JP7289670B2 - Information processing device, information processing method, and program - Google Patents

Description

The present invention relates to an information processing apparatus, an information processing method, and a program used for water treatment.

2. Description of the Related Art Conventionally, water treatment for converting raw water such as river water, dam water, or lake water (hereinafter simply referred to as “raw water”) into drinking water is known (see Patent Document 1, for example). FIG. 4 is a block diagram that schematically illustrates a water treatment system 40 used to perform water treatment. The water treatment system 40 of FIG. 4 includes a receiving well 41 for storing raw water for water treatment, and a nucleus where turbidity in the raw water is aggregated by adding a coagulant to the raw water (hereinafter referred to as "aggregation nucleus"). ), a flocculation pond 43 in which raw water to which a flocculating agent is added is slowly stirred to form flocs in which turbidity in the raw water flocculates in flocculation nuclei, and the formed flocs are precipitated. It comprises a sedimentation basin 44, a filtration basin 45 for filtering the tap water after flocs have settled, and a disinfection tank 46 for disinfecting the filtered tap water.

The flocculant is added to the raw water from the rapid mixing basin 42 after, for example, the manager of the water treatment system 40 conducts a jar test to determine the optimum pH for floc formation and the amount of flocculant to be added. At this time, if an excessive amount of flocculant is added to the raw water from the rapid mixing tank 42 with respect to the appropriate amount of original flocculant added without the jar test being performed, for example, a large amount of small flocs will be formed. The flocs formed in the flocculation basin 43 are difficult to settle in the sedimentation basin 44, and the filtration of the flocs in the filtration basin 45 is burdened. Therefore, it is necessary to form flocs of good quality in the flocculation basin 43 that have excellent sedimentation properties so as not to adversely affect the treatment in the sedimentation basin 44 and the filtration bed 45. An appropriate amount of agent to be added should be determined.

By the way, in general, whether or not good quality flocs are formed is determined by conducting a jar test using raw water sampled from the flocculation pond 43, or by observing how flocs are formed in the flocculation pond 43. Judgment is made by observation. At this time, an administrator who has a lot of experience in managing the water treatment system 40 should be able to perform the operation at an early stage after starting the jar test, or at an early stage after starting to observe the formation of flocs in the flocculation pond 43. , has a high degree of skill in determining the appropriate dosage of flocculant.

JP-A-07-185573

However, in recent years, the number of managers with advanced skills in the water treatment system 40 has decreased due to aging, etc., so flocs are formed at an early stage after the start of the jar test or in the flocculation pond 43. In some cases, it may not be possible to quickly determine an appropriate addition amount of the flocculant at an early stage after the start of observation of the state. In addition, even an administrator who does not have advanced skills of the water treatment system 40 (hereinafter referred to as "general administrator") can easily adjust the amount of coagulant to be added based on the turbidity and chromaticity of raw water. There is a simple judgment model for the amount of coagulant to be added, and general administrators can easily determine the amount of coagulant to be added to raw water based on the simple judgment model for the amount of coagulant added. The judgment model cannot be used when the turbidity of the raw water suddenly increases due to a disaster such as heavy rain. Correspondingly, the general manager carries out a jar test, but since he does not have advanced skills in the water treatment system 40, he not only carries out the jar test, but also the water flowing from the flocculation pond 43 to the sedimentation pond 44. The turbidity of the raw water is also taken into account to reliably determine whether or not good quality flocs are formed. Therefore, there is a problem that the appropriate addition amount of the flocculant cannot be determined until several hours have passed since the addition of the flocculant to the raw water.

That is, there is a problem that it is not possible to quickly and easily determine the amount of coagulant to be added for forming good quality flocs.

An object of the present invention is to provide an information processing apparatus, an information processing method, and a program capable of quickly and easily determining the addition amount of a coagulant for forming good-quality flocs.

In order to achieve the above object, an information processing apparatus of the present invention is an information processing apparatus used for clean water treatment for removing flocs formed by adding a coagulant to water containing turbidity. image data showing flocs and flocs obtained within a first time after the addition of the flocculating agent and being the nuclei when the turbidity flocculates, and the flocs to be removed; Learning means for learning teacher data that associates category information based on floc sedimentation and water turbidity , whether or not has an adverse effect on the water treatment, and the second after the flocculant is added Inputting image data acquired within 2 hours and having no categorical information of whether the removed floc adversely affects the water treatment , based on floc sedimentation and turbidity of the water. input means, wherein the input image data indicates the agglutinated nucleus ; Determination means for determining whether or not flocs included in the data have an adverse effect on the water treatment when removed from the water , the flocs to be formed in the water corresponding to the input image data in the future. and an output means for outputting the information of the determined category.

In order to achieve the above object, the information processing method of the present invention is an information processing method used in water treatment for removing flocs formed by adding a coagulant to water containing turbidity. image data showing flocs and flocs obtained within a first time after the addition of the flocculating agent and being the nuclei when the turbidity flocculates, and the flocs to be removed; A learning step of learning teacher data that associates categorical information based on floc sedimentation and water turbidity , whether or not has an adverse effect on the water treatment, and the second after the flocculant is added Inputting image data acquired within 2 hours and having no categorical information of whether the removed floc adversely affects the water treatment , based on floc sedimentation and turbidity of the water. an input step in which the input image data indicates agglutination nuclei ; and the input image based on the learning result of learning the teacher data and the input image data indicating the agglutination nuclei. a determination step of determining whether flocs included in the data, when removed from the water, adversely affect the water treatment, the flocs to be formed in the water corresponding to the input image data in the future; and an output step of outputting the determined category information.

To achieve the above object, the program of the present invention provides a computer with an information processing method for removing flocs formed by adding a coagulant to water containing turbidity, which is used for water treatment. wherein the information processing method shows image data and flocs obtained within a first time after the addition of the flocculating agent and showing flocculation nuclei, which are nuclei when the turbidity flocculates a learning step of learning teacher data that associates image data and categorical information as to whether or not the flocs to be removed adversely affect the water treatment based on floc sedimentation and water turbidity; , a category based on floc sedimentation and turbidity of the water , whether the floc obtained and removed within a second time after the flocculant is added adversely affects the water treatment. an input step of inputting image data having no information of said input image data indicating agglutination nuclei; and said input image indicating a learning result of learning said teacher data and said agglutination nuclei. a determination step of determining, based on the data, whether flocs contained in the input image data have an adverse effect on the water treatment when removed from the water , wherein the input image data and an output step of outputting the information of the determined category as to whether or not the flocs formed in the future in the water corresponding to the above will adversely affect the water treatment. characterized by

ADVANTAGE OF THE INVENTION According to this invention, the addition amount of the flocculant for forming a good quality floc can be determined rapidly and simply.

1 is a block diagram schematically showing the configuration of an information processing device according to an embodiment of the present invention; FIG. FIG. 2 is a flow chart showing a procedure of storage processing for storing pre-created teacher data in the HDD in FIG. 1; FIG. FIG. 2 is a flow chart showing a procedure of learning processing executed by a CPU in FIG. 1; FIG. 1 is a schematic block diagram of a water treatment system used to perform water treatment; FIG.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram schematically showing the configuration of an information processing apparatus 10 according to an embodiment of the invention.

The information processing apparatus 10 of FIG. 1 includes a CPU 11, a RAM 12, a ROM 13, an HDD 14, and a data input section 15, which are connected to each other via an internal bus 16. FIG. The data input unit 15 is also connected to an imaging device 17 such as a camera. The CPU 11 develops a program stored in the ROM 13 or HDD 14 in the RAM 12, which is a work memory of the CPU 11, and executes the program. Also, the imaging device 17 captures an image of a subject and acquires image data. For example, the image data is input from the data input unit 15 and stored in the HDD 14 . Note that the imaging device 17 may include a storage medium such as an HDD for storing various data, and the acquired image data may be stored in the storage medium of the imaging device 17 . The HDD 14 stores a cumulative loss function value and teacher data, which will be described later, and the CPU 11 learns the teacher data stored in the HDD 14 .

The imaging device 17 is used to photograph the formation of flocculation nuclei or flocs in the raw water to which the flocculant has been added. be. In the rapid mixing basin 42, the raw water to which the flocculant has been added is usually rapidly stirred for about 300 seconds. Note that when adding a coagulant to raw water sampled from the receiving well 41 at a location other than the rapid mixing basin 42 and quickly stirring to determine an appropriate addition amount of the coagulant, the imaging device 17 is installed at that location. may be

FIG. 2 is a flow chart showing a procedure of storage processing for storing pre-created teacher data in the HDD 14 in FIG.

In FIG. 2, first, predetermined amounts of kaolin solution and standard humic acid solution were added to 1 L of tap water to prepare a plurality of simulated river waters with different turbidities and pHs. Polyaluminum chloride (PAC) with a basicity of 50% is added to each simulated river water (S201). In this embodiment, 5 mg/L or 15 mg/L of coagulant is added to each simulated river water. Thereafter, pH, turbidity, absorbance (chromaticity) at UV-260 nm, and TOC (Total Organic Carbon) are measured as water quality data of each simulated river water (S202). Table 1 shows the measurement results for pH, turbidity, absorbance at UV-260 nm, and total organic carbon TOC of each simulated river water immediately after adding 5 mg/L or 15 mg/L flocculant.

Then, each simulated river water to which 5 mg/L or 15 mg/L of flocculant was added is rapidly stirred for 300 seconds and then slowly stirred for 1050 seconds. As a result, how each simulated river water changes according to the agitation of each simulated river water. Specifically, aggregation nuclei are formed in each simulated river water. The imaging device 17 captures a moving image of flock formation for 1350 seconds (S203). Subsequently, after the video shooting is completed, pH, turbidity, absorbance (chromaticity) at UV-260 nm, and total organic carbon TOC are measured again as water quality data of each simulated river water (S204). At this time, in addition to these water quality data, the sedimentation velocity of the formed flocs may be measured. Table 2 shows the measurement results of pH, turbidity, absorbance at UV-260 nm, and total organic carbon TOC of each simulated river water after video recording.

After that, each simulated river water is evaluated comprehensively based on how coagulation nuclei and flocs are formed in each simulated river water, the sedimentation properties of the formed flocs, changes in turbidity due to the addition of coagulants, etc. classified into two categories (S205). For example, simulated river water with good floc sedimentation and a turbidity of 0 or more and 0.10 or less after video shooting is classified into the “excellent” category. Simulated river water with a turbidity of 0.11 or more and 0.50 or less after shooting is classified into the “good” category, and the sedimentation of the formed flocs is not good, and the turbidity after shooting is 0. Simulated river water with turbidity of 51 to 1.00 is categorized as "Fair", and the sedimentation of the formed flocs is not good, and simulated river water with turbidity of 1.01 or more falls into the "bad" category.

Simulated river water categorized as "excellent", "good", or "acceptable" indicates that flocs formed in the simulated river water do not adversely affect water treatment. indicates that the amount of flocculant added to is appropriate (within an appropriate range). On the other hand, the simulated river water categorized as “bad” indicates that the flocs formed in the simulated river water have an adverse effect on water treatment, and the amount of coagulant added to the simulated river water is appropriate. indicates that it is not Table 3 shows simulated river water numbers when classified into four categories.

Then, arbitrary 3 simulated river waters are extracted from each category, and as a result, 12 simulated river waters are extracted from all categories. For each sampled simulated river water, moving images are captured for 1350 seconds. Each still image data is acquired. That is, 400 still image data are acquired from each simulated river water, 1200 still image data are acquired for each category, and 4800 still image data are acquired from all categories. Partial still image data (for example, 200 × 200 pixels) cut out from each of all the acquired still image data so that the background and stirring blades unrelated to the formation of aggregation nuclei and flocs are not included It is acquired (S206).

In the present embodiment, 4800 pieces of partial still image data are acquired. Of these, 4200 pieces of partial still image data each include water quality data of simulated river water corresponding to each piece of partial still image data and each piece of partial still image data. Teacher data is generated by associating the categories classified into the simulated river water (S207), each generated teacher data is stored in the HDD 14 (S208), and the process ends. On the other hand, the 600 partial still image data not stored in the HDD 14 as teacher data are stored in the HDD 14 for use as test data, which will be described later.

In addition, 5 mg / L or 15 mg / L of flocculant is added to multiple simulated river waters with different turbidity and pH, and the teacher data is image data acquired within 400 seconds after the start of stirring. Although it has been explained that water quality data and simulated river water are generated by associating the classified categories, and the image data acquired until flocs are formed from the aggregation nuclei in the flocculation pond 43, the water quality data of the raw water when each image data was acquired, and the categories classified into the raw water may be generated based on

FIG. 3 is a flow chart showing the procedure of learning processing executed by the CPU 11 in FIG. The learning process (learning step) in FIG. 3 is executed by the information processing apparatus 10 performing deep learning using teacher data stored in the HDD 14 . For deep learning, for example, a CNN (Convolutional Neural Network) model that recognizes a new object to be discriminated with high accuracy after the system automatically extracts and learns the features of the target image by multi-step arithmetic processing is used. In this embodiment, a so-called Alex Net is used as the CNN model.

In general, when performing deep learning using a lot of data, prepare multiple batches with a predetermined number of data (batch size), and perform deep learning on all batches. counted as going. Sufficient deep learning is rarely performed in one epoch, and deep learning is normally performed for several tens of epochs. The learning process in FIG. 3 will be described for a case where one epoch learning is performed under the condition that the batch size is 100 and the number of epochs is 50. FIG.

In FIG. 3, first, the CPU 11 calculates the cumulative error level of the learning result output based on Alex Net with respect to information such as categories associated with the teacher data (hereinafter referred to as "correct label"). A loss function value is set to 0 (S301), and all partial still image data based on teacher data are read from the HDD 14 (S302).

Next, the CPU 11 reads from the HDD 14 convolution operation processing, which is filtering processing for extracting local feature amounts of image data, and pooling operation processing for compressing data while leaving the feature amounts extracted by the convolution operation processing. Execute on image data. As a result, a plurality of feature quantity vectors based on the feature quantity of each partial still image data is obtained. Subsequently, a fully-connected arithmetic processing that combines the obtained plural feature amount vectors into one, and the combined feature amount vector is converted into one-dimensional data by an activation function (softmax function) and output. Execute the softmax calculation process. When the softmax calculation process is executed, the CPU 11 identifies the category associated with each piece of partial still image data read from the HDD 14 (S303). The categories identified here are the four categories "excellent", "good", "acceptable", or "bad" used in S205 of the storage process (FIG. 2).

In the present embodiment, the CPU 11 performs (1) convolution operation processing, (2) pooling operation processing, (3) convolution operation processing, (4) pooling operation processing, (5) three-time convolution operation processing, and (6) pooling Arithmetic processing, (7) two-time fully coupled arithmetic processing, and (8) softmax arithmetic processing are executed on the partial still image data read from the HDD 14 in this order.

After that, the CPU 11 calculates a loss function value based on the category and the correct label for each partial still image data identified in S303, and cumulatively adds the calculated loss function values to calculate a cumulative loss function value ( S304), it is determined whether or not one batch of learning has been completed (S305). As a result of the determination in S305, when the learning of one batch is not completed, the CPU 11 returns to S302, and when the learning of one batch is completed, it determines whether or not the learning of all batches prepared in advance has been completed. (S306). As a result of the determination in S306, when the learning of all batches has not been completed, the CPU 11 advances to S307 and performs learning based on the cumulative loss function value. For example, the cumulative loss function value calculated in S304 approaches 0. , and the process returns to S301. On the other hand, when the learning of all batches is completed as a result of the determination in S306, this processing ends.

The information processing apparatus 10 that has executed the learning process (learning step) in FIG. Image data showing the formation of aggregation nuclei in the pond 42 and not having information as to whether or not flocs to be formed in the future will adversely affect water treatment (hereinafter referred to as "discrimination target image data") ) is input to the data input unit 15 (input step), flocs that will be formed in the raw water corresponding to the discrimination target image data based on the learning result of learning the teacher data and the discrimination target image data are processed by water treatment. (discrimination step), and output the category classified into the raw water (hereinafter referred to as "predicted category").

By the way, it is necessary to verify whether or not the information processing apparatus 10 that has executed the learning process of FIG. 3 correctly outputs the expected category from the discrimination target image data. In this embodiment, the verification was performed based on the test data stored in the HDD 14 and the correct data described later.

The test data is 600 partial still image data based on image data taken within 400 seconds after the coagulant was added to the predetermined simulated river water and stirring was started. There is no information about the category into which the simulated river water is classified in the test data. On the other hand, the correct data is configured by associating each partial still image data constituting the test data with information about the category into which the simulated river water is classified.

After each test data is input to the data input unit 15, the information processing apparatus 10 executes S302 to S303 in the learning process of FIG. The categories into which the simulated river water corresponding to the data is classified are output. As a result of comparing each test data and output category with the correct data, 95.2% of the output categories matched the correct data. As a result, when a known amount of flocculant is added to raw water to be treated, the information processing apparatus 10, if there is image data for 400 seconds from when the flocculant is added and stirring is started, It was found to reliably and correctly determine whether flocs formed in the future adversely affect water treatment.

In the present embodiment, the information processing apparatus 10 correctly classifies the simulated river water corresponding to the test data into categories with a high probability of 95.2% based on teacher data created from 4200 pieces of partial still image data. Discrimination has been described, but the probability is expected to increase as new teacher data is learned.

According to the present embodiment, the information processing apparatus 10 includes a CPU 11 and an HDD 14, and the CPU 11 stores teaching data stored in the HDD 14, for example, from when a known amount of coagulant is added to the simulated river water and stirred. Deep learning is performed on teacher data in which each image data for 400 seconds is associated with a category indicating whether flocs to be formed in the simulated river water in the future will adversely affect water treatment. Next, the CPU 11 selects, for example, image data within 300 seconds after a known amount of flocculating agent was added to the raw water to be treated and stirred, and flocs formed in the future will adversely affect the treatment of the drinking water. When discrimination target image data, which is image data not associated with a category of whether or not, is input, a future formation in the raw water corresponding to the discrimination target image data based on the discrimination target image data and the learning result of learning the teacher data Determine the category of whether flocculation that is carried out has an adverse effect on water treatment.

That is, the information processing apparatus 10 that has performed deep learning on the teacher data will be able to form future coagulation nuclei if there is image data showing how coagulation nuclei are formed when a known amount of coagulant is added to raw water to be treated. It is determined whether the flock to be received is a good quality flock. At this time, the determination result determined by the information processing apparatus 10 may be reliable. An appropriate addition amount of the flocculant can be determined more quickly than determining the addition amount. In addition, even a general manager who cannot quickly determine the appropriate addition amount of the coagulant can use the information processing device 10 to obtain the results of the jar test and the turbidity of the raw water in the sedimentation basin 44. An appropriate addition amount of the flocculant can be quickly and easily determined without obtaining information over a long period of time. Therefore, the information processing apparatus 10 that has performed deep learning on the teacher data can quickly and easily determine the addition amount of the flocculating agent for forming good-quality flocs if there is image data showing how flocculation nuclei are formed. be able to.

In the present embodiment, the information processing apparatus 10 captures the image data showing the formation of flocculation nuclei in the rapid mixing tank 42 within 300 seconds after the addition of the coagulant to the raw water and the start of stirring. Although the expected category is output using image data, since the coagulation nuclei are formed immediately after the coagulant is added to the raw water, the coagulant is added to the raw water and the stirring is started. The captured image data may be used to output expected categories. This makes it possible to more quickly determine the appropriate addition amount of the flocculant.

Further, in the present embodiment, it is assumed that the imaging device 17 is installed in the rapid mixing basin 42, but the imaging device 17 may be installed in either the rapid mixing basin 42 or the flocculation basin 43. . As a result, for example, the information processing apparatus 10 can display not only image data for 300 seconds showing the formation of agglomeration nuclei in the rapid mixing tank 42, but also the initial state of floc formation in the flocculation tank 43. Image data for 100 seconds, that is, image data for 400 seconds after the addition of the flocculant is also used when outputting the expected category, so that 400 seconds of teacher data can be used to the maximum to obtain more accurate data. Expected categories can be output.

Although the embodiments of the present invention have been described above, the present invention is not limited to these embodiments.

The present invention supplies a program that implements one or more functions of the above-described embodiments to a system or device via a network or a storage medium, and one or more processors in the computer of the system or device read and execute the program. It can be implemented by processing, and it can also be implemented by a circuit (eg, an ASIC) that implements one or more functions.

10 information processing device 11 CPU
15 data input unit 17 imaging device

Claims (7)

An information processing device for use in water treatment that removes flocs formed by adding a flocculating agent to water containing turbidity,
Image data showing flocs and flocs obtained within a first time after the addition of the flocculating agent and being nuclei when the turbidity flocculates; a learning means for learning teacher data that associates category information based on floc sedimentation and water turbidity , whether or not water treatment is adversely affected;
a category based on floc sedimentation and turbidity of the water , whether the floc obtained and removed within a second time after the flocculant is added adversely affects the water treatment; input means for inputting image data having no information, wherein the input image data indicates agglutination nuclei ;
When the floc contained in the input image data is removed from the water based on the learning result of learning the teacher data and the input image data showing the aggregation nuclei, the water treatment is adversely affected. determining means for determining whether or not flocs to be formed in the water corresponding to the input image data will adversely affect the water treatment. means and
and output means for outputting information of the determined category . 2. An information processing apparatus according to claim 1, wherein said first time is 400 seconds. 3. An information processing apparatus according to claim 1, wherein said second time is 400 seconds. 3. An information processing apparatus according to claim 1, wherein said second time is 300 seconds. 3. An information processing apparatus according to claim 1, wherein said second time is 120 seconds. An information processing method used in water treatment for removing flocs formed by adding a flocculating agent to water having turbidity,
Image data showing flocs and flocs obtained within a first time after the addition of the flocculating agent and being nuclei when the turbidity flocculates; a learning step of learning supervised data that associates categorical information based on floc sedimentation and water turbidity , whether or not water treatment is adversely affected;
a category based on floc sedimentation and turbidity of the water , whether the floc obtained and removed within a second time after the flocculant is added adversely affects the water treatment; an input step of inputting image data having no information, wherein the input image data indicates agglutination nuclei ;
When the floc contained in the input image data is removed from the water based on the learning result of learning the teacher data and the input image data showing the aggregation nuclei, the water treatment is adversely affected. a determination step of determining whether or not the flocs formed in the water corresponding to the input image data in the future adversely affect the water treatment. a step;
and an output step of outputting the information of the determined category . A program for causing a computer to execute an information processing method used in clean water treatment for removing flocs formed by adding a coagulant to water having turbidity,
The information processing method includes:
Image data showing flocs and flocs obtained within a first time after the addition of the flocculating agent and being nuclei when the turbidity flocculates; a learning step of learning supervised data that associates categorical information based on floc sedimentation and water turbidity , whether or not water treatment is adversely affected;
a category based on floc sedimentation and turbidity of the water , whether the floc obtained and removed within a second time after the flocculant is added adversely affects the water treatment; an input step of inputting image data having no information, wherein the input image data indicates agglutination nuclei ;
When the floc contained in the input image data is removed from the water based on the learning result of learning the teacher data and the input image data showing the aggregation nuclei, the water treatment is adversely affected. a determination step of determining whether or not the flocs formed in the water corresponding to the input image data in the future adversely affect the water treatment. a step;
and an output step of outputting the information of the determined category .

Images