Method for constructing learning model, learning model, separation membrane characteristic prediction method, filtration device operation method, separation membrane characteristic prediction program, prediction system, filtration device, and recording medium

Abstract

Provided is a method for constructing a learning model for predicting a separation membrane characteristic in a filtration device that filters water to be filtered through a separation membrane to obtain treated water, said method comprising a learning model construction step for constructing a learning model by machine learning, wherein at least one of water quality information pertaining to the water to be filtered, operation condition information pertaining to the filtration device, and weather information is an explanatory variable, and a separation membrane characteristic corresponding to the explanatory variable is an objective variable.

Description

Method for constructing a learning model, learning model, method for predicting separation membrane characteristics, method for operating a filtration device, separation membrane characteristic prediction program, prediction system, filtration device, and recording medium.

[0001] The present invention relates to a method for constructing a learning model for predicting the characteristics of a separation membrane in a filtration device using a separation membrane, the learning model, a method for predicting the characteristics of the separation membrane, a method for operating the filtration device, a program for predicting the characteristics of the separation membrane, a prediction system, a filtration device, and a recording medium.

[0002] Filtration systems using membrane separation methods are gaining popularity in various fields due to their energy-saving, space-saving, and water-quality improvement features. Examples include membrane separation activated sludge methods that use microfiltration membranes or ultrafiltration membranes to treat sewage and industrial wastewater, applications in water purification processes that produce industrial water and tap water from river water, groundwater, and secondary treated water from sewage and industrial wastewater, and applications in pretreatment in reverse osmosis membrane treatment processes for seawater desalination using reverse osmosis membranes.

[0003] A common challenge with these membrane separation methods is that when water to be filtered is filtered through a membrane, the amount of contaminants adhering to the membrane surface and within the membrane pores increases with the amount of water filtered, leading to increased membrane fouling, which in turn causes a decrease in the amount of treated water or an increase in membrane filtration resistance.

[0004] To suppress this increase in membrane filtration resistance and to stably operate a filtration system using membrane separation, it is crucial to understand the fouling state of the separation membrane and perform appropriate cleaning. In other words, to operate a filtration system using membrane treatment efficiently and stably, it is important to understand the membrane filtration capacity from the state of the filtered water and find appropriate operating conditions according to the membrane filtration capacity.

[0005] Patent Document 1 proposes a method for predicting the time-dependent changes in membrane filtration resistance and intermembrane pressure by actually filtering the water to be filtered, determining membrane filtration parameters that indicate membrane filtration performance from the changes in the intermembrane pressure differential, membrane filtration flux, and membrane filtration resistance values ​​at that time.

[0006] Patent Document 2 proposes an operation support program that uses data such as water quality information of the water to be filtered, pressure supplied to the filtration device, intermembrane pressure difference in the separation membrane, permeation flux, and cleaning conditions of the filtration membrane to create a judgment model through a learning process, and then determines the cleaning conditions from the viewpoint of energy saving and membrane blockage.

[0007] Patent Document 3 proposes a filtration device that uses data related to the intermembrane differential pressure, the amount of diffused air, and the membrane filtration flow rate to perform a learning process, and then estimates the intermembrane differential pressure and controls the amount of diffused air.

[0008] International Publication No. 2009 / 054506, Japanese Patent Publication No. 2022-64517, Japanese Patent Publication No. 2022-101239

[0009] To determine appropriate operating conditions, it is necessary to understand the membrane filtration properties of the water to be filtered, which change depending on the water quality information of the water to be treated, the water quality information of the water to be filtered, the operating conditions of the filtration system, and the temperature and climate, and to consider the influence of the membrane filtration properties of the water to be filtered and the operating conditions on the separation membrane characteristics. However, while Patent Document 1 lists filtration testing as a method for understanding the membrane filtration properties of the water to be filtered, this requires actually filtering the water to be filtered using a membrane filtration testing device, which takes time for preparing the separation membrane for testing and for the membrane filtration test itself. Therefore, it is difficult to understand the membrane filtration properties of the water to be filtered, which change moment by moment, to determine operating conditions in a timely manner and to operate a filtration system using a separation membrane stably. Furthermore, Patent Document 2 describes a judgment model that evaluates the interrelationship between the actual intermembrane differential pressure of the separation membrane, the water quality of the water to be treated, and various other parameters such as washing conditions, in order to determine the optimal value of the filtration flux and washing conditions at that time. Patent Document 3 describes how the intermembrane pressure differential is estimated using a regression model equation from operating data including membrane filtration pressure and diffuser volume, and how the diffuser attached to the filtration system is controlled to achieve the optimal diffuser volume by estimating the change in intermembrane pressure over time. However, determining washing conditions that conserve energy and suppress fouling, and determining the optimal diffuser volume, requires learning a wide variety of operating data related to the water quality of the treated and filtered water, which changes due to various factors, and the intermembrane pressure of the filtration membrane due to changes in operating conditions, i.e., the impact on the operability of the filtration system using membrane separation. Acquiring this data in an actual filtration system is difficult, and its application is time-consuming.

[0010] Therefore, the objective of the present invention is to provide a method for predicting the time-dependent changes in the separation membrane characteristics during filtration operation based on the estimated separation membrane characteristics, using a learning model constructed by machine learning to estimate the separation membrane characteristics online from water quality information of the filtered water, operating conditions information of the filtration system, climate information, etc.

[0011] To achieve the above objective, one aspect of the present invention has the following configuration: [1] A method for constructing a learning model for predicting the characteristics of a separation membrane in a filtration apparatus that filters water to be filtered with a separation membrane to obtain treated water, comprising a learning model construction step of constructing a learning model by machine learning with data including at least one of water quality information of the water to be filtered, operating condition information of the filtration apparatus, and climate information as explanatory variables, and the explanatory variables and the corresponding separation membrane characteristics as objective variables. [2] The method for constructing a learning model according to [1], wherein the explanatory variables include at least water temperature, viscosity, and turbidity component index as water quality information of the water to be filtered. [3] The method for constructing a learning model according to [2], wherein the explanatory variables further include at least one selected from pH, ORP, and organic matter index as water quality information of the water to be filtered, and HRT, SRT, water level, amount of water to be filtered drawn, amount of water to be filtered supplied, amount of water to be filtered circulated, and amount of water to be filtered treated as operating condition information of the filtration apparatus. [4] A method for constructing a learning model according to any one of [1] to [3], wherein the separation membrane characteristics include membrane filterability parameters involved in calculating at least one of the following: component resistance of the filtered water, non-detachable component resistance of the filtered water, rate of progression of blockage of the separation membrane pores, and resistance due to blockage of the separation membrane pores. [5] A method for constructing a learning model according to [4], wherein the separation membrane characteristics further include membrane filterability parameters involved in calculating at least one of the following: component adhesion rate of the filtered water, component detachment rate of the filtered water, and non-detachable component formation rate of the filtered water. [6] A learning model constructed by the method for constructing a learning model according to any one of [1] to [5]. [7] A method for predicting separation membrane characteristics in a filtration apparatus that filters filtered water with a separation membrane to obtain treated water, comprising a separation membrane characteristics estimation step of estimating the change in the separation membrane characteristics over time from the input data, using the learning model according to [6], with input data including at least one of water quality information of the filtered water, operating condition information of the filtration apparatus, and climate information. [8] The separation membrane characteristic prediction method according to [7], further comprising a simulation prediction step of predicting the change over time of the separation membrane characteristics by simulation based on the membrane filterability parameters estimated in the separation membrane characteristic estimation step.[9] The separation membrane characteristics prediction method according to [8], wherein the simulation prediction step is calculated in the following calculation steps: Calculation step 1: Set initial values ​​for the separation membrane characteristics, or calculate initial values ​​for the separation membrane characteristics from the operating condition information of the filtration device. Calculation step 2: Use the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters to calculate the amount of change in the separation membrane characteristics that has changed during a predetermined time Δt. Calculation step 3: Calculate the separation membrane characteristics at an arbitrary time t + Δt from the amount of change calculated in calculation step 2.

[10] The separation membrane characteristics prediction method according to [7], wherein the learning model updates the learning model after estimating the separation membrane characteristics from the input data.

[11] The separation membrane characteristics prediction method according to [7], wherein the input data includes at least water temperature, viscosity, and turbidity component index as water quality information of the filtered water.

[12] The separation membrane characteristics prediction method according to

[11] , wherein the input data further comprises at least one selected from pH, ORP, and organic matter index as water quality information of the filtered water, and HRT, SRT, water level, amount of water to be filtered, amount of water to be filtered, amount of water to be filtered, and amount of water to be treated as operating condition information of the filtration device.

[13] A method for operating a filtration device to filter water with a separation membrane and obtain treated water, comprising the steps of: predicting the separation membrane characteristics using the separation membrane characteristics prediction method according to [8] or [9]; and controlling the operating conditions of the filtration device based on the predicted separation membrane characteristics.

[14] The method for operating a filtration device according to

[13] , wherein the operating conditions are controlled when the predicted separation membrane characteristics deviate from a predetermined standard range.

[15] The method for operating a filtration device according to

[13] , wherein the predicted date on which the separation membrane characteristics deviate from a predetermined standard range is calculated from the prediction results of the simulation prediction step, and the operating conditions are controlled when the number of days until the predicted date deviates from a predetermined standard range.

[16] A separation membrane characteristics prediction program for a filtration apparatus that filters water to be filtered with a separation membrane to obtain treated water, comprising: a learning model construction means for constructing a learning model by machine learning with data including at least one of water quality information of the water to be filtered, operating condition information of the filtration apparatus, and climate information as explanatory variables, and the separation membrane characteristics corresponding to the explanatory variables as objective variables; an input means for inputting data including at least one of water quality information of the water to be filtered, operating condition information of the filtration apparatus, and climate information; and a separation membrane characteristics estimation means for estimating the change in the separation membrane characteristics over time from the data using the learning model.

[17] The separation membrane characteristics prediction program according to

[16] , further comprising a computer for comprising a simulation prediction means for predicting the change in the separation membrane characteristics over time by simulation based on the membrane filterability parameters estimated by the separation membrane characteristics estimation means.

[18] The separation membrane characteristics prediction program according to

[17] , wherein the simulation prediction means is calculated in the following calculation steps: Calculation step 1: Set initial values ​​for the separation membrane characteristics, or calculate initial values ​​for the separation membrane characteristics from the operating condition information of the filtration apparatus. Calculation step 2: Using the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters, the amount of change in the separation membrane characteristics that has changed during a predetermined time Δt is calculated. Calculation step 3: The separation membrane characteristics at an arbitrary time t + Δt are calculated from the amount of change calculated in calculation step 2.

[19] A separation membrane characteristics prediction program according to

[16] , wherein the learning model updates the learning model after estimating the separation membrane characteristics from the input data.

[20] A separation membrane characteristics prediction program according to

[16] , wherein the explanatory variables and the input data include at least water temperature, viscosity, and turbidity component index as water quality information of the filtered water.

[21] A separation membrane characteristics prediction program according to

[20] , wherein the explanatory variables and the input data further include at least one selected from pH, ORP, and organic matter index as water quality information of the filtered water, and HRT, SRT, water level, amount of water drawn from the filtered water, amount of water supplied from the filtered water, amount of water circulated from the filtered water, and amount of water treated from the filtered water as operating condition information of the filtration device.

[22] The separation membrane characteristic prediction program according to

[16] , wherein the separation membrane characteristics include parameters involved in calculating at least one of the following: the resistance of components in the filtered water, the resistance of non-detached components in the filtered water, the rate of progression of blockage of separation membrane pores, and the resistance due to blockage of separation membrane pores.

[23] The separation membrane characteristic prediction program according to

[22] , wherein the separation membrane characteristics further include parameters involved in calculating at least one of the following: the rate of component adhesion in the filtered water, the rate of component detachment in the filtered water, and the rate of non-detached component formation in the filtered water.

[24] A prediction system implementing the separation membrane characteristic prediction program according to

[16] , comprising: communication means for receiving data from a filtration device including at least one of water quality information of the filtered water, operating condition information of the filtration device, and climate information; and transmission means for transmitting the separation membrane characteristics estimated by the separation membrane characteristic prediction program to the filtration device.

[25] A prediction system implementing the separation membrane characteristic prediction program described in

[17] , comprising: communication means for receiving data from a filtration device including at least one of water quality information of the water to be filtered, operating condition information of the filtration device, and climate information; and transmission means for transmitting to the filtration device the membrane filterability parameter estimated by the separation membrane characteristic estimation means or the change over time of the separation membrane characteristics predicted by the simulation prediction means.

[26] A filtration device for filtering water to be filtered with a separation membrane to obtain treated water, comprising: a separation membrane module; a sensor for acquiring water quality information of the water to be filtered; communication means for transmitting the water quality information acquired by the sensor to the prediction system described in

[24] or

[25] and receiving the prediction result of the separation membrane characteristics from the prediction system; and operation control means for controlling the operating conditions based on the prediction result received by the communication means.

[27] A computer-readable recording medium recording the separation membrane characteristic prediction program described in any of

[16] to

[23] .

[0012] According to the present invention, it is possible to accurately predict the changes in separation membrane characteristics over time online, thereby enabling stable operation of the filtration system over a long period of time.

[0013] Figure 1 shows an example of a filtration system when the water to be treated and the water to be filtered are different. Figure 2 shows another example of a filtration system when the water to be treated and the water to be filtered are different. Figure 3 shows an example of a filtration system when the water to be treated and the water to be filtered are the same. Figure 4 shows an example of an enlarged view of a part of the flat membrane element in the membrane separation activated sludge process. Figure 5 shows an example of a fouling state on the separation membrane surface. Figure 6 is a processing flow showing an example of the learning model construction process of the present invention. Figure 7 is a processing flow showing an example of the separation membrane characteristic estimation process of the present invention. Figure 8 is a processing flow showing an example of the simulation prediction process of the present invention. Figure 9 is an example of the separation membrane characteristic prediction program of the present invention. Figure 10 is another example of the separation membrane characteristic prediction program of the present invention. Figure 11 is a graph showing a comparison of the estimation accuracy based on the item conditions of the explanatory variables in Example 1. Figure 12 is a graph showing a comparison of the predicted result of the intermembrane differential pressure by the simulation prediction process and the measured value in Example 2.

[0014] According to the present invention, by using at least one of the following as explanatory variables—water quality information of the filtered water, operating conditions information of the filtration system, and climate information—it is possible to predict the time-dependent changes in the characteristics of the separation membrane. By accurately predicting separation membrane characteristics such as membrane filtration resistance and intermembrane differential pressure, the present invention enables stable operation of a filtration system using a separation membrane, and allows for the determination of appropriate operating conditions while controlling fouling and the treated water quality obtained by the separation membrane. Preferably, data including at least water quality information of the filtered water is used as explanatory variables. By using water quality information of the filtered water, it becomes possible to make predictions that directly reflect the state of the filtered water. In particular, when the water quality information of the filtered water includes water temperature, viscosity, and turbidity component indices, high prediction accuracy can be obtained. As turbidity component indices, MLSS (Mixed Liquor Suspended Solids), SS (Suspended Solids), VSS (Volatile Suspended Solids), etc., can be used, with MLSS being particularly preferred. Water temperature, viscosity, and turbidity component indices can be obtained online from the filtration system, allowing for accurate prediction of changes in the membrane filterability of the filtered water and the time-dependent changes in the separation membrane characteristics due to operating conditions.

[0015] Furthermore, by including information on the operating conditions of the filtration system as explanatory variables, it is possible to understand the impact of changes in operating conditions on the separation membrane characteristics. In particular, operating condition information such as HRT (hydraulic residence time), SRT (sludge residence time), water level, amount of filtered water withdrawn, amount of filtered water supplied, amount of filtered water circulated, and amount of filtered water treated reflect the retention state of the filtered water and the load on the separation membrane. Therefore, by including at least one of these as an explanatory variable, it is possible to understand in more detail the impact of operating conditions on the separation membrane characteristics. All of this operating condition information is a parameter that reflects the retention state of the filtered water and the load on the separation membrane, and is effective in predicting changes in the membrane filterability of the filtered water and the time change in separation membrane characteristics due to operating conditions. In addition, by including climate information as an explanatory variable, the impact of climate on fluctuations in the water quality of the treated water or filtered water can be reflected in the separation membrane characteristics. By using a combination of this information, more multifaceted predictions become possible, and prediction accuracy can be further improved.

[0016] Embodiments of the present invention will be described in detail below with reference to the drawings, but the present invention is not limited in any way thereto.

[0017] (Configuration of the filtration device) The filtration device in this embodiment may include, but is not limited to, a cross-flow type device that performs membrane filtration while supplying water to be filtered to a container containing a separation membrane as shown in Figure 1, an immersion type device that immerses the separation membrane in the water to be filtered as shown in Figure 2, a total filtration type device that performs membrane filtration while concentrating the water to be filtered, and a rotary type device that immerses all or part of the separation membrane in the water to be filtered and rotates the separation membrane.

[0018] The water to be treated 1 is the raw water to be treated, that is, the liquid supplied to the filtration device, and examples include river water, groundwater, seawater, treated sewage water, factory wastewater, culture medium, etc. The water to be filtered is the liquid supplied to the membrane, and if it is supplied to the separation membrane without pretreatment, the water to be treated 1 and the water to be filtered are the same, but if pretreatment is performed before treatment by the separation membrane, the water to be treated 1 and the water to be filtered are different. Here, examples of the pretreatment include removal of suspended solids, sedimentation separation, flotation separation, centrifugation, coagulation treatment, stirring treatment, aerobic treatment, anaerobic treatment, filtration treatment, sedimentation treatment, softening treatment, aeration treatment, accelerated oxidation, ion exchange, adsorption, absorption, degassing, crystallization, distillation, heat exchange, chemical solution addition, mineral addition, mixing with microbial mixed water, etc. Preferably, when the water to be filtered is modified by pretreatment, it is preferable to modify it by at least one treatment selected from removal of suspended solids, filtration treatment, sedimentation treatment, chemical solution addition, mixing with microbial mixed water, aeration treatment, and softening treatment. This reduces the accumulation of components remaining in the filtered water on the surface of the separation membrane, allowing the filtration system to operate for a longer period.

[0019] Figure 1 shows an example of a filtration system where the water to be treated and the water to be filtered are different. The filtration system consists of a pretreatment section 2 and a separation membrane module 7. The water to be treated 1 is pretreated in the pretreatment section 2 and supplied to the separation membrane module 7 as filtered water by the filtered water supply pump 4. The filtered water 9 is then reused or discharged. On the other hand, the concentrated water 8 separated by the separation membrane module 7 is returned to the system or discharged outside the system. In addition, a filtered water pressure gauge 5, a filtered water flow meter 6, etc. are generally used to control the supply of filtered water.

[0020] Figure 2, like Figure 1, is an example of a filtration system when the water to be treated and the water to be filtered are different. This filtration system comprises a water tank 10 to which the water to be treated 1 is supplied, a water supply pump 11 to which the water to be treated 1 is supplied, a water flow meter 12 to which the amount of water to be treated is supplied, an anexic tank 13 to which the water to be treated 1 is supplied, and a membrane separation tank 14 in which a separation membrane module 7 is immersed in the water to be filtered. This filtration system also comprises an aeration blower 15 to which air is supplied to the membrane separation tank 14, an air flow meter 16, a diffuser pipe 17, a suction pump 19 which is the driving source for filtration, a filtered water pressure gauge 20, a filtered water flow meter 21, and a filtered water tank 22 for storing the filtered water 9. Furthermore, this filtration system includes a water to be filtered extraction pump 23 for drawing out excess filtered water 9, a water to be treated return pump 24 for returning filtered water 9 to the anoxic tank 13, a water to be treated return flow meter 25 for measuring the flow rate of the returned water to be treated, a washing chemical tank 26, a chemical addition pump 27, and a washing chemical flow path switching valve 28. First, the water to be treated 1 is supplied to the anoxic tank 13 by the water to be treated supply pump 11. From the anoxic tank 13 to the membrane separation tank 14, it is supplied by overflow from the partition wall between the two. Next, it is filtered by the separation membrane module 7, and the filtered water 9 is stored in the filtered water tank 22 before being reused or discharged. In Figure 2, the anoxic tank 13 is equipped with a water to be treated quality sensor 3, and the membrane separation tank 14 is equipped with a water to be filtered quality sensor 18 for acquiring water quality information.

[0021] The method of supplying the water to be filtered to the separation membrane module is not limited to the method of bringing the separation membrane module 7 into contact with the water to be filtered in the membrane separation tank 14, as shown in Figure 2, but the water to be filtered may also be supplied directly to the separation membrane module 7, as shown in Figure 3. Figure 3 is an example of a filtration system when the water to be treated and the water to be filtered are the same. This filtration system comprises a water to be filtered tank 31 to which the water to be treated 1 is supplied, a water to be filtered supply pump 4 to which the water to be filtered is supplied, a water to be filtered flow meter 6 to measure the flow rate, a separation membrane module 7 to which the water to be filtered is supplied, and an aeration blower 15 to supply air to the separation membrane module 7. This filtration system also comprises an air flow meter 16 to measure the amount of air supplied, a filtered water pressure meter 20, a filtered water flow meter 21, a cleaning chemical tank 26, a chemical addition pump 27, a back pressure cleaning pump 29, and a drain line 30. First, the water to be treated 1 is supplied to the water to be filtered tank 31 and then supplied to the separation membrane module 7 by the water to be filtered supply pump 4. Next, the water is filtered by the separation membrane module 7, and the filtered water 9 is stored in the filtration tank 22 before being used for backwashing or discharged. Similar to Figure 2, the filtered water tank 31 in Figure 3 is equipped with a filtered water quality sensor 18 that acquires water quality information.

[0022] (Acquisition of Water Quality Information) Here, the water quality information of the treated or filtered water acquired by the sensor is not particularly limited, but may include viscosity, turbidity, pH, water temperature, dissolved oxygen (DO), mixed liquor suspended solids (MLSS), suspended solids (SS), total organic carbon (TOC), chemical oxygen demand (COD), biochemical oxygen demand (BOD), and dissolved organic carbon (DOC). Examples of water quality information include carbon, total oxygen demand (TOD), oxidation-reduction potential (ORP), electrical conductivity (EC), UV260, UV254, and image information obtained by optically imaging the filtered water. Furthermore, the water quality information acquired by the sensor can also be correlated data such as voltage, and is not particularly limited.

[0023] In embodiments of the present invention, the water quality information acquired by the water quality sensor of the water to be treated or filtered preferably includes at least water temperature, viscosity, and MLSS. This makes it possible to accurately predict changes in the membrane filterability of the water to be filtered and the time-dependent changes in the separation membrane characteristics due to operating conditions. More preferably, it is preferable to include at least one of turbidity, pH, DO, TOC, ORP, EC, and image information obtained by optically imaging the water to be filtered. Particularly preferable is to include pH and ORP, and further preferably include HRT as operating condition information for the filtration device. In addition, by including an organic matter index, it is possible to reflect the effect of the organic matter concentration in the water to be filtered on the separation membrane characteristics, and the prediction accuracy can be further improved. This makes it possible to determine the operating conditions of the filtration device in accordance with changes in the water to be treated or filtered.

[0024] (Separation Membrane Element) An example of a separation membrane element 32a preferably used in this embodiment is shown in Figure 4. The separation membrane element 32a consists of a base material 32c, a separation functional layer 32b, and a frame 32d. The separation membrane element 32a is not particularly limited, and any of the following may be used: a flat membrane element structure in which a flat membrane is wound in a spiral shape; a flexible flat membrane element structure that includes two flat membranes arranged so that the surfaces of the flat membrane base materials face each other, a water collection channel provided between the flat membrane base materials, and a sealing portion that seals the space between the flat membranes at the periphery of the flat membranes; or a hollow fiber membrane element structure in which a plurality of hollow fiber membranes are bundled together.

[0025] A separation membrane is a device that captures substances of a certain particle size or larger contained in the filtered water by applying pressure to the water or by suction from the permeate side. Depending on the particle size it captures, there are various types of separation membranes, such as dynamic filtration membranes, microfiltration membranes, ultrafiltration membranes, nanofiltration membranes, and reverse osmosis membranes. In this invention, microfiltration membranes and ultrafiltration membranes are preferred as separation membranes. It is preferable to select and combine one or more appropriate membranes according to the size of the substances to be separated.

[0026] The separation membrane module 7 has a structure filled with multiple separation membrane elements 32a or separation membranes 34. The separation membrane elements 32a to be filled are not limited to flat membrane elements, but may also be hollow fiber elements or spiral elements. In this embodiment, when immersing the separation membrane module 7 in the water to be filtered, multiple modules may be placed side by side, or multiple modules may be stacked in layers. When multiple modules are placed side by side, the sides parallel to the direction in which the multiple separation membrane elements 32a of the separation membrane module 7 are loaded may be placed adjacent to each other. When multiple modules are stacked in layers, the sides perpendicular to the direction in which the multiple flat membrane elements of the separation membrane module 7 are loaded may be stacked adjacent to each other in the upper and lower layers.

[0027] (Filtration Operation) The filtration time of the filtration operation is preferably set as appropriate according to the properties of the water to be filtered and the amount of treated water, but the filtration time may be continued until a predetermined transmembrane pressure difference or membrane filtration resistance is reached. Here, the amount of treated water refers to the amount of water to be treated 1 treated by the separation membrane, and the transmembrane pressure difference is the pressure difference between the primary side (the side where the water to be filtered is supplied) and the secondary side (the side of the filtered water obtained by filtering the water to be filtered with the separation membrane) of the separation membrane. Examples of the means for generating the transmembrane pressure difference include a method of pressurizing with a pump, a method of sucking from the secondary side with a pump, and a method of utilizing the water head difference between the primary side and the secondary side. Further, when calculating the transmembrane pressure difference, it is preferable to measure or calculate the pressure loss generated by the hydraulic flow and calculate by subtracting the pressure loss component from the pressure measurement value on the primary side and the pressure measurement value on the secondary side of the separation membrane.

[0028] The calculation formula for the membrane filtration resistance is shown by the following formula (1) using the transmembrane pressure difference, membrane filtration flux, and viscosity.

[0029]

[0030] Here, ΔP is the transmembrane pressure difference [Pa], μ is the viscosity of the water to be filtered [Pa·s], R is the membrane filtration resistance [1 / m], and J is the membrane filtration flux [m / s]. Here, μ may be directly measured by a viscosity sensor for the viscosity of the water to be filtered, but in this embodiment, it may be converted from the temperature according to the following formula (2).

[0031]

[0032] Here, F = 0.01257187, B = -0.005806436, C = 0.001130911, D = -0.000005723952, and T is the absolute temperature [K]. That is, when the Celsius temperature is σ [°C], it is expressed as T = σ + 273.15.

[0033] (Membrane Fouling) As shown in Fig. 5, the fouling of the separation membrane 34 can be distinguished into pore blockage 35 caused by components of the filtered water smaller than the pore diameter that have entered the pores of the separation membrane 34, and the components of the filtered water adhering to the surface of the separation membrane, including the filtered water components 36 that adhere to the surface of the separation membrane and can be peeled off by physical cleaning, and the filtered water components 37 that adhere to the surface of the separation membrane and cannot be peeled off by physical cleaning. The separable components of the filtered water adhering to the surface of the separation membrane are compacted by the applied pressure and converted into non-separable components of the filtered water. By continuing the operation of the filtration device, the above fouling progresses and the membrane filtration resistance gradually increases.

[0034] (Cleaning Means) As cleaning means for cleaning the components of the filtered water adhering to the surface and inside the pores of the separation membrane, there are physical cleaning and chemical solution cleaning. Physical cleaning is at least one cleaning operation of the membrane surface of the separation membrane, which is a flushing operation of forcibly flowing one or both of a liquid and a gas on the primary side of the separation membrane, or a reverse pressure cleaning operation of supplying a liquid from the secondary side to the primary side of the separation membrane. The gas used in the flushing operation is preferably steam in addition to air. Also, it is preferable to use gas from which oil components and mist have been removed because the risk of oil components etc. adhering to the separation membrane is reduced. The gas flow rate used in the flushing operation can be appropriately set according to the membrane module form, but the aeration air volume per unit area of the aeration air flow path is 1.3 m ３ / m ２ / min or more and 5.0 m ３ / m ２ / min or less is preferable. By setting it to 1.3 m ３ / m ２ / min or more, a sufficient cleaning effect can be obtained, the increase in membrane filtration resistance can be avoided, and by setting it to 5.0 m ３ / m ２ / min or less, the causes of damage and deterioration of the membrane module and the power cost can be suppressed.

[0035] Back pressure washing is a washing operation in which filtered water 9 is supplied from the secondary side of the separation membrane 34 to the primary side of the separation membrane 34 by a back pressure washing pump 29, as shown in Figure 3, to remove components of the filtered water that have adhered to the membrane surface and inside the membrane pores. The supplied liquid consists of filtered water 9 and / or clarified liquid, and it is effective to control the flow rate and time of back pressure washing according to the filtration flow rate. The physical cleaning intensity can be increased by increasing the flow rate of gas or liquid in the flushing operation, lengthening the processing time, increasing the flow rate of liquid in the back pressure washing operation, lengthening the processing time, or shortening the washing interval between each operation. Conversely, the physical cleaning intensity can be reduced by decreasing the flow rate of gas or liquid in the flushing operation, shortening the processing time, decreasing the flow rate of liquid in the back pressure washing operation, shortening the processing time, or lengthening the washing interval between each operation.

[0036] Chemical washing is a method of supplying a chemical solution to the separation membrane module 7 to wash the separation membrane 34. In particular, in cases where the filtered water 33 contains many microorganisms, such as in the membrane separation activated sludge method, it is known that microorganisms proliferate on the membrane surface or microbial metabolites accumulate, forming a biofilm and causing a rapid increase in the intermembrane pressure differential. To avoid this phenomenon, it is effective to perform chemical washing periodically before the intermembrane pressure differential rises rapidly. The method of injecting the chemical solution in the present invention is not limited, but in the membrane separation apparatus using the separation membrane module 7 shown in Figure 2, it is preferable to inject the chemical solution from the secondary side to the primary side of the separation membrane while the separation membrane module 7 is immersed in the filtered water after stopping the filtration operation. As for methods of contacting the separation membrane 34 with the chemical solution, there are methods such as removing the separation membrane module 7 from the membrane separation tank 14 and immersing it in the chemical washing tank, or emptying the membrane separation tank and then accumulating the chemical solution in the membrane separation tank 14 to contact the membrane, but these require large-scale auxiliary equipment and are not economical, so they are not very preferable. The chemical solution used for cleaning can be selected after appropriately setting the concentration and contact time so as not to cause fouling of the separation membrane or deterioration of the separation membrane. However, it is preferable to include at least one of the following: sodium hypochlorite, chlorine dioxide, hydrogen peroxide, ozone, etc., as this enhances the cleaning effect against organic matter. It is also preferable to include at least one of the following: hydrochloric acid, sulfuric acid, nitric acid, citric acid, oxalic acid, etc., as this enhances the cleaning effect against aluminum, iron, manganese, etc. The chemical solution concentration is preferably 5 mg / L to 10,000 mg / L. A concentration of 5 mg / L or higher provides sufficient cleaning effect, while a concentration of 100,000 mg / L or lower is economical in terms of chemical cost. It is preferable to use two or more types of chemical solutions in sequence rather than just one type. For example, it is more preferable to alternate between acid and sodium hypochlorite.

[0037] (Separation Membrane Characteristics) The present invention provides a method for predicting the time-dependent changes in separation membrane characteristics in a water treatment method using a filtration device. The separation membrane characteristics predicted in the present invention refer to the characteristics of the separation membrane and the characteristics of the water to be filtered during membrane filtration (membrane filterability parameters).

[0038] The characteristics of the separation membrane are not particularly limited, but preferably include at least one of the following: intermembrane pressure differential, filtration flux, supply pressure to the separation membrane, pressure loss on the primary side of the separation membrane module, membrane filtration resistance, and solute removal performance in the filtered water. In particular, by predicting the intermembrane pressure differential and membrane filtration resistance, the fouling state of the separation membrane can be quantitatively grasped, making it possible to determine the appropriate cleaning timing and operating conditions. Here, the pressure loss on the primary side of the separation membrane module is the difference between the pressure at which the water to be treated 1 is supplied to the separation membrane module 7 and the pressure at which the filtered water is discharged. The "primary side" is the side of the space partitioned by the separation membrane from which the water to be filtered is supplied, and the "secondary side" is the side of the filtered water that has been filtered by the separation membrane. Furthermore, the solute removal performance in the filtered water is the amount or rate of removal of solute components contained in the filtered water when they are removed by the separation membrane.

[0039] The membrane filterability parameters are not particularly limited, but preferably include parameters involved in calculating at least one of the following: the resistance of the components in the filtered water, the resistance of the non-detached components in the filtered water, the rate of progression of pore blockage in the separation membrane, and the resistance due to pore blockage in the separation membrane. These parameters directly represent the fouling state of the separation membrane and are important indicators for predicting the change in membrane filter resistance over time. More preferably, the parameters further include parameters involved in calculating at least one of the following: the rate of component adhesion in the filtered water, the rate of component detachment in the filtered water, the rate of non-detached component formation in the filtered water, and the resistance of the non-detached components in the filtered water. By including these additional parameters, the dynamic behavior of fouling can be understood in more detail, and the prediction accuracy can be further improved.

[0040] In this invention, membrane filterability parameters are estimated using machine learning, and based on these estimated parameters, changes in separation membrane characteristics such as intermembrane pressure differential and membrane filtration resistance over time can be predicted. This makes it possible to understand the influence of the membrane filterability of the water to be filtered on the characteristics of the separation membrane, and to determine the optimal operating conditions for the filtration system.

[0041] (Learning Model Construction Process) Furthermore, the method for predicting the temporal change of the separation membrane characteristics includes a learning model construction process and a separation membrane characteristics estimation process. The learning model construction process (learning model construction method) will be described using FIG. 6.

[0042] The learning model construction process uses, as explanatory variables, input data including at least any one of the water quality information of the feed water, the operating condition information of the filtration device, and the climate information, and constructs a learning model with the input data and the corresponding separation membrane characteristics as the objective variables. Here, the input data may be the time series data itself, or may be converted into feature quantities indicating features. Here, the feature quantities may include the average value, the change rate or change amount per unit time, the standard deviation, and the like. The constructed learning model is stored in a recording device.

[0043] Here, the water quality information of the feed water includes viscosity, pH, electrical conductivity, water temperature, residual chlorine, filter paper filtration amount, coliform count, organic matter concentration, inorganic matter concentration, oil content, turbidity component index, organic matter index, total nitrogen (T-N), nitrite nitrogen (NO ２ -N), nitrate nitrogen (NO ３ -N), M alkalinity, total phosphorus (T-P), phosphate phosphorus (PO ４Examples of sludge analysis metrics include, but are not limited to, P, dissolved oxygen (DO), oxidation-reduction potential (ORP), electrical conductivity (EC), silt density index (SDI), evaporation residue (TS), evaporation loss (VTS), activated sludge settling rate, sludge volume index (SVI), and image information obtained by optically imaging the filtered water. Here, the turbidity component index is an index that indicates the amount of turbidity or the degree of turbidity in the filtered water, and includes, but is not limited to, turbidity, transmittance, transparency, suspended solids concentration (SS), mixed liquor suspended solids (MLSS), volatile suspended solids (MLSVS), volatile substances (VSS), etc. Furthermore, an organic matter index is an index that indicates the amount of organic matter in filtered water or the degree of pollution by organic matter, and includes, but is not limited to, total organic carbon (TOC), biochemical oxygen demand (BOD), chemical oxygen demand (COD), dissolved organic carbon (DOC), total oxygen demand (TOD), UV260, UV254, etc.Here, the operating conditions information for the filtration system is not limited to this, but preferably includes at least one of the following: the amount of water to be filtered, the volume of the membrane separator tank, the water level, head loss, differential pressure between membranes, membrane filtration flow rate, filtration time, sludge retention time (SRT), hydraulic retention time (HRT), amount of water to be filtered withdrawn, amount of water to be treated flowing in, amount of water to be filtered circulating, F / M ratio, aeration airflow, amount of coagulant added, physical cleaning conditions, frequency of chemical cleaning, concentration, immersion time, and other chemical cleaning conditions. In particular, including at least one selected from HRT, SRT, water level, amount of water to be filtered withdrawn, amount of water to be filtered supplied, amount of water to be filtered circulating, and the amount of water to be filtered allows for accurate reflection of the influence of the water retention state and the load on the separation membrane on the separation membrane characteristics. This makes it possible to understand the amount of filtered water components adhering to the surface of the separation membrane and the amount of filtered water components accumulating within the pores of the separation membrane, which vary depending on the operating conditions.

[0044] Climate information, while not limited to those specified here, preferably includes at least one of the following: temperature, precipitation, humidity, sunshine duration, wind speed, wind direction, atmospheric pressure, sea surface temperature, tide level, and ocean currents. This allows the influence of climate on water quality fluctuations of the treated or filtered water to be reflected in the separation membrane characteristics.

[0045] Organic matter indices represent the amount of organic substances in the filtered water, allowing for quantitative identification of components that cause adsorption of organic matter onto the separation membrane surface and pore blockage. In particular, including organic matter indices such as TOC, BOD, and COD as explanatory variables makes it possible to more accurately predict changes in membrane filtration resistance due to organic matter.

[0046] Here, the learning algorithms used in machine learning can be a variety of algorithms, including but are not limited to, various regression analysis methods, random forests, decision trees, k-nearest neighbors, neural networks, support vector machines, naive Bayes, regularization, logistic regression, Markov chains, and deep learning.

[0047] (Separation membrane characteristic estimation process) Next, the separation membrane characteristic estimation process will be explained using Figure 7. The input data is input to the learning model to estimate the separation membrane characteristics. The estimated separation membrane characteristics and the learning model that estimated the separation membrane characteristics are stored in the recording device as the updated learning model. Here, updating the learning model refers to the process of optimizing the hyperparameters in the learning model.

[0048] (Simulation prediction step) Furthermore, in the present invention, if membrane filterability parameters are estimated in the separation membrane characteristics estimation step, the invention may include a simulation prediction step that predicts the change in the separation membrane characteristics of the membrane over time based on the membrane filterability parameters estimated in the separation membrane characteristics estimation step.

[0049] Next, a simulation prediction step will be described, in which the time-dependent changes in the separation membrane characteristics are predicted based on the membrane filterability parameters estimated in the separation membrane characteristics estimation step. In the simulation prediction step of the separation membrane characteristics prediction method of the present invention, the time-dependent changes in membrane filtration resistance or intermembrane pressure are predicted when membrane filtration is continued while controlling the membrane filtration flow rate to a set value. Alternatively, the time-dependent changes in membrane filtration resistance or membrane filtration flow rate (flux) are predicted when membrane filtration is continued while controlling the intermembrane pressure to a set value.

[0050] Continuing membrane filtration while controlling the membrane filtration flow rate to a set value, or continuing membrane filtration while controlling the intermembrane differential pressure to a set value, means controlling the membrane filtration flow rate or intermembrane differential pressure to a predetermined set value. This includes not only methods of controlling the membrane filtration flow rate or intermembrane differential pressure to a constant value, but also methods of periodically or intermittently stopping filtration, or methods of continuously or intermittently changing the membrane filtration flow rate or intermembrane differential pressure, which change the set values ​​over time.

[0051] Methods for continuing membrane filtration while controlling the membrane filtration flow rate to a set value include installing a suction pump on the membrane permeation side of the separation membrane to obtain the membrane filtrate and controlling the suction pump with a flow inverter. Methods for continuing membrane filtration while controlling the intermembrane differential pressure to a set value include applying the pressure necessary for membrane filtration by pressurizing the filtered water side of the separation membrane or by utilizing the head difference, and then controlling this pressure.

[0052] In the simulation prediction process of the present invention, the separation membrane characteristics at an arbitrary time are determined by performing calculation steps 1, 2, and 3, which are described below. Simulation prediction is the process of performing calculations and calculations using input values ​​(such as time-series values ​​of membrane filtration flow rate and time-series values ​​of intermembrane differential pressure) and outputting output values ​​(such as predicted values ​​of the time change of separation membrane characteristics).

[0053] In calculation step 1, initial values ​​for the separation membrane characteristics are set, or calculated from the operating conditions information of the filtration system. Here, the initial values ​​for the separation membrane characteristics may be values ​​measured during the operation of the filtration system, or virtual values ​​based on the type of separation membrane may be entered. When setting initial values ​​for the separation membrane characteristics, previously acquired separation membrane performance may be set as the initial value. For example, in the case of an unused separation membrane, the separation membrane performance at the time of manufacture may be set as the initial value. Also, when calculating initial values ​​for the separation membrane characteristics, for example, the membrane filtration flow rate is calculated using equation (1) from the supply pressure to the separation membrane and the processing flow rate.

[0054] In calculation step 2, the amount of change in the separation membrane characteristics over a predetermined time Δt is calculated using the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters.

[0055] In calculation step 3, the separation membrane characteristics at time t + Δt are calculated from the amount of change calculated in calculation step 2. Note that calculation step 2 can be performed using the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters, and can be performed using calculation step 2a, calculation step 2b, or calculation step 2c, as described below.

[0056] Figure 8 shows an example of the time-dependent change prediction process of the present invention. In the time-dependent change prediction process shown in Figure 8, the time-dependent change in membrane filtration resistance is predicted while continuing membrane filtration while controlling the membrane filtration flow rate to a set value.

[0057] In the case of constant flow filtration, where the membrane filtration flow rate (flux) is controlled to a set value, the inputs for membrane filtration prediction calculations include at least the set value of the membrane filtration flow rate (flux), the initial value of the membrane filtration resistance, membrane filtration parameters, and water quality information of the filtered water (MLSS). The initial value of the membrane filtration resistance may be a measured value such as the membrane filtration resistance value at the start of operation or after chemical washing, or it may be a virtual value based on the type of separation membrane. The membrane filtration parameters may be values ​​estimated in the parameter estimation step, or values ​​calculated from existing membrane filtration tests. MLSS may be a measured value or a virtual value. Furthermore, the time change of the MLSS of the filtered water components may be predicted using an existing simulation model such as the IWA filtered water component model. The set value of the membrane filtration flow rate (flux) is the set value of the membrane filtration flow rate or membrane filtration flux, and may be a measured value or a virtual value, and may be a constant value or a value that changes continuously or intermittently. Furthermore, the set value of the intermembrane differential pressure may be an actual measured value or a virtual value, and may be a constant value or a value that changes continuously or intermittently.

[0058] When continuing membrane filtration while controlling the membrane filtration flow rate to a set value, the membrane filtration resistance value at any given time is determined by performing at least calculation step 2a, and at least one of calculation steps 2b and 2c, and calculation step 3, as described below.

[0059] In other words, there are three possible configurations for the calculation steps to predict the membrane filtration resistance value when continuing membrane filtration while controlling the membrane filtration flow rate to a set value: (a1) Calculation step 2a, calculation step 2b, calculation step 3 described later (a2) Calculation step 2a, calculation step 2c, calculation step 3 described later (a3) ​​Calculation step 2a, calculation step 2b, calculation step 2c, calculation step 3 described later In this case, in case (a3) ​​(see Figure 8), the membrane filtration resistance value at any given time can be determined.

[0060] In this, calculation step 2a is a calculation step to calculate the change in the amount of components of the filtered water adhering to the separation membrane surface within a predetermined time, wherein the calculation formula in calculation step 2a includes a term for the rate at which the components of the filtered water adhere to the separation membrane surface and a term for the rate at which the components of the filtered water adhering to the separation membrane surface detach from the separation membrane surface, wherein the rate of adhesion to the separation membrane surface is calculated using the intermembrane pressure difference value or membrane filtration flow rate (flux) value and the components of the filtered water and / or the membrane washing force value, and the rate of detachment from the separation membrane surface is preferably calculated using the intermembrane pressure difference value or membrane filtration flow rate (flux) value and the components of the filtered water adhering to the separation membrane surface and / or the pressure density of the components of the filtered water. This makes it possible to accurately predict the components of the filtered water adhering to the separation membrane surface at any given time.

[0061] Here, "compressive density of the water to be filtered adhering to the separation membrane surface" refers to the degree to which the water to be filtered adhering to the separation membrane surface becomes compacted by the pressure applied to it. By using this compressive density to calculate the water to be filtered adhering to the separation membrane surface, it is possible to predict the water to be filtered with greater accuracy. If the water to be filtered is sufficiently compacted, the water to be filtered adhering to the separation membrane surface will not be detached from the membrane surface by washing with aeration. By defining this sufficiently compacted water to be filtered as the non-detached water to be filtered, it is possible to accurately predict the water to be filtered adhering to the separation membrane surface using the water to be filtered and the non-detached water to be filtered instead of the compressive density.

[0062] Furthermore, it is even more preferable that the formula for determining the rate at which components of the filtered water adhering to the separation membrane surface detach from the membrane surface includes a term based on the membrane cleaning force value. Here, membrane cleaning force is the stress required to detach substances adhering to the separation membrane surface. The value of the membrane cleaning force is preferably calculated based on the value of the shear force generated on the membrane surface, the flow velocity of the filtered water on the membrane surface, a value calculated based on the shear force and the flow velocity, or the power value of the cleaning means (the power value of the cleaning means is, for example, the aeration airflow rate or the output value of the aeration blower when cleaning the separation membrane is performed by aeration from the bottom of the separation membrane). Alternatively, the value of the membrane cleaning force may be calculated or estimated from the results of an actual membrane filtration test. This makes it possible to add elements that do not depend on the performance of the separation membrane, such as aeration, as factors in determining the rate at which components of the filtered water adhering to the separation membrane surface detach from the separation membrane surface, thereby making it easier to reflect the operating conditions of the separation membrane apparatus.

[0063] Examples of mathematical formulas that satisfy these conditions are formulas (3) to (5) below, and in the present invention, it is recommended to follow formulas (3) to (5). However, the scope of the present invention is not limited to formulas (3) to (5).

[0064] However, (1 - Kτ1・τ) ≥ 0 and (τ - Kτ2・ΔP) ≥ 0.

[0065] Here, Xc is the amount of components of filtered water adhering to the surface of the separation membrane per unit membrane area [gC / m²]. ２ ], t is time [s], X is the MLSS [gC / m] of the membrane separation tank. ３ ], J is the membrane filtration flux [m / d], Kτ1 is the membrane cleaning power inhibition coefficient [-], γ is the component separation coefficient of the filtered water [1 / m / s], τ is the membrane cleaning power value [-], Kτ2 is the component friction coefficient of the filtered water [1 / Pa], ΔP is the intermembrane pressure differential [Pa], and η is the reciprocal of the density of the components of the filtered water [m ３ [gC], Dmax is the maximum pressure density [-] (usually 1), D is the pressure density [-], Xc,res is the amount of unpeeled components in the filtered water [gC / m] ２]. Also, "gC" represents the weight of carbon. Here, the first term on the right-hand side of equations (3) to (5) represents the rate at which the components of the filtered water adhere to the separation membrane surface, and the second term represents the rate at which the components of the filtered water detach from the separation membrane surface. Furthermore, as will be described later, if the membrane cleaning force value τ is expressed as a function of the aeration airflow rate or the output value of the aeration blower, it is possible to convert equations (3) to (5) into calculation formulas relating to the aeration airflow rate or the output value of the aeration blower rather than the membrane cleaning force value by substituting that function into τ.

[0066] Furthermore, as described above, if the change in the components of the filtered water adhering to the surface of the separation membrane is expressed by a calculation formula that represents the difference between the rate at which the components of the filtered water adhere to the separation membrane and the rate at which they detach, it can be expressed as a differential equation relating to the components of the filtered water adhering to the separation membrane. In that case, integration methods for solving this differential equation include the Euler method, the Runge-Kutta method, and the Runge-Kutta-Gill (RKG) method.

[0067] Furthermore, in calculation step 2b, the amount of substance derived from the filtered water components present in the pores of the separation membrane at any given time (pore blockage amount) is calculated. Calculation step 2b is a calculation step that calculates the amount of change in the pore blockage amount within a predetermined time, and it is preferable that the value of the change is calculated based on the intermembrane pressure difference value and / or the membrane filtration flow rate (flux) and / or the component values ​​of the filtered water adhering to the surface of the separation membrane and / or the value of the pore blockage amount. This makes it possible to accurately predict the pore blockage amount at any given time.

[0068] Examples of mathematical formulas that satisfy these conditions include formulas (6) to (9) below, and in the present invention, it is recommended to follow formulas (6) to (9). However, the scope of the present invention is not limited to formulas (6) to (9).

[0069]

[0070] Here, Xf is the amount of pore occlusion [gC / m]. ２ ], ψ is the mass transfer rate coefficient into the pores of the separation membrane [gC / m ２ [Pa / s], ε is the inhibition coefficient for mass transfer into the pores of the separation membrane [m ４ / gC ２ ], Kf is also the inhibition coefficient for mass transfer into the pores of the separation membrane [gC ２／３ / m ４／３ ], λ is the pore closure rate coefficient [s ｍ－１ ・gC １－ｂ / m ２＋ｍ－２ｂ ], m is the pore blockage flux dependence coefficient [-], and b is the pore blockage water component dependence coefficient [-].

[0071] Furthermore, in calculation step 2c, the amount of non-detached components of the filtered water adhering to the separation membrane surface at an arbitrary time is calculated. Here, the calculation formula in calculation step 2c is a calculation step that calculates the amount of change in the non-detached components of the filtered water adhering to the separation membrane surface within a predetermined time, and it is preferable that the value of the amount of change is calculated based on the pressure applied to the components of the filtered water adhering to the separation membrane surface and / or the value of the amount of components of the filtered water adhering to the separation membrane. This makes it possible to accurately predict the amount of change in the non-detached components of the filtered water adhering to the separation membrane surface at an arbitrary time. Here, the pressure applied to the components of the filtered water adhering to the separation membrane surface may be a pressure value derived from the components of the filtered water, or a differential pressure value between membranes may be used. Alternatively, instead of the amount of non-detached components of the filtered water, the pressure density may be expressed using the pressure density of the components of the filtered water.

[0072] Examples of mathematical formulas that satisfy these conditions include formula (10) or formula (11), and in the present invention, it is recommended to follow formula (10) or formula (11). However, the scope of the present invention is not limited to formula (10) or formula (11).

[0073]

[0074] Here, D is the pressure density [-], k1 is the consolidation rate coefficient [1 / (Pa / s)], Dmax is the maximum pressure density [-] (usually 1), ΔPc is the pressure value of the components of the filtered water [Pa], k is the rate coefficient of formation of non-separated components of the filtered water [1 / Pa / s], and l is the pressure dependence coefficient of the non-separated components of the filtered water [-].

[0075] Furthermore, in calculation step 3, the membrane filtration resistance value at any given time is calculated using the components of the filtered water adhering to the separation membrane, which were determined in calculation step 2a, and / or the amount of pore blockage determined in calculation step 2b, and / or the pressure density of the components of the filtered water adhering to the separation membrane, or the non-detached components of the filtered water, which were determined in calculation step 2c.

[0076] In this case, it is preferable that in calculation step 3, the membrane filtration resistance value at any given time is calculated based on the pressure applied to the components of the filtered water adhering to the surface of the separation membrane, and / or that the calculation includes a first- or second-order higher-order equation for the amount of pore blockage. This allows for a more accurate prediction of the membrane filtration resistance value at any given time. In this case, the pressure applied to the components of the filtered water adhering to the surface of the separation membrane may be the pressure value derived from the components of the filtered water adhering to the surface of the separation membrane, or the intermembrane differential pressure value may be used.

[0077] Examples of mathematical formulas that satisfy these conditions include formulas (12) to (14) below, and in the present invention, it is recommended to follow formulas (12) to (14). However, the scope of the present invention is not limited to formulas (12) to (14).

[0078]

[0079] Here, Rc is the membrane filtration resistance [1 / m] derived from the components of the filtered water adhering to the separation membrane surface, α is the resistance coefficient of the components of the filtered water [m / gC], a is the pressure dependence coefficient of the components of the filtered water [m / gC / Pa], Rf is the membrane filtration resistance [1 / m] derived from pore blockage, and β is the pore blockage filtration resistance coefficient [m ２ / gC １．５ ], R is the membrane filtration resistance [1 / m], and Rm is the initial value of the membrane filtration resistance [1 / m].

[0080] Furthermore, in calculation step 4, the intermembrane differential pressure value at an arbitrary time is calculated using the amount of filtered water components attached to the separation membrane obtained in calculation step 2a and / or the amount of pore blockage obtained in calculation step 2b and / or the pressure density of the filtered water components attached to the separation membrane obtained in calculation step 2c. One method for doing this is to substitute equations (12) and (13) into equation (14), and then use the resulting formula obtained by substituting these into equation (1).

[0081] Table 1 summarizes the explanations of the symbols used in each formula.

[0082]

[0083] In this invention, when predicting the separation membrane characteristics while continuing membrane filtration with the membrane filtration flow rate controlled to a set value, the time-dependent changes in membrane filtration resistance and / or the time-dependent changes in intermembrane differential pressure are determined by repeatedly performing the calculation step while updating the time.

[0084] (Control of operating conditions) In water treatment using separation membranes, as the amount of water filtered through the membrane increases, the amount of contaminants adhering to the membrane surface and inside the membrane pores increases, leading to increased membrane fouling, which in turn causes a decrease in the amount of treated water or an increase in membrane filtration resistance. Furthermore, if the operating conditions are unsuitable or unstable (such as fluctuations in raw water quality or water temperature, or after chemical washing of the membrane), the water quality of the filtered water tends to deteriorate, meaning that the membrane filtration parameters, which are the characteristics of the filtered water when filtered through the membrane, deteriorate.

[0085] By using machine learning to estimate membrane filtration parameters from water quality information, operating condition information, and climate information obtained from the filtration system, and comparing the predicted membrane filtration resistance or intermembrane pressure using the membrane filtration parameters, the number of operating days and its rate of change with a pre-set reference range, the system evaluates the influence of the filtered water condition and operating conditions, and can output an alarm and / or control to optimize operating conditions before the intermembrane pressure starts to rise.

[0086] Furthermore, when predicting membrane filtration resistance or intermembrane pressure using membrane filtration parameters estimated by machine learning, it is possible to arbitrarily set the operating conditions of the filtration system, such as filtration flux, physical washing conditions, and chemical washing conditions, and perform simulation predictions. This makes it easy to perform simulations under various operating conditions, enabling optimization of operating conditions according to the membrane filtration properties of the water to be filtered, prediction of the number of operating days, and identification of the causes of sudden changes in membrane filtration resistance or intermembrane pressure. Here, the number of operating days is the number of operating days until the day (predicted day) when the set value is reached in the time-dependent changes of membrane filtration resistance and / or membrane filtration pressure and / or membrane filtration flow rate (flux) obtained by the simulation prediction.

[0087] In this embodiment, evaluation results and / or prediction results are used to detect changes in membrane filtration parameters, which represent the membrane filtrationability of the filtered water, and to control the water treatment conditions in order to prevent increases in intermembrane pressure differential, deterioration of filtered water quality, etc. The items to be controlled are at least one of the following:

[0088] A. Inflow concentration and flow rate of water to be treated B. Filtration flow rate C. Filtration time or filtration stop time D. Aeration airflow rate or aeration time E. Amount of nutrients added F. Amount of chemicals added G. Amount of water to be filtered H. Amount of water to be returned for treatment I. Operating conditions of the pretreatment process J. Operating conditions of the posttreatment process K. Conditions for adjusting the temperature of water to be filtered L. Operating conditions of the membrane element M. Conditions for cleaning the membrane element N. Conditions for cleaning the diffuser pipe

[0089] (Separation membrane characteristic prediction program) The separation membrane characteristic prediction program of the present invention is typically stored on a computer-readable recording medium together with a control and management system such as a PLC or DCS that is usually installed in a filtration device, or it is typically stored on a recording medium of an on-premise server or cloud server installed at any location, after retrieving operating data via the internet using a remote monitoring device from the control and management system. Furthermore, the separation membrane characteristic prediction program of the present invention has the following means, as shown in Figure 9 or Figure 10, for example.

[0090] Figure 9 shows an example of a separation membrane characteristic prediction program when the separation membrane characteristics are directly estimated in the separation membrane characteristic estimation process. A separation membrane characteristic prediction program 47 is installed in the computer 38 to enable it to function as an input means 39 for inputting water quality information of the water to be treated, water quality information of the water to be filtered, operating condition information of the filtration device, climate information, etc.; a learning model construction means 40 for constructing a learning model; a separation membrane characteristic estimation means 41 for estimating the change in the separation membrane characteristics over time using the learning model; a separation membrane characteristic recording means 42 for recording the estimated separation membrane characteristics; a convergence determination means 44 for determining the convergence of the calculation; a calculation result recording means 45 for recording the calculation results; and a calculation result output means 46 for outputting the calculation results.

[0091] Figure 10 shows an example of a separation membrane characteristics prediction program used in the separation membrane characteristics estimation process to estimate the separation membrane characteristics that indicate the membrane filtrationability of the filtered water. The computer 38 is equipped with a separation membrane characteristics prediction program 47 that enables it to function as an input means 39 for inputting water quality information of the water to be treated, water quality information of the filtered water, operating conditions information of the filtration device, climate information, set value of the membrane filtration flow rate, initial value of the membrane filtration resistance, etc.; a learning model construction means 40 for constructing a learning model; a separation membrane characteristics estimation means 41 for estimating the change over time of the separation membrane characteristics that indicate the membrane filtrationability of the filtered water using the learning model; a separation membrane characteristics recording means 42 for recording the estimated separation membrane characteristics; a simulation prediction means 43 for calculating the change over time of the separation membrane characteristics based on the estimated separation membrane characteristics that indicate the membrane filtrationability of the filtered water; a convergence determination means 44 for determining the convergence of the calculation; a calculation result recording means 45 for recording the calculation results; and a calculation result output means 46 for outputting the calculation results.

[0092] In particular, the learning model construction means and the separation membrane characteristic estimation means can be installed in an external prediction system (e.g., a cloud server) independent of the filtration device. The filtration device can then send and receive data such as water quality information to and from the prediction system via communication means, and receive prediction results for separation membrane characteristics from the prediction system to control operating conditions. In this case, the prediction system can build a learning model by integrating data obtained from multiple filtration devices, enabling more accurate predictions.

[0093] (Example 1) An embodiment of the present invention is shown with reference to Figures 6 and 7. Table 2 shows the item conditions for the explanatory variables used when constructing the learning model in this embodiment. The input data items in the separation membrane characteristic estimation step were the same as the item conditions in the learning model construction step described above. The target of estimation was the membrane filterability parameter, specifically the parameters necessary to calculate the component resistance of the filtered water. First, a learning model for each item condition was constructed using the processing flow in Figure 6. Furthermore, using the processing flow in Figure 7, the parameters necessary to calculate the component resistance of the filtered water were estimated using the learning model constructed for each item condition.

[0094] Figure 11 shows the estimation accuracy of the parameters. Here, estimation accuracy is an indicator that shows the error between the result estimated using the learning model and the correct value, and a higher estimation accuracy indicates a smaller error from the correct value.

[0095] Figure 11 shows that the estimation accuracy is low in conditions 1, 2, and 3, where any one of MLSS, viscosity, or water temperature is excluded. On the other hand, the estimation accuracy is greatly improved in condition 4, which includes all of these items, and it can be seen that the estimation accuracy is further improved by adding items in conditions 5, 6, and 7.

[0096] In other words, in this invention, at least three parameters—MLSS, viscosity, and water temperature—are effective in maintaining a certain level of estimation accuracy. Furthermore, the estimation accuracy can be improved by adding at least one parameter selected from pH, HRT, and ORP. Particularly preferable is the inclusion of all three parameters: pH, HRT, and ORP, which provides the highest possible estimation accuracy.

[0097] (Example 2) An example including a simulation prediction process is shown using Figures 6, 7, 8, and 10. In a filtration apparatus using a membrane separation activated sludge method as shown in Figure 2, water temperature, viscosity, MLSS water quality information obtained by the filtered water quality sensor 18, and HRT, which is the operating condition information of the filtration apparatus, are used as input data, and the objective variable is the membrane filterability parameter. Here, the membrane filterability parameter is a parameter related to calculating the component resistance of the filtered water and a parameter related to calculating the rate of progression of blockage of the separation membrane pores.

[0098] The input data and the target variable, the membrane filterability parameter, were obtained by processing time-series data acquired at 1-second intervals into a daily average value. These data were input to the input means 39 in Figure 10, and a learning model was constructed by the learning model construction means 40 according to the processing flow in Figure 6. The constructed learning model was recorded in the separation membrane characteristic recording means 42.

[0099] Next, input data not used in the learning process was input to the recorded learning model, and the membrane filterability parameters were estimated by the separation membrane characteristic estimation means 41. The estimated membrane filterability parameters were recorded in the separation membrane characteristic recording means 42, and the learning model was updated and recorded in the separation membrane characteristic recording means 42 as well. The updated learning model was used for the next estimation.

[0100] Next, the simulation prediction means 43 predicted the change in separation membrane characteristics over time based on the estimated membrane filterability parameters. In the simulation prediction step, following the processing flow in Figure 8, the membrane filtration resistance was predicted using the aforementioned (a3) ​​as the configuration for the calculation step.

[0101] In calculation step 1, initial values ​​for the separation membrane characteristics were set. Specifically, the measured value immediately after chemical washing was used as the initial value for the membrane filtration resistance.

[0102] In calculation step 2, the amount of change in the separation membrane characteristics over a predetermined time Δt was calculated using the initial values ​​set in calculation step 1 and the membrane filterability parameters estimated by the separation membrane characteristic estimation means 41. In this embodiment, calculation step 2 includes calculation step 2a, calculation step 2b, and calculation step 2c.

[0103] In calculation step 2a, the change in the amount of components of the filtered water adhering to the surface of the separation membrane was calculated using equations (3) to (5) above. In calculation step 2b, the change in the amount of substance derived from the components of the filtered water present in the pores of the separation membrane (pore blockage amount) was calculated using equations (6) to (9) above. In calculation step 2c, the change in the amount of unpeeled components of the filtered water adhering to the surface of the separation membrane was calculated using equation (10) or (11) above.

[0104] In calculation step 3, the separation membrane characteristics at an arbitrary time t + Δt were calculated from the change amount calculated in calculation step 2. Specifically, the membrane filtration resistance value at an arbitrary time was calculated using equations (12) to (14) above, using the components of the filtered water adhering to the separation membrane obtained in calculation step 2a, the amount of pore blockage obtained in calculation step 2b, and the non-detached components of the filtered water obtained in calculation step 2c.

[0105] Furthermore, in this embodiment, the calculated membrane filtration resistance was converted to the intermembrane differential pressure using equations (1) and (2) above. Calculation steps 1 to 3 were continuously repeated until the operating differential pressure of the filtration device increased by +3 [kPa] from the initial operating pressure.

[0106] Figure 12 shows a comparison of the time-dependent change in the intermembrane pressure predicted in the simulation prediction process with the measured value of the intermembrane pressure measured in the actual filtration device. From Figure 12, it can be seen that the intermembrane pressure predicted in the simulation prediction process agrees well with the measured value, confirming that the method of the present invention can accurately predict the time-dependent change in separation membrane characteristics.

[0107] Thus, according to the present invention, membrane filterability parameters can be estimated by machine learning from water quality information of the filtered water and operating conditions information of the filtration device, and the changes in separation membrane characteristics over time can be predicted by simulation based on the estimated membrane filterability parameters. This enables highly accurate predictions that reflect the state of the filtered water, which changes moment by moment, and realizes stable operation of the filtration device and determination of appropriate operating conditions.

[0108]

[0109] Although various embodiments have been described above with reference to the drawings, it goes without saying that the present invention is not limited to these examples. It is clear to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of the present invention. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the invention.

[0110] This application is based on a Japanese patent application (Patent Application No. 2025-008896) filed on January 22, 2025, the contents of which are incorporated herein by reference.

[0111] 1. Water to be treated 2. Pre-treatment section 3. Water quality sensor for treated water 4. Water to be filtered supply pump 5. Water to be filtered pressure gauge 6. Water to be filtered flow meter 7. Separation membrane module 8. Concentrated water 9. Filtered water 10. Water to be treated tank 11. Water to be treated supply pump 12. Water to be treated flow meter 13. Oxygen-free tank 14. Membrane separation tank 15. Aeration blower 16. Air flow meter 17. Diffuser 18. Water quality sensor for filtered water 19. Suction pump 20. Filtered water pressure gauge 21. Filtered water flow meter 22. Filtered water tank 23. Water to be treated extraction pump 24. Water to be treated return pump 25. Water to be treated return flow meter 26. Washing chemical tank 27. Chemical addition pump 28. Washing chemical flow path switching valve 29. Back pressure washing pump 30. Drainage line 31. Water to be filtered tank 32a Separation membrane element 32b Separation functional layer 32c Substrate 32d Frame 33 Filtered water 34 Separation membrane 35 Pore blockage 36 Filtered water components adhering to the surface of the separation membrane and removable by washing 37 Filtered water components adhering to the surface of the separation membrane and not removable by washing 38 Computer 39 Input means 40 Learning model construction means 41 Separation membrane characteristic estimation means 42 Separation membrane characteristic recording means 43 Simulation prediction means 44 Convergence determination means 45 Calculation result recording means 46 Calculation result output means 47 Separation membrane characteristic prediction program

Claims

1. A method for constructing a learning model for predicting the characteristics of a separation membrane in a filtration system that filters water to be filtered through a separation membrane to obtain treated water, comprising a learning model construction step of constructing a learning model by machine learning with data including at least one of the following: water quality information of the water to be filtered, operating condition information of the filtration system, and climate information as explanatory variables, and the explanatory variables and the corresponding separation membrane characteristics as objective variables.

2. A method for constructing a learning model according to claim 1, wherein the explanatory variables include at least water temperature, viscosity, and turbidity component index as water quality information of the filtered water.

3. A method for constructing a learning model according to claim 2, wherein the explanatory variable further includes at least one selected from pH, ORP, and organic matter index as water quality information of the filtered water, and HRT, SRT, water level, amount of water to be filtered, amount of water to be filtered, amount of water to be filtered, and amount of water to be treated as operating condition information of the filtration device.

4. A method for constructing a learning model according to any one of claims 1 to 3, wherein the separation membrane characteristics include membrane filterability parameters involved in calculating at least one of the following: the resistance of the components of the filtered water, the resistance of the non-separated components of the filtered water, the rate of progression of blockage of the separation membrane pores, and the resistance due to blockage of the separation membrane pores.

5. The method for constructing a learning model according to claim 4, wherein the separation membrane characteristics further include membrane filterability parameters involved in calculating at least one of the following: the rate of component adhesion in the filtered water, the rate of component detachment in the filtered water, and the rate of non-detachment component formation in the filtered water.

6. A learning model constructed by the method for constructing a learning model described in any one of claims 1 to 5.

7. A method for predicting the characteristics of a separation membrane in a filtration apparatus that filters water to be filtered with a separation membrane to obtain treated water, comprising a separation membrane characteristics estimation step of estimating the change in the separation membrane characteristics over time from the input data, using a learning model described in claim 6, with input data including at least one of the water quality information of the water to be filtered, the operating conditions information of the filtration apparatus, and climate information.

8. The separation membrane characteristic prediction method according to claim 7, further comprising a simulation prediction step of predicting the change in separation membrane characteristics over time by simulation based on the membrane filterability parameters estimated in the separation membrane characteristic estimation step.

9. The separation membrane characteristic prediction method according to claim 8, wherein the simulation prediction step is calculated in the following calculation steps. Calculation step 1: Set initial values ​​for the separation membrane characteristics, or calculate initial values ​​for the separation membrane characteristics from the operating condition information of the filtration device. Calculation step 2: Use the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters to calculate the amount of change in the separation membrane characteristics that occurred during a predetermined time Δt. Calculation step 3: Calculate the separation membrane characteristics at an arbitrary time t + Δt from the amount of change calculated in calculation step 2.

10. The separation membrane characteristics prediction method according to claim 7, wherein the learning model updates the learning model after estimating the separation membrane characteristics from the input data.

11. The separation membrane characteristics prediction method according to claim 7, wherein the input data includes at least water temperature, viscosity, and turbidity component index as water quality information of the filtered water.

12. The separation membrane characteristics prediction method according to claim 11, wherein the input data further comprises at least one selected from the following: pH, ORP, and organic matter index as water quality information of the filtered water, and HRT, SRT, water level, amount of water to be filtered, amount of water to be filtered, amount of water to be filtered, and amount of water to be treated as operating condition information of the filtration device.

13. A method for operating a filtration apparatus that filters water to be filtered with a separation membrane to obtain treated water, comprising the steps of: predicting the separation membrane characteristics using the separation membrane characteristics prediction method described in claim 8 or 9; and controlling the operating conditions of the filtration apparatus based on the predicted separation membrane characteristics.

14. The method for operating a filtration apparatus according to claim 13, wherein the operating conditions are controlled when the predicted separation membrane characteristics deviate from a predetermined reference range.

15. A method for operating a filtration apparatus according to claim 13, wherein the predicted date on which the separation membrane characteristics deviate from a predetermined reference range is calculated from the prediction results of the simulation prediction step, and the operating conditions are controlled when the number of days until the predicted date deviates from a predetermined reference range.

16. A program for predicting the characteristics of a separation membrane in a filtration apparatus that filters water to be filtered through a separation membrane to obtain treated water, comprising: a learning model construction means for constructing a learning model by machine learning using data including at least one of the following: water quality information of the water to be filtered, operating condition information of the filtration apparatus, and climate information as explanatory variables, and the corresponding separation membrane characteristics as objective variables; an input means for inputting data including at least one of the following: water quality information of the water to be filtered, operating condition information of the filtration apparatus, and climate information; and a separation membrane characteristic estimation means for estimating the change in the separation membrane characteristics over time from the data using the learning model.

17. The separation membrane characteristic prediction program according to claim 16, which further causes the computer to function as a simulation prediction means for predicting the change in separation membrane characteristics over time by simulation based on the membrane filterability parameters estimated by the separation membrane characteristic estimation means.

18. The separation membrane characteristic prediction program according to claim 17, wherein the simulation prediction means is calculated in the following calculation steps. Calculation step 1: Set initial values ​​for the separation membrane characteristics, or calculate initial values ​​for the separation membrane characteristics from the operating condition information of the filtration device. Calculation step 2: Use the initial values ​​set or calculated in calculation step 1 and the membrane filterability parameters to calculate the amount of change in the separation membrane characteristics that occurred during a predetermined time Δt. Calculation step 3: Calculate the separation membrane characteristics at an arbitrary time t + Δt from the amount of change calculated in calculation step 2.

19. The separation membrane characteristic prediction program according to claim 16, wherein the learning model estimates the separation membrane characteristics from the input data and then updates the learning model.

20. The separation membrane characteristic prediction program according to claim 16, wherein the explanatory variables and the input data include at least water temperature, viscosity, and turbidity component index as water quality information of the filtered water.

21. The separation membrane characteristic prediction program according to claim 20, wherein the explanatory variables and the input data further include at least one selected from the following: pH, ORP, and organic matter index as water quality information of the filtered water, and HRT, SRT, water level, amount of water to be filtered, amount of water to be filtered, amount of water to be filtered, and amount of water to be treated as operating condition information of the filtration device.

22. The separation membrane characteristic prediction program according to claim 16, wherein the separation membrane characteristics include parameters involved in calculating at least one of the following: the resistance of the components of the filtered water, the resistance of the non-separated components of the filtered water, the rate of progression of pore blockage in the separation membrane, and the resistance due to pore blockage in the separation membrane.

23. The separation membrane characteristic prediction program according to claim 22, wherein the separation membrane characteristics further include parameters involved in calculating at least one of the following: the rate of component adhesion to the filtered water, the rate of component detachment from the filtered water, and the rate of non-detachment component formation in the filtered water.

24. A prediction system implementing the separation membrane characteristic prediction program described in claim 16, comprising: communication means for receiving data from a filtration device including at least one of water quality information of the water to be filtered, operating condition information of the filtration device, and climate information; and transmission means for transmitting the separation membrane characteristics estimated by the separation membrane characteristic prediction program to the filtration device.

25. A prediction system implementing the separation membrane characteristic prediction program described in claim 17, comprising: communication means for receiving data from a filtration device including at least one of water quality information of the water to be filtered, operating condition information of the filtration device, and climate information; and transmission means for transmitting to the filtration device the membrane filterability parameter estimated by the separation membrane characteristic estimation means or the change over time of the separation membrane characteristics predicted by the simulation prediction means.

26. A filtration apparatus for filtering water to be filtered with a separation membrane to obtain treated water, comprising: a separation membrane module; a sensor for acquiring water quality information of the water to be filtered; communication means for transmitting the water quality information acquired by the sensor to a prediction system described in claim 24 or 25 and receiving prediction results of separation membrane characteristics from the prediction system; and operation control means for controlling operating conditions based on the prediction results received by the communication means.

27. A computer-readable recording medium that records a separation membrane characteristic prediction program according to any one of claims 16 to 23.

Images