JP7767246B2 - Information processing device, moisture content prediction method, and moisture content prediction program - Google Patents

Description

The present invention relates to a technology for predicting the moisture content of dehydrated cake obtained by dehydrating a liquid containing suspended solids in a dehydrator.

Sludge treatment carried out at wastewater treatment facilities such as sewage treatment plants includes a process of dehydrating the sludge using a dehydrator. For efficient sludge treatment, it is important to maintain the moisture content of the dehydrated cake obtained by dehydration within a specified range. However, when dehydration is performed under constant operating conditions for the dehydrator, the moisture content of the dehydrated cake fluctuates due to factors such as the inconsistency of the properties of the supplied sludge, making it difficult to maintain the moisture content of the dehydrated cake within the specified range.

For this reason, efforts have been made to develop technology for predicting the moisture content of dehydrated cake. If the moisture content can be predicted, it will be possible to maintain the moisture content within a predetermined range through feedforward control. For example, Patent Document 1 below discloses a technology for estimating the moisture content by creating a moisture content estimation model using multiple parameters, such as the amount of sludge supplied to a centrifugal dehydrator and values related to the centrifugal effect of the dehydrator.

Japanese Patent Application Laid-Open No. 2020-114569

Generally, the sludge treatment equipment is test-run under a wide variety of operating conditions to determine the optimal operating conditions. During this test run, the moisture content prediction model is generated using the wide variety of measured values of the multiple parameters measured under the wide variety of operating conditions as training data.

The device is then operated based on the determined operating conditions. During this operating period, the moisture content prediction model is updated using the most recent measured values of the multiple parameters as training data. This results in the moisture content prediction model being adapted to the most recent operating conditions. Therefore, under normal circumstances where there are no changes to the operating conditions, the accuracy of predictions made by the moisture content prediction model is maintained. However, there is a concern that the accuracy of predictions made by the moisture content prediction model may decrease if the operating conditions change.

In this regard, it is possible to update the moisture content prediction model using at least some of the wide variety of measurement values and the most recent measurement values as training data. However, the influence of the wide variety of measurement values will normally result in a decrease in the accuracy of predictions made by the moisture content prediction model.

One aspect of the present invention aims to mitigate any decline in the accuracy of the moisture content prediction when operating conditions change, while maintaining the accuracy of the prediction under normal circumstances.

In order to solve the above problem, an information processing device according to one aspect of the present invention includes an acquisition unit that acquires at least one of measurement data related to a liquid supplied to a coagulation tank in which a chemical agent for coagulating suspended solids is added to the liquid containing the suspended solids, measurement data related to the chemical agent supplied to the coagulation tank, measurement data related to the liquid in the coagulation tank, measurement data related to the operation of the coagulation tank, and measurement data related to the operation of a dehydrator that transports and dehydrates the liquid discharged from the coagulation tank; and a first prediction model that uses the measurement data acquired by the acquisition unit to predict the amount of water discharged from the dehydrator at a time when the residence time of the liquid in the dehydrator has elapsed since the time of measurement of the measurement data. and a prediction unit that predicts the post-dehydration moisture content at a time when the retention time has elapsed from the time when the measurement data was measured, based on a predicted value by a second prediction model of the post-dehydration moisture content, which is the moisture content of the dehydrated cake, the measurement data used for learning by the first prediction model is the measurement data acquired by the acquisition unit from a predetermined period before the learning time during the actual operation period of the dehydrator to the learning time, and the second prediction model is a prediction model trained using the measurement data acquired by the acquisition unit during an operation period in which operating conditions are different from the actual operation period as explanatory variables and the post-dehydration moisture content at a time when the retention time has elapsed from the time when the measurement data was measured as a target variable.

Furthermore, a moisture content prediction method according to one aspect of the present invention is a moisture content prediction method executed by one or more information processing devices, and includes an acquisition step of acquiring at least one of measurement data regarding a liquid supplied to a coagulation tank to which a chemical agent for coagulating suspended solids is added to the liquid containing the suspended solids, measurement data regarding the chemical agent supplied to the coagulation tank, measurement data regarding the liquid in the coagulation tank, measurement data regarding the operation of the coagulation tank, and measurement data regarding the operation of a dehydrator that dehydrates the liquid while transporting it discharged from the coagulation tank; and a first prediction model that uses the measurement data acquired in the acquisition step and a time point at which the residence time of the liquid in the dehydrator has elapsed since the time of measurement of the measurement data. and a prediction step of predicting the post-dehydration moisture content at a time when the retention time has elapsed from the time when the measurement data was measured, based on a predicted value by a second prediction model of the post-dehydration moisture content, which is the moisture content of the dehydrated cake discharged from the dehydrator. The measurement data used by the first prediction model for learning is the measurement data acquired in the acquisition step from a predetermined period before the learning time during the actual operation period of the dehydrator to the learning time, and the second prediction model is a prediction model trained using the measurement data acquired in the acquisition step during an operating period in which operating conditions are different from the actual operation period as an explanatory variable and the post-dehydration moisture content at a time when the retention time has elapsed from the time when the measurement data was measured as a target variable.

According to one aspect of the present invention, it is possible to mitigate a decrease in the accuracy of the moisture content prediction when operating conditions change, while maintaining the accuracy of the prediction under normal circumstances.

1 is a block diagram showing an example of a main configuration of an information processing device according to an embodiment of the present invention; FIG. 10 is a diagram illustrating an example of the configuration of a control system including the information processing device. 10 is a flowchart showing an example of a process for predicting the moisture content of a dehydrated cake in the information processing device. 10 is a flowchart illustrating an example of a process of updating a first prediction model in the information processing device. FIG. 10 is a diagram illustrating, in a table format, combinations of explanatory variables and dependent variables of a first prediction model in the information processing device according to the first embodiment. FIG. 10 is a diagram showing, in a table format, combinations of explanatory variables and dependent variables of the second prediction model in Example 1. FIG. 10 is a diagram showing, in tabular form, combinations of explanatory variables and response variables of a fourth comparative model in Comparative Example 4. 10 is a graph showing the variation in moisture content predicted by the first prediction model of Example 1 relative to the moisture content actually measured by the worker. 10 is a graph showing the variation in moisture content predicted values by the first to fourth comparative models in Comparative Examples 1 to 4 relative to the moisture content actually measured by the operator. FIG. 10 is a diagram showing, in tabular form, the mean absolute error, the operating time parameter importance, and the sludge flow rate parameter importance for each of Example 1 and Comparative Examples 1 to 4.

[System Configuration]
The configuration of a control system according to one embodiment of the present invention will be described with reference to Figure 2. Figure 2 is a diagram showing an example of the configuration of a control system 100. The control system 100 is a system used in a plant that adds an agent that coagulates suspended solids to a liquid to be treated in a coagulation tank to form flocs, and then performs solid-liquid separation of the liquid to be treated in which the flocs have formed. In the following, an example will be described in which the liquid to be treated is sludge, but the control system 100 can also be applied to plants that treat liquids other than sludge. Note that sludge is a liquid containing fine solids that is generated during wastewater treatment, etc., and can also be called a slurry.

As will be explained in more detail below, the control system 100 performs each step in the sludge treatment process, from the step of converting the sludge to be treated into flocculated sludge by agglomerating suspended solids in the sludge to form flocs, to the step of dewatering the flocculated sludge to obtain dehydrated sludge (also called dehydrated cake) and dehydrated filtrate. As shown in Figure 2, the control system 100 includes an information processing device 1, a control device 3, a flocculator 5, and a dehydrator 9.

The flocculator 5 is a device that adds a chemical agent that coagulates suspended solids to the liquid being treated in a coagulation tank and moderately stirs it to form flocs. Specifically, the flocculator 5 uses sludge as the liquid being treated and coagulates the suspended solids in the sludge to form flocs, thereby producing coagulated sludge. The flocculator 5 in Figure 2 is equipped with a coagulation tank 51, agitator blades 52, a motor 53, and an inspection window 54. The flocculator 5 also has a sludge inlet 55, a chemical agent inlet 56, and a discharge outlet 57.

Furthermore, a photographing device 72 and a lighting device 71 for photography are attached to the inspection window 54. The photographing device 72 needs to be capable of taking at least still images. It is preferable that the coagulation tank 51 is opaque so that the way light hits the flocs does not change while the control system 100 is in operation. Furthermore, it is preferable that the photographing device 72 and lighting device 71 are housed in a light-tight dark box that is open on the inspection window 54 side, as in the example shown.

The dehydrator 9 is a device that performs solid-liquid separation of the liquid to be treated in which flocs have formed. Specifically, the dehydrator 9 is disposed downstream of the flocculator 5 and performs solid-liquid separation by dehydrating the flocculated sludge (liquid) discharged from the flocculator 5. The dehydrator 9 in Figure 2 is a screw press type dehydrator equipped with an outer screen 91 and a screw 92. The dehydrator 9 also has a sludge inlet 93, a filtrate outlet 94, and a dehydrated cake outlet 95. Although not shown, the dehydrator 9 also includes a motor for rotating the screw 92. Of course, the dehydrator 9 is not limited to a screw press type as long as it is capable of dehydrating flocculated sludge. For example, a centrifugal dehydrator, a filter press dehydrator, or a belt press dehydrator can also be used.

In the control system 100, the sludge to be treated is continuously or intermittently supplied from the sludge inlet 55 into the coagulation tank 51 of the flocculator 5 by a supply device (not shown). The sludge supply rate may be automatically controlled by the supply device or its control device 3 according to the sludge treatment rate by the flocculator 5 and dehydrator 9.

Then, chemicals (including at least a flocculant) for flocculating the sludge are added to the sludge in the flocculation tank 51 through the chemical inlet 56. In this state, the motor 53 is driven to rotate the agitator blades 52, agitating the sludge and chemicals and forming flocs. The flocculated sludge, which is a mixture of the formed flocs and the water contained in the sludge, is discharged through the outlet 57.

The flocculated sludge is then fed into the outer screen 91 through the sludge inlet 93 of the dehydrator 9. Inside the dehydrator 9, the flocculated sludge is dehydrated under pressure by the screw 92, and the filtrate is discharged through the filtrate outlet 94. A dehydrated cake, which is a mass of dehydrated flocculated sludge, is discharged through the dehydrated cake outlet 95.

The sludge is supplied into the coagulation tank 51 through the sludge inlet 55, where it is pushed out and discharged from the outlet 57 of the coagulation tank 51, and the discharged coagulated sludge is supplied to the dehydrator 9. Therefore, the flow rate of the sludge supplied to the coagulation tank 51 and the flow rate of the sludge supplied to the dehydrator 9 are the same at the same time.

As will be explained in detail below, the information processing device 1 acquires at least one of the following: measurement data related to the liquid supplied to the coagulation tank 51, measurement data related to the chemicals supplied to the coagulation tank 51, measurement data related to the liquid in the coagulation tank 51, measurement data related to the operation of the coagulation tank 51, and measurement data related to the operation of the dehydrator 9. The information processing device 1 then predicts the moisture content of the dehydrated cake based on the acquired measurement data.

The information processing device 1 can also control the operation of various devices that are components of the control system 100 (e.g., the flocculator 5, the dehydrator 9, and sludge and chemical supply devices (not shown)) via the control device 3. The control device 3 controls the operation of various devices that are components of the control system 100. The control device 3 may be, for example, a PLC (Programmable Logic Controller).

[Device configuration]
The configuration of the information processing device 1 will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the main configuration of the information processing device 1. As shown in the figure, the information processing device 1 includes a control unit 10 that controls each unit of the information processing device 1, and a storage unit 11 that stores various data used by the information processing device 1. The information processing device 1 also includes a communication unit 12 that allows the information processing device 1 to communicate with other devices, an input unit 13 that accepts input of various data to the information processing device 1, and an output unit 14 that allows the information processing device 1 to output various data.

The control unit 10 also includes an acquisition unit 101, a prediction unit 102, and an update unit 103. The update unit 103 will be explained later in the section "About the update unit 103."

The memory unit 11 contains a first prediction model 111 and a second prediction model 112. Details of these are explained below in the sections "About the First Prediction Model" and "About the Second Prediction Model," respectively.

The acquisition unit 101 acquires at least one of the following: measurement data related to the liquid supplied to the coagulation tank 51; measurement data related to the chemicals supplied to the coagulation tank 51; measurement data related to the liquid in the coagulation tank 51; measurement data related to the operation of the coagulation tank 51; and measurement data related to the operation of the dehydrator 9. Details of each measurement data will be explained below in the section "Regarding the first prediction model."

The prediction unit 102 uses the first prediction model 111 stored in the memory unit 11 to predict the moisture content of the dehydrated cake from the measurement data acquired by the acquisition unit 101. This dehydrated cake is the dehydrated cake that is discharged from the dehydrator 9 at the point in time (hereinafter referred to as the "elapsed time") when the retention time for the flocculated sludge in the dehydrator 9 has elapsed from the time the measurement data was taken. The prediction unit 102 performs a prediction using the first prediction model 111 each time the acquisition unit 101 acquires new measurement data. Therefore, the prediction unit 102 can predict the moisture content of the dehydrated cake in real time, that is, continuously at short time intervals (for example, every minute).

[Regarding the first prediction model]
The first prediction model 111 is a prediction model trained using the measurement data acquired by the acquisition unit 101 at the time of measurement as an explanatory variable and the moisture content of the dehydrated cake discharged from the dehydrator 9 at the time when the retention time has elapsed from the time of measurement (hereinafter referred to as the "post-dehydration moisture content") as a target variable. In this embodiment, the first prediction model 111 has an additional explanatory variable, which is the predicted value of the post-dehydration moisture content from the measurement data by the second prediction model 112. The second prediction model 112 will be described later in the section "Regarding the Second Prediction Model."

The measurement data acquired by the acquisition unit 101 and used for prediction by the first prediction model 111 includes the following:

(1) Measurement data regarding the sludge supplied to the coagulation tank 51. This measurement data is, for example, at least one of the supply flow rate of the sludge per unit time and the concentration of the sludge, and is measured before the sludge inlet 55.

(2) Measurement data regarding the chemical supplied to the coagulation tank 51. This measurement data is, for example, the supply flow rate of the chemical per unit time, and is measured before the chemical inlet 56.

(3) Measurement data related to the sludge in the coagulation tank 51. This measurement data is at least one of the average floc shading value and the average unit area of the gaps between flocs, and is obtained by image processing of still images taken by the image capture device 72. The average shading value serves as an index of the sludge color tone (lightness). The average unit area serves as an index of the floc diameter.

(4) Measurement data related to the operation of the coagulation tank 51. This measurement data is the rotation speed of the agitator blades 52 in the coagulation tank 51 and is obtained from the control device 3.

(5) Measurement data related to the operation of the dehydrator 9. This measurement data is at least one of the following: the operating time of the dehydrator 9, the rotation speed of the screw of the dehydrator 9, the flow rate of flocculated sludge supplied to the dehydrator 9 per unit time, and the feeding pressure of the flocculated sludge fed to the dehydrator 9, and is acquired from the control device 3.

In this embodiment, various measurement data can be used, as shown in (1) to (5) above. Therefore, the moisture content of the dehydrated cake can be predicted with high accuracy using multifaceted explanatory variables. Of the measurement data shown in (1) to (5) above, measurement data that contribute significantly (are of high importance) to the prediction accuracy of the first prediction model 111 include the operation time, screw rotation speed, and supply flow rate shown in (5) above, and the average floc density value shown in (3) above.

[Regarding the second forecast model]
The second prediction model 112 is a prediction model trained using the measurement data acquired by the acquisition unit 101 at the time of measurement as an explanatory variable and the post-spin moisture content at the time point when the retention time has elapsed since the measurement as a target variable. In this embodiment, the second prediction model 112 is a prediction model trained during a trial operation period before the start of the actual operation period of the dehydrator 9. The measurement data that serves as the explanatory variable of the second prediction model 112 may be the same as or different from the measurement data that serves as the explanatory variable of the first prediction model 111.

As described above, the information processing device 1 of this embodiment comprises an acquisition unit 101 that acquires at least one of measurement data regarding the liquid supplied to a coagulation tank 51, where a chemical agent that coagulates suspended solids is added to the liquid containing the suspended solids, measurement data regarding the chemical agent supplied to the coagulation tank 51, measurement data regarding the liquid in the coagulation tank 51, measurement data regarding the operation of the coagulation tank 51, and measurement data regarding the operation of the dehydrator 9 that dehydrates the liquid while transporting it, and a prediction unit 102 that uses a first prediction model 111 to predict the post-dehydration moisture content, which is the moisture content of the dehydrated cake discharged from the dehydrator 9 at a point in time after the retention time has elapsed from the time the measurement data was measured, based on the measurement data acquired by the acquisition unit 101 and a predicted value by a second prediction model 112 for the post-dehydration moisture content, which is the moisture content of the dehydrated cake discharged from the dehydrator 9 at a point in time after the retention time for the liquid to remain in the dehydrator 9 from the time the measurement data was measured. The measurement data used by the first prediction model 111 for learning is measurement data acquired by the acquisition unit 101 from a predetermined period before the learning point during the actual operation period of the dehydrator 9 to the learning point. The second prediction model 112 is a prediction model trained using the measurement data acquired by the acquisition unit 101 during a trial operation period, which is an operation period in which the operating conditions are different from those of the actual operation period, as an explanatory variable, and the post-dehydration moisture content at the point in time when the retention time has elapsed since the measurement of the measurement data as a target variable.

With the above configuration, the second prediction model 112 is a prediction model trained using measurement data acquired under a wide variety of operating conditions during the test operation period of the dehydrator 9 as training data. Therefore, the predicted value by the second prediction model 112 for the post-dehydration moisture content at the point in time when the retention time has elapsed since the measurement of the measurement data is a predicted value that is suitable for a wide variety of operating conditions.

On the other hand, the first prediction model 111 is a prediction model trained using the measurement data acquired from a predetermined period before the learning point during the actual operation period of the dehydrator 9 to the learning point and the predicted value by the second prediction model 112 of the post-spin moisture content corresponding to the measurement data as explanatory variables, and the post-spin moisture content corresponding to the measurement data as the objective variable. Because the first prediction model 111 is trained using the most recent measurement data, it is a prediction model that is adapted to the most recent operating conditions. Therefore, in normal cases where there are no changes in operating conditions, the accuracy of predictions by the first prediction model 111 is maintained.

In addition, the first prediction model 111 is trained using the predicted value of the post-dehydration moisture content predicted by the second prediction model 112 using the most recent measurement data, and is therefore a prediction model that takes a wide variety of operating conditions into account. Therefore, even if the operating conditions are changed after a long period of time without any changes, it is possible to mitigate any decline in the accuracy of predictions made by the first prediction model 111.

[About the update section]
The update unit 103 performs a process of updating the first prediction model 111 during a period when the coagulation tank 51 and the dehydrator 9 are stopped (a shutdown period). Specifically, the update unit 103 uses the second prediction model 112 to calculate a predicted value of the post-dehydration moisture content at a point in time when the retention time has elapsed from the time of measurement of the measurement data acquired by the acquisition unit 101 during the operation period of the dehydrator 9. The update unit 103 then updates the first prediction model 111 using, as training data, a set of the measurement data, the predicted value of the post-dehydration moisture content, and an actual value of the moisture content of the dehydrated cake measured by an operator.

Typically, the operating period is a set period of one day. Therefore, updates by the update unit 103 are performed once a day. Furthermore, it takes time for an operator to actually measure the moisture content of the dehydrated cake. Therefore, instead of the actual moisture content measured by the operator, an estimated moisture content calculated using a different model may be used as the objective variable of the training data.

Therefore, the first prediction model 111 is updated using the latest measurement data, so that predictions using the first prediction model 111 can be adapted to the latest conditions of the coagulation tank 51 and dehydrator 9.

[Prediction processing]
3 is a flowchart showing an example of a process for predicting the moisture content of dehydrated cake (a moisture content prediction method) in the information processing device 1 having the above configuration. This prediction process is executed during operation, as described above. As shown in FIG. 3 , first, the acquisition unit 101 collects (acquires) various measurement data (S11, acquisition step). Next, the prediction unit 102 uses the first prediction model 111 to predict the moisture content at a point in time after the retention time has elapsed since the measurement of the measurement data (S12, prediction step). After that, the process returns to step S11 and repeats the above operation.

[Update process]
4 is a flowchart showing an example of the process of updating the first prediction model 111 in the information processing device 1. As described above, this update process is executed for each shutdown period.

As shown in FIG. 4, first, the update unit 103 uses the second prediction model 112 to calculate a predicted value of the post-dehydration moisture content at a point in time after the retention time has elapsed from the time of measurement of the measurement data acquired by the acquisition unit 101 during the operation of the dehydrator 9 (S21). Next, the update unit 103 updates the first prediction model 111 using a set of the measurement data, the predicted value of the post-dehydration moisture content, and the actual moisture content of the dehydrated cake measured by the operator as training data (S22). Then, the update process ends.

[Modification]
The execution entity of each process described in the above embodiment is arbitrary and is not limited to the above example. For example, each step of the moisture content prediction method shown in Figure 3 can be shared among multiple information processing devices. In other words, the moisture content prediction method may be executed by one information processing device 1 or multiple information processing devices.

[Additional Notes]
In the above embodiment, the explanatory variables of the first prediction model 111 include the predicted value of the post-spin moisture content using the second prediction model 112, but may also include predicted values of the post-spin moisture content using other models. In this way, multiple predicted values using different models may be used as explanatory variables of the first prediction model 111.

Furthermore, in the above-described embodiment, the second prediction model 112 is not updated, but this is not limited to this. For example, when unusual operating conditions occur during the actual operation period, measurement data acquired by the acquisition unit 101 may be accumulated, and the accumulated measurement data may be used to update the second prediction model 112.

Furthermore, in the above-described embodiment, a trial operation period is used as an operating period in which the operating conditions differ from those of the actual operation period. However, this different operating period is not limited to the trial operation period, and may be any operating period in which measurement data can be obtained under a wide variety of operating conditions, including the range of operating conditions in the actual operation period. For example, a test operation period performed after maintenance on the dehydrator 9 may be used as the different operating period.

[Example]
An example of the information processing device 1 having the above configuration and a comparative example will be described with reference to FIGS.

Example 1
As described above, the first prediction model 111 of this embodiment is a model that predicts the post-spin moisture content at an elapsed time point using measurement data at the time of measurement and a predicted value of the post-spin moisture content at an elapsed time point calculated from the measurement data using the second prediction model 112. The first prediction model 111 is trained using the measurement data acquired during the most recent predetermined period of the actual operation period. The second prediction model 112 is trained using the measurement data acquired during the trial operation period. Therefore, the second prediction model 112 is not updated.

Figure 5 is a diagram in tabular form showing combinations of explanatory variables and target variables of the first prediction model 111 in this embodiment. In the example of Figure 5, the measurement data at the measurement time, which serve as explanatory variables for the first prediction model 111, are the liquid (sludge) supply flow rate per unit time (sludge flow rate), the liquid (sludge) concentration, the average floc concentration value, the average unit area of gaps between flocs, the chemical supply flow rate per unit time, the rotational speed of the agitator impeller in the coagulation tank 51, the operating time of the dehydrator 9, the rotational speed of the screw of the dehydrator 9, and the input pressure of the liquid (flocculated sludge) input to the dehydrator 9.

Figure 6 is a diagram showing, in tabular form, the combinations of explanatory variables and dependent variables of the second prediction model 112. The second prediction model 112 shown in Figure 6 differs from the first prediction model 111 shown in Figure 5 in that the predicted values by the second prediction model 112 are omitted, but is otherwise similar. That is, in this embodiment, the measurement data included in the explanatory variables of the first prediction model 111 and the measurement data included in the explanatory variables of the second prediction model 112 are the same.

(Comparative Example 1)
The first comparative model, which is a comparative example of Example 1, has explanatory variables and response variables similar to those shown in Fig. 6. The first comparative model is trained using all measurement data obtained in the past. For this reason, the time required for training is longer than in other cases.

(Comparative Example 2)
A second comparative model, which is another comparative example to the first embodiment, has explanatory variables and response variables that are similar to those shown in Fig. 6. The second comparative model is trained using measurement data acquired during a predetermined period immediately preceding the actual operation period. Therefore, the second comparative model is similar to the first prediction model 111 of the first embodiment except that the predicted values by the second prediction model 112 are omitted.

(Comparative Example 3)
A third comparative model, which is yet another comparative example to Example 1, has explanatory variables and response variables similar to those shown in Fig. 6. The third comparative model is trained using a variety of measurement data, such as measurement data from a test run period, and the measurement data measured during the most recent predetermined period.

(Comparative Example 4)
7 is a table showing combinations of explanatory variables and dependent variables of a fourth comparative model, which is yet another comparative example to Example 1. As shown in FIG. 7 , the only explanatory variable of the fourth comparative model is the predicted value by the second prediction model 112. Therefore, the fourth comparative model is a univariate regression model that predicts the post-dehydration moisture content using the fourth comparative model from the predicted value of the post-dehydration moisture content predicted from the measurement data using the second prediction model 112. Furthermore, the fourth comparative model is trained using the measurement data measured during the most recent predetermined period.

(evaluation)
Fig. 8 is a graph showing the variation in moisture content predicted by the first prediction model 111 of Example 1 relative to the actual moisture content measured by different operators. Fig. 9 is a graph showing the variation in moisture content predicted by the first to fourth comparative models relative to the actual moisture content measured by different operators. In Fig. 9, the graphs for the first to fourth comparative models are shown in the order of upper left, lower left, upper right, and lower right.

Figure 10 is a diagram showing in tabular form the mean absolute error (MAE), operating time parameter importance, and sludge flow rate parameter importance for each of Example 1 and Comparative Examples 1 to 4. Here, parameter importance indicates the degree of influence that an explanatory variable in a model has on a target variable. Of the nine measurement data included as explanatory variables in the first prediction model 111 and the first to third comparative models, operating time and sludge flow rate tend to have the highest parameter importance.

Regarding operating time, the dehydrator 9 continues to discharge dehydrated cake to prevent residual cake from sticking inside the dehydrator 9 after operation is stopped. Therefore, the next time the dehydrator 9 is started, operation begins with space remaining inside. The dehydrator 9 then fills with supplied sludge, causing the pressure distribution inside the dehydrator 9 to fluctuate and then stabilize. This pressure distribution causes the dehydration of the cake to fluctuate.

Therefore, the above pressure distribution has high parameter importance. However, at present, it is difficult to include the above pressure distribution as an explanatory variable. Therefore, instead of the above pressure distribution, the operating time, which is closely related to the above pressure distribution, has high parameter importance and is included as an explanatory variable.

On the other hand, with regard to sludge flow rate, in the dehydrator 9, the sludge flow rate (amount fed) = the amount of dehydrated cake transported by the screw + the amount of filtrate. Since the amount transported remains roughly constant if the screw rotation speed is constant, the better the dehydration, the greater the amount of filtrate and the greater the sludge flow rate. Therefore, the higher the sludge flow rate, the lower the moisture content after dehydration. Therefore, the sludge flow rate is a highly important parameter.

Referring to Figures 8 to 10, it can be seen that the first prediction model 111 of Example 1 has smaller variance and the smallest mean absolute error compared to the first to fourth comparison models, and therefore has better accuracy.

Note that, referring to Figure 10, of Comparative Examples 1 to 4, Comparative Example 2 appears to have the smallest mean absolute error and good accuracy. However, the importance of the operating time parameter is twice that of the sludge flow rate parameter. This means that the second comparative model is a model that makes predictions based only on the time from start-up, indicated by the operating time, rather than the operating state indicated by the sludge flow rate, etc. The reason for this is thought to be that the second comparative model is trained using only measurement data from the most recent specified period, resulting in low diversity in the training data.

For example, consider a case where the screw rotation speed was a constant value (0.17 min ⁻¹ ) in the measurement data for the most recent predetermined period, and the screw rotation speed was changed to 0.19 min ⁻¹ on the day of operation. Normally, this would result in a worsening of the moisture content, but the second comparative model would not be able to fully take into account the effect of the screw rotation speed. Therefore, it is difficult to use the second comparative model for various controls in the control system 100.

In contrast, in Example 1, the predicted value of the spin-off rate by the second prediction model 112 deteriorates, and therefore the predicted value of the spin-off rate by the first prediction model 111 also deteriorates. In other words, the first prediction model 111 can indirectly take into account the influence of the screw rotation speed. Therefore, the first prediction model 111 can be used for various controls in the control system 100.

[Software implementation example]
The functions of the information processing device 1 (hereinafter referred to as the "device") can be realized by a program for causing a computer to function as the device, and a program (moisture content prediction program) for causing a computer to function as each control block of the device (particularly each part included in the control unit 10).

In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using this control device and storage device.

The above program may be stored non-transitory on one or more computer-readable storage media. These storage media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

Furthermore, some or all of the functions of each of the above control blocks can be realized by logic circuits. For example, integrated circuits incorporating logic circuits that function as each of the above control blocks are also included in the scope of the present invention. In addition, the functions of each of the above control blocks can also be realized by, for example, a quantum computer.

The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

REFERENCE SIGNS LIST 1 Information processing device 9 Dehydrator 10 Control unit 11 Memory unit 51 Coagulation tank 52 Stirring blade 92 Screw 101 Acquisition unit 102 Prediction unit 103 Update unit 111 First prediction model 112 Second prediction model

Claims (10)

an acquiring unit that acquires at least one of measurement data relating to a liquid supplied to a coagulation tank in which a chemical agent for coagulating suspended solids is added to the liquid containing the suspended solids, measurement data relating to the chemical agent supplied to the coagulation tank, measurement data relating to the liquid in the coagulation tank, measurement data relating to the operation of the coagulation tank, and measurement data relating to the operation of a dehydrator that dehydrates the liquid discharged from the coagulation tank while transporting it;
a prediction unit that predicts, using a first prediction model, the post-dehydration moisture content, which is the moisture content of a dehydrated cake discharged from the dehydrator at a time when a retention time has elapsed since the time when the measurement data was measured, from the measurement data acquired by the acquisition unit and a predicted value by a second prediction model for the post-dehydration moisture content, the moisture content being the moisture content of a dehydrated cake discharged from the dehydrator at a time when the retention time for the liquid to remain in the dehydrator has elapsed since the time when the measurement data was measured,
the measurement data used by the first prediction model for learning is the measurement data acquired by the acquisition unit from a predetermined period before the learning point in time during an actual operation period of the dehydrator to the learning point in time,
The second prediction model is a prediction model trained using the measurement data acquired by the acquisition unit during an operating period in which operating conditions are different from the actual operating period as an explanatory variable and the post-dehydration moisture content at a time point when the residence time has elapsed since the measurement of the measurement data as a target variable.
Information processing device. The information processing device of claim 1, wherein the measurement data regarding the liquid supplied to the coagulation tank is at least one of the supply flow rate of the liquid per unit time and the concentration of the liquid. The information processing device according to claim 1, wherein the measurement data regarding the chemical supplied to the coagulation tank is the supply flow rate of the chemical per unit time. The information processing device of claim 1, wherein the measurement data regarding the liquid in the coagulation tank is at least one of the average floc density value and the average unit area of the gaps between flocs. The information processing device according to claim 1, wherein the measurement data relating to the operation of the coagulation tank is the rotation speed of the agitator blades in the coagulation tank. the dehydrator is a screw press type dehydrator,
2. The information processing device according to claim 1, wherein the measurement data relating to the operation of the dehydrator is at least one of the operating time of the dehydrator, the supply flow rate of the liquid to the dehydrator per unit time, the rotation speed of the screw of the dehydrator, and the supply pressure of the liquid fed into the dehydrator. The information processing device according to claim 1, wherein the prediction unit performs prediction using the first prediction model each time the acquisition unit acquires new measurement data. The present invention further includes an update unit that updates the first prediction model during periods when the dehydrator is not in operation, using as training data a set of measurement data acquired by the acquisition unit during an operation period of the dehydrator, a predicted value of the post-dehydration moisture content by the second prediction model at a time when the retention time has elapsed since the measurement of the measurement data, and the post-dehydration moisture content at a time when the retention time has elapsed since the measurement of the measurement data.
The information processing device according to claim 1 . A moisture content prediction method executed by one or more information processing devices,
an acquiring step of acquiring at least one of measurement data relating to a liquid supplied to a coagulation tank in which a chemical agent for coagulating suspended solids is added to the liquid containing the suspended solids, measurement data relating to the chemical agent supplied to the coagulation tank, measurement data relating to the liquid in the coagulation tank, measurement data relating to the operation of the coagulation tank, and measurement data relating to the operation of a dehydrator that dehydrates the liquid discharged from the coagulation tank while transporting it;
a prediction step of predicting, using a first prediction model, the post-dehydration moisture content, which is the moisture content of a dehydrated cake discharged from the dehydrator at a time when a retention time has elapsed since the time when the measurement data was measured, from the measurement data acquired in the acquisition step and a predicted value by a second prediction model for the post-dehydration moisture content, which is the moisture content of a dehydrated cake discharged from the dehydrator at a time when the retention time for the liquid to remain in the dehydrator has elapsed since the time when the measurement data was measured,
the measurement data used by the first prediction model for learning is the measurement data acquired in the acquiring step from a predetermined period before the learning point in time during an actual operation period of the dehydrator to the learning point in time,
The second prediction model is a prediction model trained using the measurement data acquired in the acquisition step during an operating period in which operating conditions are different from the actual operating period as an explanatory variable, and the post-dehydration moisture content at a time point when the residence time has elapsed since the time point at which the measurement data was acquired as a response variable.
Moisture content prediction method. A moisture content prediction program for causing a computer to function as the information processing device described in claim 1, the moisture content prediction program causing a computer to function as the acquisition unit and the prediction unit.