JP7684655B2 - Method and device for predicting treated water quality at sewage treatment facilities, and method and device for predicting operating conditions - Google Patents

Description

The present invention relates to a method and device for predicting treated water quality at a sewage treatment facility, and a method and device for predicting the operating state, and in particular to a method and device for predicting treated water quality at a sewage treatment facility, which is capable of predicting the treated water quality and the operating state at a sewage treatment facility with high accuracy through image analysis using activated sludge images.

Biological treatment using microorganisms is used to treat wastewater containing organic matter. In this biological treatment, a collection of microorganisms artificially cultivated using the organic matter contained in sewage or wastewater as a substrate is called activated sludge. In this activated sludge, a variety of organisms, including bacteria (heterotrophs), filamentous fungi, yeast, protozoa, and metazoans, interact with each other to treat the organic matter, nitrogen, phosphorus, and other substances in the water.

The basic sewage treatment flow of a sewage treatment facility using this type of activated sludge method is shown in Figure 1. This sewage treatment facility mainly comprises a primary settling tank 10 for removing solid contaminants such as toilet paper and sand from the sewage flowing in from the sewer line, a biological reaction tank 20 for removing organic matter, nitrogen (ammonia), and phosphate through the action of microbial groups, and a final settling tank 30 for separating the activated sludge and treated water into solids and liquids by natural settling.

The main objects and items treated at sewage treatment plants are: organic matter, biological oxygen demand (BOD), chemical oxygen demand (COD), and suspended solids (SS); nitrogen, total nitrogen (organic nitrogen + inorganic nitrogen) T-N, ammoniacal nitrogen (NH ₄ -N), nitrate nitrogen (NO ₃ -N), and nitrite nitrogen (NO ₂ -N); and phosphorus, total phosphorus (organic phosphorus + inorganic phosphorus) TP and phosphate phosphorus PO ₄ -P.

In sewage treatment facilities that use activated sludge as described above, activated sludge monitoring is necessary for stable operation, and in many facilities, activated sludge is observed using a microscope. However, the information obtained from the images of activated sludge depends heavily on the skill of the observer, and it cannot be said that the observation results are effectively utilized in each facility. Furthermore, although optical observation using a microscope is the only way to grasp the condition of activated sludge as a routine test within a facility, it is extremely difficult to quantitatively judge biological reactions in which a variety of biological species are intricately related.

A method has been proposed in the past that uses image analysis to identify biological species in order to diagnose the treatment status from microscopic observation images of sludge (Patent Document 1).

A method has also been proposed that uses image analysis to identify and measure filamentous bacteria in flocs formed by microorganisms (groups) in activated sludge from sludge images (Patent Documents 2-4).

Furthermore, in order to optimize the amount of flocculant added, a method has been proposed in which sludge flocs are observed, features are extracted from the floc images, and the sludge flocculation rate is quantitatively evaluated (Patent Document 5).

Japanese Unexamined Patent Publication No. 8-197084 Japanese Patent Application Publication No. 5-192678 JP 2015-181374 A Japanese Unexamined Patent Publication No. 2-129765 JP 2019-52985 A

In the method described in Patent Document 1, the treatment state is estimated based on protozoa and metazoa, but these organisms grow or decrease in number following the increase or decrease in bacteria, which react more sensitively to the state of biological treatment, so there is a delay in catching up with the actual changes in the treatment state.

In addition, with regard to the methods described in Patent Documents 2-4, filamentous fungi are considered to be one indicator of abnormal performance of water treatment facilities, such as bulking, but it is difficult to explain biological treatment performance, which involves the interaction of complex reactions between bacteria and biological species, by the amount of filamentous fungi.

In addition, the method described in Patent Document 5 aims to optimize the dosage of coagulant based on floc shape and concentration, and does not contribute to the evaluation of wastewater or sewage treatment performance based on biological reactions.

Furthermore, when observing activated sludge in sewage treatment plants, it is difficult to quantitatively evaluate the performance of the activated sludge through manual observation. There are problems, such as the inability to manually determine all image features for a biological reaction system in which a wide variety of bacteria interact.

The present invention was made to solve the above-mentioned problems in the past, and its first objective is to enable highly accurate evaluation of treated water quality at sewage treatment facilities through image analysis using activated sludge images.

The second objective of the present invention is to make it possible to predict the operating status of a sewage treatment facility using the results of the predicted treated water quality.

In sewage treatment using activated sludge, bacteria such as those of the genus Zoogloea, Bacillus, Pseudomonas, and Flavobacterium are the ones that contribute most to the treatment of pollutants such as organic matter, nitrogen, and phosphorus. These bacteria are observed as flocs in optical observations such as microscope observations, which are routinely performed in sewage treatment facilities. These flocs have complex shapes with various characteristics such as size, shade, and color due to the presence of extracellular polymeric substances released by bacteria and filamentous bacteria. The present invention makes it possible to quantitatively evaluate the performance of activated sludge by performing image analysis on flocs and filamentous bacteria, which are aggregates of bacteria that account for the majority of the performance of the activated sludge.

As mentioned above, the shape of activated sludge, which is a combination of flocs and filamentous bacteria, is complex, and it is difficult for humans to comprehensively capture all of the features. For this reason, we adopted a convolutional neural network (CNN) as an image analysis method. Examples of CNNs include LeNet, AlexNet, VGG, ResNet, and Inception. Unlike general neural networks, CNNs are characterized by stacking convolutional layers and pooling layers, which allows processing that takes spatial information into account. This feature allows features in activated sludge images to be extracted without human intervention, and the features can be detected regardless of position. The image data converted in this stacked structure is converted into a multidimensional vector in the fully connected layer, and a neural network or the like is further used to obtain the output layer from the fully connected layer. In class classification, which divides images into certain classes, the output layer obtains an output value for each class, and a softmax function, for example, can be used as an activation function to obtain the output layer, which is the classification result, from the fully connected layer. As a result, when an arbitrary image is analyzed, the sum of the output values of each class is 1, and each output value can be considered as the class into which the analyzed image is classified. A sigmoid function can also be used for two-classification problems. In this invention, activated sludge images were prepared and trained on evaluation indices such as SS, which represent the treated water quality, and it was found that this class membership probability indicates the performance of the activated sludge and has a high correlation with the treated water quality.

Furthermore, even in activated sludge sampled at the same time, there are parts with good performance and parts with poor performance. For this reason, images are not judged based on only one image, but multiple images, preferably 20 or more, and more preferably 100 or more, are used to calculate the probability of each class belonging, and statistical processing is then performed on these results, making it possible to calculate the overall performance of the sampled activated sludge.

The present invention has been made based on the above findings, and solves the first problem by providing a method for predicting treated water quality at a sewage treatment facility, comprising: a step of setting standards for classifying treated water quality; an activated sludge image labeling step of determining which class of treated water quality an image belongs to based on the treated water quality linked to the activated sludge image, depending on the treated water quality collected within a set average residence time of the sewage treatment facility from the time the activated sludge was collected, and attaching the class to the image as a label , thereby labeling the image; a learning step of training an image classification algorithm using a convolutional neural network using the labeled activated sludge images as teacher images; an image analysis step of collecting activated sludge from the sewage treatment facility, collecting images, analyzing unknown sludge images with the learned algorithm, and outputting the class belonging probability for each image ; and a correlation analysis step of statistically processing the class belonging probability for each image collected from the same activated sludge sample, and analyzing the correlation with the treated water quality.

Here, at least one of the classes can be set to 80% or less of the treated water quality standard.

At least one of the classes can be set to 200% or more of the treated water quality standard.

In addition, when statistically processing the class membership probabilities in the correlation analysis step, the class membership probabilities can be averaged.

In addition, when statistically processing the class membership probabilities in the correlation analysis step, the median of the class membership probabilities can be taken.

In addition, when statistically processing the class membership probability in the correlation analysis step, a threshold value for the class membership probability can be set and the proportion of images that are above or below that threshold can be determined.

The present invention also solves the second problem by providing a method for predicting the operational status of a sewage treatment facility, which is characterized in that changes in the statistically processed value of the class membership probability calculated in the correlation analysis step of the method for predicting the treated water quality of the sewage treatment facility precede changes in the treated water quality, and use this as an indicator of changes in the value of the treated water quality .

The present invention also solves the first problem by providing a treated water quality prediction device for a sewage treatment facility, comprising: a means for setting standards for classifying treated water quality; an activated sludge image labeling means for determining which class of treated water quality an image belongs to based on the treated water quality linked to the activated sludge image, depending on the treated water quality collected within a set average residence time of the sewage treatment facility from the time the activated sludge was collected, and labeling the image by attaching the class as a label ; a learning means for learning the labeled activated sludge images using an image classification algorithm that utilizes a convolutional neural network as teacher images; an image analysis means for collecting activated sludge from the sewage treatment facility, collecting images, analyzing unknown sludge images using the learned algorithm, and outputting the class belonging probability for each image ; and a correlation analysis means for statistically processing the class belonging probability for each image collected from the same activated sludge sample, and analyzing the correlation with the treated water quality.

The present invention also solves the second problem by providing an operational status prediction device for a sewage treatment facility, which is characterized in that changes in the statistically processed value of the class membership probability calculated by the correlation analysis means of the treated water quality prediction device for the sewage treatment facility precede changes in the treated water quality, and use this as an indicator of changes in the value of the treated water quality .

According to the present invention, (I) a convolutional neural network is used as an image analysis method, (II) activated sludge images are labeled according to the treated water quality and used as teacher images, and (III) new images are evaluated using the class membership probability for each image , thereby making it possible to predict the treated water quality of a sewage treatment facility with high accuracy.

In addition, by using the results of the predicted treated water quality, it is possible to detect signs of changes in the properties of activated sludge and predict the operating status of sewage treatment facilities.

A diagram showing the basic sewage treatment flow at a sewage treatment facility FIG. 1 is a diagram showing the overall configuration of an example of a sewage treatment facility to which the present invention is applied. FIG. 1 is a block diagram, including a partial flow diagram, illustrating the concept of the present invention. A flowchart showing an example of a teacher image labeling method used in the present invention. A diagram showing an example of the probability distribution of image class membership based on statistical processing of image analysis results. FIG. 1 is a block diagram, including a partial flow diagram, illustrating the concept of an embodiment of the present invention. FIG. 1 is a diagram showing an example of the configuration of a convolutional neural network used in the present invention. A flowchart showing the procedure for labeling teacher images. FIG. 1 is a diagram showing the overall configuration of an embodiment of the present invention. A diagram comparing examples of (A) sludge with poor settling properties and (B) sludge with good settling properties. FIG. 13 is a diagram showing an example of a processing result of the embodiment when the median of class membership probability is used for statistical processing of the class membership probability. FIG. 13 is a diagram showing an example of the change in the median value of the treated water SS and class membership probability over time. FIG. 13 shows an example of the processing result when the average value of class membership probability is used for statistical processing of class membership probability. FIG. 13 shows an example of a processing result when the discrimination ratio of class membership probability is used for statistical processing of class membership probability.

The following describes in detail the embodiments of the present invention with reference to the drawings. Note that the present invention is not limited to the contents described in the following embodiments. Furthermore, the components in the embodiments described below include those that a person skilled in the art can easily imagine, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the components disclosed in the embodiments described below may be appropriately combined or appropriately selected for use.

The overall configuration of an example of a sewage treatment facility to which the present invention can be applied is shown in Figure 2.

This sewage treatment facility mainly comprises a primary sedimentation tank 10 into which sewage flows in from a sewage line 8, a biological reaction tank 20 connected to the primary sedimentation tank 10 via an inlet line 12, and a final sedimentation tank 30.

In addition, a grit basin (not shown) is provided at the inlet side of the initial sedimentation basin 10 to remove large, heavy impurities and protect pumps, etc.

The biological reactor 20 is equipped with an anaerobic tank 22 for releasing phosphorus from within the cells and synthesizing PHA within the cells, an anaerobic tank 24 for oxidatively decomposing organic matter using oxygen in nitric acid and reducing the nitric acid to nitrogen gas, and an aerobic tank 26 for oxidatively decomposing organic matter using dissolved oxygen, converting ammonia to nitrate, decomposing PHA, and removing phosphorus by excessively accumulating phosphate within the cells.

At the outlet of the final settling tank 30, there is a treated water line 32 for disinfecting the treated water and discharging it into a river or the like, a sludge return line 34 for returning a portion of the sludge to the inlet of the biological reaction tank 20, and an excess sludge line 36 for dehydrating and discharging the excess sludge.

The sludge return line 34 is equipped with a return pump 38 and a flow meter 39, and a portion of the sludge accumulated in the final settling tank 30 is returned to the inlet side of the biological reaction tank 20. The amount of returned sludge is adjusted by the return pump 38, thereby maintaining and adjusting the MLSS (sludge concentration).

The excess sludge line 36 is provided with an excess sludge pump 40 and a flow meter 41, and the excess sludge is dehydrated and discharged. The amount of excess sludge is adjusted by the excess sludge pump 40, which adjusts the amount of sludge in the system, regulates the metabolism of the sludge, and controls the sludge retention time.

The aerobic tank 26 is equipped with an air blower 42 and an air volume meter 43 to blow air into it. The volume of air from the air blower 42 is used to maintain and control the DO (dissolved oxygen concentration).

A nitrification liquid circulation line 44 is provided at the outlet of the aerobic tank 26 of the biological reaction tank 20 to circulate the nitrification liquid to the anoxic tank 24, and a nitrification liquid circulation pump 46 is provided on this nitrification liquid circulation line 44. The amount of nitrification liquid circulated by this nitrification liquid circulation pump 46 is used to adjust the amount of nitrate nitrogen supplied to the anoxic tank 24 and remove nitrogen.

The inflow line 12 is provided with a water volume sensor 14, the anaerobic tank 22 is provided with a sensor 52, for example, a DO (dissolved oxygen concentration) sensor, a pH sensor, and an ORP (oxidation-reduction potential) sensor, the anoxic tank 24 is provided with a sensor 54, for example, a DO sensor, a pH sensor, and an ORP sensor, the aerobic tank 26 is provided with a sensor 56, for example, a DO sensor, a MLSS (sludge concentration) sensor, a pH sensor, and an ORP sensor, and the treated water line 32 is provided with a water quality sensor 58.

The outputs of the sensors 14, 52, 54, 56, and 58 are input to a computer 60, and actuators such as pumps and blowers are controlled based on the results of calculations performed by the computer 60.

Here, pH is used to manage environmental conditions, and is maintained at, for example, pH 7-8 in the anaerobic tank 22 and anoxic tank 24, and at, for example, pH 6-7 in the aerobic tank 26.

The ORP (oxidation-reduction potential) is also used to manage environmental conditions, and is maintained at, for example, less than -300 to -400 mV in the anaerobic tank 22, and at, for example, less than 0 to -200 mV in the anoxic tank 24.

The DO (dissolved oxygen concentration) is also used to manage environmental conditions, and is maintained at, for example, less than 0.2 mg/L in the anaerobic tank 22 and anoxic tank 24, and at, for example, 1.0 to 2.0 mg/L in the aerobic tank 26.

The MLSS (sludge concentration) is used to manage the amount of microorganisms in the activated sludge, and is maintained at, for example, 2,000 to 3,000 mg/L overall in terms of dry weight of the centrifuged sludge.

The sludge properties analyzed include, for example, the MLSS, the activated sludge settling rate SV30, which is the scale of the sludge interface when settling for 30 minutes in a 1 L graduated cylinder, and the sludge settling target SVI (= SV30 x 10,000/MLSS), which is the volume occupied by 1 g of sludge after 30 minutes of settling.

The following describes the activated sludge determination method using image analysis according to the present invention on the computer 60 in Figure 2.

The computer 60 includes a model construction unit 100 and a sludge property diagnosis unit 200, as shown in FIG. 3.

The model construction unit 100 collects activated sludge (step 110), takes images (step 120), labels the images (step 130), performs supervised learning (step 140), and constructs a model (step 150).

Meanwhile, the sludge property diagnosis unit 200 collects new activated sludge (step 210), takes images (step 220), processes them using the trained model constructed in step 150 of the model construction unit 100 (step 230), outputs the class membership probability of each image (step 240), performs statistical processing on each output result (step 250), and outputs the diagnosis result for the collected sludge (step 260).

The method of image labeling in step 130 of the model construction unit 100 is shown in Figure 4.

First, activated sludge is collected (step 110) and an image of the sludge is taken (step 120).

Activated sludge is retained in the treatment facility for a set sludge retention time (SRT). Therefore, using the sludge collection time as a reference, the treated water quality (denoted as X) is acquired within this SRT (step 124), and the acquired treated water quality (X) is linked to the sludge image. Since activated sludge is constantly replaced, it is desirable that the treated water quality to be linked is the treated water quality collected near the sludge retention time. More desirably, it is the treated water quality collected during the hydraulic retention time (HRT).

Next, in step 130, the image is labeled based on the treated water quality associated with the image. The number of classes is set to, for example, four (classes A, B, C, and D in the figure), but the number of classes may be two or more.

For the remaining classes ( two classes B and C in the figure), intermediate values may be filled in as shown in FIG. 4, or further outer classifications may be created.

Regarding the classification of teacher images, the reference value of the treated water SS is T, and the treated water SS associated with the image is Y.
i: 0.8T<Y≦T and T≦Y<2T
ii: Y≦0.8T and T≦Y<2T
iii: 0.8T<Y≦T and 2T≦Y
iv: Y≦0.8T and 2T≦Y
When the classes were classified as above, the results of analyzing the activated sludge images using the learning models with the respective teacher images were compared and examined. The results are shown in Table 1.

When comparing the coefficient of determination ^R2 , which indicates the correlation between the sludge diagnostic results from image analysis and the treated water quality, a higher correlation was obtained when the image classification of the teacher image included a class where the treated water quality Y is 0.8T or less (ii, iv) and when the image classification included a class where the treated water quality Y is 2T or more (iii , iv) compared to when neither of these classes were included (i) . Furthermore, an even higher correlation was obtained when both classes were included (iv) .

Therefore, for at least one of the aforementioned classes (class A in the figure), when the reference value of the treated water SS is T, it is desirable to set the upper threshold value to 0.8T or less.

It is also desirable to set the lower threshold of at least one (Class D in the figure) to 2T or higher.

For the remaining classes (two classes B and C in the figure), intermediate values may be filled in as shown in FIG. 4, or further outer classifications may be created.

During the statistical processing step 250 of the sludge property diagnosis unit 200, image analysis is performed on multiple sludge images taken from the same sludge sample using a trained CNN model, and the probability of each image belonging to a class is output.

An example of the class membership probability distribution is shown in Figure 5. Since activated sludge contains sludge with various microscopic characteristics, the class membership probability of each sludge varies. The following indices can be used for the macroscopic evaluation of sludge samples.

1. The average probability of each image belonging to a class is used as the evaluation value.
2. The median of the class membership probability for each image is used as the evaluation value.
3. The ratio of images whose class belonging probability exceeds a certain threshold (e.g., 0.7) among the collected images is used as the evaluation value.
4. The ratio of images whose class belonging probability is below a certain threshold (e.g., 0.3) among the collected images is used as an evaluation value.
5. Among the collected images, the number of images whose class belonging probability is above a certain threshold is designated as A, and the number of images whose class belonging probability is below a certain threshold is designated as B. The evaluation axis is A/(A+B) or B/(A+B). Images that fall in between are deemed indistinguishable and are not used.

The reasons why statistical processing of class membership probability is necessary are as follows:

The performance of activated sludge is determined by the activity of the microbial community. In a microscopic view, such as an image observed under a microscope, sludge in good condition and sludge in bad condition are mixed. The properties of the entire activated sludge are determined by which sludge occupies the area. By calculating the class belonging probability for each image taken from the same sample, the good or bad performance from a microscopic view is calculated. By adding statistical processing to this, the overall sludge performance can be derived. That is, as shown in the coefficient of determination R ² in Table 2, the coefficient of determination R ² is low, for example, 0.360, from the class belonging probability of each image alone, so the treatment performance of the sludge cannot be evaluated. In contrast, by adding statistical processing using the median, average, or discrimination ratio as described below, the coefficient of determination R ² can be increased to, for example, 0.735, 0.746, or 0.715.

Next, an embodiment of the present invention in which treated water SS is used as an evaluation index will be described.

The model construction unit 100 and sludge property diagnosis unit 200 of this embodiment are configured as shown in FIG. 6, and when performing image labeling in step 130, if the reference value for the treated water SS is set to 100, treated water SS of 80 or less is considered normal, and treated water SS of 200 or more is considered abnormal.

In the supervised learning of step 140, a CNN having an input image 300, a repetition of a convolutional layer 310 and a pooling layer 320, a fully connected layer 330, and an output layer 340, as shown in FIG. 7, is used.

The procedure for teacher image labeling in this embodiment is shown in Figure 8. As in Figure 4, sludge is collected in step 110, and the sludge is photographed in step 120.

In step 124, the treated water SS is obtained, and the treated water quality obtained within this SRT is linked to the sludge image based on the sludge collection time.

In this embodiment, when labeling images in step 130, the standard value for treated water SS is set to 100, and images are classified into class A, where the treated water SS associated with each image is 80 or less, and class B, where the treated water SS is 200 or more. Images that are not included in this classification are not used as training data. Image analysis using CNN is used to classify images into class A and class B.

The overall configuration of the analysis procedure using CNN in this embodiment is shown in Figure 9.

First, (1) obtain images of class A with good settling properties and images of class B with poor settling properties and use them as training images.

Next, (2) use CNN to learn the training images and create a discriminant model.

Next, (3) analyze the new image, and (4) predict the treated water SS based on the analysis results, for example, class A 0.8 and class B 0.2, based on the probability that the sludge belongs to each class.

Figure 10 shows examples of (A) sludge with poor settling properties and (B) sludge with good settling properties. As shown in Figure 10 (A), sludge with poor settling properties has relatively small flocs 2, a low density of flocs 2, a large amount of filamentous bacteria 4, and a large amount of fine particles 6.

On the other hand, sludge with good settling properties has relatively large and dense flocs 2, a small amount of filamentous bacteria 4, and a small amount of microparticles 6, as shown in Figure 10 (B).

The relationship between the treated water SS and the median of the class belonging probability in this embodiment is shown in Figure 11. The coefficient of determination ^R2 was 0.7353.

Figure 12 shows an example of time-dependent change in treated water SS and median class membership probability in an embodiment. It can be seen that the median class membership probability increases prior to an increase in treated water SS. This result shows that not only the change in the median class membership probability but also the median class membership probability similarly increases prior to an increase in treated water SS, so the median class membership probability can simply be used as an indicator of the deterioration of treated water SS.

As another example of statistical processing of the class belonging probability, an example of the average value is shown in Fig. 13, and an example of the ratio of images whose class belonging probability exceeds a threshold value among the acquired images (called the discrimination ratio) is shown in Fig. 14. The coefficient of determination ^R2 in Fig. 13 was 0.7455, and the coefficient of determination ^R2 in Fig. 14 was 0.7157.

In the above embodiment, the treated water SS was the subject of prediction, but other treated water qualities managed in the sewage treatment facility can also be predicted in a similar manner. Other treated water qualities include, for example, chemical oxygen demand (COD), ammonia nitrogen, nitrate nitrogen, total nitrogen, total phosphorus, etc.

In the above embodiment, the present invention is applied to a sewage treatment facility in which the biological reaction tank 20 includes the anaerobic tank 22, the anoxic tank 24, and the aerobic tank 26 for removing organic matter, nitrogen, and phosphorus by the anaerobic-anoxic-aerobic method. However, the present invention is not limited to this application. For example, the present invention can be applied to a sewage treatment facility in which the biological reaction tank does not include an anaerobic tank for removing organic matter and nitrogen by the anoxic-aerobic method (circulating nitrification denitrification method), a sewage treatment facility in which the biological reaction tank does not include an anoxic tank, but includes an anaerobic tank and an aerobic tank for removing organic matter and phosphorus by the anaerobic-aerobic method, and a sewage treatment facility in which the biological reaction tank consists of only an aerobic tank for removing organic matter by the standard activated sludge method.

Description of the Reference Numerals 8: Sewage line 10: Primary sedimentation tank 12: Inflow line 20: Biological reaction tank 22: Anaerobic tank 24: Anoxic tank 26: Aerobic tank 30: Final sedimentation tank 32: Treated water line 34: Sludge return line 36: Excess sludge line 38: Return pump 40: Excess sludge pump 42: Air blower 60: Computer 100: Model construction section 200: Sludge property diagnosis section

Claims (14)

establishing standards for classifying treated water quality;
an activated sludge image labeling step for determining which class of treated water quality the corresponding image belongs to based on the treated water quality associated with the activated sludge image according to the treated water quality collected within a set average residence time of the sewage treatment facility from the time the activated sludge was collected, and labeling the image by attaching the class as a label;
a learning step of learning the image-labeled activated sludge image as a teacher image using an image classification algorithm utilizing a convolutional neural network;
an image analysis step of collecting activated sludge from a sewage treatment facility, collecting images, analyzing unknown sludge images using a trained algorithm, and outputting a class membership probability for each image ;
a correlation analysis step of statistically processing the class membership probability for each image collected from the same activated sludge sample and analyzing the correlation with the treated water quality;
A method for predicting treated water quality at a sewage treatment facility, comprising: 2. The method for predicting treated water quality at a sewage treatment facility according to claim 1, wherein at least one of the classes is set to be 80% or less of the treated water quality standard. 2. The method for predicting treated water quality at a sewage treatment facility according to claim 1, wherein at least one of the classes is set to be 200% or more of the treated water quality standard. A method for predicting the quality of treated water from a sewage treatment facility according to any one of claims 1 to 3, characterized in that, when statistically processing the class membership probabilities in the correlation analysis step, the class membership probabilities are averaged. A method for predicting the quality of treated water at a sewage treatment facility according to any one of claims 1 to 3, characterized in that the median of the class membership probabilities is taken during the statistical processing of the class membership probabilities in the correlation analysis step. A method for predicting the quality of treated water at a sewage treatment facility according to any one of claims 1 to 3, characterized in that, in the statistical processing of the class membership probability in the correlation analysis step, a threshold value for the class membership probability is set, and the proportion of images that are above or below the threshold value is calculated. A method for predicting the operational status of a sewage treatment facility, characterized in that changes in the statistically processed value of the class membership probability calculated in the correlation analysis step of the treated water quality prediction method for a sewage treatment facility described in any one of claims 1 to 6 precede changes in the treated water quality, and are used as an indicator of changes in the value of the treated water quality. A means for establishing standards for classifying treated water quality;
an activated sludge image labeling means for determining which class of treated water quality the corresponding image belongs to based on the treated water quality linked to the activated sludge image in accordance with the treated water quality sampled within a set average residence time of the sewage treatment facility from the time the activated sludge was sampled, and labeling the image by attaching the class as a label;
a learning means for learning the image-labeled activated sludge image as a teacher image using an image classification algorithm utilizing a convolutional neural network;
an image analysis means for collecting activated sludge from a sewage treatment facility, collecting images, analyzing unknown sludge images using a trained algorithm, and outputting a class membership probability for each image ;
A correlation analysis means for statistically processing the class belonging probability for each image collected from the same activated sludge sample and analyzing the correlation with the treated water quality;
A treated water quality prediction device for a sewage treatment facility, comprising: 9. The treated water quality prediction device for a sewage treatment facility according to claim 8, wherein at least one of the classes is set to be 80% or less of the treated water quality standard. 9. The treated water quality prediction device for a sewage treatment facility according to claim 8, wherein at least one of the classes is set to be 200% or more of the treated water quality standard. A treated water quality prediction device for a sewage treatment facility according to any one of claims 8 to 10, characterized in that, when the class belonging probabilities are statistically processed in the correlation analysis means, the class belonging probabilities are averaged. A treated water quality prediction device for a sewage treatment facility according to any one of claims 8 to 10, characterized in that the median of the class belonging probabilities is taken during statistical processing of the class belonging probabilities in the correlation analysis means. A sewage treatment facility treated water quality prediction device according to any one of claims 8 to 10, characterized in that, when the correlation analysis means statistically processes the class membership probability, a threshold value for the class membership probability is set, and the proportion of images that are above or below the threshold value is calculated. A device for predicting the operational status of a sewage treatment facility, characterized in that changes in the value of the statistically processed class membership probability calculated by the correlation analysis means of the device for predicting the treated water quality of a sewage treatment facility described in any one of claims 8 to 13 precede changes in the treated water quality, and are used as an indicator of changes in the value of the treated water quality.

Images