in a regional range in which the abnormality affects after the occurrence time of the abnormality as the influence data based on the estimated abnormality attribute. An abnormality detection device characterized by.
The abnormality detection device according to claim 4,
The prediction determination unit provides a plurality of time intervals that define the data range, determines the abnormality, and obtains a certainty degree representing the certainty of the abnormality determination result,
The method selection unit selects the prediction determination method when the certainty is less than a predetermined threshold,
The abnormality detection apparatus, wherein the determination integration unit outputs the abnormality determination result to the output unit when the certainty level is equal to or higher than the predetermined threshold.
The abnormality detection device according to claim 4,
In accordance with the selection of the prediction determination method, the method selection unit instructs the measurement value collection unit that collects the measurement values to collect the measurement values included in the data range used by the selected prediction determination method. An abnormality detection device characterized by that.
The abnormality detection device according to claim 1,
The prediction determination unit outputs a plurality of the abnormality determination results using the prediction determination method with a short determination cycle and the prediction determination method with a long determination cycle,
The determination integration unit reverses the determination result of the prediction determination method with a long determination cycle and determines the prediction determination with a short determination cycle when the abnormality is determined in the determination of the prediction determination method with a short determination cycle. An abnormality detection apparatus, wherein the abnormality determination result is integrated by selecting the abnormality determination result of a method.
The abnormality detection device according to claim 7,
The monitored process is a water distribution process, and the type of the abnormal attribute includes occurrence of water leakage,
The sensor includes any one of a flow meter and a pressure gauge,
The influence data selection unit uses the measurement value of the sensor in the regional range from the hydraulic connection relationship of the water distribution process as the influence data,
The abnormality determination device, wherein the prediction determination unit includes a determination that the occurrence of water leakage is abnormal based on a night water distribution flow rate.
The abnormality detection device according to claim 1,
When the prediction determination unit determines that the monitored process is abnormal, it estimates an abnormality attribute including at least one of the occurrence time, type, and occurrence location of the abnormality,
The influence data selection unit sets the second measurement value of the sensor in the regional range affected by the abnormality after the occurrence time of the abnormality as the influence data based on the estimated abnormality attribute. An abnormality detection device characterized by.
The abnormality detection device according to claim 9,
The monitored process is a water distribution process, and the type of the abnormal attribute includes occurrence of water leakage,
The sensor includes any one of a flow meter and a pressure gauge,
The said influence data selection part makes the said measured value of the said sensor in the said regional range the said influence data from the hydraulic connection relation of the said water distribution process, The abnormality detection apparatus characterized by the above-mentioned.
The abnormality detection device according to claim 1,
The abnormality determination apparatus, wherein the prediction determination unit is configured such that the data range is a time zone that is a predetermined time before the determination time, and each of the plurality of prediction determination methods has the predetermined time different from each other. .
An anomaly detection system that monitors processes,
A measurement value collection device for collecting measurement values of sensors installed in the process, and
Using each of the plurality of prediction determination methods, the predicted value of the first measurement value of the sensor at the determination time is determined based on the measurement value of the sensor of the process in the data range corresponding to the plurality of prediction determination methods. Predicting, (1) a difference between the predicted value and the first measured value at the determination time, and (2) the first measured value outside the range between the upper limit value and the lower limit value of the predicted value, A prediction determination unit that determines abnormality of the process based on at least one of the plurality of prediction determination methods and outputs an abnormality determination result of determining the abnormality of the plurality of prediction determination methods;
An influence data selection unit that selects, as the influence data range, a range including the second measurement value affected by the abnormality included in the abnormality determination result as the influence data;
Among the abnormality determination results, a determination integration unit that lowers the reliability of the abnormality determination results based on the influence data included in the influence data range and integrates the abnormality determination results of the plurality of prediction determination methods, and An abnormality detection system comprising an output unit for outputting the integrated abnormality determination result.
An anomaly detection method by an anomaly detector that monitors a process, the anomaly detector is
A predicted value of the first measured value of the sensor at the determination time based on the measured value of the sensor of the monitoring target process in the data range corresponding to the multiple predicted determination methods using each of the plurality of prediction determination methods. Predict
(1) at least one of the difference between the predicted value and the first measured value at the determination time; and (2) the first measured value outside the range between the upper limit value and the lower limit value of the predicted value. To determine whether the monitored process is abnormal,
A range including the second measurement value affected by the abnormality including the abnormality determination result of determining the abnormality of the plurality of prediction determination methods as the influence data is selected as the influence data range,
Of the abnormality determination results, lowering the reliability of the abnormality determination results based on the influence data included in the influence data range, and integrating the abnormality determination results of a plurality of the prediction determination methods,
An abnormality detection method comprising outputting the integrated abnormality determination result.