Control device, water treatment method, and water treatment program

Abstract

The present invention provides a control device, a water treatment method, and a water treatment program for efficient and appropriate wastewater treatment. [Solution] The state determination unit uses an estimation model 14 that has learned the relationship between the measurement information and the floc overflow state, based on the measurement information obtained by an ultrasonic sensor placed in a sedimentation tank for settling and separating inorganic flocs in wastewater treatment and the result of identifying the floc overflow state, to confirm the floc overflow state at a predetermined timing, which is output from the estimation model 14 based on the first measurement information obtained by the ultrasonic sensor at a predetermined timing. The control unit outputs information related to wastewater treatment control based on the floc overflow state confirmed by the state determination unit.

Description

Technical Field

[0004]

[0001] The present invention relates to a control device, a water treatment method, and a water treatment program. [[ID=]6]

Background Art

[0002] Conventionally, in order to reduce the environmental impact caused by industrial wastewater and the like, wastewater treatment has attracted attention. For example, a technique has been proposed for determining the internal state from a supernatant water image by using a learning model that has learned the relationship between the supernatant water image showing the supernatant water inside the solid-liquid separation layer and the state inside the solid-liquid separation layer. In addition, a control device has been proposed that controls the addition amount of a flocculant based on an image of the state of flocs in water to which the flocculant has been added and at least one measurement result of the pressure at the inlet of the filtration membrane, the filtration flow rate, or the water level in the filtration membrane. Further, a technique has been proposed for obtaining the degree of aggregation and concentration from a photographed image of flocs by using a machine learning model that has learned the relationship between the image features extracted from the photographed image of flocs and the degree of aggregation and concentration of the flocs. In addition, a technique has been proposed for calculating the predicted treated water turbidity from an image of the water to be treated by using a pass / fail determination model created using images of the water to be treated taken at different locations and measurement results of the turbidity of the water to be treated and the treated water.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Patent Document 2

Patent Document 3

Patent Document 4

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, the water quality of wastewater from the production process fluctuates from day to day, and homogenization in the raw water tank is insufficient to absorb these fluctuations, making efficient and appropriate wastewater treatment difficult. Furthermore, efficient and appropriate wastewater treatment is difficult with any of the conventional technologies mentioned above.

[0005] The disclosed technology was made in view of the above and aims to provide a control device, a water treatment method, and a water treatment program for efficient and appropriate wastewater treatment. [Means for solving the problem]

[0006] In one embodiment of a water treatment method and a water treatment program, the control device and state determination unit disclosed in this application determine the floc overflow state at a predetermined timing based on first measurement information acquired by the ultrasonic sensor at a predetermined timing and information specifying the floc overflow state, using an estimation model that has learned the relationship between the measurement information and the floc overflow state. The control unit outputs information related to the control of the wastewater treatment based on the floc overflow state determined by the state determination unit. [Effects of the Invention]

[0007] In one respect, the present invention enables efficient and appropriate wastewater treatment. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 is a schematic diagram showing the overall wastewater treatment system. [Figure 2] Figure 2 is a block diagram of the control system. [Figure 3] Figure 3 shows an example of the arrangement of ultrasonic sensors and an ultrasonic image. [Figure 4] Figure 4 shows an example of an ultrasonic image for each label of the overflow state according to the embodiment. [Figure 5] Figure 5 shows an example of the estimation results. [Figure 6] Figure 6 shows an example of a control pattern classification table. [Figure 7] Figure 7 is a diagram showing the overall configuration of the control system. [Figure 8] Figure 8 is a flowchart of the learning process for the estimation model by the learning processing unit. [Figure 9] Figure 9 is a flowchart of the wastewater treatment process by the wastewater treatment system. [Figure 10] Figure 10 is a flowchart of the control process performed by the control device. [Figure 11] Figure 11 is a flowchart showing the control pattern selection process. [Figure 12] Figure 12 shows a comparison between the control device according to the embodiment and the case where manual checks of the overflow state are used. [Figure 13] Figure 13 shows a comparison of the ultrasonic image and the water surface image used in the control processing according to the embodiment. [Figure 14] Figure 14 is a hardware configuration diagram of a computer. [Modes for carrying out the invention]

[0009] Embodiments of the control device, water treatment method, and water treatment program disclosed in this application will be described in detail below with reference to the drawings. However, the control device, water treatment method, and water treatment program disclosed in this application are not limited to the embodiments described below.

[0010] (Wastewater treatment system) FIG. 1 is a schematic diagram showing the overall wastewater treatment system. In the inorganic treatment process of wastewater treatment, a flocculant or polymer is injected into the raw water according to the amount of suspended solids (SS) in the raw water, the chemical oxygen demand (COD), and the concentration of removal substances such as heavy metal ions and inorganic phosphorus. As the flocculant, for example, aluminum sulfate or polyaluminum chloride is used. For example, by allowing the injected flocculant to stay in the raw water for about 3 to 6 hours, the concentration is equalized. More specifically, in the reaction flocculation process during the inorganic treatment process, the flocculant is injected into the raw water, and the removal substances in the raw water are flocculated. The flocculated removal substances are separated from the raw water by solid-liquid separation in a solid-liquid separation tank.

[0011] Here, in wastewater treatment, the quality of the wastewater from the production process varies depending on the date and time, and it is difficult to absorb the fluctuations in water quality by equalization in the raw water tank. Therefore, in order to reduce the occurrence of poor flocculation due to the influence of fluctuations, it is common to inject more flocculant. When more flocculant is injected, the excess flocculant becomes sludge, which may increase the running cost of the wastewater treatment plant. Also, even if more flocculant is injected, if wastewater containing more removal substances than the amount that can be treated by the injected flocculant is discharged from the production process, there is a risk that the flocculant will be insufficient and the flocs will overflow into the next process without being completely separated in the solid-liquid separation tank. Therefore, the wastewater treatment system 100 performs the following wastewater treatment.

[0012] As shown in FIG. 1, in the wastewater treatment system 100, each treatment in the raw water process S1, reaction process S2, chemical process S3, sedimentation process S4, filtration activated carbon process S5, and discharge process S6 is performed.

[0013] In the raw water process S1, the factory wastewater is received and stored in the raw water tank 101. The wastewater is equalized by storing it in the raw water tank 101. The equalized wastewater is called raw water.

[0014] In reaction step S2, the raw water is transferred from the raw water tank 101 to the reaction coagulation tank 102. Then, the inorganic substances in the raw water react with the chemicals, and the pollutants are suspended. In other words, the inorganic substances in the raw water are coagulated, and inorganic flocs are formed. Flocs are substances that have accumulated and solidified from fine particles in wastewater.

[0015] In the chemical process S3, chemicals such as flocculants and polymers, including aluminum sulfate and polyaluminum chloride, are stored. The stored chemicals are then injected into the reaction flocculation tank 102 by a pump.

[0016] In the sedimentation process S4, the raw water containing suspended pollutants is transferred from the reaction coagulation tank 102 to the sedimentation tank 103. The suspended flocs then settle in the sedimentation tank 103, separating the solid and liquid. However, while flocs of a certain size will settle in the sedimentation tank 103 due to their own weight, smaller flocs will float inside the tank. The interface between the water and the settled, stacked flocs is called the interface. The settled and separated flocs are withdrawn from the sedimentation tank 103, dried, and then discarded.

[0017] In the activated carbon filtration process S5, the treated water from which the flocs have been separated flows out of the sedimentation tank 103 and is received. The treated water is then subjected to activated carbon filtration treatment to remove residual substances. The phenomenon of treated water flowing out of the sedimentation tank 103 into the activated carbon filtration process S5 is called "overflow." In this context, the overflow of flocs from the sedimentation tank 103 into the activated carbon filtration process S5 is called "floc overflow."

[0018] In the discharge process S6, treated water that has had residual substances removed in the filtration and activated carbon process S5 and that meets legal requirements is discharged.

[0019] (Control system) In the wastewater treatment system 100 described above, in reaction step S2, fluctuations in the quality of the raw water may cause an excess or deficiency of chemicals used to coagulate inorganic substances and generate flocs. If there is an excess of chemicals, the chemicals themselves become sludge, requiring treatment to remove the sludge. If there is a shortage of chemicals, there is a risk of flocs overflowing into the next step, the activated carbon filtration step S5. In the activated carbon filtration step S5, the flocs are removed by a filter, but if a large amount of flocs overflow, this filter is subjected to a heavy load. In addition, cleaning of the filter is necessary to maintain it in proper condition, which increases the workload of cleaning.

[0020] Therefore, the control device 1 controls the amount of chemical injected into the reaction flocculation tank 102 based on ultrasonic images obtained using the ultrasonic sensor 2, display device 3, and camera 4 located in the sedimentation tank 103, in order to ensure that flocs are properly settled and to suppress the generation of sludge from the chemical itself. The control system 10, including the control device 1, will be described below.

[0021] Figure 2 is a block diagram of the control system. The control system 10 includes a control device 1, an ultrasonic sensor 2, a display device 3, a camera 4, and a learning processing device 5. The control system 10 has two operation phases: a learning phase in which the relationship between ultrasonic images and the overflow state is learned by the estimation model 14, and an estimation phase in which the learned estimation model 14 estimates the overflow state from a given ultrasonic image. The estimation phase is the operation phase when the control system 10 is in operation.

[0022] In the learning phase, the learning processing device 5 trains the estimation model 14 using ultrasonic images, labels indicating the floc overflow state identified from the presence or absence of sludge in the ultrasonic images and the presence or absence of floc overflow at the time the ultrasonic images were taken. In the estimation phase, the control device 1 estimates the floc overflow state of inorganic materials using the trained estimation model 14. Subsequently, the control device 1 controls the chemical process system 20 according to the estimation results. Here, "floc overflow state" refers to each state related to the occurrence of floc overflow, such as whether floc overflow has not occurred and there is no risk of occurrence, whether floc overflow has not occurred but there is a risk of occurrence, or whether floc overflow has occurred.

[0023] (Ultrasound image) Figure 3 shows an example of the arrangement of ultrasonic sensors and an ultrasonic image. Referring to Figure 3, the ultrasonic image used by estimation model 14 for estimation will be explained.

[0024] The sedimentation tank 103 is a tank for settling and separating inorganic flocs in wastewater during wastewater treatment. The sedimentation tank 103 comprises an inner cylinder 131, a rake 132, and a pit 133. Raw water containing suspended pollutants flows into the inner cylinder 131. The raw water is then discharged from the bottom of the inner cylinder 131, i.e., from an opening leading to the pit 133. The flocs agitated in the raw water settle by their own weight and accumulate on the bottom surface 134 of the sedimentation tank 103. This allows for solid-liquid separation within the sedimentation tank 103. The rake 132 collects the accumulated flocs in the pit 133. The flocs collected in the pit 133 are then drawn out of the pit 133. The supernatant water of the treated water from which the flocs have been separated flows out from a predetermined outlet and is sent to the next filtration activated carbon process S5. The supernatant water is the wastewater near the water surface in the sedimentation tank 103 after the flocs have been separated.

[0025] However, if the flocs are small and light, they may not settle but float in the raw water. These floating flocs may flow out with the supernatant water of the treated water and be sent to the next filtration activated carbon process S5, potentially causing floc overflow.

[0026] The ultrasonic sensor 2 only needs to be able to determine the approximate location of flocs floating in the raw water and does not need to have high resolution. The ultrasonic sensor 2 may be, for example, a general-purpose fish finder.

[0027] The ultrasonic sensor 2 is placed, for example, on the surface of the raw water stored in the sedimentation tank 103. The ultrasonic sensor 2 emits ultrasonic pulses toward the bottom surface 134 of the sedimentation tank 103 and acquires ultrasonic data from the reflected waves. The ultrasonic sensor 2 continuously acquires ultrasonic data at predetermined time intervals.

[0028] The ultrasonic sensor 2 does not need to be in contact with the water surface, as long as it can acquire reflected waves between the water surface and the bottom surface 134. For example, the ultrasonic sensor 2 may be away from the water surface, or it may be near the water surface or submerged.

[0029] Furthermore, near the inner cylinder 131 from which the raw water from the sedimentation tank 103 is discharged, the movement of flocs is not settled, making it difficult to accurately grasp the sedimentation of flocs. For this reason, given that the ultrasonic sensor 2 collects ultrasonic data to estimate the floc overflow state, it is preferable to position it in a location where it can acquire data from the area surrounding a predetermined outlet. However, if the floc overflow state is estimated using information from a location too close to the predetermined outlet, the control of the amount of chemical injected using the estimation result may not be timely, increasing the likelihood of floc overflow. Therefore, it is preferable to position the ultrasonic sensor 2 at a certain distance from the predetermined outlet, regardless of the size of the sedimentation tank 103. Moreover, it is preferable to position the ultrasonic sensor 2 at a location where the distance from the predetermined outlet is shorter than the distance from the inner cylinder 131. For example, the ultrasonic sensor 2 is positioned at a depth of 3600 mm to the bottom surface 134, 700 mm from the predetermined outlet, and 3300 mm from the inner cylinder 131. In this way, the ultrasonic sensor 2 is positioned in the area surrounding the place where wastewater flows out of the sedimentation tank 103.

[0030] The ultrasonic data acquired by the ultrasonic sensor 2 is sent to the display device 3. The display device 3 generates an ultrasonic image from the acquired ultrasonic data and displays the generated ultrasonic image on a monitor or the like. Screen 30 in Figure 3 is the display screen of the ultrasonic image on the display device 3. Position 31 in the ultrasonic image displayed on screen 30 corresponds to the water surface of the raw water in the sedimentation tank 103. Position 32 corresponds to the bottom surface 134 of the sedimentation tank 103. Furthermore, the displayed object 33 represents flocs.

[0031] Camera 4 is, for example, a network camera. Camera 4 captures the ultrasound image displayed by the display device 3. The ultrasound image captured by camera 4 is then transmitted to the control device 1. Hereafter, the ultrasound image captured by camera 4 will simply be referred to as the "ultrasound image".

[0032] In this embodiment, estimation is performed using ultrasonic images captured by camera 4, but the data used for estimation is not limited to this. For example, image data displayed by display device 3 may be used directly for estimation. Alternatively, ultrasonic data output from ultrasonic sensor 2 may be used directly for estimation. Furthermore, ultrasonic images may be in color or in black and white. Using ultrasonic images makes it easier to label the data used for training. Also, using ultrasonic data output from ultrasonic sensor 2 makes it easier to track the time-series changes in the data used for training, thus facilitating estimation along the time series.

[0033] Furthermore, although one ultrasonic sensor 2 was used in this embodiment, if the size of the sedimentation tank 103 is large, it is possible to place two or more ultrasonic sensors 2 in one sedimentation tank 103. By using multiple ultrasonic sensors 2, the amount of information used for estimation increases, making it possible to improve the estimation accuracy. For example, two ultrasonic sensors 2 may be placed close together to acquire ultrasonic data at almost the same time, or they may be placed at positions somewhat separated from the center of the sedimentation tank 103 in the direction toward the drainage area, and ultrasonic data may be acquired with a time lag.

[0034] (Estimated model) Figure 2 illustrates two estimation models 14, both of which are the same machine learning model. The estimation model 14 in the learning processing device 5 becomes a trained estimation model 14 by training it in the learning processing device 5 from an untrained state. The control device 1 holds the trained estimation model 14 that has been trained in the learning processing device 5.

[0035] Estimation model 14 is a machine learning model that receives ultrasonic images obtained from camera 4 as input and estimates the floc overflow state of inorganic matter in the sedimentation tank 103. For example, estimation model 14 uses a neural network (NN) to perform convolution on the ultrasonic images, extracts features from the ultrasonic images, and estimates the floc overflow state according to those features. More specifically, estimation model 14 uses training images, which are excerpts of the region from the water surface to a certain depth in the ultrasonic images, to estimate which of the predetermined labels the floc overflow state corresponds to.

[0036] In this embodiment, labels such as Normal, Caution, Abnormal, and Other are used to indicate the classification of the ultrasonic image. Normal indicates a state where no sludge is visible in the training image, which is a cropped region from the water surface to a certain depth, and no floc overflow is occurring. Here, the presence of sludge in the training image means that a predetermined amount or more of flocs is visible in the training image. Caution indicates a state where sludge is visible in the training image, which is a cropped region from the water surface to a certain depth, but no floc overflow is occurring. Caution is considered a state where, if the condition persists, it may become abnormal after a predetermined time, such as 15 or 30 minutes. Abnormal indicates a state where sludge is visible in the training image, which is a cropped region from the water surface to a certain depth, and floc overflow is occurring. Other indicates, for example, a state where there is no output from the ultrasonic sensor 2, such as when the wastewater treatment system 100 is stopped, or when the camera 4 is not operating.

[0037] Estimation model 14 estimates the flock overflow state for the input ultrasound image as either normal, caution, abnormal, or other.

[0038] However, the method of classifying ultrasound images is not limited to this. For example, there is no particular limit to the number of groups into which ultrasound images assigned different labels can be classified. Ultrasound images may be classified into more detailed groups. Furthermore, there are no particular restrictions on what information is used for the labels, as long as it is information that can distinguish and represent the groups of ultrasound images in relation to the overflow state. For example, ultrasound images may be classified using labels corresponding to the amount of chemical injected that corresponds to the overflow state. In this case, by dividing the groups of ultrasound images into smaller groups, the estimation model 14 can estimate the overflow state corresponding to the precisely defined amount of chemical injected.

[0039] A label assigned based on measurement information and information on whether or not floc overflow occurs, where flocs flow out of the sedimentation tank 103 along with the wastewater, is an example of a "final result of identifying the floc overflow state." In other words, in the learning phase, the estimation model 14 learns the relationship between measurement information and the floc overflow state based on the measurement information acquired by the ultrasonic sensor 2 placed in the sedimentation tank 103 and the final result of identifying the floc overflow state. In the estimation phase, the estimation model 14 determines the floc overflow state as a normal state in which flocs do not flow out of the sedimentation tank 103, an abnormal state in which flocs flow out of the sedimentation tank 103, and a caution state in which flocs do not flow out of the sedimentation tank 103 but there is a risk of an abnormal state occurring. For example, a normal state is when no sludge is visible and there is no floc overflow from the sedimentation tank 103. A caution state is when sludge is visible and there is no floc overflow from the sedimentation tank 103. Furthermore, the presence of sludge and the occurrence of floc overflow from the sedimentation tank 103 indicate an abnormal condition.

[0040] (Learning processing device) Next, the learning processing unit 5 will be described. The learning processing unit 5 includes an estimation model 14, a data acquisition unit 51, and a learning processing unit 52.

[0041] The data acquisition unit 51 acquires and stores ultrasonic images transmitted from the camera 4 during the learning phase. The data acquisition unit 51 collects a sufficient number of ultrasonic images to train the estimation model 14. The data acquisition unit 51 also receives input for each ultrasonic image, which is a label identified based on the presence or absence of sludge in the ultrasonic image and the presence or absence of floc overflow at the time each ultrasonic image was taken, and associates and stores the label with each ultrasonic image.

[0042] Figure 4 shows an example of ultrasonic images for each label of the overflow state according to the embodiment. Image 201 is an ultrasonic image to which the label "normal" is assigned because no floc overflow occurred and no flocs are visible at this time. Image 202 is an ultrasonic image to which the label "caution" is assigned because no floc overflow occurred and flocs are visible at this time. Image 203 is an ultrasonic image to which the label "abnormal" is assigned because floc overflow occurred at this time. Image 204 is an ultrasonic image of the ultrasonic sensor 2 in a stopped state, among the ultrasonic images to which other labels are assigned. The user can, for example, check ultrasonic images such as images 201 to 204, assign the labels "normal," "caution," "abnormal," and "other" to each, and input the information of the labels assigned to each ultrasonic image into the data acquisition unit 51.

[0043] Furthermore, the data acquisition unit 51 may receive label input for all images, or it may determine other labels based on the input of a specific label. For example, the data acquisition unit 51 may use an ultrasound image from a specific time prior as a reference for an ultrasound image that has been labeled as abnormal, and then determine the label of ultrasound images within a predetermined time range as "caution" relative to that reference ultrasound image.

[0044] The data acquisition unit 51 associates each of the accumulated ultrasound images with a specified label and outputs it to the learning processing unit 52.

[0045] The learning processing unit 52 receives an input of an ultrasound image with associated labels from the data acquisition unit 51. Next, the learning processing unit 52 extracts a region of the ultrasound image near the water surface to use as a training image. For example, the learning processing unit 52 extracts a region of the ultrasound image from the water surface to a depth of 2m to use as a training image.

[0046] The wastewater treatment system 100 according to this embodiment is configured to discharge supernatant water from the upper end of the sedimentation tank 103. The important information for the estimation model 14 to estimate the state of floc overflow is information about the region where the separated fluid is generated, and in this embodiment, it is information about flocs present from the water surface to a certain depth. Therefore, it is preferable to concentrate the estimation on the region from the water surface to a certain depth for the estimation model 14 to perform estimation in order to improve estimation accuracy. Here, it is preferable that the certain depth is determined according to the possibility of floc overflow occurring. Therefore, the learning processing unit 52 improves the estimation accuracy of the estimation model 14 by cutting out the region from the water surface to a certain depth from the ultrasonic image and having the estimation model 14 learn from it.

[0047] However, the learning processing unit 52 only needs to use the area near the region where the supernatant water in the sedimentation tank 103 is discharged, and the portion below that region, as the learning image. For example, when draining water through an opening in the wall of the sedimentation tank 103, the learning processing unit 52 can crop the portion of the ultrasound image below the opening and use that as the learning image. Alternatively, the learning processing unit 52 can also use the entire ultrasound image to train the estimation model 14.

[0048] The learning processing unit 52 uses the learning image and labels determined based on the presence or absence of sludge and the presence or absence of floc overflow in the ultrasonic image as learning data to have the estimation model 14 estimate the overflow state relative to the ultrasonic image, and trains the estimation model 14 with parameters according to the error between the estimation result and the label. In this way, the learning processing unit 52 generates a trained estimation model 14 that has learned the relationship between the ultrasonic image and the label. In other words, the learning processing unit 52 has the estimation model 14 learn the relationship between the measurement information and the floc overflow state based on the measurement information and the result of identifying the floc overflow state at the time of acquisition of the measurement information, which is determined from the measurement information and the presence or absence of floc overflow at the time of acquisition of the measurement information.

[0049] In this embodiment, the estimation model 14 is trained to estimate that an image is "other" when it does not fall under any of the following categories: normal, caution, or abnormal. This is because the ultrasound image will appear blank when there is no output from the ultrasound sensor 2 or when the camera 4 is not operating. Therefore, the learning processing unit 52 trains the estimation model 14 to estimate that an image is "other" when a blank learning image is input. More specifically, the estimation model 14 is trained to calculate the probability that the learning image falls under normal, caution, abnormal, or other categories. For example, in this embodiment, the estimation model 14 is trained to output a probability of "100%" for normal and a probability of "0%" for caution, abnormal, or other categories when an image corresponding to normal is input.

[0050] Thus, in the learning phase, the estimation model 14 performs learning using the labels and ultrasonic images determined by a comprehensive judgment of whether or not floc overflow occurs and the buoyancy state of the flocs obtained from the ultrasonic images. In other words, the estimation model 14 can perform learning using the information from the water surface to a predetermined depth among the measurement information obtained by ultrasonic waves emitted from the water surface to the bottom of the sedimentation tank 103. Here, the range from the water surface to a depth of 2m is an example of "from the water surface to a predetermined depth".

[0051] In this embodiment, the learning processing unit 52 trained the estimation model 14 to estimate the flock overflow state at the time the ultrasound image was taken for a single ultrasound image. However, the learning content is not limited to this, as long as the flock overflow state can be estimated. For example, the learning processing unit 52 may train the estimation model 14 to estimate the flock overflow state after a predetermined time for an ultrasound image. More specifically, the learning processing unit 52 may train the estimation model 14 using an ultrasound image at a specific timing, an ultrasound image taken a predetermined time after that specific timing, and labels assigned to each ultrasound image. In such a case, the estimation model 14 can estimate whether flock overflow will occur after a predetermined time from an ultrasound image showing the flock overflow state at a certain timing. The learning processing unit 52 may also perform training using, for example, multiple ultrasound images created according to changes over time and their labels. In this case, the estimation model 14 learns to estimate the flock overflow state after a predetermined time has elapsed from the timing when the ultrasound image was taken, based on the time-series changes in the flock overflow state shown by the given multiple ultrasound images. These processes are variations, and in this embodiment, the estimation model 14 is made to estimate the floc overflow state at a certain timing, and drug injection is controlled based on the estimation result or the time-series change of the estimation result.

[0052] Alternatively, a recurrent neural network (RNN) or a long-short-term memory (LSTM) network can also be used as the estimation model 14. In this case, the learning processing unit 52 may train the estimation model 14 to estimate the overflow state based on changes over time, using multiple ultrasound images in a time series as training data. Furthermore, after training the estimation model 14 using ultrasound images, the learning processing unit 52 may further train it using the required amount of chemicals to be injected for each overflow state.

[0053] Furthermore, the learning processing unit 52 may also include water temperature, temperature, humidity, etc., in the learning data. In addition, the learning processing unit 52 may train the estimation model 14 with various parameters such as the size, shape, and depth of the sedimentation tank 103, the discharge status of wastewater, pH, amount of sludge, and type of chemicals.

[0054] Furthermore, although the control device 1 and the learning processing device 5 are separate devices in this embodiment, they may be combined into a single device. For example, the control device 1 may include a data acquisition unit 51 and a learning processing unit 52.

[0055] (Control device) Next, the control device 1 will be described. As shown in Figure 2, the control device 1 includes an image acquisition unit 11, a state determination unit 12, a control unit 13, and an estimation model 14. The control device 1 downloads and stores the trained estimation model 14 from the training processing device 5.

[0056] The image acquisition unit 11 is connected to the camera 4. The image acquisition unit 11 acquires the ultrasound image captured by the camera 4, which will be described below. The image acquisition unit 11 then outputs the acquired ultrasound image to the state determination unit 12.

[0057] The state determination unit 12 receives the ultrasonic image to be estimated as input from the image acquisition unit 11. The state determination unit 12 then extracts the region near the water surface from the acquired ultrasonic image and inputs it into the trained estimation model 14, obtaining the flock overflow state at the time the ultrasonic image was taken as the estimated result. In this embodiment, the state determination unit 12 obtains one of the following flock overflow states as the inference result: normal, caution, abnormal, or other.

[0058] More specifically, the estimation model 14 calculates probabilities for each of the four states of an ultrasound image: normal, caution, abnormal, or other. In this case, the sum of the probabilities for each of the four states equals 100%. The estimation model 14 then estimates that a state is a flock overflow state if there is a state whose probability exceeds a predetermined estimation threshold. For example, the estimation model 14 can use 60% as the judgment threshold. That is, the estimation model 14 determines a state to be a flock overflow state if its probability is between 70% and 80%. For example, if the probabilities for each of the four states are 80%, 15%, 5%, and 0%, the estimation model 14 determines that the flock overflow state is normal. Also, for example, if the probabilities for each of the four states are 25%, 25%, 25%, and 25%, and none of them exceed the judgment threshold, the estimation model 14 does not output any estimation results and proceeds with the next estimation.

[0059] Figure 5 shows an example of the estimation results. Table 310 shows the estimation results for the sedimentation tank 103 in three states, #1 to #3.

[0060] For example, when the overflow state is estimated using an ultrasonic image of sedimentation tank 103 in state #1, the estimation results are 98% "normal", 1% "caution", 1% "abnormal", and 0% "other". Therefore, in this case, estimation model 14 outputs "normal". Also, when the overflow state is estimated using an ultrasonic image of sedimentation tank 103 in state #2, the estimation results are 1% "normal", 82% "caution", 17% "abnormal", and 0% "other". Therefore, in this case, estimation model 14 outputs "caution". Furthermore, when the overflow state is estimated using an ultrasonic image of sedimentation tank 103 in state #3, the estimation results are 0% "normal", 21% "caution", 79% "abnormal", and 0% "other". Therefore, in this case, estimation model 14 outputs "abnormal". All of these judgment results are consistent with the results that a human would get when judging based on the respective ultrasonic images, indicating that accurate estimation is being performed.

[0061] Thus, in the estimation phase, the estimation model 14 estimates the flock overflow state based on conditions such as which of the four states—normal, caution, abnormal, or other—is most likely, or in other words, which has the highest probability. This allows the estimation model 14 to perform real-time labeling on the estimated image.

[0062] Next, the state determination unit 12 stores 0 as the value indicating the flock overflow state if the estimation result is "normal". The state determination unit 12 also stores 1 as the value indicating the flock overflow state if the estimation result is "caution". The state determination unit 12 also stores 2 as the value indicating the flock overflow state if the estimation result is "abnormal". The state determination unit 12 also stores 0 as the value indicating the flock overflow state if the estimation result is "other". Note that here, we focused on the flock overflow state, so we set it to 0 for both normal and other cases, but this value may be different.

[0063] Here, the ultrasonic image is generated sequentially from ultrasonic data acquired continuously at predetermined time intervals by the ultrasonic sensor 2. The state determination unit 12 stores a value indicating the current flock overflow state and a value indicating the previous flock overflow state, which are consecutive in time.

[0064] If the state determination unit 12 obtains a new value indicating a flock overflow state, it updates the value indicating the current flock overflow state with the newly obtained value indicating a flock overflow state.

[0065] Next, the state determination unit 12 multiplies the value indicating the current flock overflow state by the value indicating the previous flock overflow state. The state determination unit 12 then outputs the multiplication result to the control unit 13 as a value indicating the confirmation result of the flock overflow state. Hereafter, the value indicating the confirmation result of the flock overflow state will be referred to as the "flock overflow confirmation value". Subsequently, the state determination unit 12 updates the value of the previous flock overflow state with the value of the current flock overflow state.

[0066] For example, if the flock overflow state is normal for two consecutive times, the state determination unit 12 calculates the flock overflow confirmation value as 0 × 0 = 0. Also, if either of the two consecutive flock overflow states is normal, the state determination unit 12 calculates the flock overflow confirmation value as 0 × (1 or 2) = 0 or (1 or 2) × 0 = 0. Also, if the flock overflow state is caution for two consecutive times, the state determination unit 12 calculates the flock overflow confirmation value as 1 × 1 = 1. Also, if the flock overflow state is abnormal for two consecutive times, the state determination unit 12 calculates the flock overflow confirmation value as 2 × 2 = 4.

[0067] During operation when the control device 1 is operating in the estimation phase, the ultrasonic data acquired by the ultrasonic sensor 2 is an example of "first measurement information," and the timing at which the ultrasonic sensor 2 acquires ultrasonic data for determining the overflow state is an example of "predetermined timing." That is, the state determination unit 12 uses the estimation model 14 to determine the floc overflow state based on the first measurement information acquired by the ultrasonic sensor 2 at the predetermined timing. The state determination unit 12 also inputs the information from the water surface to a predetermined depth from the first measurement information into the estimation model 14 to perform estimation. Furthermore, the state determination unit 12 uses the estimation model 14 to determine the floc overflow state as a normal state in which the floc is not flowing out of the sedimentation tank 103, an abnormal state in which the floc is flowing out of the sedimentation tank 103, and a caution state in which the floc is not flowing out of the sedimentation tank 103 but there is a risk of an abnormal state occurring.

[0068] The control unit 13 stores information on control patterns corresponding to individual floc overflow confirmation values. For example, in this embodiment, the control unit 13 stores a first control pattern that controls the amount injected from the chemical injection pump to a standard amount when the floc overflow confirmation value is 0. The control unit 13 also stores a second control pattern that controls the amount of chemical injected from the chemical injection pump into the reaction condensation tank 102 to be 10% more than the standard amount when the floc overflow confirmation value is 1 or 2. The control unit 13 also stores a third control pattern that controls the amount of chemical injected from the chemical injection pump into the reaction condensation tank 102 to be 20% more than the standard amount when the multiplication result is 4. In this way, the control unit 13 interprets the estimated floc overflow state obtained from the ultrasonic image in a time series and selects a control pattern.

[0069] The control unit 13 receives the confirmation value of the floc overflow from the state determination unit 12. Then, the control unit 13 selects one of the first to third control patterns for the chemical injection pump according to the confirmation value of the floc overflow. Specifically, if the confirmation value of the floc overflow is 0, the control unit 13 selects the first control pattern. If the confirmation value of the floc overflow is 1, the control unit 13 selects the second control pattern. If the confirmation value of the floc overflow is 4, the control unit 13 selects the third control pattern. After that, the control unit 13 notifies the chemical process system 20, which executes various processes in the chemical process, of the information of the selected control pattern.

[0070] The chemical process system 20 controls the amount of chemical injected from the chemical injection pump into the reaction coagulation tank 102 according to the notified control pattern. Here, there is an elapsed time, for example 15 minutes, from the injection of chemical into the reaction coagulation tank 102 to the sedimentation of flocs in the sedimentation tank 103. Therefore, there is a time lag between changing the amount of chemical injected into the reaction coagulation tank 102 and the occurrence of floc sedimentation in the sedimentation tank 103 corresponding to that amount. However, in the control device 1 according to this embodiment, a caution state is set as a floc overflow state, so by increasing the amount of chemical injected when the caution state is set, it is possible to suppress floc overflow before actual floc overflow occurs.

[0071] In this embodiment, the accuracy of selecting a control pattern is improved compared to making a decision based on a single image by using two pieces of information: a value indicating the previous flock overflow state and a value indicating the current flock overflow state. However, the values ​​indicating the flock overflow state used for determination do not have to be two sets of values. For example, the state determination unit 12 and the control unit 13 may use three or more pieces of information, including the value indicating the flock overflow state two sets prior, or they may select a control pattern from a single value indicating the flock overflow state. Furthermore, when using values ​​indicating multiple flock overflow states, the state determination unit 12 and the control unit 13 may use the average of the multiple values ​​or the time-window average. In addition, the state determination unit 12 and the control unit 13 may select a control pattern using a majority vote or a label ratio on the estimation result.

[0072] Furthermore, in this embodiment, the control pattern was determined using a value indicating the flock overflow state, but the determination method is not limited to this. For example, the estimation model 14 may output normal, caution, abnormal, and other information as the overflow state, and the control unit 13 may determine the control pattern using that overflow state.

[0073] In this embodiment, the control device 1 estimates the flock overflow state using ultrasonic images such as images obtained by a fish finder, but the information used for estimation is not limited to this. For example, ultrasonic images are data created by stacking multiple ultrasonic scan lines, but the control device 1 may perform estimation using ultrasonic data, which is the individual scan line, rather than image data. In this case, the control device 1 will change the estimation result according to the changes in each scan line. Here, the information obtained from the change in ultrasonic data over time can be considered as planar information, and the information obtained from the change in ultrasonic images over time can be considered as three-dimensional information, and it is preferable to use information suitable for the operating conditions by considering the difference between the two. Alternatively, for example, the control device 1 may perform estimation by using the image data displayed by the display device 3 directly for estimation without taking images with the camera 4. Also, the ultrasonic image may be in color or in black and white.

[0074] Figure 6 shows an example of a control pattern classification table. The control unit 13 can determine a control pattern from the previous overflow state and the current overflow state by using the control pattern classification table 300 shown in Figure 6. The control pattern classification table 300 shows the amount to change from the standard amount of chemical to be injected for the combination of the previous floc overflow state and the current floc overflow state. "No change" in the control pattern classification table 300 indicates that the standard amount of chemical will be injected. "10% increase" in the control pattern classification table 300 indicates that 10% more chemical than the standard amount will be injected. "20% increase" in the control pattern classification table 300 indicates that 20% more chemical than the standard amount will be injected.

[0075] For example, if the previous floc overflow condition was normal and the current floc overflow condition is caution, the control unit 13 refers to the control pattern classification table 300 and selects a first control pattern in which the amount of chemical injected is the standard amount. If the previous floc overflow condition was caution and the current floc overflow condition is abnormal, the control unit 13 refers to the control pattern classification table 300 and selects a second control pattern in which the amount of chemical injected is increased by 10% from the standard amount. If the previous floc overflow condition was abnormal and the current floc overflow condition is abnormal, the control unit 13 refers to the control pattern classification table 300 and selects a third control pattern in which the amount of chemical injected is increased by 20% from the standard amount.

[0076] Thus, the control unit 13 increases the amount of drug injected when the same warning result is predicted consecutively, such as when the previous inference result was warning and the current inference result is also warning. Furthermore, the control unit 13 increases the amount of drug injected more than when warnings are predicted consecutively when the same abnormality result is predicted consecutively, such as when the previous inference result was abnormal and the current inference result is also abnormal. In addition, the control unit 13 increases the amount of drug injected in the same way as when warnings are predicted consecutively when warnings and abnormalities are predicted consecutively between the previous and current inference results. Here, the phrase "warning or abnormality is predicted consecutively" is used to include cases where warnings and abnormalities are predicted consecutively.

[0077] On the other hand, if the sequence of warnings or abnormalities is interrupted, that is, if the inference result is normal or otherwise obtained midway through, the control unit 13 uses the drug injection amount as a reference amount and waits for the measurement result from the ultrasonic sensor 2 to be obtained again. Subsequently, if the sequence of warnings or abnormalities occurs again, the control unit 13 increases the drug injection amount again.

[0078] In other words, as shown in Figure 6, the control unit 13 interprets the inference results obtained from the previous ultrasonic image and the inference results obtained from the current ultrasonic image in a time series to determine the control pattern for the amount of chemical injected. By interpreting the inference results in a time series in this way, the control unit 13 can more accurately control the amount of chemical injected according to the state of the sedimentation tank 103 compared to simply controlling the amount of chemical injected based on a single inference result, thereby reducing the occurrence of floc overflow.

[0079] Furthermore, although the floc overflow state was used as the label for the ultrasonic image in this embodiment, it is also conceivable to use the amount of chemical injected as the label for the ultrasonic image. In that case, the control unit 13 may notify the chemical process system 20 of a control pattern that corresponds to the estimated amount of chemical injected. Also, although the injection amount was increased by 10% in this embodiment, the control pattern is not limited to this. For example, the control unit 13 may change the amount of chemical injected more finely in response to changes in the floc overflow state. In addition, the control unit 13 may control the timing and amount of chemical injection in accordance with the time-series changes of each probability value output by the estimation model 14, for example.

[0080] The control unit 13 outputs information regarding the control of wastewater treatment based on the floc overflow state confirmed by the state determination unit 12. Here, the confirmed floc overflow value, which is the result of multiplying the value indicating the previous floc overflow state and the value indicating the current floc overflow state obtained by the state determination unit 12, is an example of the "floc overflow state confirmed by the state determination unit 12". The control unit 13 outputs information to control the injection amount of chemicals that are injected into the wastewater to generate inorganic flocs. In addition, if an abnormal or cautionary state is detected, the control unit 13 outputs information to increase the injection amount of chemicals compared to the normal state.

[0081] Furthermore, the control unit 13 repeatedly acquires the floc overflow state determined by the state determination unit 12, and if it acquires a warning state or an abnormal state as the floc overflow state, it outputs information to increase the amount of drug injected compared to the normal state if the next floc overflow state is also a warning state or an abnormal state. In addition, if the control unit 13 acquires an abnormal state as the floc overflow state, it increases the amount of drug injected compared to when the warning state is continuous, if the next floc overflow state is also an abnormal state.

[0082] Furthermore, the overflow state obtained based on the current overflow state and the previous overflow state is an example of the overflow state at one timing in "overflow state at multiple timings". In other words, the control unit 13 outputs information regarding wastewater treatment control based on the flock overflow state at each of the multiple consecutive timings. Also, if the flock overflow state at a specific timing in the multiple consecutive timings is the warning state or abnormal state, and the flock overflow state at the next timing is the warning state or abnormal state, the control unit 13 outputs information regarding the wastewater treatment control to be executed in the warning state. Also, if the flock overflow state at a specific timing in the multiple consecutive timings is an abnormal state, and the flock overflow state at the next timing is an abnormal state, the control unit 13 outputs information regarding the wastewater treatment control to be executed in the abnormal state. Also, if the flock overflow state at a specific timing in the multiple consecutive timings is the warning state or abnormal state, and the flock overflow state at the next timing is not the warning state or abnormal state, the control unit 13 outputs information regarding the wastewater treatment control to be executed in the normal state.

[0083] (Operation of the entire control system) Figure 7 is a diagram of the entire control system. Next, we will refer to Figure 7 to summarize and explain the overall overview of learning, inference, and control.

[0084] In the learning phase, the learning processing device 5 collects ultrasonic images generated by the camera 4 capturing ultrasonic images displayed on the display device 3. Furthermore, the learning processing device 5 receives input of labels indicating the overflow state for each ultrasonic image, which are determined based on the information of the presence or absence of sludge and the occurrence of floc overflow in each ultrasonic image. The learning processing device 5 then associates and stores the ultrasonic images with the labels determined based on the information of the presence or absence of sludge and the occurrence of floc overflow in each ultrasonic image (step S11). In Figure 7, ultrasonic images with the label "normal" are shown as normal images, ultrasonic images with the label "caution" are shown as caution images, and ultrasonic images with the label "abnormal" are shown as abnormal images. Ultrasonic images with other labels are omitted from the illustration.

[0085] Next, the learning processing device 5 performs a learning process to train the estimation model 14 using the ultrasound images and labels (step S12).

[0086] The learning processing device 5 has the trained estimation model 14 perform estimations and determines whether the number of misclassifications included in the estimation results falls within an acceptable range. If it does not fall within an acceptable range, steps S11 and S12 are repeated (step S13). If the number of misclassifications falls within an acceptable range, the learning processing device 5 terminates the training of the estimation model 14.

[0087] The control device 1 downloads the trained estimation model 14 from the training processing device 5 and stores it as the estimation model 14 (step S14).

[0088] During operation, the control device 1 acquires an ultrasound image from the camera 4 (step S15).

[0089] Next, the control device 1 performs a control process to select a control pattern based on the ultrasonic image (step S16). The control process includes the following steps: The control device 1 inputs the acquired ultrasonic image into the estimation model 14 to perform estimation and performs an estimation process to obtain the floc overflow state, which is the estimation result (step S17). Next, the control device 1 uses the previous floc overflow state and the current floc overflow state obtained from the estimation result to select a control pattern for the chemical injection pump (step S18).

[0090] Subsequently, the control device 1 notifies the chemical process system 20 of the selected control pattern and controls the amount of chemical injected into the reaction condensation tank 102 by the chemical injection pump (step S19).

[0091] (Learning process) Figure 8 is a flowchart of the learning process for the estimation model by the learning processing unit. Next, the learning process flow of estimation model 14 will be explained with reference to Figure 8.

[0092] The ultrasonic sensor 2 emits ultrasonic pulses toward the bottom surface 134 of the sedimentation tank 103 (step S101).

[0093] The display device 3 generates and displays an ultrasonic image from ultrasonic data obtained from reflected waves by the ultrasonic sensor 2 (step S102).

[0094] Camera 4 captures the ultrasonic image displayed on the display device 3 and transmits it to the control device 1 (step S103).

[0095] The data acquisition unit 51 of the learning processing device 5 receives and collects ultrasonic images sent from the camera 4 (step S104). The data acquisition unit 51 collects a sufficient number of ultrasonic images to train the estimation model 14. The data acquisition unit 51 also acquires information on whether or not floc overflow occurred at the time each ultrasonic image was captured and associates it with the ultrasonic image.

[0096] Next, the data acquisition unit 51 receives input of the floc overflow state, which is to be the label assigned to each ultrasonic image, and classifies the ultrasonic images into groups according to the label (step S105). Here, this label is determined from the information of whether or not sludge is visible in the ultrasonic image and whether or not floc overflow has occurred. More specifically, the data acquisition unit 51 labels ultrasonic images in which sludge is not visible and no floc overflow has occurred as normal. The data acquisition unit 51 also labels ultrasonic images in which sludge is visible and no floc overflow has occurred as caution. The data acquisition unit 51 also labels ultrasonic images in which sludge is visible and floc overflow has occurred as abnormal. The data acquisition unit 51 also labels ultrasonic images in which nothing is visible as other. In this way, the data acquisition unit 51 performs labeling based on a comprehensive judgment of whether or not floc overflow has occurred and whether or not sludge is visible in the ultrasonic image.

[0097] The learning processing unit 52 generates a learning image by cropping each ultrasonic image collected by the data acquisition unit 51 from the water surface to a depth of 2m. Next, the learning processing unit 52 uses the learning image and labels as learning data and adjusts the parameters according to the results estimated by the estimation model 14 to perform learning (step S106). That is, the learning processing unit 52 performs learning on the estimation model 14 based on the state of the flocs shown in the learning image, based on the labeling assigned based on a comprehensive judgment of the presence or absence of floc overflow and the presence or absence of sludge shown in the ultrasonic image.

[0098] Subsequently, the learning processing unit 52 extracts the portion of the ultrasonic image for evaluation from the water surface to a depth of 2m and inputs it into the estimation model 14 to perform estimation of the overflow state of the ultrasonic image for evaluation (step S107).

[0099] Then, the learning processing unit 52 determines whether the number of inference errors falls within an acceptable range (step S108).

[0100] If the number of misclassifications does not fall within the acceptable range (step S108: No), the learning processing unit 52 notifies the user to perform retraining. The user reclassifies each ultrasound image and inputs the label for each ultrasound image to the data acquisition unit 51 according to the classification result. The learning process then returns to step S105.

[0101] In contrast, if the misclassification falls within an acceptable range (step S108: Yes), the learning processing unit 52 terminates the learning process of the estimation model 14.

[0102] (Wastewater treatment) Here, we will summarize again the overall flow of wastewater treatment by the wastewater treatment system 100. Figure 9 is a flowchart of the wastewater treatment by the wastewater treatment system.

[0103] The wastewater treatment system 100 receives the factory wastewater, stores it in the raw water tank 101, and performs a damping process to equalize it (step S111).

[0104] Next, in reaction step S2, the wastewater treatment system 100 transfers the raw water from the raw water tank 101 to the reaction coagulation tank 102, and carries out a reaction step in which inorganic substances contained in the raw water react with chemicals to suspend pollutants and generate flocs (step S112).

[0105] Next, the wastewater treatment system 100 stores chemicals such as coagulants and polymers, and uses the chemical process system 20 to perform a chemical process in which the stored chemicals are injected into the reaction coagulation tank 102 by pump (step S113). The amount of chemicals injected in this chemical process is determined by the control device 1 using the estimation model 14 to estimate the floc overflow state and according to the control pattern selected according to the confirmation value of floc overflow calculated based on the estimation result. In other words, since the amount of chemicals injected is determined according to the floc overflow state, the amount of chemicals necessary to ensure that the flocs settle in the sedimentation tank 103 in the next sedimentation process can be used without excess or deficiency.

[0106] Next, the wastewater treatment system 100 transfers the raw water containing suspended pollutants from the reaction coagulation tank 102 to the sedimentation tank 103 and performs a sedimentation process to separate the solids and liquids by allowing the flocs to settle in the sedimentation tank 103 (step S114). Since the chemical process injects the exact amount of chemicals necessary to settle the flocs, the wastewater treatment system 100 can improve the probability of settling the flocs and reduce the amount of flocs that overflow. In the sedimentation process, the wastewater treatment system 100 withdraws the settled and separated flocs from the sedimentation tank 103, dries them, and then discards them.

[0107] Next, the wastewater treatment system 100 performs a filtration activated carbon process to remove residual substances by allowing the treated water from which the flocs have been separated to overflow from the sedimentation tank 103 and applying a filtration activated carbon treatment to the overflowed treated water (step S115). Thus, in the inorganic treatment process, the treated water from which the flocs have been separated is subjected to superactivated carbon treatment, so the separation of flocs is not necessarily required. However, in order to keep the filter in proper condition, the filter needs to be cleaned, and considering the workload of cleaning, it is not practical to filter the entire amount of flocs by the filter, so it is desirable to separate the flocs in steps S113 and S114. Against this backdrop, the wastewater treatment system 100 according to this embodiment injects the exact amount of chemicals necessary to settle the flocs in the chemical process, thereby reducing the amount of flocs that cause overflow. This reduces the load on the filter and avoids malfunctions, as well as reducing the workload of cleaning the filter.

[0108] Next, the wastewater treatment system 100 performs a discharge process (step S116) in which treated water is discharged after residual substances have been removed in the filtration activated carbon process and the water has met legal requirements.

[0109] (Control processing) Figure 10 is a flowchart of the control process by the control device. Next, the flow of the control process by the control device 1 will be explained with reference to Figure 10.

[0110] The ultrasonic sensor 2 emits ultrasonic pulses toward the bottom surface 134 of the sedimentation tank 103 (step S201).

[0111] The display device 3 generates and displays an ultrasonic image from ultrasonic data obtained from reflected waves by the ultrasonic sensor 2 (step S202).

[0112] Camera 4 captures the ultrasonic image displayed on the display device 3 and transmits it to the control device 1 (step S203).

[0113] The image acquisition unit 11 of the control device 1 receives ultrasonic images sent from the camera 4. The state determination unit 12 then acquires the ultrasonic images from the image acquisition unit 11. Next, the state determination unit 12 extracts the portion of the acquired ultrasonic image from the water surface to a depth of 2m, inputs it into the estimation model 14, and has the estimation model 14 estimate the overflow state, and obtains the estimation result (step S204).

[0114] Next, the state determination unit 12 stores a value indicating the flock overflow state corresponding to the current estimation result as the value indicating the current flock overflow state (step S205).

[0115] Next, the state determination unit 12 calculates a confirmation value for flock overflow by multiplying the value indicating the previous flock overflow state by the value indicating the current flock overflow state (step S206).

[0116] The control unit 13 obtains a confirmation value for floc overflow from the state determination unit 12. Next, the control unit 13 selects a control pattern for the chemical injection pump according to the confirmation value for floc overflow (step S207). After that, the control unit 13 notifies the chemical process system 20 of the selected control pattern for the chemical injection pump.

[0117] The state determination unit 12 updates the value of the previous flock overflow state with the value of the current flock overflow state (step S208).

[0118] The control unit 13 determines whether or not to stop controlling the chemical injection pump based on whether or not there is an input from the user to stop operation (step S209). If control is not stopped (step S209: No), the control process returns to step S201. On the other hand, if control is stopped (step S209: Yes), the control unit 13 stops the operation that performs the control processing.

[0119] (Selection process for control pattern) Figure 11 is a flowchart illustrating the control pattern selection process. Next, the flow of the control pattern selection process will be explained with reference to Figure 11. In reality, the control unit 13 multiplies the values ​​indicating the previous and current flock overflow states to obtain a flock overflow confirmation value and determines the control pattern from the obtained flock overflow confirmation value. However, for the sake of clarity, it will be explained here that the control pattern is selected according to the information on the flock overflow state. The control pattern selection process described here is equivalent to the process when selecting a control pattern using a value indicating the flock overflow state.

[0120] Measurement is performed using ultrasonic sensor 2 (step S301).

[0121] The control unit 13 acquires information on the previous flock overflow state and the current flock overflow state (step S302).

[0122] The control unit 13 determines whether the previous floc overflow condition was normal and whether the current floc overflow condition is normal (step S303). If both the previous and current floc overflow conditions are normal (step S303: Yes), the control unit 13 determines the amount of chemical to be injected as the standard amount (step S304).

[0123] Then, the control unit 13 selects a first control pattern in which the official's injection amount is a standard amount (step S305).

[0124] In response to this, if either the previous or current floc overflow condition is not normal (step S303: No), the control unit 13 determines whether both the previous and current floc overflow conditions are warning (step S306). If both the previous and current floc overflow conditions are warning (step S306: Yes), the control unit 13 decides to increase the amount of chemical injected by 10% from the standard amount (step S307).

[0125] Then, the control unit 13 selects a second control pattern in which the amount injected by the official is 10% more than the standard amount (step S308).

[0126] In response to this, if neither the previous nor the current floc overflow condition is a warning (step S306: No), the control unit 13 determines whether the previous floc overflow condition was a warning and the current floc overflow condition is abnormal (step S309). If the previous floc overflow condition was a warning and the current floc overflow condition is abnormal (step S309: Yes), the control unit 13 decides to increase the amount of chemical injected by 10% from the standard amount (step S307).

[0127] Then, the control unit 13 selects a second control pattern in which the amount injected by the official is 10% more than the standard amount (step S308).

[0128] In contrast, if the conditions that the previous floc overflow state was a warning and the current floc overflow state is abnormal are not met (Step S309: No), the control unit 13 determines whether the previous floc overflow state was abnormal and whether the current floc overflow state is abnormal (Step S310). If both the previous and current floc overflow states are abnormal (Step S310: Yes), the control unit 13 decides to increase the amount of chemical injected by 20% from the standard amount (Step S311).

[0129] Then, the control unit 13 selects a second control pattern in which the amount injected by the official is 20% more than the standard amount (step S312).

[0130] In response to this, if neither the previous nor the current floc overflow condition is abnormal (step S310: No), the control unit 13 determines whether the previous floc overflow condition was abnormal and the current floc overflow condition is a warning (step S313). If the previous floc overflow condition was abnormal and the current floc overflow condition is a warning (step S313: Yes), the control unit 13 decides to increase the amount of chemical injected by 10% from the standard amount (step S307).

[0131] Then, the control unit 13 selects a second control pattern in which the amount injected by the official is 10% more than the standard amount (step S308).

[0132] In contrast, if the previous floc overflow condition was abnormal and the current floc overflow condition does not meet the condition of requiring caution (step S313: No), the control unit 13 determines the amount of chemical to be injected as the standard amount (step S304).

[0133] Then, the control unit 13 selects a first control pattern in which the official's injection amount is a standard amount (step S305).

[0134] After selecting a control pattern, the control unit 13 determines whether or not to stop the control (step S314). If the control is not to be stopped (step S314: No), the control pattern selection process returns to step S301. Conversely, if the control is to be stopped (step S314: Yes), the control unit 13 stops the control pattern selection process.

[0135] (effect) As described above, the control device 1 according to this embodiment estimates the state of inorganic floc overflow in the sedimentation tank 103 using an estimation model 14 based on ultrasonic data obtained from ultrasonic pulses emitted from the water surface toward the bottom surface 134 of the sedimentation tank 103. Then, the control device 1 controls the amount of chemical injected by the chemical injection pump according to the estimated overflow state.

[0136] Furthermore, by using ultrasonic data, the underwater conditions in the sedimentation tank 103, which are difficult to confirm visually or through surface imaging, can be accurately grasped. This allows the control device 1 to improve the accuracy of estimating the overflow state of inorganic flocs in the sedimentation tank 103. As a result, the control device 1 can use the exact amount of chemicals necessary to ensure that inorganic flocs settle in the sedimentation tank 103, without excess or deficiency, thereby suppressing the injection of unnecessary chemicals and the occurrence of inorganic floc overflow. Therefore, the control device 1 enables efficient and appropriate wastewater treatment.

[0137] Figure 12 is a diagram comparing the control device according to the embodiment with the case where manual confirmation of the overflow state is used. Table 400 in Figure 12 shows the monitoring timing when manual confirmation is performed and when the control device 1 is used, within a 24-hour period. The numbers lined up at the top of the page in Table 400 represent the time within a 24-hour period, and the patterned areas in Table 400 indicate the period during which the sedimentation tank 103 is being monitored.

[0138] In the case of manual monitoring, it is difficult to constantly monitor the sedimentation tank 103, and for example, checks are performed every 8 hours. In that case, manual monitoring of the sedimentation tank 103 is limited to short periods within a 24-hour period, such as a fixed period from 8:00, a fixed period from 16:00, and a fixed period from 24:00.

[0139] In contrast, when using the control device 1, the ultrasonic sensor 2 can continuously generate time-series ultrasonic data for 24 hours, and the control device 1 constantly monitors the sedimentation tank 103. Therefore, it is possible to reduce the chances of overflow being overlooked compared to when a person makes a visual judgment. Furthermore, even if a problem such as floc overflow occurs, the problem can be addressed in a shorter time compared to when a person makes a visual judgment, enabling stable operation of the wastewater treatment system 100.

[0140] Figure 13 is a diagram showing a comparison between an ultrasonic image and a water surface image used in the control processing according to this embodiment. Next, referring to Figure 13, a comparison between the determination of the overflow state using an ultrasonic image and the determination of the overflow state using a water surface image by the control device 1 according to this embodiment will be explained. Here, the overflow determination using a water surface image also includes the case where a person visually inspects the sedimentation tank 103 from above.

[0141] Since water surface images 401-403 are simply images of the water surface, it is difficult to obtain significant differences in the state of flocculation between the normal water surface image 401, the cautionary water surface image 402, and the abnormal water surface image 403.

[0142] In contrast, for the same normal sedimentation tank 103 as in water surface image 401, ultrasonic image 411 confirms that flocs have accumulated to an estimated extent, and that no flocs are floating near the water surface. Furthermore, for the same cautionary condition as in water surface image 402, ultrasonic image 412 confirms that a certain amount of flocs are floating near the water surface, increasing the likelihood of overflow. Also, for the same abnormal condition as in water surface image 403, ultrasonic image 413 confirms that a very large number of flocs have reached the water surface and are floating, indicating that overflow is occurring.

[0143] Thus, ultrasonic imaging allows for easy and accurate confirmation of the water conditions in the sedimentation tank 103, which are difficult to observe using surface images or visual inspection. By using such ultrasonic imaging to estimate the overflow state, the estimation accuracy can be further improved. Therefore, efficient and appropriate wastewater treatment becomes possible.

[0144] (Hardware configuration) Figure 14 is a hardware configuration diagram of the computer. Next, with reference to Figure 14, an example of a hardware configuration for realizing each function of the control device 1 and the learning processing device 5 will be described. Both the control device 1 and the learning processing device 5 can be realized by the computer 90 shown in Figure 14.

[0145] As shown in Figure 14, the computer 90 includes, for example, a CPU (Central Processing Unit) 91, memory 92, a hard disk 93, and a network interface 94. The CPU 91 is connected to the memory 92, the hard disk 93, and the network interface 94 via a bus.

[0146] The network interface 94 is an interface for communication between the control device 1 or the learning processing device 5 and an external device. For example, the network interface 94 relays communication between the camera 4 and the CPU 91.

[0147] The hard disk 93 is an auxiliary storage device. In the case of the control device 1, the hard disk 93 may store the estimation model 14 illustrated in Figure 2. The hard disk 93 also stores various programs, including programs for realizing the functions of the image acquisition unit 11, state determination unit 12, and control unit 13, as illustrated in Figure 2. On the other hand, in the case of the learning processing device 5, the hard disk 93 may store the estimation model 14 illustrated in Figure 2. The hard disk 93 also stores various programs, including programs for realizing the functions of the data acquisition unit 51 and learning processing unit 52, as illustrated in Figure 2.

[0148] Memory 92 is the main memory. Memory 92 can be, for example, DRAM (Dynamic Random Access Memory).

[0149] The CPU 91 reads various programs from the hard disk 93, loads them into memory 92, and executes them. In the case of the control device 1, the CPU 91 implements the functions of the image acquisition unit 11, the state determination unit 12, and the control unit 13, as illustrated in Figure 2. On the other hand, in the case of the learning processing device 5, the CPU 91 implements the functions of the data collection unit 51 and the learning processing unit 52, as illustrated in Figure 2. [Explanation of symbols]

[0150] 1. Control device 2. Ultrasonic sensor 3 Display device 4 cameras 5. Learning Processing Unit 10 Control Systems 11 Image acquisition unit 12 State determination unit 13 Control Unit 14 Estimated Models 20. Chemical Processing Systems 51 Data Collection Unit 52 Learning Processing Unit 100 Wastewater Treatment Systems 101 Raw water tank 102 Reaction and Coagulation Tank 103 Sedimentation tank

Claims

1. A state determination unit that uses an estimation model that has learned the relationship between the measurement information and the floc overflow state, based on measurement information obtained by an ultrasonic sensor placed in a sedimentation tank for settling and separating inorganic flocs in wastewater in wastewater treatment, and the result of identifying the floc overflow state, to confirm the floc overflow state at a predetermined timing based on first measurement information obtained by the ultrasonic sensor at a predetermined timing, Based on the floc overflow state determined by the state determination unit, a control unit outputs information related to the control of the wastewater treatment. A control device characterized by being equipped with

2. The estimation model performs the learning process using the information from the water surface to a predetermined depth, which is obtained from the measurement information obtained by ultrasonic waves emitted from the water surface towards the bottom of the sedimentation tank. The state determination unit inputs the information from the water surface to a predetermined depth from the first measurement information into the estimation model to perform estimation. The control device according to feature 1.

3. The control device according to claim 1, characterized in that the estimation model performs the learning using the measurement information from the ultrasonic sensor located in the area surrounding the location where wastewater flows out of the sedimentation tank.

4. The control device according to claim 1, characterized in that the control unit outputs information for controlling the amount of chemical injected into the wastewater to generate inorganic flocs.

5. The control device according to claim 1, characterized in that the estimation model determines, as the floc overflow state, a normal state in which the flocs are not flowing out of the sedimentation tank, an abnormal state in which the flocs are flowing out of the sedimentation tank, and a caution state in which the flocs are not flowing out of the sedimentation tank but there is a risk that the abnormal state may occur.

6. The control device according to claim 5, characterized in that the control unit outputs information to increase the amount of chemical injected into the wastewater to generate inorganic flocs when the abnormal condition or the warning condition is met, compared to the normal condition.

7. The control device according to claim 6, characterized in that the control unit repeatedly acquires the floc overflow state determined by the state determination unit, and when the floc overflow state is the warning state or the abnormal state, outputs information to increase the amount of the drug injected compared to the normal state if the next floc overflow state is the warning state or the abnormal state.

8. The control device according to claim 7, characterized in that, when the control unit acquires the abnormal state as the floc overflow state, if the next floc overflow state is the abnormal state, it increases the amount of chemical injected compared to when the caution state is continuous.

9. The state determination unit confirms the flock overflow state at the predetermined timing, The control device according to claim 1, characterized in that the control unit outputs information regarding the control of the wastewater treatment based on the flock overflow state at each of a plurality of consecutive timings.

10. The state determination unit uses the estimation model to determine the floc overflow state as a normal state in which no flocs are flowing out of the sedimentation tank, an abnormal state in which flocs are flowing out of the sedimentation tank, and a cautionary state in which no flocs are flowing out of the sedimentation tank but there is a risk of the abnormal state occurring. The control device according to claim 9, wherein the control unit outputs information regarding the control of the wastewater treatment to be performed in the warning state if the flock overflow state at a specific timing of the plurality of consecutive timings is the warning state or an abnormal state, and the flock overflow state at the next timing is the warning state or an abnormal state.

11. The control device according to claim 10, wherein the control unit outputs information regarding the control of the wastewater treatment to be performed in the abnormal state if the floc overflow state at a specific timing of the plurality of consecutive timings is the abnormal state, and the floc overflow state at the next timing is the abnormal state.

12. The control device according to claim 10, wherein the control unit outputs information regarding the control of the wastewater treatment to be performed in the normal state if the flock overflow state at a specific timing of the plurality of consecutive timings is the warning state or the abnormal state, and the flock overflow state at the next timing is not the warning state or the abnormal state.

13. The control device according to claim 1, further comprising a learning processing unit that causes the estimation model to learn the relationship between the measurement information and the floc overflow state based on the measurement information and the result of identifying the floc overflow state at the time of acquisition of the measurement information, which is determined from the presence or absence of floc overflow at the time of acquisition of the measurement information, where flocs flow out of the sedimentation tank.

14. The control device A state determination step in which, based on measurement information obtained by an ultrasonic sensor placed in a sedimentation tank for settling and separating inorganic flocs in wastewater in wastewater treatment, and the result of identifying the floc overflow state, an estimation model that has learned the relationship between the measurement information and the floc overflow state is used to determine the floc overflow state at a predetermined timing based on first measurement information obtained by the ultrasonic sensor at a predetermined timing, A control step that outputs information regarding the control of the wastewater treatment based on the floc overflow state confirmed in the state determination step. A water treatment method characterized by performing the following.

15. A state determination step in which, based on measurement information obtained by an ultrasonic sensor placed in a sedimentation tank for settling and separating inorganic flocs in wastewater in wastewater treatment, and the result of identifying the floc overflow state, an estimation model that has learned the relationship between the measurement information and the floc overflow state is used to determine the floc overflow state at a predetermined timing based on first measurement information obtained by the ultrasonic sensor at a predetermined timing, A control step that outputs information regarding the control of the wastewater treatment based on the floc overflow state confirmed in the state determination step. A water treatment program characterized by having a computer execute it.

Images