JP2021102184A - Abnormality determination method for immersion type membrane separation device and abnormality determination apparatus for immersion type membrane separation device - Google Patents

Abstract

To provide an abnormality determination method for an immersion type membrane separation device which can suitably determine whether or not abnormality occurs by taking into account a history of the immersion type membrane separation device that changes over time.SOLUTION: An abnormality determination method for an immersion type membrane separation device includes: modeling processing for previously generating a time-series model with a transmembrane pressure difference as a characteristic value of the immersion type membrane separation device on the basis of the transmembrane pressure difference and an actual measurement value of a permeate flow rate of the immersion type membrane separation device for a prescribed period that change over time; and abnormality determination processing for determining whether or not abnormality occurs in the immersion type membrane separation device on the basis of comparison between a characteristic value calculated by the time-series model from the characteristic value and the actual measurement value of the permeate flow rate with a characteristic value actually measured by the immersion type membrane separation device.SELECTED DRAWING: Figure 2

Description

The present invention relates to an abnormality determination method for an immersion type membrane separation device and an abnormality determination device for an immersion type membrane separation device, and is suitable for, for example, an immersion type membrane separation device used in a biological treatment for purifying organic wastewater using microorganisms. The present invention relates to an abnormality determination method for an immersion type membrane separation device and an abnormality determination device for an immersion type membrane separation device.

Patent Document 1 discloses a remote monitoring device for a membrane installation facility using an inexpensive general-purpose personal computer. The remote monitoring device of the membrane installation facility detects predetermined data that is an index of membrane contamination of the membrane separation device immersed in the treatment tank by the membrane contamination detecting means and displays it on the display means of the remote monitoring device. The display means is configured to start and stop in conjunction with the start and stop of intermittent filtration of the membrane separation device. As the membrane contamination detecting means, a pressure detecting means for detecting the pressure on the suction side of the drainage pump and a liquid level detecting means for detecting the liquid level of the liquid to be treated in the tank are used.

Patent Document 2 proposes a remote monitoring and control system that enables the entire membrane separation device to be continuously monitored and controlled in an optimum state by remotely monitoring and controlling the air diffuser of the membrane separation device. The remote monitoring and control system is a local data storage device that captures and stores the operation data of the membrane separation device, a communication device that transmits and receives the stored operation data, a calculation device that receives and calculates the operation data transmitted from the communication device, and a calculation. It is equipped with a control device that controls the membrane separation device based on the obtained data, and is configured to remotely control the operation of the air diffuser. For example, the air diffuser is provided with a remotely controllable pressure gauge, and is configured to wash the air diffuser when the obtained pressure rises by 3 kPa or more from the initial value.

Japanese Unexamined Patent Publication No. 10-314733 Japanese Unexamined Patent Publication No. 2015-231591

However, in the conventional remote monitoring device disclosed in Patent Document 1, the pressure detecting means for detecting the pressure on the suction side of the drainage pump and the liquid level detecting means for detecting the liquid level of the liquid to be treated in the tank are used instantaneously. It was only possible to monitor the degree of membrane contamination based on the output, and it was not possible to evaluate the appropriate degree of contamination that reflected the operating status of the membrane separation device.

Further, the remote monitoring and control system disclosed in Patent Document 2 merely cleans the air diffuser when the pressure loss value of the separation membrane becomes high, and cannot appropriately determine the state of the separation membrane. It was.

An object of the present invention is an abnormality determination method for an immersion type membrane separation device capable of appropriately determining the presence or absence of an abnormality in consideration of the history of the immersion type membrane separation device that changes over time in view of the above-mentioned prior art. The point is to provide an abnormality determination device for an immersion type membrane separation device.

In order to achieve the above object, the first characteristic configuration of the abnormality determination method of the immersion type membrane separation device according to the present invention is the abnormality determination method of the immersion type membrane separation device, which is the above-mentioned method for a predetermined period of time, which changes in time series. Based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device, a modeling process for generating in advance a time-series model with the intermembrane differential pressure of the immersion type membrane separation device as the characteristic value, and the characteristic value Whether or not an abnormality has occurred in the immersion type membrane separation device based on the comparison between the characteristic value calculated by the time series model from the actual measurement value of the permeate flow rate and the characteristic value actually measured by the immersion type membrane separation device. The point is that it includes an abnormality determination process for determining whether or not it is present.

The time series model is a model that analyzes the time fluctuation of the phenomenon and predicts the future on the assumption that the observed values at each time are influenced from the previous time point and are not independent data. In the modeling process, a time-series model in which the intermembrane differential pressure of the immersion type membrane separation device is used as a characteristic value based on the intermembrane differential pressure of the immersion type membrane separation device and the permeate flow rate that change over time is used. Will be generated. Then, after the time series model is generated, the abnormality determination process is performed. In the abnormality determination process, it is determined whether or not the state is abnormal based on the comparison between the characteristic value actually measured by the immersion type membrane separation device and the characteristic value calculated by the time series model, that is, the predicted value. If there is a difference between the characteristic value calculated by the time series model and the actual characteristic value, it can be determined with high probability that a factor not assumed at the time of generating the time series model has some influence.

The second characteristic configuration is a method for determining an abnormality of the immersion type membrane separation device, which is based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series. Machine learning was performed to generate a time-series model in advance using the intermembrane differential pressure of the immersion type membrane separation device as a characteristic value, and the time-series model was calculated from the characteristic values and the measured values of the permeate flow rate. The immersion type membrane is subjected to a comparison calculation process for comparing the characteristic value and the characteristic value actually measured by the immersion type membrane separation device in time series, and whether or not the evaluation value calculated by the comparison calculation process is equal to or higher than a predetermined threshold value. The point is that it includes an abnormality determination process for determining whether or not an abnormality has occurred in the separation device.

In the modeling process, a time-series model in which the intermembrane differential pressure of the immersion type membrane separation device is used as a characteristic value is created by machine learning using the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period as a teacher signal. Will be generated. In the comparison calculation process, the characteristic value calculated by the time series model, that is, the predicted value and the actually measured characteristic value are compared, and when the evaluation value exceeds a predetermined threshold value, an abnormality has occurred in the immersion type membrane separation device. Is determined.

In the third feature configuration, in addition to the second feature configuration described above, the characteristic values actually measured by the immersion type membrane separation device are normal corresponding to each time when the evaluation value is calculated by the comparison calculation process. Alternatively, the characteristic value at each time of the time-series model is regarded as abnormal or normal based on whether the evaluation value calculated by the comparison calculation process is equal to or higher than the provisional threshold and the labeling process for labeling any of the abnormalities. Calculated for a tentative determination process for tentative determination, a confusion matrix generation process for generating a binary confusion matrix based on the results of the labeling process and the tentative determination process, and a binary confusion matrix generated by the matrix generation process. A matrix setting process for obtaining a tentative threshold that maximizes the F value as the predetermined threshold, and
Is set at the point where the predetermined threshold value is set.

The above-mentioned predetermined threshold value can be appropriately obtained by performing an F value operation using a binary confusion matrix. First, in the comparison calculation process, the characteristic value calculated by the time series model, that is, the predicted value and the actually measured characteristic value are compared. If it can be determined that an abnormality has occurred in the immersion type membrane separation device based on the evaluation value, the measured value should be labeled as an abnormality, and an abnormality has occurred in the immersion type membrane separation device based on the evaluation value. If it can be determined that it is normal, the measured value is labeled as normal. Next, the characteristic value of the time-series model corresponding to the case where the evaluation value is equal to or more than the provisional threshold is tentatively determined as abnormal, and the characteristic value of the time-series model corresponding to the case where the evaluation value is less than the provisional threshold is normal. Tentatively determined. In other words, the tentative judgment can be said to be the labeling process of the characteristic values of the time series model. Based on such a processing result, a binary confusion matrix is generated and an F value is calculated. A temporary threshold value is set to a value different from the previous value, a plurality of similar processes are performed, and a temporary threshold value that maximizes the calculated F value is set as a predetermined threshold value. The F value is a harmonic mean (F = 2 · R · P / (R + P)) of the detection rate R (Recall) and the precision rate P (Precision), and both evaluation values are at the maximum F value. It is a high equilibrium value.

In the fourth feature configuration, in addition to the above-mentioned second or third feature configuration, a notification process for notifying when it is determined that an abnormality has occurred by the abnormality determination process continuously for a predetermined time is performed. It is in the point of execution.

When the evaluation value is continuously equal to or higher than the predetermined threshold value by the abnormality determination process, the alarm processing is performed to prompt a specific action.

The fifth feature configuration is that, in addition to any of the first to fourth feature configurations described above, the time series model is a state space model.

The state space model is a statistical model in which a time series is represented by two equations called a state equation and an observation equation.

The first characteristic configuration of the abnormality determination device of the immersion type membrane separation device according to the present invention is the abnormality determination device of the immersion type membrane separation device, which changes between the membranes of the immersion type membrane separation device for a predetermined period of time. A modeling processing unit that performs machine learning based on the measured values of the differential pressure and the permeate flow rate and pre-generates a time-series model using the intermembrane differential pressure of the immersion type membrane separation device as the characteristic value, and the characteristic value and permeation. Calculated by the comparison calculation processing unit and the comparison calculation processing unit that compare the characteristic value calculated by the time series model from the measured value of the liquid flow rate and the characteristic value measured by the immersion type membrane separation device in time series. The point includes an abnormality determination processing unit for determining whether or not an abnormality has occurred in the immersion type membrane separation device depending on whether or not the evaluation value is equal to or higher than a predetermined threshold value.

In the second feature configuration, in addition to the first feature configuration described above, the characteristic values actually measured by the immersion type membrane separation device are obtained corresponding to each time when the evaluation value is calculated by the comparison calculation processing unit. The characteristic value of each time of the time series model is abnormal based on whether the evaluation value calculated by the labeling processing unit that labels either normal or abnormal and the evaluation value calculated by the comparison calculation processing unit is equal to or higher than a tentative threshold value. Alternatively, a tentative determination processing unit for tentatively determining normal, a confusion matrix generation processing unit for generating a binary confusion matrix based on the results of the labeling processing unit and the tentative determination processing unit, and two generated by the matrix generation processing unit. The point is that it includes a threshold value setting processing unit for obtaining a tentative threshold value at which the F value calculated for the value confusion matrix is maximized as the predetermined threshold value.

In the third feature configuration, in addition to the first or second feature configuration described above, a notification process for notifying that an abnormality has occurred is continuously determined by the abnormality determination processing unit for a predetermined time. The point is that it has a part.

As described above, according to the present invention, an abnormality determination method and an immersion type of an immersion type membrane separation device capable of appropriately determining the presence or absence of an abnormality in consideration of the history of the immersion type membrane separation device that changes over time. It has become possible to provide an abnormality determination device for a membrane separation device.

Explanatory drawing of immersion type membrane separation apparatus Functional block configuration diagram of the abnormality determination device of the immersion type membrane separation device (A) is a characteristic diagram of predicted value and measured value based on a time series model, (b) is a characteristic diagram of difference, and (c) is a characteristic diagram binarized with a predetermined threshold value. Explanatory diagram of correlation between threshold value and F value Explanatory diagram of the state space model Explanatory diagram of binary confusion matrix

The abnormality determination method of the immersion type membrane separation device and the abnormality determination device of the immersion type membrane separation device according to the present invention will be described below. The abnormality determination method and device are abnormality determination methods and devices that enable abnormality determination of the immersion type membrane separation device at a computer terminal installed at a remote location of a wastewater treatment facility where the immersion type membrane separation device is installed. Preferably, the measurement data such as the intermembrane differential pressure and the permeated water flow rate for the immersion type membrane separation device installed in the wastewater treatment facility are collected by the cloud server via the communication device, and the computer terminal connected to the cloud server. It can be configured to determine an abnormality in the immersion type membrane separation device. The computer terminal is not limited to the mode in which it is installed in a remote place, but is an abnormality determination method and device that enables an abnormality determination of the immersion type membrane separation device in the facility with the computer terminal installed in the wastewater treatment facility. You may.

[Explanation of immersion type membrane separation device]
As shown in FIG. 1, the immersion type membrane separation device 3 is immersed and arranged in the membrane separation tank 1 of a wastewater treatment facility adopting the membrane separation active sludge method, and flat membrane type filtration is performed on both sides of a flat filter plate. It is provided with a membrane module in which a plurality of membrane elements on which membranes are arranged are arranged in a vertical posture at predetermined intervals. Examples of the filtration membrane include an organic filtration membrane in which a support made of a non-woven fabric made of PET is coated and impregnated with a porous resin to form a microporous membrane having an average pore size of about 0.2 μm.

By operating the air diffuser 4 and the pump 5 installed below the immersion type membrane separation device 3, the water to be treated in the membrane separation tank 1 is filtered by the immersion type membrane separation device 3 and the permeated water at a set flow rate. In order to prevent fouling of the filtration membrane of the immersion type membrane separation device 3 while maintaining the properties of the active sludge in the tank, for example, when the suction pressure is increased periodically or during the filtration operation. , A relaxation operation is executed in which only the air diffuser 4 is operated with the pump 5 stopped.

Further, when the suction pressure becomes high periodically or after the relaxation operation, a chemical solution cleaning step of injecting a chemical solution from the secondary side of the immersion type membrane separation device 3 to clean the immersion type membrane separation device 3 is executed. Air is supplied to the air diffuser 4 from the blower fan B. A pressure gauge P for measuring the differential pressure between membranes and a flow meter Q for measuring the flow rate of the permeate are arranged in the conduit between the immersion type membrane separation device 3 and the pump 5.

[Explanation of abnormality determination device]
As shown in FIG. 2, the abnormality determination device 10 of the immersion type membrane separation device according to the present invention is a machine that uses the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series as teacher signals. The modeling processing unit 11 that performs learning and generates a time-series model in advance using the intermembrane differential pressure of the immersion type membrane separation device as the characteristic value, and the characteristic value and immersion calculated by the time-series model generated by the modeling processing unit 11. Immersion type membrane separation depending on whether the evaluation value calculated by the comparison calculation processing unit 12 and the comparison calculation processing unit 12 that calculates the difference from the characteristic values actually measured in time series of the type membrane separation device is equal to or higher than a predetermined threshold value. An abnormality determination processing unit 13 that determines whether or not an abnormality has occurred in the device and an abnormality determination processing unit 13 that continuously determines that an abnormality has occurred for a predetermined time set in advance issues a notification. The information processing unit 14 is provided.

The time series model is a model that analyzes the time fluctuation of the phenomenon and predicts the future on the assumption that the observed values at each time are influenced from the previous time point and are not independent data. In the modeling process, when the intermembrane differential pressure of the immersion type membrane separation device is used as the characteristic value based on the measured values of the intermembrane differential pressure of the immersion type membrane separation device and the permeate flow rate that change over time. A series model is generated. Then, after the time series model is generated, the abnormality determination process is performed. In the abnormality determination process, whether or not the immersion type membrane separation device is in an abnormal state based on the difference between the characteristic value actually measured in time series and the characteristic value calculated by the time series model in consideration of the measured value of the permeate flow rate. Is determined. If there is a difference between the characteristic value calculated by the time series model and the actual characteristic value, it can be determined with high probability that a factor not assumed at the time of generating the time series model has some influence.

As shown in FIG. 5, the actual state of the immersion type membrane separation device at each time (t1, t2, t3, ...), For example, the state variables x1, x2, x3, ... It changes under the influence of the previous state. The state can be observed with the observed values y1, y2, y3, ...

The equation of state can be expressed by the following equation.
x _{t + 1} = _ft (x _t , v _t ), v _t can be expressed by a function that depends on the state of the t period in the random variable t + 1 period and the random variable v _{t in which some explanatory variable and noise are added.}
The observation equation can be expressed by the following mathematical formula.
y _t = g _t (x _t , w _t ), w _t is a random variable

A state space model can be obtained by estimating the state variable x based on the above observation equation. That is, a time-series model, which is a state-space model, can be obtained by machine learning the intermembrane differential pressure and the permeate flow rate within a predetermined period sampled in time series as a teacher signal. Hereinafter, in the present embodiment, in order to perform the filtration operation of the immersion type membrane separation device 3 under the condition that the flow rate of the permeate is constant, an example of generating a time series model in which the differential pressure between the membranes is a characteristic value will be described. When the filtration operation is performed under a constant differential pressure, it is also possible to generate a time-series model in which the permeate flow rate is a characteristic value.

In FIG. 3A, the characteristic values calculated by the time-series model generated by the modeling processing unit 11 and the actually measured intermembrane differential pressure of the immersion type membrane separation device are plotted so as to overlap. .. The characteristic values calculated by the time series model are shown by broken lines, and the measured values are shown by solid lines. The area where both are superimposed is recognized as a solid line.

FIG. 3 (b) shows the difference calculated by the comparison calculation processing unit 12, and FIG. 3 (c) shows the characteristic of binarizing the difference of FIG. 3 (b) with a predetermined threshold value. ing. It is 1 when the evaluation value is larger than the predetermined threshold value, and 0 when the evaluation value is less than the predetermined threshold value.

It can be seen that the reliability of the abnormality determination depends on how the predetermined threshold value is set. Therefore, the abnormality determination device 10 is provided with a calculation unit for determining an appropriate threshold value.

That is, the labeling processing unit in which a person labels the characteristic values of the time series actually measured by the immersion type membrane separation device as either normal or abnormal in advance corresponding to each time when the difference is calculated by the comparison calculation processing unit 15. 16 and the provisional determination processing unit 17 that provisionally determines that the characteristic value of the time-series model is abnormal or normal based on whether or not the difference calculated by the comparison calculation processing unit 12 is equal to or greater than the provisional threshold, and the labeling processing unit. The maximum F value calculated for the confusion matrix generation processing unit 18 that generates a binary confusion matrix based on the results of 16 and the tentative determination processing unit 17 and the binary confusion matrix generated by the confusion matrix generation processing unit 18. It is provided with a threshold setting processing unit 19 for obtaining a temporary threshold value as a predetermined threshold value.

First, the comparison calculation processing unit 12 calculates the difference between the characteristic value calculated by the time series model and the actually measured characteristic value. The labeling processing unit 16 labels the actually measured value as an abnormality when it can be determined by a person that an abnormality has occurred in the immersion type membrane separation device based on the actually measured value at the time when the difference occurs.

Further, if a person can determine that the immersion type membrane separation device is normal without any abnormality based on the measured value at the time when the difference occurs, the measured value is labeled as normal.

Next, the tentative determination processing unit 17 tentatively determines the characteristic value of the corresponding time series model as abnormal when the evaluation value is equal to or higher than the tentatively set threshold value, and when the evaluation value is less than the tentative threshold value, Temporarily determine that the characteristic value of the corresponding time series model is normal. In other words, the tentative judgment process can be said to be a labeling process for the characteristic values of the time series model.

Based on such a processing result, a binary confusion matrix is generated and an F value is calculated.
FIG. 6 shows a binary confusion matrix. The area judged to be abnormal by both the measured value and the time series model (state space model) is TP (True Positive), and the area judged to be normal by both the measured value and the time series model (state space model) is TN ( True Negative), the measured value is abnormal, the area judged to be normal by the time series model (state space model) is judged to be normal by FN (False Negative), the measured value is normal, and the time series model (state space model) is judged to be abnormal. The area becomes FP (False Positive).

In order to improve the judgment accuracy, the detection rate R (Recall: R = TP / (TP + FN)), which correctly estimates abnormal data as abnormal without missing data, and the data classified as abnormal are actually abnormal. It is desirable that both the precision ratio P (Precision: P = TP / (TP + FP)), which is the ratio of the data, is high.

Therefore, the threshold value setting processing unit 19 calculates an F value (F = 2 · R · P / (R + P)) which is a harmonic mean of the detection rate R and the precision rate P when the threshold value is changed in various ways, and F. It is configured to set a threshold value when the value of the value becomes maximum as a predetermined threshold value.

FIG. 4 shows a characteristic diagram in which the threshold value is plotted on the horizontal axis, the F value, the detection rate R, and the precision rate P are plotted on the vertical axis. In the example of FIG. 3B, the F value becomes maximum when the predetermined threshold value is set to “6”.

The state space model is a statistical model in which a time series is represented by two equations called a state equation and an observation equation.

Adopting the intermembrane differential pressure of the immersion type membrane separation device as a characteristic value is preferable in that the state of membrane clogging of the immersion type membrane separation device can be appropriately detected. However, as described above, it is also possible to adopt the permeate flow rate of the immersion type membrane separation device as a characteristic value.

As described above, the method for determining an abnormality of the immersion type membrane separation device according to the present invention is based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series. Modeling process to generate a time-series model in advance with the intermembrane differential pressure of the immersion type membrane separation device as the characteristic value, the characteristic value calculated by the time-series model from the characteristic value and the measured value of the permeate flow rate, and the immersion type membrane separation It includes an abnormality determination process for determining whether or not an abnormality has occurred in the immersion type membrane separation apparatus based on a comparison between the characteristic value actually measured by the apparatus.

In addition, machine learning is performed using the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change over time as a teacher signal, and the intermembrane differential pressure of the immersion type membrane separation device is characterized. Compare the modeling process that generates a time-series model as a value in advance with the characteristic value calculated by the time-series model from the characteristic value and the measured value of the permeate flow rate and the characteristic value actually measured by the immersion type membrane separation device in time series. It includes a comparison calculation process and an abnormality determination process for determining whether or not an abnormality has occurred in the immersion type membrane separation device depending on whether or not the evaluation value calculated by the comparison calculation process is equal to or higher than a predetermined threshold value.

A labeling process that labels the characteristic values measured by the immersion type membrane separation device to either normal or abnormal corresponding to each time when the evaluation value is calculated by the comparison calculation process, and an evaluation value calculated by the comparison calculation process. A tentative judgment process for tentatively determining the characteristic value of the time series model as abnormal or normal based on whether or not is equal to or higher than the tentative threshold value, and a binary confusion matrix are generated based on the results of the labeling process and the tentative judgment process. By executing the confusion matrix generation process and the threshold value setting process for obtaining a tentative threshold value that maximizes the F value calculated for the binary confusion matrix generated in the matrix generation process as a predetermined threshold value, a predetermined threshold value can be obtained. Set.

In the above-described embodiment, the difference value between the characteristic value calculated by the time series model and the measured characteristic value is used as the evaluation value for determining the abnormality of the immersion type membrane separation device 3, but the evaluation value is In addition, it may be the ratio of two characteristic values or the difference value of the rate of change of the two characteristic values.

In the above-described embodiment, the case where the alarm processing unit 14 issues a report when the abnormality determination processing unit 13 determines that an abnormality has occurred continuously for a predetermined time has been described, but the abnormality is reported for a predetermined period. An upper limit may be set for the number of times of abnormality determination in the above and the cumulative value of the amount exceeding the threshold value, and a notification may be issued when the upper limit value is exceeded.

In the above-described embodiment, the case where the immersion type membrane separation device is a flat membrane type membrane separation device used in a sewage treatment facility in which the membrane separation active sludge method is adopted has been described, but the present invention is not limited to the flat membrane. It can also be applied to hollow fiber membranes and porous ceramic membranes.

The present invention can also be applied to an immersion type membrane separation device used in a water treatment facility such as a water purification plant, a sewage treatment plant, and an industrial water distribution treatment plant.

It goes without saying that the above-described embodiment is merely an example of the present invention, and the specific configuration of each functional block can be appropriately modified and designed within the range in which the action and effect according to the present invention are exhibited.

1: Membrane separation tank 3: Immersion type membrane separation device 4: Air diffuser 5: Pump 10: Abnormality determination device 11: Modeling processing unit 12, 15: Comparison calculation processing unit 13: Abnormality determination processing unit 14: Notification processing unit 16: Labeling processing unit 17: Temporary judgment processing unit 18: Confusion matrix generation processing unit 19: Threshold setting processing unit

Claims (8)

This is a method for determining abnormalities in the immersion type membrane separation device.
Based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series, a time series model in which the intermembrane differential pressure of the immersion type membrane separation device is used as a characteristic value is created. Modeling process generated in advance and
An abnormality occurred in the immersion type membrane separator based on a comparison between the characteristic value calculated by the time series model from the measured value of the characteristic value and the permeate flow rate and the characteristic value actually measured by the immersion type membrane separation device. Abnormality judgment processing to determine whether or not it is
Abnormality determination method of the immersion type membrane separation device including. This is a method for determining abnormalities in the immersion type membrane separation device.
Machine learning is performed based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series, and the intermembrane differential pressure of the immersion type membrane separation device is used as a characteristic value. Modeling process to generate a time series model in advance
Comparison calculation processing that compares the characteristic value calculated by the time series model from the measured value of the characteristic value and the permeate flow rate with the characteristic value actually measured by the immersion type membrane separation device in time series.
An abnormality determination process for determining whether or not an abnormality has occurred in the immersion type membrane separation device based on whether or not the evaluation value calculated by the comparison calculation process is equal to or higher than a predetermined threshold value.
Abnormality determination method of the immersion type membrane separation device including. A labeling process that labels the characteristic values actually measured by the immersion type membrane separation device to either normal or abnormal corresponding to each time when the evaluation value is calculated by the comparison calculation process.
A tentative determination process for tentatively determining the characteristic value at each time of the time series model as abnormal or normal based on whether or not the evaluation value calculated by the comparison calculation process is equal to or higher than the tentative threshold value.
A confusion matrix generation process that generates a binary confusion matrix based on the results of the labeling process and the tentative determination process, and
A threshold setting process for obtaining a tentative threshold value at which the F value calculated for the binary confusion matrix generated in the matrix generation process is maximized as the predetermined threshold value, and
The method for determining an abnormality of the immersion type membrane separation device according to claim 2, wherein the predetermined threshold value is set by executing the above method. The abnormality determination method for an immersion type membrane separation device according to claim 2 or 3, wherein an alarm processing is executed to notify when it is determined that an abnormality has occurred by the abnormality determination processing continuously for a predetermined time. The method for determining an abnormality in the immersion type membrane separation device according to any one of claims 1 to 4, wherein the time series model is a state space model. It is an abnormality judgment device for the immersion type membrane separation device.
Machine learning is performed based on the measured values of the intermembrane differential pressure and the permeate flow rate of the immersion type membrane separation device for a predetermined period that change in time series, and the intermembrane differential pressure of the immersion type membrane separation device is used as a characteristic value. A modeling processing unit that generates a time-series model in advance,
A comparison calculation processing unit that compares the characteristic value calculated by the time-series model from the measured value of the characteristic value and the permeate flow rate with the characteristic value actually measured by the immersion type membrane separation device in time series.
An abnormality determination processing unit that determines whether or not an abnormality has occurred in the immersion type membrane separation device based on whether or not the evaluation value calculated by the comparison calculation processing unit is equal to or higher than a predetermined threshold value.
Abnormality determination device for immersion type membrane separation device including. A labeling processing unit that labels the characteristic values actually measured by the immersion type membrane separation device to either normal or abnormal corresponding to each time when the evaluation value is calculated by the comparison calculation processing unit.
A provisional determination processing unit that provisionally determines whether the characteristic value at each time of the time series model is abnormal or normal based on whether or not the evaluation value calculated by the comparison calculation processing unit is equal to or higher than the provisional threshold value.
A confusion matrix generation processing unit that generates a binary confusion matrix based on the results of the labeling processing unit and the provisional determination processing unit, and
A threshold setting processing unit that obtains a tentative threshold value at which the F value calculated for the binary confusion matrix generated by the matrix generation processing unit is maximum as the predetermined threshold value.
6. The abnormality determination device for the immersion type membrane separation device according to claim 6. The abnormality determination device for the immersion type membrane separation device according to claim 7 or 8, further comprising an alarm processing unit that notifies when the abnormality determination processing unit determines that an abnormality has occurred continuously for a predetermined time.

Images