JP2019215743A - Method for manufacturing database, and water treatment or sludge treatment system - Google Patents

Abstract

To provide a method for manufacturing database which constructs a model capable of returning highly accurate output including image information in a short time as database used when a model for acquiring an arbitrary prediction value in water treatment or sludge treatment using a mechanical learning algorithm is constructed, and to provide a water treatment or sludge treatment system.SOLUTION: A method for manufacturing database for learning data is used when a model for acquiring an arbitrary prediction value in water treatment or sludge treatment using a mechanical learning algorithm is constructed. A batch or continuous test device using actual raw water or actual sludge acquires learning data including a static or continuous image acquired by at least an image acquisition device, and uses the learning data as database. The data obtained by operation of an actual plant is also added to the above database.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a method for manufacturing a database including images obtained by a camera and the like, which is suitable for use in constructing a model for obtaining a predicted value in water treatment or sludge treatment using a machine learning algorithm, and water treatment. Or, it relates to a sludge treatment system.

  In recent years, with the dramatic improvement in computer computing capacity, the improvement of the communication infrastructure environment, and the evolution of analysis algorithms, the third artificial intelligence boom has emerged. With the development of the communication infrastructure environment, it is possible to easily acquire a large amount of data, and unlike the artificial intelligence boom, expectations for its use in the real world are increasing.

  Machine learning / artificial intelligence is expected to be applied to various fields, and in the field of water treatment / sludge treatment, there is a search for a method of utilizing the same. One method of utilizing machine learning / artificial intelligence in the field of water treatment / sludge treatment is to predict a target variable in an unknown environment, for example, in the future in terms of time. One of the methods for predicting a target variable in an unknown environment is an analytical approach, which analyzes a physical phenomenon theoretically and formulates it. On the other hand, prediction of objective variables in unknown environments by machine learning / artificial intelligence is based on statistical analysis of the correlation between inputs and outputs without analyzing the theoretical relationship between the inputs and outputs in accordance with natural laws. In this case, a regression equation is created as an example of a relational equation.

  In the field of water treatment / sludge treatment, it is sometimes difficult to theoretically analyze the phenomena that occur, and machine learning / artificial intelligence that can predict the output from the input without elucidating the theoretical background is It is expected that there will be many aspects that can be effectively used.

  As mentioned above, the artificial intelligence boom itself has reached its third boom this time. The prominent algorithm of the second artificial intelligence boom that occurred in the late 1980s was neural networks. During the first artificial intelligence boom, the addition of a hidden layer to the perceptron, which consisted of only the input and output layers, and the establishment of a learning method called the "backpropagation method," Artificial intelligence boom. Eventually, the second artificial intelligence boom will be abolished, but it is said that one of the factors was the lack of satisfactory prediction accuracy depending on the task. In the third artificial intelligence boom, Deep Learning, which has multiple layers of hidden layers in the neural network, plays a leading role. That is part of the boom. In the present specification, a neural network composed of only three layers, an input layer, one hidden layer, and an output layer, is classified as a neural network, and a layer composed of two or more hidden layers is classified as deep learning. In artificial intelligence and machine learning, artificial intelligence is a broader concept, and there is machine learning as one of the means for creating artificial intelligence.

JP 2013-94686 A JP-A-6-328092

  Even in the field of water treatment / sludge treatment, utilization of machine learning / artificial intelligence has the above advantages. When creating a model that obtains an output from an input with machine learning / artificial intelligence, it is important to avoid overfitting that gives high accuracy only to learning data. Here, the learning data is synonymous with the teacher data / training data. What is required of machine learning / artificial intelligence is that high prediction accuracy can be obtained for data other than learning data (test data), and a sufficiently large amount of learning data is prepared as one of the methods to avoid over-learning It is mentioned. Artificial intelligence has once again attracted attention due to deep learning, but one of the factors is its high prediction accuracy, which has made it easier to collect large amounts of data by improving the communication infrastructure environment, etc. This has also contributed greatly.

  In other words, it can be said that building a database having a large amount of data is indispensable in order to build a practical-level model with machine learning / artificial intelligence that has been attracting attention again. Taking Deep Learning as an example, the hidden layers of the neural network are multi-layered, so the number of parameters that need to be optimized is increasing. If the learning data is small, over-learning may occur. Therefore, the need for a database having a large amount of data can be said to be a new problem that was not found in the age of the second artificial intelligence boom.

  However, the recent increase in the degree of attention to machine learning / artificial intelligence and the enhancement of the degree of communication infrastructure environment have been recent years, so we have not assumed the use of machine learning / artificial intelligence in recent years. In the field of water treatment / sludge treatment, the data required for model building is not collected or is collected, but the frequency is extremely low (for example, once / day). Is sufficient, but it is often not possible to create high-precision regression equations with machine learning / artificial intelligence under the current conditions, such as the lack of data sets due to the short days of data storage. .

Further, from the viewpoint of creating a prediction formula (= regression formula) from the data, the problems of the conventional invention will be described below.
Patent Literature 1 describes that in order to determine an optimal coagulant injection rate in a water purification plant, raw water quality is measured, and a multiple regression analysis is performed using the coagulant injection rate as an objective variable. At that time, it is shown that the past water quality data and the coagulant injection rate are accumulated in the database. However, this invention has two problems. The first is that the machine learning algorithm is a multiple regression analysis. In the multiple regression analysis, it is generally known that the obtained regression equation is not stable when there is collinearity (correlation) between explanatory variables. Further, since the multiple regression analysis is a linear regression method, if there is a non-linear relationship between the explanatory variable group and the objective variable, there is a problem that a highly accurate regression equation cannot be obtained. Furthermore, the present invention has a problem that it takes a lot of time to store sufficient data for performing multiple regression analysis. Since the regression equation is updated while accumulating data in an actual plant, for example, there is a problem that a newly installed facility cannot be controlled at the beginning of operation because the database itself does not exist.

  Patent Document 2 discloses a method of accumulating neural network teacher data (learning data) using an actual plant. However, this invention also has a problem that it takes time to accumulate data because an actual plant is used, and cannot be controlled when a new plant is in operation because there is no data in the first place. Furthermore, since the machine learning algorithm is a neural network having a single hidden layer, there is a problem in its accuracy as described above.

  In addition, another issue that must be considered when using machine learning / artificial intelligence in the field of water treatment / sludge treatment is described. In the maintenance and management of a plant in the field of water treatment / sludge treatment, there are cases where the treatment state of a plant is judged not only by numerical values that can be obtained by sensors and analytical instruments, but also by the senses of the operator, particularly by eyes.

  For example, when judging excess or deficiency of the flocculant from the size and shape of the flocculant generated when the inorganic flocculant is injected in the water treatment, or in the sludge flocculation floc generated when the inorganic flocculant or the polymer is injected in the sludge treatment In some cases, it is determined whether the amount of the medicine is excessive or insufficient based on the size and shape of the medicine. In addition, when the activated sludge method is used in water treatment, the biota of the activated sludge changes due to over-aeration or overload, and the activated sludge is dispersed or dismantled (does not cause bulking) In order to judge the above, there is a case where the amount and state of small flocculated flocs of activated sludge having poor sedimentation are checked by looking at the overflow water in the sedimentation tank.

  Furthermore, in water treatment using microorganisms, there is a biological carrier method for attaching microorganisms using activated carbon or a plastic material as a carrier for the purpose of efficient biological reaction and solid-liquid separation. The suitability of the attachment / growth state may be confirmed visually.

  That is, in the field of water treatment / sludge treatment, even if only the numerical data of the sensors and the like are learned, there is a possibility that the operation management level that currently relies on human visual information and experience cannot be achieved.

  The present invention has been made in view of the above points, its purpose is to use a machine learning algorithm to build a model for obtaining any predicted value in water treatment or sludge treatment, as a database used when building a model, It is an object of the present invention to provide a database manufacturing method and a water treatment or sludge treatment system capable of constructing a model capable of returning a highly accurate output including image information in a short time.

The present invention is a method for manufacturing a database for learning data used when constructing a model for obtaining an arbitrary predicted value in water treatment or sludge treatment using a machine learning algorithm, comprising: real raw water or real sludge. It is characterized in that learning data including at least still images or continuous images acquired by an image acquisition device is acquired by a batch or continuous test device using.
That is, in the present invention, a batch test apparatus using actual raw water or sludge, or a continuous test apparatus installed near an actual plant or the like and operated using actual raw water or sludge is used in advance to construct a database. Get the data of
As learning data to be acquired, it is preferable to use measurement data from various sensors and the like in addition to still or continuous images from the image acquiring device.
Here, an important parameter in the process can be selected as a prediction target from which the prediction value is to be obtained. If it is a water treatment system, sludge treatment process such as coagulant injection rate, flocculated floc shape, turbidity of treated water, chromaticity of treated water, methane gas generation, TOC (total organic carbon) or COD _{Cr of} treated water, etc. If so, methane gas generation amount, flocculent floc shape, flocculant injection rate, sludge concentration, water content of dewatered cake, etc. The prediction target may be any parameter that is valuable to know in advance in the process, and can be freely set each time according to the target process or the prediction purpose.
By using the present invention, many settings can be tried by batch or continuous test equipment, and a large amount of data can be acquired in a short time as compared with storing data in an actual plant. it can.
In addition, since a database is created using batch or continuous test equipment, many conditions can be used at once in batch tests, and a wide range of conditions can be applied without worrying about the effect on processing performance because it is not an actual plant. Can be set.
That is, in a situation where the number of data sets is not sufficient, a database having a sufficient number of data sets can be manufactured. Here, the data set refers to a pair of a target variable and an explanatory variable group, and a set of the data sets is a database. Various data are elements that constitute a data set.
As a result, the extrapolation range of the learning data can be reduced, and a model that returns a highly accurate output in more situations can be constructed.
That is, one of the fundamental problems of machine learning / artificial intelligence is that it is almost impossible to predict extrapolated data. The extrapolated data is data beyond the range of the learning data. For example, if the range adopted as the learning data of a certain parameter is 0 to 10, -1 and 11 are outside the range of the learning data and extrapolated. Data. Alternatively, even if the range of the learning data is 0 to 10, if there is no data in the range of 5 to 7, the data of 5 to 7 are also extrapolated data. If an attempt is made to construct a database for learning data for model construction using only data obtained from an actual plant, extrapolated data is likely to be generated, and actual raw water and actual sludge will be in a situation where no data has been obtained before. If it does, the output will be less reliable. In recent years, sudden changes in the weather typified by guerrilla downpours are common, and it is inevitable in the era that in addition to control during normal times, provision should be made so that good control can be performed even in the event of sudden abnormalities. According to the present invention, the possibility can be increased as described above.
Further, the present invention includes a model including a phenomenon that has been visually judged by a plant operator by including still or continuous image data acquired by an image acquisition device such as a camera in the database. Becomes possible.

Further, in addition to the above features, the present invention is characterized in that data obtained from the operation of the actual plant for treating the actual raw water or the actual sludge is also added to the database.
By adding data obtained from an actual plant to this database, the accuracy can be further improved.

In addition, the present invention is characterized in that, in addition to the above-described features, among the items to be measured or set as conditions by the batch or continuous test device, at least items that can be measured by an image acquisition device or a sensor installed in an actual plant are included. And
The learning data obtained by the batch or continuous test apparatus matches the data obtained from the operation of the actual plant, and the two can be combined to form a more effective database.
By the way, as a machine learning algorithm used when constructing a model using the database, SVR method (support vector regression method), PLS method (Partial Least Squares method), Deep Learning method, random forest method, Alternatively, a decision tree method or the like is preferable. If the above methods are used without using a conventional neural network having a single hidden layer or multiple regression analysis, it is possible to construct a model that returns a highly accurate output. In particular, the Deep Learning method is suitable for analyzing image data.

Further, in addition to the above features, the present invention is characterized in that the still or continuous image acquired by the batch or continuous test apparatus is for a flocculated floc.
This makes it possible to construct a model that returns a more accurate output, for example, in determining whether the amount of the inorganic coagulant or polymer injected in water treatment or sludge treatment is excessive or insufficient.

Further, in addition to the above features, the present invention is characterized in that the still or continuous image acquired by the batch or continuous test apparatus is intended for a water treatment carrier.
This makes it possible to construct a model that returns a more accurate output, for example, in water treatment or sludge treatment using microorganisms.

  The present invention also resides in a water treatment or sludge treatment system having a control unit for controlling a model constructed based on the database manufactured by the above-described database manufacturing method.

  According to the method for manufacturing a database according to the present invention, as a database used when constructing a model for obtaining an arbitrary predicted value in water treatment or sludge treatment using a machine learning algorithm, the accuracy including image information, It is possible to obtain a database that can build a model that can return a high output in a short time.

It is a figure which shows the database preparation procedure of the learning data for model construction by a batch test. 1 is a schematic configuration diagram of a water purification plant (water treatment system) 1-1 configured using a built model. It is a figure which shows the database preparation procedure of the learning data for model construction by a continuous test apparatus. It is a figure which shows the database preparation procedure of the learning data for model construction by a continuous test apparatus. FIG. 2 is a schematic configuration diagram of a wastewater treatment plant 1-2 using a carrier-input methane fermentation method configured using a constructed prediction model.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[First Embodiment]
FIG. 1 is a diagram illustrating a method of manufacturing a database for learning data according to the first embodiment of the present invention, and is a diagram illustrating a procedure for creating a database of learning data for model construction based on a batch test. Specifically, an example of a procedure of creating a database for learning data used when constructing a model for predicting an optimal coagulant injection rate in a water purification plant is shown.

  In the present embodiment, first, actual raw water (actual raw water) to be introduced into a landing well of an actual water purification plant is obtained (step 1-1).

  Next (or prior to obtaining the actual raw water), a number of conditions are set for the actual raw water using parameters that are candidates for explanatory variables. As parameters to be set, many parameters considered to affect aggregation, for example, water temperature, M-alkalinity, pH, turbidity, chromaticity, TOC, ultraviolet absorbance, fluorescence intensity, weather information, etc. are set. (Step 1-2).

  Next, a jar test is performed using each of the setting conditions. At this time, the coagulated floc generated during the jar test is photographed using a camera (step 1-3).

  The optimum coagulant injection rate under each condition is determined based on the state of the coagulated floc and the quality of the treated water (eg, turbidity, residual aluminum concentration, etc.) after the jar test (step 1-4). Here, in the jar test, for example, actual raw water is put into a plurality of beakers, and an agglutination test is performed using a jar tester under the same stirring conditions and the same addition timing of the aggregating agent. This is performed in any combination of the addition amount of the flocculant, the stirring conditions, and the like. As a result, the optimum coagulant injection rate under many conditions can be determined.

  Next, the set conditions of the explanatory variables determined above and the optimum coagulant injection rate under those conditions are added to the database of the learning data for model construction as one data set (step 1-5).

  The actual raw water is collected again at a certain frequency (for example, every 24 hours) or at a timing when it is considered that the raw water quality has sufficiently changed (step 1-6).

  Using the fresh raw water collected again, the above-mentioned Step 1-2 to Step 1-5 are performed again, and the operation of accumulating the data sets to be stored in the database of the learning data for model construction is repeated (Step 1-7).

  By storing a predetermined amount of data sets, a database of learning data for model construction is completed (step 1-8).

  Then, a prediction model is constructed by a machine learning algorithm using the database completed by the batch test (step 1-9). As a machine learning algorithm used at this time, an SVR method, a PLS method, a Deep Learning method, a random forest method, a decision tree method, or the like is used. Then, the constructed prediction model is actually introduced into the control unit for controlling the coagulant injection rate in the water purification plant, and the control of the coagulant injection rate in the actual water purification plant is started. Various actual data (for example, raw water quality and treated water quality) obtained by controlling the actual coagulant injection rate are also accumulated in the database. As a result, more accurate control can be performed. Even after the actual control is started, if the batch test is continuously performed, the obtained data set is added to the database, and if the model is updated, the accuracy is continuously high. Control can be performed.

(Example)
In order to determine the optimal coagulant injection rate in the water treatment plant, we constructed a model that outputs the optimal coagulant injection rate and tried to use it for control.

  The construction of the database using the jar test was started three months before the scheduled control start date by the machine learning model. Raw water (the actual raw water) was collected every 24 hours or even within 24 hours from the previous sampling if the water quality fluctuated significantly due to the weather or the like. In addition, the value of turbidity was obtained from an industrial water quality meter at a water purification plant as the quality of raw water at the time of water sampling.

  The collected raw water is divided into a plurality, and has M-alkalinity of 2 types (10 mg / L and 20 mg / L) and pH of 3 types (6.7, 7.0 and 7.3). Sample water was prepared. That is, for example, when the M-alkalinity of the raw water is 10 mg / L, a carbonate salt is added thereto to prepare a sample water having an M-alkalinity of 20 mg / L. If any, an acid or alkali was added thereto to prepare sample water having a pH of 6.7 or 7.3. The jar test was performed for each sample water with the coagulant injection rate set at 5 points (5 mg / L, 10 mg / L, 15 mg / L, 20 mg / L, and 25 mg / L). However, when the raw water quality was poor and a further coagulant was required, the injection rate score was increased.

  PAC (polyaluminum chloride) was used as the coagulant. After the same rapid stirring and slow stirring as in the water purification plant, the mixture was allowed to stand for a certain period of time, and the turbidity of the supernatant water and the residual aluminum concentration were measured. The state of the flocculated floc at the end of the rapid stirring was photographed with a camera. The optimum coagulant injection rate under each sample water condition was determined from the coagulated floc shape and the quality of the supernatant water.

  One raw water sampling obtained the optimum coagulant injection rate for each of the six conditions. From the above jar test, raw water turbidity, M-alkalinity, pH, and a flocculated floc image after rapid stirring were obtained as explanatory variables, and an optimal flocculant injection rate was obtained as an objective variable. The combination of the explanatory variable and the objective variable under each condition becomes a data set.

  The data set includes, in addition to the three conditions (raw water turbidity, M-alkalinity, pH), other properties of raw water such as raw water temperature, raw water TOC concentration, and raw water chromaticity, It is more preferable to use the flocculated floc image after slow stirring as an explanatory variable in constructing a model that returns a highly accurate output in more situations. Note that it is preferable that the items set in the jar test, and hence the explanatory variables of the model, are items that can be measured by sensors or cameras installed in the actual plant. Conversely, as long as a highly accurate output can be returned with fewer explanatory variables (for example, only the aggregated floc image), the number of explanatory variables may be reduced. In this case, the number of devices in the plant can be reduced.

  As the above-mentioned objective variable, a judgment value of the coagulant injection rate may be adopted instead of the optimum coagulant injection rate. For example, when the jar test is performed at five points of the coagulant injection rate of 5 mg / L, 10 mg / L, 15 mg / L, 20 mg / L, and 25 mg / L, and the optimum coagulant injection rate is determined to be 15 mg / L. , The objective variable in the data set of the jar test of 15 mg / L is set to “0” (optimal). From this point, when the objective variable is 20 mg / L, “1” (somewhat large) and 25 mg / L At this time, "2" (excess), at 10 mg / L, "-1" (slightly less), and at 5 mg / L, "-2" (under). In this case, in addition to the raw water turbidity, M-alkalinity, pH, and the flocculated floc image after rapid stirring as the explanatory variables, the data set is the flocculant injection rate. A judgment value (in the above example, any of -2, -1, 0, 1, 2) is given.

  By repeating this work, a database for determining the optimal coagulant injection rate was completed, this database was used as learning data for model construction, and a model was constructed using a Deep Learning method or the like as a machine learning algorithm.

  FIG. 2 is an overall schematic configuration diagram of a water purification plant (water treatment system) 1-1 configured using the constructed model. As shown in the figure, raw water taken from dams and rivers is introduced into a landing well 11 of a water purification plant. Next, in the landing well 11, the turbidity sensor 23A, the M-alkalinity sensor 23B, and the pH sensor 23C installed in the well 11 detect the turbidity of the raw water, the M-alkalinity, and the numerical value of pH, respectively. . Each detected value is transmitted to the control unit 25 that has introduced the constructed model.

  The raw water transferred from the landing well 11 to the rapid stirring tank 13 is injected with a coagulant by the coagulant injection pump 27 to form fine coagulated flocs. At this time, the flocculated floc image is captured by the flocculated floc observation camera (camera, image acquisition device) 29 and transmitted to the control unit 25. The turbidity of the processing liquid in the sedimentation basin 17 described below is also detected by the turbidity sensor 31, and the data is also transmitted to the control unit 25. The control unit 25 issues a command to the coagulant injection pump 27 from the data transmitted to these control units 25 by using the constructed model so that the optimum coagulant injection rate for the raw water is obtained.

  The raw water stirred with the flocculant in the rapid stirring tank 13 is then slowly stirred in the slow stirring tank 15 to coarsen the flocculated floc, and then the coarse floc is precipitated in the sedimentation tank 17. Then, the water is filtered in the filtration pond 19 to remove small dirt, and then transferred to the water purification pond 21.

After the control in the water purification plant 1-1 was started by the control unit 25 using the constructed model, the prediction accuracy for one month was as follows. Using the coefficient of determination R ² value as an evaluation of the prediction accuracy. The actual optimum coagulant injection rate was determined as a correct answer by a jar test using actual raw water.

[After control starting, ^{R 2} value of 1 month]: 0.83

Wherein the coefficient of determination R ² is an index for evaluating the predictive accuracy of the model, defined by the equation of the following "Equation 1". The value of the coefficient of determination R ² is 1 the maximum value, the closer to 1, so that the model prediction accuracy of high.

  From the above results, control using the model became possible from the day of the control start date. If the present invention is not used, since only the actual operating conditions of the actual plant and the information obtained from the sensors of the actual plant are used for constructing the database, the control cannot be performed on the scheduled control start date in the first place. . According to the present invention, in addition to the data obtained in an actual plant, a jar test (batch test) previously performed for constructing a database enables early and accurate control start-up.

[Second embodiment]
FIG. 3 is a diagram illustrating a method for manufacturing a learning data database according to the second embodiment of the present invention, and is a diagram illustrating a procedure for creating a database of learning data for model construction by a continuous test apparatus. Specifically, an example of a procedure of creating a database for learning data used when constructing a model for predicting an optimal coagulant injection rate in a water purification plant is shown.

  In the present embodiment, first, the actual raw water introduced into the landing well of the actual water purification plant is introduced into a continuous test apparatus having a plurality of systems (step 2-1).

  Next (or prior to the introduction of the actual raw water), a plurality of conditions are set for the injection rate of the flocculant to be injected into the actual raw water of the continuous test apparatus (step 2-2), and the flocculant is injected. . In addition, the properties of the actual raw water (water temperature, M-alkalinity, pH, turbidity, chromaticity, TOC, ultraviolet absorbance, fluorescence intensity, weather information, etc.) are also measured.

  Next, a photograph of the flocculation floc of the rapid stirring tank in each series of the continuous test apparatus and the turbidity (sedimentation water) in each series after the respective residence time of the rapid stirring tank, the slow stirring tank, and the sedimentation basin have elapsed. (Turbidity) is measured (step 2-3).

  Next, the properties of the actual raw water, each set coagulant injection rate and the flocculated floc image are used as explanatory variables, and the measured sediment water turbidity is used as an objective variable. (Step 2-4).

  When the raw water state has changed significantly, the above-mentioned steps 2-2 to 2-4 are repeated to accumulate the data sets to be stored in the learning data for model construction (step 2-5).

  By storing a predetermined amount of data sets, a database of learning data for model construction is completed (step 2-6).

  Then, a model is constructed by a machine learning algorithm using the database completed by the continuous test apparatus (step 2-7). As a machine learning algorithm used at this time, an SVR method, a PLS method, a Deep Learning method, a random forest method, a decision tree method, or the like is used. Then, the constructed model is actually introduced into the control unit for controlling the coagulant injection rate of the water purification plant, and for example, the control of the coagulant injection rate in the actual water purification plant shown in FIG. 2 is started. Actual various data obtained by controlling the actual coagulant injection rate are also accumulated in the database. As a result, more accurate control can be performed. Even after the actual control is started, if the test is continuously performed by the continuous test apparatus, the obtained data set is added to the database, and the model is updated, the accuracy is continuously improved. High control can be performed.

[Third Embodiment]
FIG. 4 is a diagram illustrating a method of manufacturing a learning data database according to the third embodiment of the present invention, and is a diagram illustrating a procedure of creating a database of learning data for model construction by a continuous test apparatus. Specifically, an example of a procedure for creating a database for learning data used in constructing a model for predicting an optimal alkali agent injection rate in a wastewater treatment plant using a carrier-input methane fermentation method for water treatment is shown. I have.

  Here, the carrier for water treatment refers to a carrier capable of supporting microorganisms and allowing microorganisms to propagate on the surface of the carrier, and is not limited in shape and material, but a porous body to which microorganisms are easily attached, such as activated carbon and polyvinyl. Alcohol, ethylene glycol and the like are preferred.

  In the present embodiment, first, actual raw water (actual raw water) to be introduced into the methane fermentation tank of an actual wastewater treatment plant is obtained (step 3-1). Alternatively, when the plant is not yet completed, for example, it is possible to obtain wastewater to be fed into the preceding acid fermentation tank, perform acid fermentation in a laboratory under predetermined conditions, and create water equivalent to actual raw water. good.

Next (or prior to obtaining the actual raw water), a number of conditions are set for the actual raw water using parameters that are candidates for explanatory variables. As parameters to be set, many parameters considered to affect the gas generation amount, for example, residence time (HRT), carrier concentration, water temperature, pH, alkalinity, TOC or COD _Cr , volatile lower fatty acid, raw water amount , Treated water volume, circulating water volume, acid fermentation tank volume, methane fermentation tank volume, etc. are set (step 3-2).

  Next, a methane fermentation test using a vial is performed using each of the above set conditions. At this time, the amount of gas generated under each condition is measured, and the carrier to which the microorganisms are attached is photographed using a camera (step 3-3).

  One data set is created using the setting conditions and the image of the carrier as explanatory variables and the gas generation amount as an objective variable, and added to a database of learning data for model construction (step 3-4).

  When the properties of the actual raw water have changed, the actual raw water is obtained again (step 3-5).

  Steps 3-2 to 3-4 are performed again using the fresh raw water that has been collected again, and the operation of accumulating a data set to be stored in the database of the learning data for model construction is repeated (step 3-6).

  By storing a predetermined amount of data set, a database of learning data for model construction is completed (step 3-7).

  Then, a prediction model is constructed by a machine learning algorithm using the database completed by the batch test (step 3-8). As a machine learning algorithm used at this time, an SVR method (support vector regression method), a PLS method, a Deep Learning method, a random forest method, a decision tree method, or the like is used.

  FIG. 5 is a schematic configuration diagram illustrating an example in which the prediction model according to the third embodiment is actually introduced into a wastewater treatment plant (sludge treatment system) 1-2 using a carrier-input methane fermentation method. In the wastewater treatment plant 1-2 shown in the figure, the wastewater (non-treated water) is introduced into the acid fermentation tank 51 and subjected to acid fermentation, and then transferred to the carrier-input methane fermentation tank 53 using this fermented liquid as raw water. The methane gas is fermented and produced by the microorganisms carried on the carrier. Next, the methane gas and the treated water generated in the carrier-injection type methane fermentation tank 53 are respectively led to the next step.

  The prediction model according to the third embodiment is introduced into the control unit 55 for controlling the alkali agent injection rate of the wastewater treatment plant 1-2. Control of the injection rate is started.

  Specifically, the temperature sensor 57A, the COD sensor 57B, the M-alkalinity sensor 57C, and the pH sensor 57D installed near the outlet of the acid fermentation tank 51 into which the wastewater has been introduced are used to fermentation liquor near the outlet of the acid fermentation tank 51. (Raw water) water temperature, COD (chemical oxygen demand), M-alkalinity (total alkalinity), and pH are detected, and each detected value is controlled by introducing the constructed model. To the unit 55. At the same time, the current carrier state is photographed by the carrier observation camera (camera, image acquisition device) 59 in the carrier-injection type methane fermentation tank 53, and the image data is transmitted to the control unit 55. At the same time, the flow rate and the concentration of the methane gas derived from the carrier input type methane fermentation tank 53 are detected by the flow rate sensor 63A and the methane concentration meter 63B, respectively, and the data are transmitted to the control unit 55.

  The control unit 55 predicts the raw water quality that can maximize the gas generation amount from the current raw water quality of the methane fermentation tank measured at the outlet of the acid fermentation tank 51. The control unit 55 calculates an alkali agent injection rate for achieving the predicted raw water quality (specifically, alkalinity and pH), transmits a control signal to the alkali agent injection pump 61, and injects the alkali agent. By doing so, the amount of generated gas is maximized.

  Further, actual various data (for example, raw water quality and gas generation amount) obtained from the various sensors and cameras by controlling the actual alkali agent injection rate are also accumulated in the database.

  As a result, more accurate control can be performed. Even after the actual control is started, the vial test is continuously performed, the obtained data set is added to the database, and the model is updated. Control can be performed.

  Although the embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various modifications may be made within the scope of the claims and the technical idea described in the specification and the drawings. It is possible. It should be noted that any configuration or material not directly described in the specification and the drawings is within the technical idea of the present invention as long as the effects and effects of the present invention are exhibited. For example, in the above embodiment, an example of a water treatment plant is shown as an example of a water treatment or sludge treatment system. However, the present invention is applied to, for example, water treatment of a food factory or coagulation treatment equipment for dewatered sludge of a sewage treatment plant. can do. In addition, other examples of the wastewater treatment plant using the carrier input type methane fermentation method are shown, but the present invention can be applied to any facility using a carrier to which microorganisms are attached regardless of aerobic or anaerobic. it can. Further, the embodiments described in the above description and in the respective drawings can be combined with each other as long as there is no inconsistency in the object and the configuration. In addition, the above description and the contents of each drawing may be independent embodiments even if they are a part thereof, and the embodiment of the present invention is one embodiment in which the above description and each drawing are combined. However, the present invention is not limited to this.

1-1 Water purification plant (water treatment system)
DESCRIPTION OF SYMBOLS 11 Landing well 13 Rapid stirring tank 15 Slow stirring tank 17 Sedimentation basin 19 Filtration pond 21 Purification tank 23A Turbidity sensor 23B M-alkalinity sensor 23C pH sensor 25 Control part 27 Coagulant injection pump 29 Camera for observation of flocculation floc , Image acquisition device)
31 Turbidity sensor 1-2 Wastewater treatment plant (sludge treatment system)
51 Acid fermentation tank 53 Carrier input type methane fermentation tank 55 Control unit 57A Temperature sensor 57B COD sensor 57C M-alkaliness sensor 57D pH sensor 59 Camera for carrier observation (camera, image acquisition device)
61 Alkaline agent injection pump 63A Flow sensor 63B Methane concentration meter

Claims (6)

Using a machine learning algorithm, a method for manufacturing a database for learning data used when constructing a model for obtaining any predicted value in water treatment or sludge treatment,
A method for manufacturing a database, wherein learning data including at least still or continuous images obtained by an image obtaining apparatus is obtained by a batch or continuous test apparatus using real raw water or real sludge. The method for manufacturing a database according to claim 1, wherein
A method for manufacturing a database, wherein data obtained from operation of an actual plant for treating the actual raw water or the actual sludge is also added to the database. It is a manufacturing method of the database of Claim 1 or 2, Comprising:
A method of manufacturing a database, wherein the items set as the measurement or the conditions by the batch or continuous test apparatus include at least items that can be measured by an image acquisition device or a sensor installed in an actual plant. A database manufacturing method according to any one of claims 1 to 3,
A method of manufacturing a database, wherein the still or continuous image acquired by the batch or continuous test apparatus is for a flocculated floc. A database manufacturing method according to any one of claims 1 to 3,
A method for manufacturing a database, wherein the still or continuous image acquired by the batch or continuous test apparatus is for a water treatment carrier.   A water treatment or sludge treatment system comprising a control unit for controlling a model constructed using a database produced by the method for producing a database according to any one of claims 1 to 5.

Images