Method and apparatus for determining the addition rate of chemicals for water treatment.

Abstract

To provide a method for accurately determining a rate of addition of chemicals for water treatment at water treatment facilities.SOLUTION: A method for determining a rate of addition of chemicals for water treatment includes: inputting predictive condition data including at least a first measured value of water quality of water to be treated and a second measured value of water quality of water added with chemicals, to a model constructed by machine learning using, as training data, data including at least past first measurement data of water quality of water to be treated, past second measurement data of water quality of water added with chemicals and an actual rate of addition of chemicals associated with the past first measurement data and the past second measurement data; and outputting the rate of addition of chemicals from the model.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a technique for determining the addition rate of chemicals used in a water treatment system.

Background Art

[0002] In a water purification plant, in the process of each water purification process, sodium hypochlorite is added to the water to be treated to control the residual chlorine concentration to a predetermined concentration (see, for example, Patent Document 1). The residual chlorine concentration in the treated water of the water purification plant is controlled so that it falls within the range of the management target values set in each water purification plant according to the use concentration of the supply destination (for example, for household use, for industrial use). For example, the residual chlorine concentration in the clear water tank, which is the final facility in the water purification plant, is measured. The residual chlorine concentration in the clear water tank is controlled so that sodium hypochlorite is added in the pipeline before flowing into the clear water tank to achieve a residual chlorine concentration of a predetermined concentration.

[0003] Patent Document 1 describes a method for setting the chlorine injection rate when injecting chlorine into the water to be treated in a water purification plant. The method is an approximate formula obtained based on past measurement data of the water temperature of the water to be treated and the residual chlorine concentration of the water to be treated. The approximate formula approximates the difference between the daytime residual chlorine concentration and the nighttime residual chlorine concentration of the water to be treated with an exponential function having the water temperature of the water to be treated as a variable. Based on the approximate formula and the measured value of the water temperature of the water to be treated, a difference calculation step for calculating the difference between the daytime residual chlorine concentration and the nighttime residual chlorine concentration of the water to be treated having the measured temperature, and an injection rate setting step for adding the difference to the nighttime residual chlorine concentration to set the daytime residual chlorine concentration are provided.

[0004] When performing such automatic control of the residual chlorine concentration, for example, it can be realized by measuring the residual chlorine concentration at the outlet of the filtration tank and performing feed-forward control to inject the insufficient amount into the residual chlorine in the clear water tank. Also, it is possible to measure the residual chlorine concentration in the clear water tank and perform feedback control on the post-chlorine injection rate with respect to the management target value.

[0005] Patent Document 2 describes a membrane separation apparatus for separating permeate from raw water, comprising: an oxidizing agent addition device for adding an oxidizing agent to water supplied to the separation membrane; a water temperature measuring device for continuously or intermittently measuring the temperature of the water supplied to the separation membrane; an oxidizing agent concentration measuring device for measuring the concentration of the oxidizing agent remaining in the water downstream of the point where the oxidizing agent was added; and a calculation / control device that continuously or intermittently calculates the optimal residual oxidizing agent concentration based on the water temperature measured by the water temperature measuring device according to the set correlation between the temperature of the water supplied to the separation membrane and the optimal residual oxidizing agent concentration, and controls the amount of oxidizing agent added by the oxidizing agent addition device so that the residual oxidizing agent concentration measured by the oxidizing agent concentration measuring device becomes the optimal residual oxidizing agent concentration.

[0006] Patent Document 3 describes a pretreatment method for a pure water production apparatus in which chlorine or a chlorine-based oxidizing agent such as hypochlorous acid is added to water to be treated, and then the water to be treated is supplied to a reverse osmosis apparatus to obtain pure water, wherein the chlorine concentration in the produced water obtained from the reverse osmosis apparatus is continuously detected and the amount of chlorine-based oxidizing agent injected into the water to be treated is feedback controlled so that the concentration reaches a predetermined value.

[0007] Patent Document 4 describes a method for disinfecting rainwater sewage, which contains organic matter, ammonia or ammonium ions, and rainwater, and is discharged into public water bodies through a rainwater sewer of a separate sewer system during rainy weather. The method includes the steps of: adding and dissolving a solid disinfectant consisting of 1-bromo-3-chloro-5,5-dimethylhydantoin in water to obtain disinfectant water; and adding the disinfectant water to the sewage for disinfection. Patent Document 4 further describes that this method can efficiently disinfect wastewater such as rainwater sewage. Furthermore, it describes that disinfection is possible even when the residual halogen concentration is 0.4 mg / L or less, which is below the LC50 value, and that by detecting the residual halogen concentration, the supply of disinfectant or disinfectant water can be reduced or shut off if the residual halogen concentration exceeds the control value, thus demonstrating environmental consideration. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] Patent No. 6466213 [Patent Document 2] Patent No. 3251145 [Patent Document 3] Japanese Patent Application Publication No. 62-30599 [Patent Document 4] Patent No. 4628132 [Overview of the project] [Problems that the invention aims to solve]

[0009] However, a challenge in automatic control of residual chlorine concentration at water treatment plants is when water quality fluctuates. When water quality fluctuates, feedforward and feedback data lack immediacy, resulting in a time lag and preventing real-time tracking, sometimes leading to significant deviations from target control values. In such cases, operators often have to make manual adjustments based on their experience, which frequently increases their workload.

[0010] Furthermore, in the water supply business, including the operation of water treatment plants, the decrease in experienced staff has made it difficult to pass on water treatment plant operation techniques. In addition, the outsourcing of water treatment plant operation and management to private companies is progressing, and these private companies need to manage chlorine injection based on the characteristics of the water treatment plant they are contracted to manage. However, manually setting the chlorine injection rate by operators is a significant burden, and there is a challenge in stabilizing the residual chlorine concentration of the treated water.

[0011] Furthermore, at sewage treatment plants, sewage undergoes a series of processes in this order: grit chambers to remove sand and other debris, solid-liquid separation to remove suspended solids (SS), activated sludge treatment, and then disinfection before being discharged into public waters such as rivers, lakes, harbors, and coastal waters. Disinfection is generally carried out using chlorine gas or chlorine-based disinfectants. This is because sewage, human waste, and industrial wastewater can contain pathogenic bacteria that can cause infectious diseases. Generally, chlorine-based disinfectants are added to reduce the number of coliform bacteria to 3,000 or less per milliliter. In some cases, ultraviolet irradiation or ozone addition is used instead of chlorine-based disinfectants, but this is limited in its application due to the enormous amount of equipment required.

[0012] However, applying technologies used in normal sewage treatment to sewage treatment during rainy weather presents the following problems. First, since ammonia and amines coexist in sewage during rainy weather, chemical reactions occur between them, causing activated chlorine to change into chloramine, and reducing the sterilization effect to less than 1 / 10. Therefore, controlling the concentration of the oxidizing agent becomes crucial.

[0013] In all methods, devices, and systems, predictions are limited to values ​​detected by past sensors, and information that was missed by those sensors could not be obtained, making it a challenge to improve prediction accuracy.

[0014] Therefore, the present invention has been made in view of the above-mentioned conventional problems, and its purpose is to provide a method and apparatus for determining the addition rate of water treatment chemicals in a water treatment facility, which can accurately predict the addition rate of water treatment chemicals and further reduce chemical costs. [Means for solving the problem]

[0015] In one embodiment, a method is provided for determining the chemical addition rate for water treatment, characterized in that prediction condition data, which includes at least a first measurement of the water quality of the water to be treated and a second measurement of the water quality of the water after the chemical has been added, is input to a model constructed by machine learning using training data that includes at least a first measurement of the water quality of the water to be treated and a second measurement of the water quality of the water after the chemical has been added, and the model outputs the chemical addition rate.

[0016] In one embodiment, the first measurement value included in the prediction condition data is the most recent first time-series measurement data including current and past measurement values ​​of the water quality of the water to be treated; the second measurement value included in the prediction condition data is the most recent second time-series measurement data including current and past measurement values ​​of the water quality after the chemical has been added; the past first measurement data included in the training data is the past first time-series measurement data of the water quality of the water to be treated; and the past second measurement data included in the training data is the past second time-series measurement data of the water quality after the chemical has been added.

[0017] In one embodiment, a method is provided for determining the addition rate of a chemical for water treatment, characterized in that the i-1 model is constructed by machine learning using the i-1 past measurement data of the water quality of the water to be treated before the addition of the i-1 chemical (i is a natural number of 2 or more), the i past measurement data of the water quality of the water after the addition of the i-1 chemical, the i past calculation data of the addition rate of the i-1 chemical, and the i-1 prediction condition data, which includes at least the i-1 measurement value of the water quality, the i past measurement value of the water quality after the addition of the i-1 chemical, and the i past model is outputting the i-1 chemical addition rate.

[0018] In one embodiment, the i-1 measurement included in the i-1 prediction condition data (where i is a natural number greater than or equal to 2) is the most recent i-1 time series measurement data, including current and past measurements of the water quality before the i-1 chemical was added; the i-1 prediction condition data and the i measurement included in the i prediction condition data are the most recent i time series measurement data, including current and past measurements of the water quality after the i-1 chemical was added; and the i+1 measurement included in the i prediction condition data is the current and past measurements of the water quality after the i chemical was added and further added. The most recent (i+1)th time-series measurement data includes past measurements, the past (i-1)th measurement data included in the (i-1)th training data is the past (i-1)th time-series measurement data of the water quality before the (i-1)th chemical was added, the past (i)th measurement data included in the (i-1)th training data and the (i)th training data is the past (i)th time-series measurement data of the water quality after the (i)th chemical was added, and the past (i+1)th measurement data included in the (i)th training data is the past (i+1)th time-series measurement data of the water quality after the (i)th chemical was added and then further added.

[0019] In one embodiment, a chemical addition rate determination device for determining the addition rate of chemicals for water treatment is provided, characterized in that it inputs prediction condition data, which includes at least a model constructed by machine learning, a first measurement of the water quality of the water to be treated, and a second measurement of the water quality after the chemicals have been added, into the model, and has a chemical addition rate output means that outputs the chemical addition rate from the model.

[0020] In one aspect, a measurement value acquisition means for obtaining a first i model (where i is a natural number of 2 or more) constructed by machine learning, a first i - 1 measurement value of the water quality of the water to be treated before the addition of the first i - 1 chemical, a first i measurement value of the water quality of the water after the addition of the first i - 1 chemical, and a first i + 1 measurement value of the water quality of the water after the addition of the first i chemical and before the further addition of the first i chemical after the addition of the first i - 1 chemical, inputs first i prediction condition data including at least the first i measurement value and the first i + 1 measurement value into the first i model, and outputs the addition rate of the first i chemical from the first i model; a first i - 1 model constructed by machine learning, inputs first i - 1 prediction condition data including the first i - 1 measurement value, the first i measurement value, and the addition rate of the first i chemical into the first i - 1 model, and outputs the addition rate of the first i - 1 chemical from the first i - 1 model. There is provided an apparatus for determining the addition rate of a chemical for water treatment, characterized by having a first i - 1 chemical addition rate output means.

[0021] In one aspect, there is provided a method for determining the addition rate of a chemical for water treatment, which inputs prediction condition data including at least recent time - series measurement data including the current measurement value and past measurement values of the water quality of the water to be treated into a model constructed by machine learning, and outputs the addition rate of the chemical from the model based on the control target value of the water quality of the water after the addition of the chemical.

[0022] In one aspect, the model is a trained model constructed by machine learning using training data, and the training data includes at least past time - series measurement data of water quality and the actual addition rate of the chemical associated with the past time - series measurement data. In one aspect, the method further includes adjusting the addition rate output from the model based on the measurement value of the water quality of the water before the addition of the chemical.

[0023] In one aspect, provided is a method for determining an addition rate of a chemical for water treatment, which includes inputting prediction condition data including at least a first measurement value of the water quality of water to be treated and a second measurement value of the water quality of the water after the chemical is added into a model constructed by machine learning, and outputting the addition rate of the chemical from the model.

[0024] In one aspect, the method further includes adjusting the addition rate output from the model based on a measurement value of the water quality of the water after the chemical is added. In one aspect, the model is a trained model constructed by machine learning using training data, and the training data includes at least past first measurement data of the water quality of water to be treated, past second measurement data of the water quality of the water after the chemical is added, and the actual addition rate of the chemical associated with the past first measurement data and the past second measurement data. In one aspect, the first measurement value included in the prediction condition data is recent first time-series measurement data including current and past measurement values of the water quality of the water to be treated, the second measurement value included in the prediction condition data is recent second time-series measurement data including current and past measurement values of the water quality after the chemical is added, the past first measurement data included in the training data is past first time-series measurement data of the water quality of the water to be treated, and the past second measurement data included in the training data is past second time-series measurement data of the water quality after the chemical is added.

[0025] In one embodiment, a method is provided for determining the addition rate of chemicals for water treatment, characterized by obtaining a first measurement of the water quality of the water to be treated, a second measurement of the water quality of the water after a first chemical has been added, and a third measurement of the water quality of the water after a second chemical has been further added after the first chemical has been added; inputting second prediction condition data, which includes at least the second and third measurement values, into a second model constructed by machine learning; outputting the addition rate of the second chemical from the second model; and inputting first prediction condition data, which includes at least the first measurement value, the second measurement value, and the addition rate of the second chemical, into a first model constructed by machine learning; and outputting the addition rate of the first chemical from the first model.

[0026] In one embodiment, the method further includes adjusting the addition rate of the first chemical based on a measurement of the water quality of the water to which the first chemical has been added. In one embodiment, the method further includes adjusting the addition rate of the second chemical based on a measurement of the water quality of the water to which the second chemical has been added. In one embodiment, the first model is a trained model constructed by machine learning using first training data, the first training data includes at least past first measurement data of the water quality of the water to be treated, past second measurement data of the water quality after the first chemical has been added, past calculation data of the addition rate of the second chemical, and the actual addition rate of the first chemical associated with the past first measurement data, the past second measurement data, and the past calculation data of the addition rate of the second chemical. In one embodiment, the second model is a trained model constructed by machine learning using second training data, the second training data comprising at least the past second measurement data, past third measurement data of the water quality of the water to which the second chemical has been further added after the first chemical has been added, and the actual addition rates of the second chemical associated with the past second measurement data and the past third measurement data. In one embodiment, the first measurement value included in the first prediction condition data is the most recent first time-series measurement data including current and past measurements of the water quality before the first chemical is added; the second measurement value included in the first and second prediction condition data is the most recent second time-series measurement data including current and past measurements of the water quality after the first chemical is added; the third measurement value included in the second prediction condition data is the most recent third time-series measurement data including current and past measurements of the water quality after the second chemical is added after the first chemical has been added; the past first measurement value included in the first training data is the past first time-series measurement data of the water quality before the first chemical was added; the past second measurement value included in the first and second training data is the past second time-series measurement data of the water quality after the first chemical has been added; and the past third measurement value included in the second training data is the past third time-series measurement data of the water quality after the second chemical is added after the first chemical has been added.

[0027] In one embodiment, a chemical addition rate determination device for determining the addition rate of chemicals for water treatment is provided, characterized in that it comprises a model constructed by machine learning, and is configured to input prediction condition data, which includes at least the most recent time-series measurement data including current and past measurements of the water quality of the water to be treated, into the model, and to output the addition rate of chemicals from the model.

[0028] In one embodiment, the model is a trained model constructed by machine learning using training data, the training data including at least past time-series measurement data of water quality and the actual addition rates of chemicals associated with the past time-series measurement data. In one embodiment, the chemical addition rate determination device is further configured to adjust the addition rate output from the model based on a measurement of the water quality of the water to which the chemical has been added.

[0029] In one embodiment, a chemical addition rate determination device for determining the addition rate of chemicals for water treatment is provided, which includes a model constructed by machine learning, and is configured to input prediction condition data including at least a first measurement of the water quality of the water to be treated and a second measurement of the water quality after the chemicals have been added to the model, and to output the addition rate of the chemicals from the model.

[0030] In one embodiment, the chemical addition rate determination device is further configured to adjust the addition rate output from the model based on a measurement of the water quality of the water to which the chemical has been added. In one embodiment, the model is a trained model constructed by machine learning using training data, the training data including at least past first measurement data of the water quality of the water to be treated, past second measurement data of the water quality after the chemical has been added, and the actual addition rates of the chemical associated with the past first measurement data and the past second measurement data. In one embodiment, the first measurement value included in the prediction condition data is the most recent first time series measurement data including current and past measurement values ​​of the water quality of the water to be treated; the second measurement value included in the prediction condition data is the most recent second time series measurement data including current and past measurement values ​​of the water quality after the chemical has been added; the past first measurement data included in the training data is the past first time series measurement data of the water quality of the water to be treated; and the past second measurement data included in the training data is the past second time series measurement data of the water quality after the chemical has been added.

[0031] In one embodiment, a chemical addition rate determination device for determining the addition rate of chemicals for water treatment is provided, comprising a first model constructed by machine learning and a second model constructed by machine learning, wherein the device acquires a first measurement value of the water quality of the water to be treated, a second measurement value of the water quality after a first chemical has been added, and a third measurement value of the water quality after a second chemical has been further added after the first chemical has been added, inputs second prediction condition data including at least the second measurement value and the third measurement value into the second model, outputs the addition rate of the second chemical from the second model, and inputs first prediction condition data including at least the first measurement value, the second measurement value and the addition rate of the second chemical into the first model, outputs the addition rate of the first chemical from the first model.

[0032] In one embodiment, the chemical addition rate determination device is further configured to adjust the addition rate of the first chemical based on a measurement of the water quality of the water to which the first chemical has been added. In one embodiment, the chemical addition rate determination device is further configured to adjust the addition rate of the second chemical based on a measurement of the water quality of the water to which the second chemical has been added. In one embodiment, the first model is a trained model constructed by machine learning using first training data, the first training data includes at least past first measurement data of the water quality of the water to be treated, past second measurement data of the water quality after the first chemical has been added, past calculation data of the addition rate of the second chemical, and the actual addition rate of the first chemical associated with the past first measurement data, the past second measurement data, and the past calculation data of the addition rate of the second chemical. In one embodiment, the second model is a trained model constructed by machine learning using second training data, the second training data comprising at least the past second measurement data, past third measurement data of the water quality of the water to which the second chemical has been further added after the first chemical has been added, and the actual addition rates of the second chemical associated with the past second measurement data and the past third measurement data. In one embodiment, the first measurement value included in the first prediction condition data is the most recent first time-series measurement data including current and past measurements of the water quality before the first chemical is added; the second measurement value included in the first and second prediction condition data is the most recent second time-series measurement data including current and past measurements of the water quality after the first chemical is added; the third measurement value included in the second prediction condition data is the most recent third time-series measurement data including current and past measurements of the water quality after the second chemical is added after the first chemical has been added; the past first measurement value included in the first training data is the past first time-series measurement data of the water quality before the first chemical was added; the past second measurement value included in the first and second training data is the past second time-series measurement data of the water quality after the first chemical has been added; and the past third measurement value included in the second training data is the past third time-series measurement data of the water quality after the second chemical is added after the first chemical has been added. [Effects of the Invention]

[0033] According to the present invention, the drug addition rate can be accurately determined using time-series measurement data and a model constructed by machine learning. Furthermore, according to the present invention, since both a first measurement of water quality before the addition of the chemical and a second measurement of water quality after the addition of the chemical are input to the model, both feedforward control and feedback control are performed within the model, and as a result, the chemical addition rate can be determined with high accuracy. [Brief explanation of the drawing]

[0034] [Figure 1] This is a schematic diagram of one embodiment of a water treatment system in which chemicals are added for water treatment. [Figure 2] Figure 1 is a flowchart illustrating one embodiment of the operation of the water treatment system shown. [Figure 3]This is a schematic diagram of another embodiment of the water treatment system. [Figure 4] Figure 3 is a flowchart illustrating one embodiment of the operation of the water treatment system shown. [Figure 5] This flowchart illustrates one embodiment of the operation of a water treatment system, combining the embodiments described with reference to Figures 1 and 2 with the embodiments described with reference to Figures 3 and 4. [Figure 6] This is a schematic diagram of yet another embodiment of the water treatment system. [Figure 7] This is the first half of a flowchart illustrating one embodiment of the operation of the water treatment system shown in Figure 6. [Figure 8] This is the second half of the flowchart illustrating one embodiment of the operation of the water treatment system shown in Figure 6. [Figure 9] This is the first half of a flowchart illustrating one embodiment of the operation of a water treatment system, combining the embodiments described with reference to Figures 1 and 2 with the embodiments described with reference to Figures 6 to 8. [Figure 10] This is the latter half of a flowchart illustrating one embodiment of the operation of a water treatment system, combining the embodiments described with reference to Figures 1 and 2 with the embodiments described with reference to Figures 6 to 8. [Figure 11] This graph shows the operating results of the embodiment described with reference to the water treatment system shown in Figure 3 and the flowchart shown in Figure 5. [Modes for carrying out the invention]

[0035] Embodiments of the present invention will be described below with reference to the drawings. Figure 1 is a schematic diagram of one embodiment of a water treatment system to which chemicals are added for water treatment. The water treatment system of this embodiment is applicable to facilities such as water purification and sewage treatment. Examples of chemicals used in water treatment include oxidizing agents such as sodium hypochlorite in the case of water purification.

[0036] As shown in Figure 1, the water treatment system of this embodiment includes a water quality meter 2 for measuring the water quality of the water to be treated, a pump 3 as a chemical supply device for adding chemicals to the water, and a chemical addition rate determination device 10 for determining the rate at which chemicals are added to the water. The water quality meter 2 and the pump 3 are connected to the chemical addition rate determination device 10. The water quality measurements obtained by the water quality meter 2 are sent to the chemical addition rate determination device 10. The operation of the pump 3 as a chemical supply device (i.e., the amount of chemicals added to the water) is controlled by the chemical addition rate determination device 10. The chemical addition rate is the amount of chemical added relative to the water flow rate or unit volume of water. In one embodiment, the chemical addition rate determination device 10 determines the chemical addition rate to the water based on a control target value for the water quality after the chemicals have been added.

[0037] In this embodiment, pump 3 is used as the chemical supply device, but it is not limited to a pump as long as it can supply chemicals to water. For example, a screw feeder may be used as the chemical supply device.

[0038] The location where the water quality meter 2 measures water quality (hereinafter referred to as the water quality measurement location) is downstream in the direction of water flow from the location where the chemical is added to the water by the pump 3 (hereinafter referred to as the chemical addition location). In other words, the water quality measurement location and the chemical addition location are aligned along the direction of water flow. Water treatment elements such as a filter or mixing tank may be located downstream of the chemical addition location.

[0039] The drug addition rate determination device 10 includes a storage device 10a that stores a program for building a model by performing machine learning, as described later, and a processing unit 10b that performs calculations according to the instructions included in the program. The storage device 10a includes storage devices such as hard disk drives (HDDs) and solid-state drives (SSDs). Examples of processing units 10b include CPUs (central processing units) and GPUs (graphics processing units). However, the specific configuration of the drug addition rate determination device 10 is not limited to these examples.

[0040] The chemical addition rate determination device 10 consists of at least one computer. This at least one computer may be one server or multiple servers. The chemical addition rate determination device 10 may be an edge server connected to the water quality meter 2 and pump 3 by a communication line, or it may be a cloud server or fog server connected to the water quality meter 2 and pump 3 by a communication network such as the Internet or a local area network.

[0041] The drug addition rate determination device 10 may consist of multiple servers connected by a communication network such as the Internet or a local area network. For example, the drug addition rate determination device 10 may consist of a combination of edge servers and cloud servers. The storage device 10a and the processing unit 10b may be located in multiple computers installed in separate locations.

[0042] Next, the functions of the chemical addition rate determination device 10 will be described. The chemical addition rate determination device 10 is configured to calculate the addition rate of chemicals to be added to water by the pump 3 using a model constructed by machine learning (i.e., a trained model). The model is stored in the storage device 10a. The calculation unit 10b of the chemical addition rate determination device 10 is configured to input prediction condition data into the model constructed by machine learning and output the chemical addition rate from the model by performing calculations according to the model's algorithm. The prediction condition data includes at least the most recent time-series measurement data, including current and past water quality measurements obtained by the water quality measuring instrument 2. The chemical addition rate determination device 10 has the function of a chemical addition rate output means that inputs prediction condition data into a model constructed by machine learning and outputs the chemical addition rate from the model, and also has the function of a measurement value acquisition means that acquires water quality measurements.

[0043] The most recent time-series measurement data includes recent water quality measurements for a specified number of days. For example, it includes water quality measurements for the most recent 5 days. The specified number of days can be changed in the settings. Since the most recent time-series measurement data includes not only current water quality measurements but also recently acquired historical measurements, the trend in water quality can be reflected in the chemical addition rate.

[0044] The specific physical quantities of water quality vary depending on the type of water being treated, but examples include water pH, water turbidity, residual concentration of chemicals in the water, and chemical oxygen demand (COD).

[0045] The model is created using machine learning with training data. Explanatory variables are input to the model, and the model outputs the target variable. Explanatory variables: Variables used to derive the variable we want to predict (dependent variable). Target variable: The variable to be predicted. In this embodiment, the target variable is the rate at which chemicals are added to water by pump 3.

[0046] The explanatory variables included in the training data include at least historical time-series measurement data of water quality. This historical time-series measurement data of water quality is acquired by the water quality meter 2 and stored in the storage device 10a. The dependent variable included in the training data is the actual addition rate of the chemicals associated with the above historical time-series measurement data as explanatory variables.

[0047] The chemical addition rate determination device 10 creates a model according to a machine learning algorithm using past time-series measurement data of water quality and training data including the actual addition rate of chemicals. The model constructed by machine learning is stored in the storage device 10a as a trained model.

[0048] Examples of machine learning algorithms include SVR (Support Vector Regression), PLS (Partial Least Squares), deep learning, random forest, and decision tree. In this embodiment, deep learning is employed for machine learning. In particular, in this embodiment, the model consists of a neural network comprising long short-term memory (LSTM). However, the model used in the present invention is not limited to an LSTM type model, as long as time-series data can be used.

[0049] The operation of the water treatment system shown in Figure 1 will be explained with reference to the flowchart in Figure 2. In Step 1, the water quality meter 2 measures the water quality at predetermined time intervals and sends the measured values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the water quality measured values ​​sent from the water quality meter 2 in the storage device 10a. Each time the chemical addition rate determination device 10 receives water quality measured values, it stores those values ​​in the storage device 10a, associating them with time. Therefore, the storage device 10a stores time-series data of water quality measured values ​​(i.e., time-series water quality measurement data).

[0050] In step 2, the chemical addition rate determination device 10 reads the most recent time-series measurement data of water quality from the storage device 10a and inputs it into the model. In step 3, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the model and outputs the drug addition rate from the model. In step 4, the chemical addition rate determination device 10 controls the operation of the pump 3 so that the chemical is added to the water at the addition rate output from the model. As a result, the chemical is added to the water and the water is treated with the chemical.

[0051] As described above, the model in this embodiment is a Long Short-Term Memory (LSTM) model. The Long Short-Term Memory (LSTM) model is a type of model that can handle time-series data as an extension of the Recurrent Neural Network (RNN). The Long Short-Term Memory (LSTM) model is realized by replacing the intermediate layer units of the Recurrent Neural Network (RNN) with blocks that have memory called LSTM blocks and three gates (input gate, output gate, and forget gate).

[0052] The most significant feature of the Long-Short-Term Memory (LSTM) model is its ability to learn long-term dependencies, which conventional RNNs could not learn. While regular RNNs can handle short-term time series data of a few dozen steps, they could not learn long-term time series data of 1000 steps or more. The Long-Short-Term Memory (LSTM) model can produce appropriate output even for such long-term time series data.

[0053] This embodiment utilizes the ability of a long-short-term memory model (LSTM) to learn and store long-term time-series data to predict the chemical addition rate in a water treatment process. The chemical addition rate determination device 10 of this embodiment, which is equipped with a long-short-term memory model (LSTM), can use long-term time-series data to calculate an appropriate chemical addition rate based on recent water quality trends.

[0054] In one embodiment, the prediction condition data input to the model may further include, in addition to the most recent time-series measurement data of water quality, the most recent time-series measurement data of operating conditions, which includes at least one of water flow rate, water temperature, and meteorological data. Meteorological data may include rainfall, temperature, etc. The training data used for machine learning of the model in this embodiment includes at least historical time-series measurement data of water quality, historical time-series data of operating conditions, and the actual addition rates of chemicals associated with the historical time-series measurement data of water quality and historical time-series data of operating conditions. During the operation of the water treatment system, the chemical addition rate determination device 10 inputs both the most recent time-series measurement data of water quality and the most recent time-series data of operating conditions into the model and outputs the chemical addition rate from the model.

[0055] Next, schematic diagrams of other embodiments of the water treatment system will be described with reference to Figure 3. The configuration and operation of these embodiments, which are not specifically described, are the same as those of the embodiments described with reference to Figures 1 and 2, so redundant explanations will be omitted. As shown in Figure 3, the water treatment system includes a first water quality meter 2 for measuring the water quality of the water to be treated, a pump 3 as a chemical supply device for adding chemicals to the water, a second water quality meter 5 for measuring the water quality of the water to which chemicals have been added, and a chemical addition rate determination device 10 for determining the rate at which chemicals are added to the water.

[0056] The first water quality meter 2, the second water quality meter 5, and the pump 3 are connected to the chemical addition rate determination device 10. The first water quality measurement obtained by the first water quality meter 2 and the second water quality measurement obtained by the second water quality meter 5 are sent to the chemical addition rate determination device 10. In addition, the operation of the pump 3 (i.e., the amount of chemical added to the water) is controlled by the chemical addition rate determination device 10.

[0057] The location where the second water quality meter 5 measures water quality (hereinafter referred to as the second water quality measurement location) is downstream in the direction of water flow from the location where the first water quality meter 2 measures water quality (hereinafter referred to as the first water quality measurement location). The location where chemicals are added to the water by the pump 3 (hereinafter referred to as the chemical addition location) is between the first water quality measurement location and the second water quality measurement location. In other words, the first water quality measurement location, the chemical addition location, and the second water quality measurement location are aligned along the direction of water flow. Therefore, the first water quality meter 2 measures the water quality before chemicals are added from the pump 3, and the second water quality meter 5 measures the water quality after chemicals are added from the pump 3. A water treatment element such as a filter or mixing tank may be placed between the chemical addition location and the second water quality measurement location.

[0058] The calculation unit 10b of the chemical addition rate determination device 10 is configured to input prediction condition data into a model constructed by machine learning and output the chemical addition rate from the model. The prediction condition data includes at least a first measurement of the water quality of the water to be treated and a second measurement of the water quality of the water after the chemical has been added.

[0059] The model in this embodiment is also a trained model constructed by machine learning using training data. The training data includes at least past first measurement data of the water quality of the water to be treated, past second measurement data of the water quality after chemicals have been added, and the actual addition rates of the chemicals associated with the past first and past second measurement data.

[0060] The explanatory variables included in the training data include at least the past first measurement data of the water quality of the water to be treated and the past second measurement data of the water quality after the chemical has been added. The past first measurement data of water quality is acquired by the first water quality meter 2 and stored in the storage device 10a. The past second measurement data of water quality is acquired by the second water quality meter 5 and stored in the storage device 10a. The dependent variable included in the training data is the actual addition rate of the chemical associated with the above past first measurement data and the above past second measurement data, which are used as explanatory variables.

[0061] The chemical addition rate determination device 10 creates a model according to a machine learning algorithm using past first measurement data of water quality, past second measurement data of water quality, and training data including the actual addition rate of chemicals. The model constructed by machine learning is stored in the storage device 10a as a trained model.

[0062] Examples of machine learning algorithms include SVR (Support Vector Regression), PLS (Partial Least Squares), deep learning, random forest, or decision tree. In this embodiment, deep learning is employed for machine learning. In the embodiment described with reference to Figures 1 and 2, the model is a long-short-term memory (LSTM) model, but the model in this embodiment does not have to be an LSTM type model.

[0063] According to this embodiment, both a first measurement of water quality before chemicals are added from pump 3 and a second measurement of water quality after chemicals are added from pump 3 are input to the model. As a result, both feedforward control and feedback control are performed, and the chemical addition rate can be determined with high accuracy.

[0064] In one embodiment, the chemical addition rate determination device 10 may adjust the addition rate output from the model based on the post-addition measurement value, which is a measurement of the water quality of the water after the chemical has been added. In one example, the post-addition measurement value is the second measurement of water quality at the second water quality measurement location, which is input into the model for calculating the chemical addition rate. More specifically, the chemical addition rate determination device 10 calculates the difference between the post-addition measurement value and the target water quality value and adjusts the addition rate in a direction that reduces this difference. For example, if the water quality is the residual concentration of chemicals in the water, and the post-addition measurement value (measured residual concentration) is higher than the target water quality value (target residual concentration), the chemical addition rate determination device 10 will decrease the addition rate output from the model. Through such operation, the water quality of the water after the chemical has been added can be brought closer to the target water quality value.

[0065] Figure 4 is a flowchart illustrating one embodiment of the operation of the water treatment system shown in Figure 3. In step 1, the first water quality measuring instrument 2 measures the water quality at the first water quality measuring position upstream of the chemical addition position and obtains the first measurement value, which is then sent to the chemical addition rate determination device 10. At the chemical addition position, the pump 3 adds the chemical to the water at a preset addition rate. In step 2, the second water quality meter 5 measures the water quality at a second water quality measurement location downstream of the chemical addition location to obtain a second measurement value, and sends the second measurement value to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the first and second measurement values ​​of water quality sent from the first water quality meter 2 and the second water quality meter 5 in the storage device 10a. Steps 1 and 2 described above may be performed simultaneously, or in a different order than the example shown in Figure 4.

[0066] In step 3, the chemical addition rate determination device 10 reads the first and second water quality measurements from the storage device 10a and inputs them into the model. In step 4, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the model and outputs the drug addition rate from the model. In step 5, the chemical addition rate determination device 10 controls the operation of the pump 3 so that the chemical is added to the water at the addition rate output from the model.

[0067] In step 6, the chemical addition rate determination device 10 adjusts the addition rate output from the model based on the post-addition measurement (for example, the second measurement obtained in step 2 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the post-addition measurement and the target water quality value, and adjusts the addition rate in a direction that reduces this difference. In step 7, the chemical addition rate determination device 10 controls the operation of the pump 3 so that the chemical is added to the water at the adjusted addition rate.

[0068] If the difference between the measured value after addition and the target water quality value is small, that is, if the accuracy of the chemical addition rate output from the model is high, steps 6 and 7 above may be omitted.

[0069] In one embodiment, the prediction condition data input to the model may further include current operating condition data, including at least one of the following: water flow rate, water temperature, and meteorological data, in addition to the first and second measured values ​​of water quality. Meteorological data may include rainfall, temperature, etc. The training data used for machine learning of the model in this embodiment includes at least the past first measured water quality data obtained by the first water quality meter 2, the past second measured water quality data obtained by the second water quality meter 5, the past operating condition data, and the actual chemical addition rate. During the operation of the water treatment system, the chemical addition rate determination device 10 inputs the first measured value of the water quality of the water to be treated, the second measured value of the water quality of the water to which the chemical has been added, and the current operating condition data into the model, and outputs the chemical addition rate from the model.

[0070] The embodiments described with reference to Figures 1 and 2 may be combined with the embodiments described with reference to Figures 3 and 4. Hereinafter, embodiments combining time-series measurement data, feedforward control, and feedback control will be described.

[0071] In this embodiment, the prediction condition data input to the model includes at least the most recent first time-series measurement data, which includes current and past measurements of the water quality of the water to be treated, and the most recent second time-series measurement data, which includes current and past measurements of the water quality after chemicals have been added from pump 3. The most recent first time-series measurement data is acquired by the first water quality meter 2 shown in Figure 3, and the most recent second time-series measurement data is acquired by the second water quality meter 5 shown in Figure 3. Each of the most recent first time-series measurement data and the most recent second time-series measurement data includes the most recent measurements of water quality for a predetermined number of days.

[0072] The first water quality measuring instrument 2 and the second water quality measuring instrument 5 measure water quality at predetermined time intervals and send the first and second measured values ​​of water quality to the chemical addition rate determination device 10. Each time the chemical addition rate determination device 10 receives the first and second measured values ​​of water quality, it stores them in the storage device 10a, associating them with time. Therefore, the storage device 10a stores time-series data of the first measured value of water quality (i.e., first time-series measurement data of water quality) and time-series data of the second measured value of water quality (i.e., second time-series measurement data of water quality).

[0073] The model used in this embodiment is a trained model constructed by machine learning using the training data described below. The training data includes past first time series measurement data of the water quality of the water to be treated, past second time series measurement data of the water quality after chemicals are added from pump 3, and the actual addition rates of chemicals associated with the past first time series measurement data and past second time series measurement data. The past first time series measurement data is acquired by the first water quality meter 2 shown in Figure 3, and the past second time series measurement data is acquired by the second water quality meter 5 shown in Figure 3.

[0074] The chemical addition rate determination device 10 creates a model according to a machine learning algorithm using past first time-series measurement data of water quality, past second time-series measurement data of water quality, and training data including the actual addition rate of chemicals. The model constructed by machine learning is stored in the storage device 10a as a trained model. In this embodiment, the model consists of a neural network comprising long short-term memory (LSTM). However, the model used in the present invention is not limited to an LSTM type model, as long as time-series data is available.

[0075] Figure 5 is a flowchart illustrating one embodiment of the operation of a water treatment system, combining the embodiments described with reference to Figures 1 and 2 with the embodiments described with reference to Figures 3 and 4. In Step 1, the first water quality meter 2 measures the water quality at a first water quality measurement location upstream of the chemical addition location at predetermined time intervals to obtain multiple first measurement values, and sends these first measurement values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the first measurement values ​​of water quality sent from the first water quality meter 2 in the storage device 10a. Each time the chemical addition rate determination device 10 receives a first measurement value of water quality, it stores that first measurement value in the storage device 10a in relation to time. Therefore, the storage device 10a stores time-series data of the first measurement values ​​of water quality (i.e., first time-series measurement data of water quality). At the chemical addition location, the pump 3 adds chemicals to the water at a predetermined addition rate.

[0076] In step 2, the second water quality meter 5 measures the water quality at a second water quality measurement location downstream of the chemical addition location at predetermined time intervals to obtain multiple second measurement values, and sends these second measurement values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the second measurement values ​​of water quality sent from the second water quality meter 5 in the storage device 10a. Each time the chemical addition rate determination device 10 receives a second measurement value of water quality, it stores that second measurement value in the storage device 10a in relation to time. Therefore, the storage device 10a stores time-series data of the second measurement values ​​of water quality (i.e., second time-series measurement data of water quality). Steps 1 and 2 described above may be performed simultaneously, or in a different order than the example shown in Figure 5.

[0077] In step 3, the chemical addition rate determination device 10 reads the most recent first time-series measurement data and the most recent second time-series measurement data of water quality from the storage device 10a and inputs them into the model. In step 4, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the model and outputs the drug addition rate from the model. In step 5, the chemical addition rate determination device 10 controls the operation of the pump 3 so that the chemical is added to the water at the addition rate output from the model.

[0078] In step 6, the chemical addition rate determination device 10 adjusts the addition rate output from the model based on the post-addition measurement (for example, the most recent of the multiple second measurement values ​​obtained in step 2 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the post-addition measurement and the target water quality value, and adjusts the addition rate in a direction that reduces this difference. In step 7, the chemical addition rate determination device 10 controls the operation of the pump 3 so that the chemical is added to the water at the adjusted addition rate.

[0079] If the difference between the measured value after addition and the target water quality value is small, that is, if the accuracy of the chemical addition rate output from the model is high, steps 6 and 7 above may be omitted.

[0080] The embodiment shown in Figure 5 allows for more precise determination of the drug addition rate through a combination of time-series measurement data, feedforward control, and feedback control.

[0081] Next, a schematic diagram of yet another embodiment of the water treatment system will be described with reference to Figure 6. The configuration and operation of this embodiment, which are not specifically described, are the same as those of the embodiment described with reference to Figure 3, so a redundant explanation will be omitted. As shown in Figure 6, the water treatment system includes a first water quality meter 2 for measuring the water quality of the water to be treated, a first pump 3 as a first chemical supply device for adding a first chemical to the water, a second water quality meter 5 for measuring the water quality of the water to which the first chemical has been added, a second pump 7 as a second chemical supply device for further adding a second chemical to the water to which the first chemical has been added, a third water quality meter 6 for measuring the water quality of the water to which the second chemical has been added, and a chemical addition rate determination device 10 for determining the addition rate of the first chemical and the addition rate of the second chemical to the water. The chemical addition rate determination device 10 may consist of a single computer or multiple computers connected by a communication network.

[0082] The first and second chemicals may be the same type of chemical or different types of chemicals. For example, both the first and second chemicals may be oxidizing agents such as sodium hypochlorite. In this embodiment, pumps 3 and 7 are used as the first and second chemical supply devices, but the system is not limited to pumps as long as it can supply the first and second chemicals to water. For example, screw feeders may be used as the first and second chemical supply devices.

[0083] The first water quality meter 2, the second water quality meter 5, the third water quality meter 6, the first pump 3, and the second pump 7 are connected to the chemical addition rate determination device 10. The first water quality measurement obtained by the first water quality meter 2, the second water quality measurement obtained by the second water quality meter 5, and the third water quality measurement obtained by the third water quality meter 6 are sent to the chemical addition rate determination device 10. In addition, the operation of the first pump 3 and the second pump 7 (i.e., the addition rate of the first chemical and the addition rate of the second chemical to the water) is controlled by the chemical addition rate determination device 10.

[0084] The location where the second water quality meter 5 measures water quality (hereinafter referred to as the second water quality measurement location) is downstream in the direction of water flow from the location where the first water quality meter 2 measures water quality (hereinafter referred to as the first water quality measurement location). The location where the first chemical is added to the water by the first pump 3 (hereinafter referred to as the first chemical addition location) is between the first water quality measurement location and the second water quality measurement location. The location where the third water quality meter 6 measures water quality (hereinafter referred to as the third water quality measurement location) is downstream in the direction of water flow from the second water quality measurement location. The location where the second chemical is added to the water by the second pump 7 (hereinafter referred to as the second chemical addition location) is between the second water quality measurement location and the third water quality measurement location.

[0085] In other words, the first water quality measurement position, the first chemical addition position, the second water quality measurement position, the second chemical addition position, and the third water quality measurement position are arranged along the direction of water flow. The first water quality meter 2 measures the water quality before the first chemical is added, and the second water quality meter 5 measures the water quality after the first chemical has been added but before the second chemical is added. The third water quality meter 6 measures the water quality after the second chemical has been added after the first chemical has been added.

[0086] A water treatment element such as a filter or mixing tank may be placed between the first chemical addition location and the second water quality measurement location. Similarly, a water treatment element such as a filter or mixing tank may be placed between the second chemical addition location and the third water quality measurement location.

[0087] The chemical addition rate determination device 10 has a first model for determining the addition rate of a first chemical to be added to water by a first pump 3, and a second model for determining the addition rate of a second chemical to be added to water by a second pump 7. The first and second models are trained models constructed by machine learning and are stored in the memory device 10a.

[0088] The chemical addition rate determination device 10 is configured to input first prediction condition data into a first model and output the addition rate of the first chemical from the first model. The first prediction condition data includes at least a first measurement of water quality before adding the first chemical, a second measurement of water quality after adding the first chemical but before adding the second chemical, and the addition rate of the second chemical output from the second model.

[0089] Furthermore, the chemical addition rate determination device 10 is configured to input second prediction condition data into the second model and output the addition rate of the second chemical from the second model. The second prediction condition data includes at least a second measurement of water quality and a third measurement of water quality of water to which the first and second chemicals have been added. This addition rate of the second chemical is input into the first model as described above. In other words, the calculation results of the second model are fed back into the calculation of the addition rate of the first chemical in the first model.

[0090] The first model is a trained model constructed by machine learning using the first training data. The first training data includes at least past first measurement data of the water quality of the water to be treated, past second measurement data of the water quality after the first chemical has been added but before the second chemical has been added, past calculation data of the addition rate of the second chemical output from the second model, and the actual addition rate of the first chemical associated with the past first measurement data, the past second measurement data, and the past calculation data of the addition rate of the second chemical.

[0091] The explanatory variables included in the first training data include at least past first measurement data of water quality, past second measurement data of water quality, and past calculation data of the second chemical addition rate. Past first measurement data of water quality is acquired by the first water quality meter 2 and stored in the storage device 10a. Past second measurement data of water quality is acquired by the second water quality meter 5 and stored in the storage device 10a. Past calculation data of the second chemical addition rate is output from the second model and stored in the storage device 10a. The dependent variable included in the first training data is the actual addition rate of the first chemical associated with the above past first measurement data, past second measurement data, and past calculation data of the second chemical addition rate.

[0092] The drug addition rate determination device 10 uses the first training data described above to create a first model according to a machine learning algorithm. The first model constructed by machine learning is stored in the storage device 10a as a trained model.

[0093] The second model is a trained model constructed by machine learning using the second training data. The second training data includes at least past second measurement data of water quality after the first chemical has been added but before the second chemical has been added, past third measurement data of water quality after the second chemical has been added after the first chemical has been added, and the actual addition rate of the second chemical associated with the past second measurement data and the past third measurement data.

[0094] The explanatory variables included in the second training data include at least the past second measurement data and the past third measurement data of water quality. The past second measurement data of water quality is acquired by the second water quality meter 5 and stored in the storage device 10a. The past third measurement data of water quality is acquired by the third water quality meter 6 and stored in the storage device 10a. The dependent variable included in the second training data is the actual addition rate of the second chemical associated with the above past second measurement data and the above past third measurement data.

[0095] The drug addition rate determination device 10 uses the second training data described above to create a second model according to a machine learning algorithm. The second model constructed by machine learning is stored in the memory device 10a as a trained model.

[0096] In the operation of the water treatment system of this embodiment, the chemical addition rate determination device 10 acquires a first measurement of the water quality before the first chemical is added by the first pump 3, a second measurement of the water quality after the first chemical is added by the first pump 3, and a third measurement of the water quality after the second chemical is further added by the second pump 7. The chemical addition rate determination device 10 inputs second prediction condition data, which includes at least the second and third measurement values, into the second model, and outputs the addition rate of the second chemical from the second model. Furthermore, the chemical addition rate determination device 10 inputs first prediction condition data, which includes at least the first measurement value, the second measurement value, and the addition rate of the second chemical output from the second model, into the first model, and outputs the addition rate of the first chemical from the first model.

[0097] In one embodiment, the chemical addition rate determination device 10 may adjust the addition rate output from the first model based on a first post-addition measurement value, which is a measurement of water quality at a second water quality measurement location after the first chemical has been added. In one example, the first post-addition measurement value is the second measurement of water quality input into the first model for calculating the addition rate of the first chemical. More specifically, the chemical addition rate determination device 10 calculates the difference between the first post-addition measurement value and the first target water quality value at the second water quality measurement location, and adjusts the addition rate of the first chemical in a direction that reduces this difference. Similarly, the chemical addition rate determination device 10 may adjust the addition rate output from the second model based on a second post-addition measurement value, which is a measurement of water quality at a third water quality measurement location after the second chemical has been added. In one example, the second post-addition measurement value is the third measurement of water quality input into the second model for calculating the addition rate of the second chemical. More specifically, the chemical addition rate determination device 10 calculates the difference between the second measurement value after addition and the second target water quality value at the third water quality measurement location, and adjusts the addition rate of the second chemical in a direction that reduces this difference.

[0098] Figures 7 and 8 are flowcharts illustrating one embodiment of the operation of the water treatment system shown in Figure 6. In step 1, the first water quality measuring instrument 2 measures the water quality at the first water quality measurement position upstream of the first chemical addition position to obtain the first measurement value, and sends the first measurement value to the chemical addition rate determination device 10. At the first chemical addition position, the first pump 3 adds the chemical to the water at a preset addition rate. In step 2, the second water quality measuring instrument 5 measures the water quality at the second water quality measurement position downstream of the first chemical addition position to obtain a second measurement value, and sends the second measurement value to the chemical addition rate determination device 10. In step 3, the third water quality measuring instrument 6 measures the water quality at the third water quality measuring position downstream of the second chemical addition position to obtain the third measurement value, and sends the third measurement value to the chemical addition rate determination device 10. At the second chemical addition position, the second pump 7 adds the chemical to the water at a preset addition rate. Steps 1-3 above may be performed simultaneously, or in a different order than shown in the example in Figure 7.

[0099] In step 4, the chemical addition rate determination device 10 acquires the first, second, and third water quality measurements sent from the first water quality meter 2, the second water quality meter 5, and the third water quality meter 6, and stores them in the storage device 10a. In step 5, the chemical addition rate determination device 10 reads the second and third water quality measurements from the storage device 10a and inputs them into the second model. In step 6, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the second model and outputs the addition rate of the second drug from the second model. In step 7, the chemical addition rate determination device 10 inputs the first water quality measurement, the second water quality measurement, and the addition rate of the second chemical into the first model. In step 8, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the first model and outputs the addition rate of the first drug from the first model.

[0100] In step 9, the chemical addition rate determination device 10 controls the operation of the first pump 3 so that the first chemical is added to the water at the addition rate output from the first model. In step 10, the chemical addition rate determination device 10 adjusts the addition rate of the first chemical output from the first model based on the first post-addition measurement value (for example, the second measurement value obtained in step 2 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the first post-addition measurement value and the first target water quality value at the second water quality measurement location, and adjusts the addition rate of the first chemical in a direction that reduces this difference. In step 11, the chemical addition rate determination device 10 controls the operation of the first pump 3 so that the first chemical is added to the water at the adjusted addition rate.

[0101] In step 12, the chemical addition rate determination device 10 controls the operation of the second pump 7 so that the second chemical is added to the water at the addition rate output from the second model. In step 13, the chemical addition rate determination device 10 adjusts the addition rate of the second chemical output from the second model based on the second post-addition measurement value (for example, the third measurement value obtained in step 3 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the second post-addition measurement value and the second target water quality value at the third water quality measurement location, and adjusts the addition rate of the second chemical in a direction that reduces this difference. In step 14, the chemical addition rate determination device 10 controls the operation of the second pump 7 so that the second chemical is added to the water at the adjusted addition rate.

[0102] If the difference between the first measurement after addition and the first target water quality value is small, that is, if the accuracy of the addition rate of the first chemical output from the first model is high, steps 10 and 11 above may be omitted. Similarly, if the difference between the second measurement after addition and the second target water quality value is small, that is, if the accuracy of the addition rate of the second chemical output from the second model is high, steps 13 and 14 above may be omitted.

[0103] According to this embodiment, in multi-stage chemical addition, the addition information at the later addition point in the water treatment process can be fed back to the earlier addition point, further improving the accuracy of predicting the addition rate at the earlier stage. In the embodiment shown in Figure 6, two chemical addition positions and three water quality measurement positions are provided, but the present invention is not limited to this embodiment, and three or more chemical addition positions and four or more water quality measurement positions may be provided. In that case, three or more models constructed by machine learning may be provided.

[0104] In one embodiment, the first prediction condition data and first training data used for the first model may further include current operating condition data, including at least one of water flow rate, water temperature, and meteorological data, in addition to the first water quality measurement, the second water quality measurement, and the addition rate of the second chemical. Similarly, in one embodiment, the second prediction condition data and second training data used for the second model may further include current operating condition data, including at least one of water flow rate, water temperature, and meteorological data, in addition to the second water quality measurement and the third water quality measurement. Meteorological data may include rainfall, temperature, etc.

[0105] The embodiments described with reference to Figures 1 and 2 may be combined with the embodiments described with reference to Figures 6 to 8. Below, embodiments of multi-stage drug addition combining time-series measurement data, feedforward control, and feedback control will be described.

[0106] In this embodiment, the first prediction condition data input to the first model includes at least the most recent first time-series measurement data, which includes current and past measurements of water quality before the first chemical is added; the most recent second time-series measurement data, which includes current and past measurements of water quality after the first chemical is added; and the addition rate of the second chemical output from the second model. The most recent first time-series measurement data is acquired by the first water quality meter 2 shown in Figure 6, and the most recent second time-series measurement data is acquired by the second water quality meter 5 shown in Figure 6. Each of the most recent first time-series measurement data and the most recent second time-series measurement data includes the most recent measurements of water quality for a predetermined number of days.

[0107] The second prediction condition data input to the second model includes at least the most recent second time-series measurement data and the most recent third time-series measurement data, which includes current and past measurements of water quality after the first chemical has been added and the second chemical has been added further. The most recent third time-series measurement data is acquired by the third water quality measuring instrument 6 shown in Figure 6.

[0108] The first water quality meter 2, the second water quality meter 5, and the third water quality meter 6 measure water quality at predetermined time intervals and acquire multiple first measurement values, multiple second measurement values, and multiple third measurement values. These multiple first measurement values, multiple second measurement values, and multiple third measurement values ​​are sent to the chemical addition rate determination device 10. Each time the chemical addition rate determination device 10 receives the first, second, and third measurement values ​​of water quality, it stores them in the storage device 10a, associating them with time. Therefore, the storage device 10a stores time-series data of the first measurement value of water quality (i.e., first time-series measurement data of water quality), time-series data of the second measurement value of water quality (i.e., second time-series measurement data of water quality), and time-series data of the third measurement value of water quality (i.e., third time-series measurement data of water quality).

[0109] The first model used in this embodiment is a trained model constructed by machine learning using the first training data described below. The first training data includes past first time-series measurement data of water quality before the first chemical is added from the first pump 3, past second time-series measurement data of water quality after the first chemical is added from the first pump 3, past calculation data of the addition rate of the second chemical, and the actual addition rate of the first chemical associated with the past first time-series measurement data, the past second time-series measurement data, and the past calculation data of the addition rate of the second chemical. The past first time-series measurement data is acquired by the first water quality meter 2 shown in Figure 6, and the past second time-series measurement data is acquired by the second water quality meter 5 shown in Figure 6.

[0110] The second model used in this embodiment is a trained model constructed by machine learning using the second training data described below. The second training data includes the above-mentioned past second time-series measurement data, past third time-series measurement data of water quality after the first chemical has been added and then the second chemical has been further added by the second pump 7, and the actual addition rate of the second chemical associated with the above-mentioned past second time-series measurement data and the above-mentioned past third time-series measurement data. The past third time-series measurement data is acquired by the third water quality meter 6 shown in Figure 6.

[0111] The drug addition rate determination device 10 creates a first model using the first training data according to a machine learning algorithm, and creates a second model using the second training data according to a machine learning algorithm. The first and second models constructed by machine learning are stored in the memory device 10a as trained models. In this embodiment, the first and second models are each composed of a neural network consisting of long short-term memory (LSTM). However, the first and second models used in the present invention are not limited to LSTM type models, as long-series data can be used.

[0112] Figures 9 and 10 are flowcharts illustrating one embodiment of the operation of a water treatment system, combining the embodiments described with reference to Figures 1 and 2 with the embodiments described with reference to Figures 6 to 8. In Step 1, the first water quality meter 2 measures the water quality at a first water quality measurement position upstream of the first chemical addition position at predetermined time intervals to obtain multiple first measurement values, and sends these first measurement values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the first measurement values ​​of water quality sent from the first water quality meter 2 in the storage device 10a. Each time the chemical addition rate determination device 10 receives a first measurement value of water quality, it stores that first measurement value in the storage device 10a in relation to time. Therefore, the storage device 10a stores time-series data of the first measurement values ​​of water quality (i.e., first time-series measurement data of water quality). At the first chemical addition position, the first pump 3 adds chemicals to the water at a predetermined addition rate.

[0113] In step 2, the second water quality meter 5 measures the water quality at a second water quality measurement location downstream of the first chemical addition location and upstream of the second chemical addition location at predetermined time intervals to obtain multiple second measurement values, and sends these second measurement values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the second measurement values ​​of water quality sent from the second water quality meter 5 in the storage device 10a. Each time the chemical addition rate determination device 10 receives a second measurement value of water quality, it stores that second measurement value in the storage device 10a in relation to time. Therefore, the storage device 10a stores time-series data of the second measurement values ​​of water quality (i.e., second time-series measurement data of water quality).

[0114] In step 3, the third water quality meter 6 measures the water quality at the third water quality measurement position downstream of the second chemical addition position at predetermined time intervals to obtain multiple third measurement values, and sends these third measurement values ​​to the chemical addition rate determination device 10. The chemical addition rate determination device 10 stores the third measurement values ​​of water quality sent from the third water quality meter 6 in the storage device 10a. Each time the chemical addition rate determination device 10 receives a third measurement value of water quality, it stores that third measurement value in the storage device 10a in relation to time. Therefore, the storage device 10a stores time-series data of the third measurement values ​​of water quality (i.e., time-series measurement data of the third water quality). At the second chemical addition position, the second pump 7 adds chemicals to the water at a predetermined addition rate. Steps 1-3 above may be performed simultaneously, or in a different order than shown in Figure 9.

[0115] In step 4, the chemical addition rate determination device 10 reads the second and third time-series measurement data of water quality from the storage device 10a and inputs it into the second model. In step 5, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the second model and outputs the addition rate of the second drug from the second model. In step 6, the chemical addition rate determination device 10 inputs the first time-series measurement data of water quality, the second time-series measurement data of water quality, and the addition rate of the second chemical into the first model. In step 7, the drug addition rate determination device 10 performs calculations according to the algorithm defined by the first model and outputs the addition rate of the first drug from the first model.

[0116] In step 8, the chemical addition rate determination device 10 controls the operation of the first pump 3 so that the first chemical is added to the water at the addition rate output from the first model. In step 9, the chemical addition rate determination device 10 adjusts the addition rate of the first chemical output from the first model based on the first post-addition measurement (for example, the most recent of the multiple second measurement values ​​obtained in step 2 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the first post-addition measurement and the first target water quality value at the second water quality measurement location, and adjusts the addition rate of the first chemical in a direction that reduces this difference. In step 10, the chemical addition rate determination device 10 controls the operation of the first pump 3 so that the first chemical is added to the water at the adjusted addition rate.

[0117] In step 11, the chemical addition rate determination device 10 controls the operation of the second pump 7 so that the second chemical is added to the water at the addition rate output from the second model. In step 12, the chemical addition rate determination device 10 adjusts the addition rate of the second chemical output from the second model based on the second post-addition measurement value (for example, the most recent of the multiple third measurement values ​​obtained in step 3 above). More specifically, the chemical addition rate determination device 10 calculates the difference between the second post-addition measurement value and the second target water quality value at the third water quality measurement location, and adjusts the addition rate of the second chemical in a direction that reduces this difference. In step 13, the chemical addition rate determination device 10 controls the operation of the second pump 7 so that the second chemical is added to the water at the adjusted addition rate.

[0118] If the difference between the first measurement after addition and the first target water quality value is small, that is, if the accuracy of the addition rate of the first chemical output from the first model is high, steps 9 and 10 above may be omitted. Similarly, if the difference between the second measurement after addition and the second target water quality value is small, that is, if the accuracy of the addition rate of the second chemical output from the second model is high, steps 12 and 13 above may be omitted.

[0119] In the embodiments shown in Figures 9 and 10, two chemical addition locations and three water quality measurement locations are provided. However, the present invention is not limited to these embodiments, and three or more chemical addition locations and four or more water quality measurement locations may be provided. In that case, three or more models constructed by machine learning may be provided.

[0120] Next, the operating results of the water treatment system according to the above-described embodiment will be explained. Figure 11 is a graph showing the operating results of the embodiment described with reference to the water treatment system shown in Figure 3 and the flowchart shown in Figure 5. In this operation, sodium hypochlorite was used as the chemical, and water quality was measured upstream and downstream of the chemical addition point. The training data used for machine learning of the model included past first time series measurement data and past second time series measurement data obtained from two years of operation.

[0121] In Figure 11, the vertical axis represents the chemical addition rate, and the horizontal axis represents time. As shown in Figure 11, the error between the actual chemical addition rate by skilled workers and the addition rate calculated using the model was very small. From these operational results, it can be seen that the chemical addition rate determination device 10, equipped with a model built using machine learning, can predict the chemical addition rate with very high accuracy. [Explanation of symbols]

[0122] 2 Water quality measuring device, 1st water quality measuring device 3 pumps, pump 1 5 Second water quality meter 6 Third water quality meter 7. Pump No. 2 10. Chemical Addition Rate Determination Device

Claims

1. A trained model constructed by machine learning using training data that includes at least the past first time series measurement data of the water quality of the water to be treated, past second time series measurement data of the water quality of the water after chemicals have been added, and the actual addition rate of chemicals associated with the past first time series measurement data and the past second time series measurement data is input into a trained model, which includes at least the most recent first time series measurement data including the current and past measurements of the water quality of the water to be treated, and the most recent second time series measurement data including the current and past measurements of the water quality of the water after chemicals have been added. The training model outputs the addition rate of the drug, This includes controlling the operation of the pump so that the chemical is added to the water at the chemical addition rate output from the trained model, The aforementioned machine learning is deep learning, A method for adding chemicals for water treatment, characterized in that the trained model is a model constructed by deep learning.

2. The method for adding chemicals for water treatment according to claim 1, characterized in that the trained model is a long- and short-term memory model.

3. The i-th model, constructed by machine learning using the i-th model, is input to the i-th model, which includes the i-th measured value of the water quality of the water to be treated before the addition of the i-th chemical (i is a natural number greater than or equal to 2), the i-th measured value of the water quality of the water after the addition of the i-th chemical, the i-th calculated value of the addition rate of the i-th chemical, and the i-th predicted condition data, which includes at least the i-th measured value of the water quality, the i-th measured value of the water quality after the addition of the i-th chemical, and the addition rate of the i-th chemical. A method for determining the addition rate of chemicals for water treatment, characterized by outputting the addition rate of the i-1 chemical from the i-1 model.

4. The method further includes obtaining the i-1 measurement, the i measurement, and the i+1 measurement of the water quality after the i-1 chemical has been added and the i chemical has been further added, inputting the i-th prediction condition data, which includes at least the i-th measurement and the i+1 measurement, into an i-th model constructed by machine learning, and outputting the addition rate of the i-th chemical from the i-th model. The addition rate of the i-chemical included in the i-1 prediction condition data is the addition rate of the i-chemical output from the i-model. The i-th model is a trained model constructed by machine learning using the i-th training data, wherein the i-th training data includes at least the past i-th measurement data, the past i+1 measurement data of the water quality of the water to which the i-th chemical has been further added after the i-1 chemical has been added, and the actual addition rate of the i-th chemical associated with the past i-th measurement data and the past i+1 measurement data, the method for determining the addition rate of a chemical for water treatment according to claim 3.

5. The i-1 measurement value included in the i-1 prediction condition data (where i is a natural number greater than or equal to 2) is the most recent i-1 time series measurement data, which includes current and past measurements of the water quality before the i-1 chemical was added. The i-1 prediction condition data and the i-measured values ​​included in the i-prediction condition data are the most recent i-th time-series measurement data, including current and past measurements of the water quality after the i-1 chemical has been added. The i+1 measurement value included in the i prediction condition data is the most recent i+1 time series measurement data, which includes current and past measurements of the water quality of the water to which the i chemical has been further added after the i-1 chemical has been added. The past i-1 measurement data included in the i-1 training data is the past i-1 time-series measurement data of the water quality before the i-1 chemical was added, The i-1 training data and the i-th past measurement data included in the i-th training data are the i-th past time-series measurement data of the water quality after the i-1 chemical was added. The method for determining the addition rate of chemicals for water treatment according to claim 4, characterized in that the past (i+1) measurement data included in the i training data is the past (i+1) time-series measurement data of the water quality after the i-1 chemical has been added and the i chemical has been further added.

6. The i-th model (where i is a natural number greater than or equal to 2) constructed using machine learning, Measurement acquisition means for acquiring the i-1 measurement of the water quality of the water to be treated before the addition of the i-1 chemical, the i measurement of the water quality of the water after the addition of the i-1 chemical, and the i+1 measurement of the water quality of the water after the addition of the i-1 chemical and further addition of the i chemical. An i-th drug addition rate output means inputs i-th prediction condition data, which includes at least the i-th measurement value and the i+1 measurement value, into the i-th model and outputs the i-th drug addition rate from the i-th model, The i-1 model constructed using machine learning, An apparatus for determining the addition rate of chemicals for water treatment, characterized by having an i-1 chemical addition rate output means that inputs the i-1 measured value, the i measured value, and i-1 prediction condition data including the addition rate of the i chemical to the i-1 model, and outputs the addition rate of the i-1 chemical from the i-1 model.

Images