Activated carbon injection control system, activated carbon injection control method, and water treatment system

Abstract

To provide an activated carbon injection control system, an activated carbon injection control method, and a water treatment system which are capable of inexpensively and quickly identifying moldy odors generated from treated water, and determining an activated carbon injection rate without excess or deficiency, thereby stably suppressing moldy odors generated from the treated water for a long period of time.SOLUTION: An activated carbon injection control system includes picoplankton number prediction means 201 that predicts a number of picoplankton in raw water by machine learning using raw water quality data and meteorological data, storage method 103 that stores water quality data, meteorological data, and predicted values of picoplankton number, moldy odor substance concentration prediction means 202 that predicts the concentration of moldy odor substances in raw water by machine learning using the predicted values of picoplankton number and water quality data, and activated carbon injection control means 204 that controls the activated carbon injection rate of activated carbon injected into raw water based on the activated carbon injection rate determined based on the predicted results of the concentration of moldy odor substances.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an activated carbon injection control system, an activated carbon injection control method, and a water treatment system, and more particularly to an activated carbon injection control system, an activated carbon injection control method, and a water treatment system suitable for activated carbon treatment in which odorous substances in raw water such as river water, dam lakes, and groundwater are adsorbed and removed by activated carbon. [Background technology]

[0002] In the treatment of raw water such as river water or groundwater, insoluble components such as turbidity components and algae are removed by solid-liquid separation techniques such as coagulation sedimentation and sand filtration. In the process of coagulation and filtration of raw water, coagulation inhibition may occur due to various factors, and the presence of picoplankton has become a problem in recent years as one of the causes. Picoplankton is a type of plankton with a size of 0.2 to 2.0 μm, and it has been reported that it is a factor in increasing the turbidity of filtered water (see Non-Patent Document 1).

[0003] Picoplankton are classified into two groups based on the mechanism of nutrient uptake. One is heterotrophic picoplankton, which is mainly composed of bacteria that break down organic matter to obtain nutrients, and the other is phytopicoplankton, which performs photosynthesis. Picoplankton contain pigments such as chlorophyll and can be observed under a fluorescent microscope. It has been confirmed that they are present in raw water or treated water in water purification treatment plants.

[0004] In recent years, the Ministry of Health, Labor and Welfare has been instructing the management of filtered water turbidity at 0.1 degrees or less as a measure against infectious diseases caused by Cryptosporidium transmitted through tap water (see Non-Patent Document 2). For this reason, water suppliers in each region have positioned the management of filtered water turbidity at 0.1 degrees or less at all times as one of the most important water quality management items.

[0005] However, in recent years, picoplankton outbreaks due to eutrophication have become frequent in tap water sources all over the country. Picoplankton are minute, measuring 0.2 to 2.0 μm in size, and leak out of filtration basins, causing the turbidity of the filtered water to rise to 0.1 degrees or more. Water utilities are currently struggling with treatment measures. In addition, some picoplankton emit odorous substances, which are the cause of an unpleasant odor and taste in tap water. To prevent such problems, a technique is known that continuously monitors the amount of picoplankton present in water and controls the amount of chemicals such as coagulants and chlorine to be injected depending on the monitoring results (see, for example, Patent Document 1).

[0006] In many water purification plants, powdered activated carbon is injected to remove odorous substances and dissolved organic matter in the raw water, and coagulants are injected to remove turbidity. In this type of water treatment system, the injection rate of powdered activated carbon must be determined according to the quality of the raw water.

[0007] The jar test (hereinafter also referred to as the "beaker test") is often used as a method for determining the injection rate of powdered activated carbon. In the jar test, raw water is collected in multiple beakers, and different amounts of powdered activated carbon are injected into each of the collected raw water samples to evaluate the removal rate of odorous substances and soluble organic matter, and the minimum injection rate of powdered activated carbon required to reduce the odor concentration and soluble organic matter after treatment to below the target concentration is determined.

[0008] However, in the method of determining the injection rate of powdered activated carbon by the jar test, it is very difficult to inject powdered activated carbon in accordance with changes in the quality of the raw water, and there is a risk of over- or under-injection. In addition, since it takes time to determine the optimal injection rate of powdered activated carbon in the jar test, the quality of the raw water may have already changed by the time the optimal injection rate of powdered activated carbon is obtained. In fact, when the injection rate of powdered activated carbon to be added is determined, the injection rate is determined on the safe side compared to the injection rate obtained by the jar test, resulting in over-injection.

[0009] Powdered activated carbon has a very high unit price compared to other chemicals such as coagulants, sulfuric acid, and sodium hypochlorite. Therefore, excessive injection of powdered activated carbon may lead to a sharp rise in chemical costs. From an economical point of view, a method is desired that suppresses excessive injection of powdered activated carbon and controls the optimal injection rate of powdered activated carbon according to changes in the quality of the raw water.

[0010] As a method for controlling the activated carbon injection rate, for example, an odor sensor is used to detect the concentration of odorous substances in the raw water in the powdered activated carbon treatment tank, and at the same time, the injection rate of powdered activated carbon required for removing odorous substances in the raw water is calculated based on at least one of the ultraviolet absorbance, the fluorescence intensity, and the soluble organic carbon concentration of the raw water in the powdered activated carbon treatment tank and the coagulant injection rate, and the powdered activated carbon is controlled based on this powdered activated carbon injection rate (see, for example, Patent Document 2).

[0011] Furthermore, a method is also known in which treated water quality values, including raw water turbidity and raw water musty odor value, are predicted from the coagulant injection rate and activated carbon injection rate, and optimal coagulant injection rate and activated carbon injection rate are controlled based on the prediction results (see, for example, Patent Document 3). [Prior art documents] [Patent documents]

[0012] [Patent Document 1] Patent No. 5430788 [Patent Document 2] Patent Publication No. 2022-026969 [Patent Document 3] JP 2015-157239 A [Non-patent literature]

[0013] [Non-Patent Document 1] Japan Water Works Association, Water Purification Treatment Guide to Prevent Biological Hazards, March 2006, p. 57 [Non-Patent Document 2] Ministry of Health, Labor and Welfare Notification, Yakuseisuihatsu 0529 No. 1, "Guidelines for Countermeasures against Cryptosporidium, etc. in Water Supply," Appendix 2, page 3, May 29, 2019 Summary of the Invention [Problem to be solved by the invention]

[0014] However, none of the conventional techniques described in Patent Documents 1 to 3 has yet established a method that can monitor odorous substances in raw water inexpensively, simply, and in real time, and determine the appropriate activated carbon injection rate.

[0015] For example, in the method described in Patent Document 1, a biological particle counter is used to monitor the number of biological particles in water, and the monitoring results are used to control the injection amount of chemicals such as coagulants and chlorine. It takes a certain amount of time for the concentration of the chemicals to be adjusted to an appropriate range and for odorous substances to be removed. Therefore, it is difficult to respond quickly when a sudden increase in biological particles occurs. In addition, Patent Document 1 requires the introduction and maintenance of special equipment, including a biological particle counter, which is costly and time-consuming.

[0016] The method described in Patent Document 2 requires real-time measurement of moldy odor substances in raw water. The minimum concentration at which odor complaints about moldy odors are received is, for example, 20 ng / L or more for 2-methylisoborneol (2-MIB) and 30 to 40 ng / L or more for geosmin, but it is difficult to measure such odors in real time without error. In addition, the introduction of a highly sensitive odor sensor requires high installation costs and maintenance and management efforts, so the barrier to introduction is high. The method described in Patent Document 3 also requires the introduction of special sensors such as an odor sensor when measuring the raw water mold odor value contained in the raw water in predicting the coagulant injection rate and the activated carbon injection rate, so it requires high installation costs and maintenance and management efforts, just like Patent Document 2.

[0017] In view of the above problems, the present invention provides an activated carbon injection control system, an activated carbon injection control method, and a water treatment system that can inexpensively and quickly grasp the mold odor generated from the treated water and determine the activated carbon injection rate without excess or deficiency, thereby stably suppressing the mold odor generated from the treated water for a long period of time. [Means for solving the problem]

[0018] As a result of intensive research aimed at solving the above-mentioned problems, the inventors discovered that it would be effective to use machine learning to predict the number of picoplankton, which has a strong positive correlation with mold-odor substances, and then use the predicted results of the picoplankton number to further predict the concentration of the mold-odor substances using machine learning.

[0019] In order to solve the above problems, in one aspect, the present invention is an activated carbon injection control system comprising: a picoplankton count prediction means for predicting the number of picoplankton in raw water through machine learning using water quality data and meteorological data of the raw water; a storage means for storing predicted values ​​of water quality data, meteorological data, and picoplankton count; a moldy odor substance concentration prediction means for predicting the moldy odor substance concentration in the raw water through machine learning using the predicted value of the picoplankton count and the water quality data; and an activated carbon injection control means for controlling the activated carbon injection rate of activated carbon injected into the raw water based on the activated carbon injection rate determined based on the predicted result of the moldy odor substance concentration.

[0020] In one embodiment of the activated carbon injection control system according to the present invention, the picoplankton count prediction means calculates a predicted value of the picoplankton count using a machine learning model constructed by a machine learning algorithm that uses water quality data including at least information on the pH, conductivity, turbidity, and M-alkalinity of the raw water and meteorological data including at least information on the amount of solar radiation as explanatory variables, and the picoplankton count as a response variable.

[0021] In another embodiment of the activated carbon injection control system according to the present invention, the mold odor substance concentration prediction means calculates a predicted value of the mold odor substance concentration using a machine learning model constructed by a machine learning algorithm that uses the predicted value of the picoplankton count and water quality data including at least information on the electrical conductivity, turbidity, and M-alkalinity of the raw water as explanatory variables, and the mold odor substance concentration as the objective variable.

[0022] In yet another embodiment, the activated carbon injection control system according to the present invention includes an activated carbon injection rate prediction means for calculating a predicted value of the activated carbon injection rate using a machine learning model constructed by a machine learning algorithm in which a predicted value of the mold odor substance concentration, water quality data including the temperature of the raw water, and a set value of the activated carbon injection rate associated with the mold odor substance concentration are explanatory variables, and the predicted value of the activated carbon injection rate is an objective variable.

[0023] In yet another embodiment of the activated carbon injection control system according to the present invention, the water quality data further includes information on the water level of a storage tank in which the raw water is stored, the amount of raw water flowing into the storage tank, and the temperature of the raw water.

[0024] In yet another embodiment of the activated carbon injection control system according to the present invention, the meteorological data further includes data on the outside air temperature, the hours of sunshine, and the amount of rainfall around the raw water.

[0025] In another aspect, the present invention is an activated carbon injection control method comprising: inputting measured values ​​of raw water quality data and weather data into a machine learning model constructed by machine learning using learning data including past water quality data and past weather data of the raw water, and outputting a predicted value of the future picoplankton count in the raw water; inputting a predicted value of future picoplankton count and measured water quality data into a machine learning model constructed by machine learning using the past predicted values ​​of picoplankton count and the past water quality data, and outputting a predicted value of future moldy odor substance concentrations in the raw water; and controlling the activated carbon injection rate of activated carbon to be injected into the raw water based on the activated carbon injection rate determined based on the future prediction result of the moldy odor substance concentration.

[0026] In yet another aspect, the present invention is a water treatment system comprising: a picoplankton number prediction means for predicting the number of picoplankton in the raw water by machine learning using water quality data and meteorological data of the raw water; a moldy odor substance concentration prediction means for predicting the concentration of moldy odor substances in the raw water by machine learning using the predicted value of the picoplankton number and the water quality data; an activated carbon injection means for injecting activated carbon into the raw water; and an activated carbon injection control means for controlling the activated carbon injection rate of activated carbon injected by the activated carbon injection means based on the activated carbon injection rate determined on the basis of the prediction result of the moldy odor substance concentration. Effect of the Invention

[0027] According to the present invention, it is possible to provide an activated carbon injection control system, an activated carbon injection control method, and a water treatment system that can inexpensively and quickly grasp the mold odor emanating from the treated water and determine the activated carbon injection rate appropriately, thereby enabling the mold odor emanating from the treated water to be stably suppressed for a long period of time. [Brief description of the drawings]

[0028] [Figure 1] 1 is a block diagram showing an example of a water treatment system including an activated carbon injection control system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a block diagram showing an example of the configuration of a server. [Diagram 3] FIG. 13 is an image diagram showing an example of a transition in the predicted value of picoplankton count. [Figure 4] FIG. 13 is an image diagram showing an example of the transition of predicted values ​​of picoplankton count and musty odor substance concentration. [Diagram 5] 1 is an image diagram showing an example of changes in predicted values ​​of picoplankton count, mold odor substance concentration, and activated carbon injection rate. [Figure 6] FIG. 4 is a flow diagram illustrating an example of a method for controlling activated carbon injection according to an embodiment of the present invention. [Figure 7] This is a graph showing the relationship between the predicted results and the measured values ​​when predicting picoplankton counts using the random forest (RF) method as a machine learning algorithm. [Figure 8] 1 is a graph showing the relationship between the predicted results and the measured values ​​when the number of picoplankton is predicted using an artificial neural network (ANN) method as a machine learning algorithm. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0029] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the following description of the drawings, the same or similar parts are denoted by the same or similar reference numerals. Note that the embodiment shown below is an example of an apparatus and method for embodying the technical idea of ​​the present invention, and the technical idea of ​​the present invention does not limit the structure, arrangement, etc. of the components to those described below.

[0030] A water treatment system equipped with an activated carbon injection control system according to an embodiment of the present invention includes a water treatment plant 10, a server 100, an activated carbon injection rate control device 200, and an activated carbon injection facility (activated carbon injection means) 300, as shown in FIG.

[0031] For example, water purification equipment such as a water purification plant can be used as the water treatment plant 10. In Fig. 1, a water purification plant including a receiving well 1, a coagulation sedimentation basin 2, a sand filtration basin 3, and a distributing reservoir 4 is used as the water treatment plant 10. However, the water treatment plant 10 of the present invention is not limited to the example shown in Fig. 1, and may of course be equipped with various other water treatment equipment not shown.

[0032] The receiving well 1 is a storage tank that stores raw water conveyed from a water intake tower or the like, and typically regulates the amount and level of the raw water taken in and monitors the quality of the raw water. The receiving well 1 can be equipped with a water quality data measuring device 11 for measuring water quality data of the raw water, a meteorological data measuring device 12 for measuring meteorological data of the raw water, and an activated carbon injection facility 300.

[0033] The water quality data measuring device 11 may include, but is not limited to, a water level meter for measuring the water level in the receiving well 1, a pH meter, a conductivity meter, a turbidity meter, an alkalinity meter, a flow meter for measuring the flow rate of raw water to the receiving well 1, a water thermometer, a color meter, a particle counter for measuring the number of picoplankton, etc. In addition, the water quality data measuring device 11 may include various measuring instruments for measuring the total organic carbon (TOC), chemical oxygen demand (COD), ultraviolet absorbance, organic matter fraction, light intensity, etc. of the raw water in the receiving well 1.

[0034] As the meteorological data measuring device 12, various sensors for measuring meteorological data around the water treatment plant 10 can be used. For example, measuring instruments for measuring precipitation, sunshine hours, solar radiation, and outside air temperature can be used as the meteorological data measuring device 12. If the meteorological data measuring device 12 is not installed inside the water treatment plant 10, meteorological data from multiple municipalities around the water treatment plant 10 may be used. For example, it is preferable to obtain multiple meteorological data including precipitation, sunshine hours, solar radiation, and outside air temperature from multiple municipalities and use the meteorological data from points with high correlation.

[0035] The water quality data and meteorological data obtained by the water quality data measuring device 11 and the meteorological data measuring device 12 are output to and stored in a storage means 103 (see Figure 2) which stores various data within the server 100, or in a storage means (not shown) consisting of a storage medium for storing various data provided in the activated carbon injection rate control device 200 described later.

[0036] The activated carbon injection equipment 300 is equipment that injects activated carbon into raw water. Depending on the properties of the activated carbon, the activated carbon injection equipment 300 may be equipment that injects wet powdered activated carbon (wet carbon), dry powdered activated carbon (dry carbon), or granular or powdered activated carbon. In this embodiment, it is preferable to use, as the activated carbon injection equipment 300, injection equipment that can always inject the right amount of activated carbon and that can inject granular or powdered activated carbon.

[0037] The type of activated carbon is not limited. For example, as activated carbon capable of efficiently removing moldy odors, it is preferable to use activated carbon that is made by reacting coconut shell, coal, or charcoal as the main raw material with steam at 800 to 1000°C to activate it. Among them, activated carbon that has fine pores and has a diameter of about 800 to 2500 m is preferable. 2 Activated carbon having a specific surface area of ​​about 1 / g can be used. By adsorbing odorous substances such as geosmin and 2-MIB, or organic matter that causes dirt in the raw water and treated water, into the pores of such activated carbon, the odorous substances and organic matter can be efficiently removed.

[0038] The particle size of the activated carbon used to remove mold odor substances in this embodiment is not particularly limited, but in this embodiment, from the viewpoint of ease of handling, it is preferable to use powdered activated carbon having a particle size of 0.15 mm or less and pores with an average pore size of about 1 to 20 nm. The particle size of the activated carbon is evaluated by the effective diameter (10% passing diameter) based on the activated carbon test method of JIS K1474:2014. The particle size of the activated carbon may be adjusted and changed as appropriate according to the picoplankton count of the raw water in order to efficiently remove the mold odor.

[0039] In the example of FIG. 1, the water quality data measuring device 11, the meteorological data measuring device 12, and the activated carbon injection equipment 300 are connected to the receiving well 1. However, it goes without saying that the water quality data measuring device 11, the meteorological data measuring device 12, and the activated carbon injection equipment 300 can be applied to a storage tank where activated carbon treatment is performed, or any equipment where the number of picoplankton is to be measured. For example, the water quality data measuring device 11, the meteorological data measuring device 12, and the activated carbon injection equipment 300 may be connected to any equipment such as the coagulation sedimentation basin 2, the sand filtration basin 3, and the distribution reservoir 4. In other words, the raw water described in the present invention is used to include not only the raw water supplied to the receiving well 1, but also various treated waters stored in various equipment to which the water quality data measuring device 11, the meteorological data measuring device 12, and the activated carbon injection equipment 300 are connected.

[0040] The system may include one or more servers 100. The server 100 does not necessarily have to be located within the water treatment plant 10, and may be located at a location other than the water treatment plant 10 and connected to the activated carbon injection rate control device 200 via a communication network (not shown). The server 100 may be omitted, and the activated carbon injection rate control device 200 may be used instead of the server 100.

[0041] As shown in Figure 2, the server 100 is equipped with, for example, a control unit 101 that calculates and processes various information based on various programs, a communication unit 102 that exchanges various information with client devices, etc., and a storage means 103 that can store raw water quality data, meteorological data, and predicted values ​​of picoplankton numbers calculated by the picoplankton number prediction means 201, various control programs, etc., and these are connected by a bus or the like.

[0042] 1 may be configured with a general hardware configuration, and may include a processor, a memory (e.g., RAM, etc.), a storage medium (e.g., HDD, SSD, etc.), and a communication module. The activated carbon injection rate control device 200 may be configured as a single piece of hardware, or may be distributed across multiple pieces of hardware.

[0043] The activated carbon injection rate control device 200 comprises picoplankton number prediction means 201 which predicts the number of picoplankton in the raw water by machine learning using water quality data and meteorological data of the raw water, moldy odor substance concentration prediction means 202 which predicts the concentration of moldy odor substances in the raw water by machine learning using the predicted value of the picoplankton number by the picoplankton number prediction means 201 and the water quality data of the raw water, activated carbon injection rate determination means 203 which determines the injection rate of activated carbon to be injected into the raw water in the receiving well 1, typically based on the prediction result of the moldy odor substance concentration by the moldy odor substance concentration prediction means 202, and activated carbon injection control means 204 which controls the injection rate of activated carbon to be injected into the raw water based on the activated carbon injection rate determined based on the prediction result of the moldy odor substance concentration.

[0044] The picoplankton number prediction means 201 uses a machine learning model constructed to acquire specified explanatory variables and output specified objective variables using a machine learning algorithm to calculate a predicted value of the picoplankton number in the raw water, and stores the calculation result in the storage means 103 of the server 100 or in a storage means (not shown) provided in the activated carbon injection rate control device 200.

[0045] In this embodiment, the "picoplankton number" refers to, for example, the number of picoplankton present per unit volume, and specifically, is expressed as the number of picoplankton with a cell diameter of 0.2 to 2.0 μm present per 1 ml of raw water.

[0046] The machine learning algorithm is not particularly limited, but may be a random forest (RF) method, an artificial neural network (ANN) method, a support vector regression (SVR) method, a partial least squares (PLS) method, a decision tree method, a very short-term memory neural network (LSTM) method, etc. In the present embodiment, it is particularly preferable to use a machine learning model constructed using the ANN method, the RF method, or the LSTM method.

[0047] As explanatory variables, various data obtained by the water treatment plant 10 can be used, but in this embodiment, at least raw water quality data and meteorological data are used.

[0048] For example, in the case of a water purification plant, water quality data may include pH, conductivity, turbidity, alkalinity (M-alkalinity), color, transparency, total organic carbon (TOC), chemical oxygen demand (COD), ultraviolet absorbance, organic matter fraction, particulate count, nutrient concentrations (total nitrogen (TN), total phosphorus (TP)), light intensity, and dissolved oxygen.

[0049] Various types of data can be used for water quality data, but it is preferable to preferentially use data that has a strong positive correlation with the increase or decrease in picoplankton count as an explanatory variable. For example, turbidity is an example of water quality data that has a correlation with picoplankton count, and it has a large effect on the increase or decrease in picoplankton count. By using turbidity as an explanatory variable for picoplankton count as learning data for machine learning, it is possible to accurately predict the picoplankton count.

[0050] As the amount of CO2 in the raw water increases, the number of picoplankton tends to increase. When the amount of CO2 in the raw water increases, the raw water becomes slightly acidic and the pH decreases. Therefore, it is also preferable to use pH as an explanatory variable for the number of picoplankton in the learning data during machine learning.

[0051] As the concentration of nutrients in the raw water increases, the number of picoplankton increases. As the concentration of nutrients in the raw water increases, the electrical conductivity and M-alkalinity also increase, so it is also preferable to use electrical conductivity and M-alkalinity as explanatory variables for the number of picoplankton in the learning data during machine learning.

[0052] The more explanatory variables there are, the more accurate the prediction of the picoplankton count will be, but if there are too many, huge computational resources will be required for verification, etc. Also, if the picoplankton count in the raw water is significantly higher than the standard value, or if there is a sudden change in water quality or weather, the prediction accuracy of the picoplankton count may decrease.

[0053] As explanatory variables for easily and quickly grasping the musty odor generated from the raw water or treated water while increasing the prediction accuracy of the picoplankton count and determining the appropriate activated carbon injection rate, it is preferable to include water quality data including at least information on pH, conductivity, turbidity, and alkalinity (M-alkalinity) and meteorological information including the amount of sunlight. Furthermore, it is preferable to further include as explanatory variables data on the water level of a storage tank such as the receiving well 1 in which the raw water or treated water to be predicted for the picoplankton count is stored, the amount of raw water flowing into the storage tank, and water temperature.

[0054] When the water level in the receiving well 1 drops, mold and other microorganisms may grow in the raw water. Also, when the amount of water flowing into the storage tank decreases, the water level in the receiving well 1 drops, and there is a risk that the concentration of moldy odor substances in the raw water may increase. When the water temperature in the receiving well 1 rises due to strong sunlight or other factors, picoplankton may grow. Therefore, by including data on the water level of the storage tank that stores the raw water or treated water, the amount of raw water flowing into the storage tank, and water temperature in addition to the above water quality data, the accuracy of predicting the number of picoplankton can be further improved.

[0055] It is more preferable that the meteorological data further includes information on the amount of solar radiation, as well as information on the outside temperature, the hours of sunlight, and the amount of rainfall. The amount of solar radiation is measured using a solar radiation sensor or the like to measure any amount of solar radiation, such as direct solar radiation, diffuse solar radiation, and global solar radiation, around the receiving well 1. If there is no sensor for measuring meteorological data around the storage tank that stores the raw water or treated water, the average value of the meteorological data around the receiving well 1 may be calculated from the meteorological data of multiple municipalities around the receiving well 1.

[0056] When using the water quality data and meteorological data as learning data for a machine learning model, it is preferable to perform preprocessing such as deleting or complementing missing values ​​contained in the water quality data and meteorological data and performing preprocessing (correction) to delete outliers. In addition, it is preferable to use time-series data including not only instantaneous values ​​but also trends in past data as the water quality data and meteorological data. From the viewpoint of accurate prediction, it is preferable for the time-series data to include at least several months, and more preferably to include about six months to one year.

[0057] The moldy odor substance concentration prediction means 202 calculates a predicted value of the moldy odor substance concentration using a machine learning model constructed by a machine learning algorithm that uses the predicted value of the picoplankton count predicted by the picoplankton count prediction means 201 and water quality data including at least information on the electrical conductivity, turbidity, and M-alkalinity of the raw water as explanatory variables, and the moldy odor substance concentration as the objective variable.

[0058] There is no need to measure moldy odor substances directly from the raw water because the moldy odor substance concentration prediction means 202 calculates a predicted value for the moldy odor substance concentration through machine learning using the predicted value of the picoplankton count and water quality data. As a result, there is no need for a sensor to measure moldy odor substances, making it possible to downsize and simplify the equipment, and maintenance work for the sensor can also be omitted, and the moldy odor substance concentration in the raw water can be grasped quickly and in real time.

[0059] Explanatory variables used in predicting the musty odor substance concentration include water quality data including at least the predicted value of picoplankton count, raw water electrical conductivity, turbidity (preferably ultraviolet absorbance at 260 nm (E260)), and M-alkalinity, as well as meteorological data including at least information on the amount of solar radiation and water temperature, and may further include information on outside air temperature and hours of sunshine. In order to further improve the prediction accuracy of the musty odor substance concentration, it is preferable to further include water quality data such as TOC, water level in the storage tank, amount of raw water flowing into the storage tank, and hours of sunshine, and any of the water quality data of pH, precipitation (rainfall), TN, and TP may further be added as explanatory variables.

[0060] Musty odor substances are often produced when picoplankton proliferates, and have a strong positive correlation with the number of picoplankton. In other words, the concentration of musty odor substances can be expressed as [picoplankton number x mold proliferation rate]. When predicting the concentration of musty odor substances, the accuracy of predictions can be improved by taking into account factors that inhibit the proliferation of picoplankton numbers and factors that promote it.

[0061] For example, electrical conductivity is effective in expressing the total amount of organic acids, TN, and TP required for phytoplankton growth with a certain degree of accuracy. Iron, a trace element essential for phytoplankton, exists as dissolved organic complex iron. Therefore, turbidity in raw water can be said to be an effective factor in that it can grasp the increase or decrease of phytoplankton. In addition, since phytoplankton grow by photosynthesis, if the other factors considered (electrical conductivity, turbidity, water temperature) exceed a certain value and the amount of solar radiation is high, the number of picoplankton will increase rapidly, which will lead to an increase in moldy odor substances. M-alkalinity can be used as an indicator of the concentration of CO2 dissolved in the water, and the higher the value, the easier it is for picoplankton to grow and the higher the concentration of moldy odor substances. Water temperature depends on the type of mold, but in many cases mold growth rates tend to increase when it exceeds 20°C. The lower the water level in the storage tank, the higher the concentration of phytoplankton per unit volume, which leads to an increase in the concentration of moldy odor substances. The smaller the amount of raw water flowing into the storage tank, the longer the retention time of the raw water and the longer the time for which the concentration of moldy odor substances will accumulate. Also, the more hours of sunlight there are, the easier it is for phytoplankton to grow and the higher the concentration of moldy odor substances will be. pH can be an indicator of the CO2 concentration and dissolved organic matter in the raw water, and when phytoplankton grows, the CO2 concentration in the raw water decreases, so it tends to become alkaline. Rainfall is an effective factor in that the concentration of moldy odor substances is diluted by the amount of raw water increased by rainfall.

[0062] Considering these circumstances, when predicting the concentration of moldy odor substances, it is possible to make an appropriate prediction by using information on electrical conductivity, solar radiation, turbidity, M-alkalinity, and water temperature as learning parameters, and it is preferable to further take into consideration the water level of the storage tank, the amount of raw water flowing into the storage tank, and the hours of sunlight, and it is even more preferable to take into consideration pH and precipitation. Furthermore, at least one of the nutrient salt concentration and light intensity may be taken into consideration.

[0063] The activated carbon injection rate determination means 203 determines the activated carbon injection rate so as to be the optimum activated carbon injection rate for the moldy odor substance concentration predicted value predicted by the moldy odor substance concentration prediction means 202. The activated carbon injection rate determination means 203 may perform calculations using a correlation equation, a physical model, or the like that is prepared in advance for determining the activated carbon injection rate.

[0064] Alternatively, the activated carbon injection rate determination means 203 may include an activated carbon injection rate prediction means for calculating a predicted value of the activated carbon injection rate using a machine learning model constructed by a machine learning algorithm that uses the predicted value of the mold odor substance concentration, water quality data including the temperature of the raw water, and a set value of the activated carbon injection rate associated with the mold odor substance concentration as explanatory variables, and the predicted value of the activated carbon injection rate as a target variable. The activated carbon injection control means 204 outputs a control signal to the activated carbon injection facility 300 for controlling the injection rate of activated carbon injected from the activated carbon injection facility 300, based on the predicted value of the activated carbon injection rate by the activated carbon injection rate determination means 203.

[0065] FIG. 3 is an image diagram showing the transition of the predicted value of the picoplankton count in the raw water at the current time and in the future (n minutes later) by the picoplankton count prediction means 201, FIG. 4 is a image diagram showing the transition of the predicted value of the mold odor substance concentration in the raw water by the mold odor substance concentration prediction means 202, and FIG. 5 is an image diagram showing the transition of the result of the determination of the activated carbon injection rate by the activated carbon injection rate determination means 203. The mold odor substance concentration increases and decreases according to the increase and decrease of the picoplankton count. Therefore, by adjusting the activated carbon injection rate to an optimal injection rate based on the predicted value of the mold odor substance concentration, it is possible to determine the activated carbon injection rate without excess or deficiency. Furthermore, according to the water treatment device using the activated carbon injection control system of this embodiment, the predicted value of the mold odor substance concentration in the future (n minutes later) can be predicted, so that the activated carbon injection control can be performed more appropriately, for example, even when the picoplankton count in the raw water increases suddenly in a short time or when the fluctuation of water quality or weather changes suddenly. As a result, it is possible to stably suppress the mold odor generated from the treated water for a long period of time.

[0066] An example of an activated carbon injection control method using the water treatment system shown in Fig. 1 will be described with reference to the flowchart in Fig. 6. A meteorological data measuring device 12 and an activated carbon injection facility 300 are installed at a point where it is desired to measure the number of picoplankton. For example, as shown in Fig. 1, a water quality data measuring device 11 for measuring at least information on pH, conductivity, turbidity, and M-alkalinity as water quality data, a meteorological data measuring device 12 for measuring at least information on the amount of solar radiation, and an activated carbon injection facility 300 are installed in a water-receiving well 1. In step S11, the water quality data measuring device 11 and the meteorological data measuring device 12 acquire water quality data and meteorological data of the raw water. The acquired results are stored in the server 100.

[0067] In step S12, the picoplankton number prediction means 201 calculates a predicted value of the picoplankton number using a machine learning model constructed by a machine learning algorithm in which water quality data including at least information on the pH, electrical conductivity, turbidity, and M-alkalinity of the raw water and meteorological data including at least information on the amount of solar radiation are used as explanatory variables, and the picoplankton number is used as the objective variable. For example, the picoplankton number prediction means 201 inputs the measured values ​​of the water quality data and meteorological data of the raw water into a machine learning model constructed by machine learning using learning data including past water quality data of the raw water and past meteorological data, and outputs a predicted value of the future (n minutes later) picoplankton number in the raw water. The calculation result of the picoplankton number is stored in the server 100.

[0068] In step S13, moldy odor substance concentration prediction means 202 calculates a predicted value of the moldy odor substance concentration using a machine learning model constructed by a machine learning algorithm that uses the predicted value of the picoplankton count predicted by picoplankton count prediction means 201 and water quality data including at least information on the electrical conductivity, turbidity, and M-alkalinity of the raw water as explanatory variables, and the moldy odor substance concentration as the objective variable. For example, moldy odor substance concentration prediction means 202 inputs a predicted value of the future picoplankton count and measured weather data into a machine learning model constructed by machine learning using past predicted values ​​of the picoplankton count and past water quality data, and outputs a predicted value of the future moldy odor substance concentration in the raw water (n minutes later). The predicted value of the moldy odor substance concentration is saved in server 100.

[0069] In step S14, the activated carbon injection rate determination means 203 calculates a predicted value of the activated carbon injection rate using a machine learning model constructed by a machine learning algorithm that uses the predicted value of the mold odor substance concentration, water quality data including the temperature of the raw water, and a set value of the activated carbon injection rate associated with the mold odor substance concentration as explanatory variables, and the predicted value of the activated carbon injection rate as a target variable. For example, the activated carbon injection rate determination means 203 outputs a predicted value of the injection rate of activated carbon to be injected into the raw water in the future (n minutes later) based on the predicted value of the mold odor substance concentration in the future (n minutes later). The predicted value of the activated carbon injection rate is stored in the server 100. Note that the predicted value of the activated carbon injection rate may be calculated by using a predetermined physical model or the like from the relationship between the predicted value of the mold odor substance concentration and a preset activated carbon injection rate, without using machine learning.

[0070] In step S15, the activated carbon injection control means 204 outputs a control signal to the activated carbon injection equipment 300 for controlling the injection rate of activated carbon injected from the activated carbon injection equipment 300 based on the predicted value of the activated carbon injection rate by the activated carbon injection rate determination means 203.

[0071] According to the activated carbon injection control method of the embodiment of the present invention, it is possible to appropriately predict the current or future mold odor occurring from raw water or treated water by using machine learning without using a special sensor. This makes it possible to constantly control the injection rate of activated carbon for removing mold odor within an appropriate range, and to stably suppress the mold odor that may occur from raw water or treated water for a long period of time.

[0072] An example of a method for creating a machine learning model used in the activated carbon injection control method according to the embodiment of the present invention will be described below. Note that the method for creating the machine learning model shown below is only one example, and various methods other than the example shown below may be used for creating the model. Of course, if the prediction error is within an appropriate range, some of the steps shown below can be omitted.

[0073] First, time series data of raw water quality data and meteorological data are acquired by the water quality data measuring device 11 and the meteorological data measuring device 12. Next, the number of picoplankton measured in the receiving well 1 is used as the objective variable, explanatory variables for executing machine learning processing are selected from the water quality data and meteorological data obtained by the water quality data measuring device 11 and the meteorological data measuring device 12, and a machine learning model is created using a predetermined machine learning algorithm.

[0074] In creating a machine learning model, it is preferable to select hyperparameters to improve the prediction accuracy of machine learning. In selecting hyperparameters, for example, the ratio of training data for machine learning to test data is set to 70:30 to 90:10, typically 80:20, and the prediction accuracy is determined based on the prediction accuracy for the test data.

[0075] For example, when using the RF method as a machine learning algorithm, the number of decision trees, depth, number of explanatory variables used in the decision trees, etc. are adjusted. The more decision trees there are, the higher the prediction accuracy, but if there are too many, it takes time to perform calculations. If the depth of the decision tree is too shallow, accurate prediction cannot be performed, while if it is too deep, overlearning is likely to occur. Although not limited to the following, it is preferable that the number of decision trees is about 10 to 500, more preferably 50 to 120, the number of types of data used as explanatory variables is about 3 to 15, more preferably 5 to 12, and the depth is None, at least in the prediction of the number of picoplankton and the concentration of moldy odor substances.

[0076] When using the ANN method as a machine learning algorithm, the number of layers to be installed, the number of neurons to be installed in one layer, the dropout rate, etc. are adjusted. The more layers there are, the more accurate predictions can be made for complex events, but learning can become more difficult. Similarly, the more neurons there are, the more complex learning can be performed, but if there are too many, overlearning is likely to occur. Depending on the type of activation function, the output will differ. If the learning rate is too large, the learning speed will increase, but the optimal solution may not be obtained properly. Although not limited to the following, at least in predicting the number of picoplankton and the concentration of moldy odor substances, it is preferable to set the number of layers to about 1 to 10, more preferably 2 to 5, and to appropriately optimize the number of neurons, dropout rate, activation function, and learning rate to set the hyperparameters.

[0077] Figure 7 shows the characteristics of the prediction results for picoplankton numbers when the RF method was used as a machine learning algorithm. The correlation coefficient was 0.934, indicating that the picoplankton numbers were predicted with high accuracy.

[0078] Figure 8 shows the characteristics of the prediction results for picoplankton numbers when the ANN method was used for machine learning. The correlation coefficient was 0.941, which shows that the ANN method can also predict picoplankton with high accuracy.

[0079] In this way, the activated carbon injection control system, the activated carbon injection control method, and the water treatment system according to the embodiments of the present invention can inexpensively and quickly grasp the mold odor generated from the treated water without using expensive sensors, etc., and can determine the appropriate activated carbon injection rate. Furthermore, future water quality fluctuations can be predicted by prediction using a machine learning model, making it possible to stably suppress the mold odor generated from the raw water and the treated water obtained by treating the raw water for a long period of time.

[0080] Although the present invention has been described by the above embodiment, the description and drawings forming a part of this disclosure should not be understood as limiting the present invention. The present disclosure is not limited to the above embodiment, and the components can be combined and modified to be embodied without departing from the spirit of the present disclosure. [Explanation of symbols]

[0081] 1…Waterfall well 2…Flocculation and settling tank 3. Sand filter pond 4…Distribution reservoir 10. Water treatment plant 11...Water quality data measuring device 12...Weather data measuring device 100…Server 101...Control unit 102…Communications Department 103...Storage means 200... Activated carbon injection rate control device 201...Method of predicting picoplankton numbers 202...Mold odor substance concentration prediction means 203...activated carbon injection rate determination means 204... Activated carbon injection control means 300…Activated carbon injection equipment

Claims

1. A picoplankton number prediction means for predicting the number of picoplankton in the raw water by machine learning using water quality data and meteorological data of the raw water; A storage means for storing the water quality data, the meteorological data, and the predicted value of the picoplankton count; a moldy odor substance concentration prediction means for predicting a moldy odor substance concentration in the raw water by machine learning using the predicted value of the picoplankton count and the water quality data; an activated carbon injection control means for controlling an activated carbon injection rate of the activated carbon to be injected into the raw water based on the activated carbon injection rate determined based on the predicted result of the musty odor substance concentration; An activated carbon injection control system comprising:

2. The picoplankton number prediction means The activated carbon injection control system according to claim 1, further comprising: calculating a predicted value of the picoplankton count using a machine learning model constructed by a machine learning algorithm in which the water quality data, which includes at least information on the pH, electrical conductivity, turbidity, and M-alkalinity of the raw water, and the meteorological data, which includes at least information on the amount of solar radiation, are explanatory variables, and the picoplankton count is a response variable.

3. The mold odor substance concentration prediction means The activated carbon injection control system according to claim 1, further comprising: calculating a predicted value of the concentration of moldy odor substances using a machine learning model constructed by a machine learning algorithm in which the predicted value of the picoplankton count and the water quality data including at least information on the electrical conductivity, turbidity, and M-alkalinity of the raw water are used as explanatory variables, and the concentration of the moldy odor substances is used as a target variable.

4. 2. The activated carbon injection control system according to claim 1, further comprising an activated carbon injection rate prediction means for calculating a predicted value of the activated carbon injection rate using a machine learning model constructed by a machine learning algorithm in which the predicted value of the moldy odor substance concentration, water quality data including a temperature of the raw water, and the set value of the activated carbon injection rate associated with the moldy odor substance concentration are explanatory variables, and the predicted value of the activated carbon injection rate is an objective variable.

5. The activated carbon injection control system according to any one of claims 2 to 4, characterized in that the water quality data further includes data on the water level of a storage tank in which the raw water is stored, the amount of the raw water flowing into the storage tank, and the water temperature of the raw water.

6. 5. The activated carbon injection control system according to claim 2, wherein the meteorological data further includes data on outside temperature, hours of sunshine, and amount of rainfall.

7. A machine learning model is constructed by machine learning using learning data including past water quality data of the raw water and past meteorological data, and the measured values ​​of the water quality data and meteorological data of the raw water are inputted, and a predicted value of the future number of picoplankton in the raw water is outputted; The predicted value of the future picoplankton count and the measured value of the water quality data are input to a machine learning model constructed by machine learning using the past predicted value of the picoplankton count and the past water quality data, and a predicted value of the future musty odor substance concentration in the raw water is output; Controlling the activated carbon injection rate of the activated carbon injected into the raw water based on the activated carbon injection rate determined based on the future prediction result of the mold odor substance concentration.

13. A method for controlling activated carbon injection, comprising:

8. A picoplankton number prediction means for predicting the number of picoplankton in the raw water by machine learning using water quality data and meteorological data of the raw water; a moldy odor substance concentration prediction means for predicting a moldy odor substance concentration in the raw water by machine learning using the predicted value of the picoplankton count and the water quality data; An activated carbon injection means for injecting activated carbon into the raw water; an activated carbon injection control means for controlling the activated carbon injection rate of the activated carbon injected by the activated carbon injection means based on the activated carbon injection rate determined based on the predicted result of the musty odor substance concentration; A water treatment system comprising:

Images