Odor prediction system, odor prediction method, learning method for learning model, and processing system

Abstract

To provide an odor prediction system, an odor prediction method, and a learning model learning method that accurately predict an odor generation state from treated water.SOLUTION: There is provided a control device configured to input second condition information indicating at least one of a second state condition of new water to be treated and a second environmental condition of the water treatment system when treating the new water to be treated, into a learning model generated by machine learning using multiple teacher data including first odor information regarding water to be treated for which treatment is performed in the water treatment system, and first condition information indicating at least one of a first state condition regarding a state of the water to be treated and a first environmental condition of the water treatment system when treating the water to be treated, and outputs odor information output from the learning model in response to input of the second condition information as second odor information for the new water to be treated.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to an odor prediction system and an odor prediction method. 、 How the learning model is trained and processing system Regarding. [Background technology]

[0002] For example, in a water purification treatment system installed in a water supply plant, suspended solids (SS) contained in raw water such as river water (hereinafter simply referred to as raw water) are removed. Also, in a sewage treatment system installed in a sewage plant, organic matter and suspended matter contained in sewage containing sludge (hereinafter simply referred to as sewage) are removed (see Patent Documents 1 and 2). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2004-073903 [Patent Document 2] Patent Publication No. 2021-079335 Summary of the Invention [Problem to be solved by the invention]

[0004] In the above-described water purification system and sewage treatment system (hereinafter, these are also collectively referred to as simply water treatment system), for example, it is necessary to suppress odors generated from raw water and sewage (hereinafter, these are also collectively referred to as simply water to be treated) in each facility within the water treatment system. Therefore, in the above-described water treatment system, it is desired to accurately predict, for example, the state of odor generation from the water to be treated. [Means for solving the problem]

[0005] In order to accurately predict the odor generation conditions as described above, the odor prediction system of the present invention includes a control device that inputs second condition information indicating at least one of a second state condition of new water to be treated and a second environmental condition of the water treatment system when treating the new water to a learning model generated by machine learning using multiple training data each including first odor information of the water to be treated in a water treatment system and first condition information indicating at least one of a first state condition for the state of the water to be treated and a first environmental condition of the water treatment system when treating the water to be treated, and outputs the odor information output from the learning model in response to the input of the second condition information as the second odor information of the new water to be treated. [Effects of the Invention]

[0006] Odor prediction system and odor prediction method according to the present invention 、 How the learning model is trained and processing system According to this, it is possible to accurately predict the state of odor generation from the water to be treated. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram illustrating an example of the configuration of an odor measurement system 800 in a comparative example. [Figure 2] FIG. 2 is a diagram illustrating an example of the configuration of the odor prediction system 100 according to the first embodiment. [Figure 3] FIG. 3 is a diagram illustrating an example of the configuration of the information processing device 40 according to the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of the configuration of the control device 30 in the first embodiment. [Figure 5] FIG. 5 is a flowchart illustrating the model generation process. [Figure 6] FIG. 6 is a diagram illustrating a first specific example of the teacher data DT3. [Figure 7] FIG. 7 is a diagram illustrating a second specific example of the training data DT3. [Figure 8] FIG. 8 is a flowchart illustrating the deodorization control process. [Figure 9] FIG. 9 is a diagram illustrating a modified example of the odor prediction system 100 in the first embodiment. [Figure 10] FIG. 10 is a diagram illustrating a modified example of the odor prediction system 100 in the first embodiment. [Figure 11] FIG. 11 is a diagram illustrating a modified example of the odor prediction system 100 in the first embodiment. [Figure 12] FIG. 12 is a diagram illustrating an example of the configuration of an odor measurement system 900 in a second comparative example. [Figure 13] FIG. 13 is a diagram illustrating an example of the configuration of an odor prediction system 200 according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. However, the technical scope of the present invention is not limited to these preferred embodiments.

[0009] [Odor Measurement System 800 in First Comparative Example] First, a description will be given of an odor measurement system 800 in a first comparative example. Figure 1 is a diagram illustrating an example of the configuration of an odor measurement system 800 in a first comparative example.

[0010] The odor measurement system 800 includes, for example, the water treatment system 10.

[0011] The water treatment system 10 is a water purification system installed in a water supply plant, and includes, for example, a gritification basin 11, a receiving well 12, a mixing basin 13, a flocculation basin 14, a sedimentation basin 15, a filtration basin 16, a purified water basin 17, and a distribution basin 18.

[0012] The settling basin 11 is, for example, a tank into which the water to be treated (raw water) first flows, and is a tank in which sedimentation and removal of sediment and the like contained in the water to be treated is performed.

[0013] The receiving well 12 is, for example, a tank that adjusts the amount of water to be treated supplied from the settling basin 11 and supplies it to the mixing basin 13.

[0014] The mixing basin 13 is, for example, a tank in which a flocculant is injected into the water to be treated supplied from the receiving well 12 and the water is stirred.

[0015] The flocculation basin 14 is a tank in which suspended solids contained in the water to be treated supplied from the mixing basin 13 are coagulated by an injected coagulant to form flocs.

[0016] The settling tank 15 is, for example, a tank in which flocs contained in the water to be treated supplied from the flocculation tank 14 are allowed to settle and are separated from the water to be treated.

[0017] The filtration basin 16 is a tank that filters the water to be treated supplied from the settling basin 15 by using a filter body (not shown) made of, for example, sand, gravel, or the like.

[0018] The purified water reservoir 17 is, for example, a tank that temporarily stores the treated water supplied from the filtration reservoir 16 (for example, the treated water after being disinfected with chlorine downstream of the filtration reservoir 16) and supplies it to the distribution reservoir 18.

[0019] The distributing reservoir 18 temporarily stores the water to be treated (treated water) supplied from the purified water reservoir 17, for example, and supplies it to homes and the like (not shown).

[0020] The water treatment system 10 may include other facilities in addition to the grit basin 11, the receiving well 12, the mixing basin 13, the flocculation basin 14, the sedimentation basin 15, the filtration basin 16, the purified water basin 17, and the distribution basin 18. Specifically, the water treatment system 10 may include, for example, a thickening tank (not shown) that thickens the flocs separated in the sedimentation basin 15.

[0021] The odor measuring system 800 also includes, for example, a deodorizing device 20 and a control device 830.

[0022] The deodorizing device 20 is, for example, a device that deodorizes the water to be treated in the water treatment system 10. Specifically, the deodorizing device 20 deodorizes the water to be treated (for example, the water to be treated stored in the receiving well 12) by using, for example, activated carbon or a deodorizing filter.

[0023] The control device 830 controls the deodorization device 20 based on, for example, information indicating the odor generation status in the water treatment system 10 (hereinafter also referred to as odor information). The control device 830 is, for example, a computer device having a CPU (Central Processing Unit) and memory. The odor information is, for example, information indicating at least one of the odor index, odor intensity, type of odor, presence or absence of odor, odor constituents, and the concentration of each odor constituent of the water to be treated in the water treatment system 10. Below, a case where the odor information is an odor index will be described.

[0024] Specifically, as shown in FIG. 1, the operator OP obtains water to be treated from at least one of the receiving well 12, the settling tank 15, and the distributing tank 18. Then, the operator OP measures the odor index of the obtained water to be treated, for example, by the operator OP's sense of smell. Furthermore, the operator OP inputs the measured odor index into the control device 830. Thereafter, the control device 30 controls the deodorizing device 20 (for example, controls the amount of activated carbon supplied and the setting of the deodorizing filter, etc.) based on the odor index input by the operator OP.

[0025] However, when an operator OP measures the odor index as described above, not only does the operator OP have to spend more time working, but the operator OP is also restricted from using items that may contaminate the odor (such as soap or cigarettes), which increases the operator OP's workload. Furthermore, when multiple operators OP share the odor index measurement, there is a possibility that individual differences in the measurement results may occur among the multiple operators OP, and the odor index measurement may not be performed accurately and stably. Furthermore, when the location where the water to be treated is obtained and the location where the odor index is measured are different, the odor index measurement cannot be performed continuously, and the odor index measurement cannot be performed efficiently.

[0026] In contrast, for example, an analyzer for the water to be treated (not shown) may be used in odor measurement system 800. This makes it possible for odor measurement system 800 to reduce the workload of the operator OP required to measure the odor index, for example, and also to identify the constituent substances of the odor in the water to be treated and measure the concentration of each constituent substance.

[0027] However, the above-mentioned analyzers may not be able to measure the odor index of the water to be treated with the same accuracy as when the odor index is measured by the operator's sense of smell. Furthermore, when measuring using an analyzer, for example, the purchase cost and maintenance cost of the analyzer are required.

[0028] Therefore, the inventors conducted extensive research to enable efficient acquisition of odor information for the water to be treated, and as a result, found that there is a correlation between the odor information of the water to be treated that was actually measured and the conditions under which the water to be treated was treated.

[0029] For example, this corresponds to the correlation between the odor index of the water to be treated and various measurements of the water measured by a turbidity meter, pH meter, etc. Specifically, the odor index of the water to be treated tends to increase as the turbidity of the water to be treated increases, for example, when the turbidity is high, the water to be treated cannot be sufficiently treated due to the large amount of sludge particles and the large amount of nutrients adhering to the sludge particles, resulting in a deterioration of water quality. The odor index also tends to decrease as the turbidity of the water to be treated decreases. Furthermore, when the turbidity is high, bacteria are protected by the turbidity particles, reducing the disinfection effect of chlorine, and the chlorine demand and chlorine concentration of the water to be treated also increase. Therefore, when the chlorine demand of the water to be treated is high, the odor index increases. Furthermore, the odor index of the water to be treated varies depending on, for example, the pH of the water to be treated. The pH of the water to be treated is often approximately 6.5 to 8.5 (nearly neutral), and when biological treatment is good, the pH is 7 or higher (alkaline). Here, if the pH of the water to be treated rises (for example, pH = 9), the amount of ammonia nitrogen increases, resulting in the generation of a large amount of ammonia, and the volatility of ammonia also rises sharply, resulting in a high odor index. In this case, the generation of hydrogen sulfide and methyl mercaptan is suppressed, and their volatility also decreases. On the other hand, if the pH of the water to be treated falls (for example, pH = 4), the generation of hydrogen sulfide and methyl mercaptan increases, resulting in a high odor index.

[0030] Another example of this correlation is the correlation between odor information about the water being treated and at least one of the date and time, precipitation, temperature, and humidity of the water treatment system 10 when the water is being treated. Specifically, for example, the amount of sewage flowing into a sewage treatment facility increases during the rainy season compared to other seasons. When the amount of sewage increases, ponds and other facilities with fixed retention times are unable to provide sufficient retention time, resulting in insufficient particle settling, chemical reactions with chemicals, and biological treatment, resulting in a deterioration in the quality of the treated water. Furthermore, during the rainy season, for example, humidity and temperatures are high due to heavy rainfall. High humidity increases the amount of water-soluble ammonia and hydrogen sulfide in the water vapor, which encourages the proliferation of organisms such as mold and the production of large amounts of odorous substances. Furthermore, high temperatures activate organisms, increasing the amount of gas produced from the water being treated. This increases the movement of water molecules in the water being treated and the generated gas molecules, facilitating the chemical reactions that generate odors and the volatilization and diffusion of odor molecules. Therefore, for example, the odor index of the water to be treated during the rainy season is higher than that during other times of the year. Furthermore, for example, the amount of sewage flowing into a sewage treatment facility during the daytime is greater than that during early morning or late night hours, and the temperature during the daytime is higher than that during early morning or late night hours. Therefore, for example, the odor index of the water to be treated during the daytime is higher than that during early morning or late night hours. Furthermore, for example, in Japan, humidity and temperature are higher during the summer and lower during the winter. Therefore, for example, the odor index of the water to be treated during the summer is higher than that during the winter.

[0031] The inventors then came up with the idea of ​​predicting (outputting) odor information for the water to be treated by using a machine learning model (hereinafter simply referred to as a learning model) that has learned the above correlation. Below, we will explain the odor prediction system 100 that uses the above learning model.

[0032] [Odor Prediction System 100 in the First Embodiment] Fig. 2 is a diagram illustrating an example of the configuration of an odor prediction system 100 in the first embodiment. Fig. 3 is a diagram illustrating an example of the configuration of an information processing device 40 in the first embodiment. Fig. 4 is a diagram illustrating an example of the configuration of a control device 30 in the first embodiment. Below, differences from odor measurement system 800 in the first comparative example will be described.

[0033] The odor prediction system 100 in this embodiment generates, for example, a learning model that predicts odor information of water to be treated in the water treatment system 10. Specifically, the odor prediction system 100 generates, for example, a learning model that predicts the odor index or odor intensity of water to be treated in the water treatment system 10. The odor prediction system 100 then controls the deodorization device 20 based on, for example, the odor information predicted by the learning model. Below, a case where the learning model predicts an odor index will be described.

[0034] Specifically, the odor prediction system 100 includes, for example, a control device 30 and an information processing device 40, as shown in FIG.

[0035] In addition, as shown in FIG. 2, the water treatment system 10 is provided with measuring devices M1, M2, M3, M4, M5, M6, M7, M8, and M9 (hereinafter also referred to as measuring device M1, etc.) installed in the inlet pipe L1 that supplies the water to be treated to the settling tank 11, the settling tank 11, the receiving well 12, the mixing tank 13, the flocculation tank 14, the sedimentation tank 15, the filtration tank 16, the purified water tank 17, and the distribution tank 18, respectively.

[0036] Each of measuring devices M1, M2, and M3 may be, for example, at least one of a turbidity meter, pH meter, electrical conductivity meter, alkalinity meter, water thermometer, mold odor meter, ammonia nitrogen meter, residual chlorine meter, chlorine demand meter, dissolved oxygen concentration meter, oil content meter, and oil pressure meter. Each of measuring devices M4, M5, M6, and M7 may be, for example, at least one of a turbidity meter, total organic carbon meter, organic pollutant measuring device, residual chlorine meter, and chlorine demand meter. Furthermore, each of measuring devices M8 and M9 may be, for example, at least one of a residual chlorine meter, mold odor meter, electrical conductivity meter, trihalomethane meter, turbidity meter, colorimeter, water thermometer, pH meter, organic pollutant measuring device, and total organic carbon meter.

[0037] The control device 30 controls the deodorizing device 20 based on, for example, measurements taken in the water treatment system 10. The control device 30 is, for example, a computer device having a CPU and a memory.

[0038] Specifically, the control device 30 acquires, for example, a measurement value measured by at least one of the measuring devices M1, etc., as shown in Fig. 2. Then, the control device 30 controls the deodorizing device 20 based on, for example, the acquired measurement value.

[0039] In addition, in the water treatment system 10, for example, other measuring devices (not shown) than the measuring device M1 may be installed in facilities other than the settling basin 11, the receiving well 12, the mixing basin 13, the flocculation basin 14, the sedimentation basin 15, the filtration basin 16, the purified water reservoir 17, and the distributing reservoir 18. Specifically, in the water treatment system 10, for example, another measuring device may be installed in a thickening tank (not shown). Then, the control device 30 may, for example, acquire measured values ​​measured by the other measuring devices and control the deodorizing device 20 based on the acquired measured values.

[0040] The information processing device 40 performs, for example, a process (hereinafter also referred to as a model generation process) for generating a learning model that predicts odor information of water to be treated in the water treatment system 10. The information processing device 40 is, for example, a computer device having a CPU and a memory. Note that the control device 30 and the information processing device 40 may be, for example, a single computer device.

[0041] Specifically, the information processing device 40 generates a learning model that outputs odor information for the water to be treated in the water treatment system 10, in response to input of information (hereinafter also referred to as condition information) indicating at least one of the state conditions for the state of the water to be treated in the water treatment system 10 and the environmental conditions of the water to be treated. The state conditions for the water to be treated are, for example, conditions for the state of the water treatment system 10 indicated by measurements taken by at least one of the measuring devices M1, etc. The environmental conditions for the water to be treated are, for example, at least one of the date and time, precipitation, temperature, and humidity of the water treatment system 10 when the water to be treated is treated.

[0042] More specifically, as shown in Fig. 3, the operator OP measures odor information DT1 (hereinafter also referred to as first odor information) of the water to be treated in advance at multiple timings. The odor information DT1 is, for example, the odor index or odor intensity of the water to be treated. Furthermore, the control device 30 acquires, in advance, condition information DT2 (hereinafter also referred to as first condition information) indicating at least one of the state condition of the water to be treated (hereinafter also referred to as first state condition) and the environmental condition of the water to be treated (hereinafter also referred to as first environmental condition) at each timing, for example, at each of the multiple timings at which the odor information DT1 of the water to be treated was measured.

[0043] Then, the information processing device 40 generates, for each of a plurality of timings, teacher data DT3 including a combination of odor information DT1 corresponding to each timing and condition information DT2 corresponding to each timing, and stores the generated teacher data DT3 in the storage device 41. The storage device 41 is, for example, a storage device having a storage medium such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0044] Thereafter, the information processing device 40 generates a learning model MD by performing machine learning using, for example, the plurality of pieces of teacher data DT3 stored in the storage device 41, and stores the generated learning model MD in the storage device 42. The storage device 42 is, for example, a storage device accessible by the control device 30, and is a storage device having a storage medium such as an HDD or SSD.

[0045] The storage device 41 and the storage device 42 may be, for example, devices different from the control device 30 and the information processing device 40, respectively, or may be devices built into the control device 30 or the information processing device 40. Furthermore, the storage device 41 and the storage device 42 may be, for example, a single storage device.

[0046] 4, the control device 30 acquires, for example, at regular intervals, condition information DT11 (hereinafter also referred to as second condition information) indicating at least one of the state conditions of the water to be treated (hereinafter also referred to as second state conditions) and the environmental conditions of the water to be treated (hereinafter also referred to as second environmental conditions), and inputs the acquired condition information DT11 to the learning model MD stored in the memory device 42. Then, the control device 30 acquires, for example, a value output from the learning model MD in response to the input of the condition information DT11, as odor information DT12 of the water to be treated (hereinafter also referred to as second odor information) at each interval.

[0047] As described above, the odor prediction system 100 in this embodiment is equipped with a control device 30 that inputs condition information DT11 indicating at least one of the second state conditions of new water to be treated and the second environmental conditions of the water treatment system 10 when treating the new water to be treated into a learning model MD generated by machine learning using multiple training data DT3, each of which includes odor information DT1 of the water to be treated in the water treatment system 10 and condition information DT2 indicating at least one of the first state conditions for the state of the water to be treated and the first environmental conditions of the water treatment system 10 when treating the water to be treated, and outputs the odor information output from the learning model MD in response to the input of the condition information DT11 as odor information DT12 of the new water to be treated.

[0048] That is, there is a certain correlation between odor information DT1 and condition information DT2 acquired (measured) at the same time in the water treatment system 10. Therefore, the information processing device 40 in this embodiment generates a learning model MD by performing machine learning using training data DT3 including odor information DT1 and condition information DT2 acquired (measured) at past times. Then, the control device 30 predicts odor information DT12 (current odor information) of the water to be treated in the water treatment system 10, for example, by using the learning model MD generated by the information processing device 40.

[0049] This allows the control device 30 in this embodiment to predict, for example, odor information DT12 of water to be treated during operation of the water treatment system 10 using the learning model MD generated based on past performance of the water treatment system 10. Therefore, the control device 30 can accurately predict, for example, odor information DT12 of water to be treated in the water treatment system 10.

[0050] In the above example, the case where the teacher data DT3 is stored in the storage device 41 has been described, but this is not limiting. For example, the information processing device 40 may store the generated teacher data DT3 in another storage device provided outside the odor prediction system 100 (for example, another storage device provided in an external cloud environment). When generating the learning model MD, the information processing device 40 may acquire the teacher data DT3 by, for example, accessing the other storage device.

[0051] In the above example, the learning model MD is stored in the storage device 42, but this is not limiting. The information processing device 40 may, for example, store the generated learning model MD in another storage device provided outside the odor prediction system 100 (for example, another storage device provided in an external cloud environment). The control device 30 may, for example, input condition information DT11 to the learning model MD stored in the other storage device.

[0052] [Model generation process] Next, the model generation processing in the information processing device 40 will be described. Fig. 5 is a flowchart illustrating the model generation processing. Figs. 6 and 7 are diagrams illustrating the model generation processing. The model generation processing is executed, for example, in the information processing device 40 by a program loaded into memory from a storage medium and a CPU working together.

[0053] The information processing device 40 acquires odor information DT1 of the water to be treated at multiple times (step S1 in FIG. 5). Specifically, the information processing device 40 acquires, for example, multiple pieces of odor information DT1 measured by the operator OP using their sense of smell at multiple times.

[0054] Furthermore, the information processing device 40 acquires, for example, condition information DT2 at multiple times (multiple times when odor information DT1 was acquired) (step S1 in FIG. 5). Specifically, the information processing device 40 acquires, for example, multiple pieces of condition information DT2 acquired by the control device 30 at each of the multiple times.

[0055] Then, the information processing device 40 generates, for example, for each of a plurality of timings, training data DT3 including a combination of odor information DT1 and condition information DT2 corresponding to each timing (step S2 in FIG. 5). Specific examples of the plurality of training data DT3 will be described below.

[0056] [First example of training data DT3] First, a first specific example of the teacher data DT3 will be described. Fig. 6 is a diagram illustrating the first specific example of the teacher data DT3. The teacher data DT3 shown in Fig. 6 (hereinafter also referred to as teacher data DT3a) is teacher data DT3 that includes, as odor information DT1, an odor index measured in the water-receiving well 12, and, as condition information DT2, the turbidity, pH, and chlorine demand measured in the water-receiving well 12. In other words, the teacher data DT3a shown in Fig. 6 is teacher data DT3 that includes, as condition information DT2, the state conditions of the water to be treated stored in the water-receiving well 12.

[0057] Specifically, in the training data DT3a shown in Figure 6, the first row of data has the "odor index" set to "20", the "turbidity" set to "1 (NTU)", the "pH" set to "6.5", and the "chlorine demand" set to "0.1 (L / min)".

[0058] In addition, in the training data DT3a shown in Fig. 6, the data in the second row has "30" set as the "odor index," "2 (NTU)" set as the "turbidity," "7.0" set as the "pH," and "0.3 (L / min)" set as the "chlorine demand." Explanation of the other data included in Fig. 6 will be omitted.

[0059] [Second example of training data DT3] Next, a second specific example of the teacher data DT3 will be described. Fig. 7 is a diagram illustrating the second specific example of the teacher data DT3. The teacher data DT3 shown in Fig. 7 (hereinafter also referred to as teacher data DT3b) is teacher data DT3 that includes, as odor information DT1, an odor index measured in the water-receiving well 12, and, as condition information DT2, the date, time, precipitation, temperature, and humidity at the time when the water to be treated was treated in the water-receiving well 12. In other words, the teacher data DT3b shown in Fig. 7 is teacher data DT3 that includes, as condition information DT2, the environmental conditions of the water treatment system 10.

[0060] Specifically, in the training data DT3b shown in Figure 7, the first row of data has the "odor index" set to "20", the "date" set to "4 / 1", the "time" set to "8:00", the "precipitation" set to "0 (mm)", the "temperature" set to "5 (℃)", and the "humidity" set to "10 (%)".

[0061] 7, the data in the second row has "30" set as the "odor index," "4 / 1" set as the "date," "10:00" set as the "time," "0 (mm)" set as the "precipitation," "12 (°C)" set as the "temperature," and "10 (%)" set as the "humidity." Explanation of the other data included in FIG. 7 will be omitted.

[0062] Returning to FIG. 5, the information processing device 40 generates a learning model MD by performing machine learning using a plurality of pieces of training data DT3 generated for each of a plurality of timings (step S3 in FIG. 5).

[0063] In this way, the information processing device 40 in this embodiment acquires a plurality of pieces of training data DT3 each including, for example, first odor information on the water to be treated in the water treatment system 10 and condition information DT2 indicating at least one of a first state condition regarding the state of the water to be treated and a first environmental condition of the water treatment system 10 when treating the water to be treated. Specifically, the plurality of pieces of training data DT3 is training data including at least one data (set) of the training data DT3a and training data DT3b described in FIGS. 6 and 7. Then, the information processing device 40 generates, for example, by machine learning using the acquired plurality of pieces of training data DT3, a learning model MD that outputs odor information DT12 for new water to be treated in response to input of condition information DT11 indicating at least one of a second state condition of the new water to be treated and a second environmental condition of the water treatment system 10 when treating the new water to be treated.

[0064] This enables the information processing device 40 in the present embodiment to generate a learning model MD that can accurately predict the odor information DT12 of the water to be treated in the water treatment system 10, for example.

[0065] In addition, the information processing device 40 may perform machine learning of multiple training data DT3 by using any of multiple techniques including linear regression, GBDT (Gradient Boosting Decision Tree), decision tree, regression tree, and k-nearest neighbor method, for example, in step S3.

[0066] [Deodorization control treatment] Next, a description will be given of the deodorization control process in the control device 30. Fig. 8 is a flowchart illustrating the deodorization control process. Note that the deodorization control process is a process executed, for example, in the control device 30 by a program loaded into memory from a storage medium and a CPU working together.

[0067] The control device 30 acquires the condition information DT11, for example, at regular intervals (step S11 in FIG. 8).

[0068] Specifically, the control device 30 acquires, for example, odor information of the same type as the condition information DT2 included in the teacher data DT3 generated in step S2 in Fig. 5 from the water treatment system 10. That is, for example, if the condition information DT2 included in the teacher data DT is the turbidity of the water to be treated in the receiving well 12 (turbidity measured by the measuring device M3), the control device 30 acquires the current turbidity of the water to be treated in the receiving well 12 as the condition information DT11.

[0069] Then, the control device 30 inputs the condition information DT11 acquired in step S11 into the learning model MD. After that, the control device 30 predicts, for example, the value output from the learning model MD as the current odor information DT12 of the water to be treated in the water treatment system 10 (step S12 in FIG. 8).

[0070] Next, the control device 30 controls the deodorizing device 20 in accordance with the odor information DT12 predicted in step S12 (step S13 in FIG. 8), for example.

[0071] Specifically, for example, if the odor information DT12 (e.g., odor index or odor intensity) predicted in step S12 exceeds a predetermined threshold (hereinafter also referred to as a first threshold), the control device 30 starts the deodorization device 20 to begin deodorizing the water to be treated (e.g., the water to be treated stored in the receiving well 12) being treated in the water treatment system 10. On the other hand, for example, if the odor information DT12 predicted in step S12 falls below the first threshold, the control device 30 stops the deodorization device 20 to end the deodorization of the water to be treated being treated in the water treatment system 10.

[0072] Furthermore, for example, if the odor information DT12 predicted in step S12 exceeds a predetermined threshold (hereinafter also referred to as a second threshold), the control device 30 controls the deodorizing device 20 to increase the amount of activated carbon supplied to the water treatment system 10, the cleaning frequency, or the cleaning time (e.g., the amount of activated carbon supplied to the water receiving well 12, the cleaning frequency, or the cleaning time), the amount of air (or oxygen) supplied (e.g., the rotation speed of a fan that sucks in or blows odors), the amount of water supplied (e.g., the amount of water supplied to the cleaning tower), or the amount of chemical solution, deodorant, or ozone supplied (e.g., the pressure of a pump that sprays deodorant or ozone water). The second threshold is, for example, a threshold higher than the first threshold. On the other hand, for example, if the odor information DT12 predicted in step S12 falls below the second threshold, the control device 30 controls the deodorizing device 20 to decrease the amount of activated carbon, air, water, ozone, or the like supplied to the water treatment system 10.

[0073] In step S13, the control device 30 may, for example, display the odor information DT12 predicted in step S12 on an output device (not shown) of the control device 30. Then, the operator OP may, for example, view the odor information DT12 displayed on the output device and manually control the deodorizing device 20.

[0074] As described above, the odor prediction system 100 in this embodiment includes, for example, a deodorization device 20 that deodorizes the water to be treated in the water treatment system 10. The control device 30 in this embodiment controls the deodorization device 20 based on odor information DT12 predicted by using, for example, the learning model MD.

[0075] That is, the control device 30 in this embodiment controls, for example, the deodorization device 20 in the water treatment system 10 by using the odor information DT12 with high prediction accuracy output from the learning model MD.

[0076] As a result, the odor prediction system 100 of this embodiment can, for example, efficiently deodorize the water to be treated in the water treatment system 10. Specifically, the odor prediction system 100 can control the deodorizing device 20 without, for example, measuring odor information using the olfactory sense of the operator OP or performing analysis using an analytical device. Therefore, the odor prediction system 100 can, for example, reduce the workload of the operator OP associated with measuring odor information of the water to be treated and reduce the costs incurred when using an analytical device. Furthermore, the odor prediction system 100 can, for example, reduce the occurrence of individual differences in measurement results among multiple operators OP and the occurrence of a decrease in measurement efficiency due to differences between the location where the water to be treated is obtained and the location where the odor information of the water to be treated is measured.

[0077] Furthermore, by using odor information DT12 with high prediction accuracy output from the learning model MD, for example, the odor prediction system 100 can deodorize the water to be treated by the deodorizing device 20 at the timing when deodorization of the water to be treated is necessary. Therefore, the odor prediction system 100 can improve the deodorizing effect of the water to be treated, for example.

[0078] In step S3, the information processing device 40 may generate a plurality of learning models MD by using, for example, a plurality of types of teacher data DT3 having different condition information DT2.

[0079] Specifically, the information processing device 40 may generate, for example, a learning model MD using multiple pieces of teacher data DT3a described in Figure 6, and a learning model MD using multiple pieces of teacher data DT3b described in Figure 7.

[0080] Then, the control device 30 may, for example, in a test stage (before the deodorization control process begins), predict odor information DT12 by using each of the multiple learning models MD that have been generated, and select the learning model MD that can be determined to have the highest accuracy as the learning model MD to be used in the deodorization control process.

[0081] This enables the control device 30 to, for example, identify training data DT3 suitable for predicting odor information DT12 of the water to be treated in each water treatment system 10, thereby further improving the prediction accuracy of odor information DT12 of the water to be treated.

[0082] In addition, in step S3, the information processing device 40 may generate multiple learning models MD using different methods from among multiple methods (e.g., linear regression, GBDT, decision tree, regression tree, k-nearest neighbor method, etc.).

[0083] This enables the control device 30 to, for example, identify a machine learning method suitable for predicting odor information DT12 of the water to be treated in each water treatment system 10, thereby further improving the prediction accuracy of odor information DT12 of the water to be treated.

[0084] Furthermore, in step S3, the information processing device 40 may generate a plurality of learning models MD from the same type of teacher data DT3 (for example, teacher data DT3 including the same type of environmental condition as the condition information DT2).

[0085] Specifically, the information processing device 40 may, for example, classify each of the multiple teacher data DT3b described in Figure 7 into multiple groups based on the date range contained in each data or the time range contained in each data, and generate a learning model MD using the teacher data DT3b classified into each group for each of the multiple classified groups.

[0086] More specifically, the information processing device 40 may generate, for example, a learning model MD using teacher data DT3b with dates between January and March, a learning model MD using teacher data DT3b with dates between April and June, a learning model MD using teacher data DT3b with dates between July and September, and a learning model MD using teacher data DT3b with dates between October and December. In this case, the information processing device 40 may further generate learning models MD corresponding to, for example, the rainy season, long holidays, etc.

[0087] Then, in step S12, the control device 30 may, for example, select a learning model MD from the multiple generated learning models MD that corresponds to the date on which the deodorization control process will be performed, and predict the odor information DT12.

[0088] This enables the control device 30 to accurately predict the odor information DT12 even when the tendency of the odor information DT12 of the water to be treated varies greatly depending on the operating environment (e.g., season) of the odor prediction system 100.

[0089] Furthermore, in step S3, the information processing device 40 may generate a plurality of learning models MD from the same type of teacher data DT3 (for example, teacher data DT3 including the same type of state condition as the condition information DT2).

[0090] Specifically, the information processing device 40 may, for example, classify each of the multiple teacher data DT3a described in Figure 6 into multiple groups based on the turbidity range contained in each data, and generate a learning model MD for each of the multiple classified groups using the teacher data DT3a classified into each group.

[0091] More specifically, the information processing device 40 may generate, for example, a learning model MD using teacher data DT3a having a turbidity of less than 5 (NTU) and a learning model MD using teacher data DT3a having a turbidity of 5 (NTU) or more.

[0092] Then, for example, in step S12, the control device 30 may select a learning model MD corresponding to the current turbidity of the treated water from among the multiple learning models MD generated, and predict the odor information DT12.

[0093] This enables the control device 30 to accurately predict the odor information DT12 even when the tendency of the odor information DT12 of the treated water varies greatly depending on, for example, the operating conditions of the odor prediction system 100 (e.g., the turbidity of the treated water).

[0094] [Modification of odor prediction system 100 in the first embodiment] Next, a description will be given of a modified example of the odor prediction system 100 in the first embodiment. Figures 9 to 11 are diagrams for explaining a modified example of the odor prediction system 100 in the first embodiment.

[0095] For example, in step S13, the control device 30 may use the odor information DT12 predicted in step S12 to make further predictions about other odor information (hereinafter simply referred to as other odor information) other than the odor information DT12. The control device 30 may then control the deodorizing device 20 based on the odor information DT12 and the other odor information.

[0096] Specifically, the control device 30 may use a calibration curve graph generated in advance to predict other odor information from the odor information DT12 predicted in step S12. The process of step S13 when using a calibration curve graph will be described below.

[0097] [Modification (1) of step S13] Next, a first modified example of the process of step S13 will be described. Fig. 9 is a diagram illustrating the first modified example of the process of step S13. Note that the following description will be given assuming that the odor information predicted in step S12 is an odor index.

[0098] For example, in a test stage (before the deodorization control process is started), the control device 30 acquires, for each of a plurality of timings, a combination of the odor index measured by the operator OP's sense of smell and the concentration of constituent substances corresponding to each type of odor (for example, the concentration measured by an analyzer).The control device 30 then generates, as a calibration curve graph G11, an approximate straight line or an approximate curve calculated from the combination acquired for each of the plurality of timings.

[0099] 9(A), the control device 30 generates a calibration curve graph G11 including, for example, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a malodor, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a moldy odor, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a chlorine odor, and a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a fruity odor. Note that the calibration curve graph G11 may further include, for example, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a putrid odor, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a fecal odor, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to an ozone odor, a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a burnt odor, or a calibration curve graph showing the relationship between the odor index and the concentration of a constituent substance corresponding to a floral scent.

[0100] Thereafter, as shown in FIG. 9(A), for example, in step S13, the control device 30 refers to the calibration curve graph G11 and identifies the concentration of each type corresponding to X1, which is the odor index predicted in step S12.

[0101] Specifically, the control device 30, for example, identifies Y11 (ppm) as the concentration of the constituent corresponding to the bad odor, identifies Y12 (ppm) as the concentration of the constituent corresponding to the mold odor, identifies Y13 (ppm) as the concentration of the constituent corresponding to the chlorine odor, and identifies Y14 (ppm) as the concentration of the constituent corresponding to the fruity odor.

[0102] Then, when the control device 30 determines that the number of values ​​exceeding a predetermined threshold for each odor information, for example, among X1, which is the odor index predicted in step S12, Y11 (ppm), which is the concentration of the constituent substance corresponding to the bad odor, Y12 (ppm), which is the concentration of the constituent substance corresponding to the mold odor, Y13 (ppm), which is the concentration of the constituent substance corresponding to the chlorine odor, and Y14 (ppm), which is the concentration of the constituent substance corresponding to the fruity odor, exceeds a predetermined number, the control device 30 starts deodorizing the treated water using the deodorizing device 20.

[0103] In this way, the control device 30 in the first variant calculates a first concentration corresponding to the predicted odor information DT12 using the learning model MD based on information (e.g., a calibration curve graph G11) showing the relationship between the odor information of the water to be treated and the concentration of a substance corresponding to each of one or more types of odor in the water to be treated (hereinafter also referred to as the first concentration), and outputs the calculated first concentration.

[0104] As a result, the control device 30 in this embodiment can accurately predict other odor information (e.g., the first concentration) besides the odor information DT12, for example, by using the odor information DT12 predicted using the learning model MD and the calibration curve graph G11 generated in advance. Therefore, by controlling the deodorizing device 20 using other odor information, for example, the control device 30 can further improve the control accuracy of the deodorizing device 20 and further improve the deodorizing effect of the water to be treated.

[0105] The control device 30 may display, for example, on an output device (not shown) of the control device 30, a graph showing the time series change (trend) of at least one of the odor index X1 predicted in step S12 and the concentrations Y11, Y12, Y13, and Y14 identified in step S13.

[0106] Specifically, the control device 30 may display, on an output device, a time series graph G12 showing the time series change in the concentration of the constituent substance corresponding to the malodor (hereinafter also referred to as the malodor concentration), as shown in Figure 9(B).

[0107] The operator OP may then manually control the deodorizing device 20 by viewing the time series graph G12 displayed on the output device, for example.

[0108] In the above example, the first concentration is calculated using a calibration curve graph, but the present invention is not limited to this. For example, the control device 30 may calculate the first concentration using information (e.g., a table) other than a graph showing the relationship between odor information of the water to be treated and the first concentration.

[0109] [Modification (2) of step S13] Next, a second modified example of the process of step S13 will be described. Fig. 10 is a diagram illustrating the second modified example of the process of step S13. Note that, although the following description will be given assuming that the concentration of each constituent substance corresponding to a malodor among the types of odors shown in Fig. 9(A) is predicted, the concentration of each constituent substance corresponding to an odor other than a malodor (for example, the concentration of each constituent substance corresponding to a moldy odor) may also be predicted. Furthermore, the following description will be given assuming that the calibration curve graph G31 has been generated in advance, as in the case of the calibration curve graph G21.

[0110] As shown in FIG. 10(A), the control device 30 refers to, for example, a calibration curve graph G31 showing the relationship between the odor index and the concentration of each constituent substance corresponding to the malodor, and identifies the concentration of each constituent substance corresponding to X1, which is the odor index predicted in step S12.

[0111] Specifically, in this case, the control device 30 identifies, for example, Y21 (ppm) as the concentration of ammonia, Y22 (ppm) as the concentration of hydrogen sulfide, Y23 (ppm) as the concentration of methyl sulfide, Y24 (ppm) as the concentration of methyl disulfide, and Y25 (ppm) as the concentration of methyl mercaptan.

[0112] Then, when the control device 30 determines that the number of values ​​exceeding a predetermined threshold for each odor information, for example, among X1, which is the odor index predicted in step S12, Y21 (ppm), which is the ammonia concentration, Y22 (ppm), which is the hydrogen sulfide concentration, Y23 (ppm), which is the methyl sulfide concentration, Y24 (ppm), which is the methyl disulfide concentration, and Y25 (ppm), which is the methyl mercaptan concentration, exceeds a predetermined number, the control device 30 starts deodorizing the treated water using the deodorizing device 20.

[0113] In this way, the control device 30 in the second variant calculates a second concentration corresponding to odor information DT12 based on information (e.g., calibration curve graph G21) showing the relationship between the odor information of the treated water and the respective concentrations (hereinafter also referred to as second concentrations) of one or more substances corresponding to specific types of odor in the treated water, and outputs the calculated second concentration.

[0114] As a result, the control device 30 in this embodiment can accurately predict other odor information (e.g., the second concentration) besides the odor information DT12, for example, by using the odor information DT12 predicted using the learning model MD and the calibration curve graph G21 generated in advance. Therefore, by controlling the deodorizing device 20 using other odor information, for example, the control device 30 can further improve the control accuracy of the deodorizing device 20 and further improve the deodorizing effect of the water to be treated.

[0115] The control device 30 may display, for example, a graph showing time series changes of at least one of the odor index X1 predicted in step S12 and the concentrations Y21, Y22, Y23, Y24, and Y25 identified in step S13 on an output device (not shown) of the control device 30.

[0116] Specifically, the control device 30 may display, on an output device, a time series graph G22 showing the time series change in the concentration of a constituent substance corresponding to methyl sulfide (hereinafter also referred to as methyl sulfide concentration), as shown in Figure 10(B).

[0117] The operator OP may then manually control the deodorizing device 20 by viewing the time series graph G22 displayed on the output device, for example.

[0118] In addition, the control device 30 may refer to a calibration curve graph (not shown) showing the relationship between the odor index and the contribution rate of each constituent substance corresponding to the bad odor (the contribution rate of each constituent substance when the worker OP detects the bad odor), and identify the contribution rate of each constituent substance corresponding to the odor index X1 predicted in step S12.

[0119] Then, the control device 30 may start deodorizing the treated water by the deodorizing device 20, for example, when it determines that the number of values ​​that exceed a predetermined threshold for each odor information, among the odor index X1 predicted in step S12, the contribution rate of methyl mercaptan, the contribution rate of methyl sulfide, the contribution rate of methyl sulfide, the contribution rate of hydrogen sulfide, and the contribution rate of ammonia, exceeds a predetermined number.

[0120] In the above example, the second concentration is calculated using a calibration curve graph, but the present invention is not limited to this. For example, the control device 30 may calculate the second concentration using information other than a graph showing the relationship between odor information of the water to be treated and the second concentration (for example, a table, etc.).

[0121] [Modification (3) of step S13] Next, a third modified example of the process of step S13 will be described. Fig. 11 is a diagram illustrating the third modified example of the process of step S13. Note that the following description will be given assuming that the calibration curve graph G31 has been generated in advance, as in the case of the calibration curve graph G21.

[0122] As shown in FIG. 11(A), the control device 30 refers to, for example, a calibration curve graph G31 showing the relationship between the odor index and the odor intensity, and identifies the odor intensity corresponding to X1, which is the odor index predicted in step S12.

[0123] Specifically, in this case, the control device 30 specifies, for example, the odor intensity Y31(TON) as the odor intensity.

[0124] Then, when the control device 30 determines that the number of values ​​of X1, the odor index predicted in step S12, and Y31(TON), the odor intensity, that exceed a predetermined threshold value for each odor information exceeds a predetermined number, the control device 30 starts deodorizing the water to be treated by the deodorizing device 20.

[0125] In addition, the control device 30 may start deodorizing the treated water by the deodorizing device 20, for example, when it determines that the number of values ​​exceeding a predetermined threshold for each odor information, including the odor index X1 predicted in step S12, the odor intensity Y31(TON), and the concentrations described in Figures 9 and 10, exceeds a predetermined number.

[0126] In addition, the control device 30 may display, for example, a graph showing the time series change of at least one of X1, which is the odor index predicted in step S12, and Y31(TON), which is the odor intensity, on an output device (not shown) of the control device 30.

[0127] Specifically, the control device 30 may display, for example, a time series graph G32 showing the time series change in odor intensity on the output device, as shown in FIG. 11(B).

[0128] The operator OP may then manually control the deodorizing device 20 by viewing the time series graph G32 displayed on the output device, for example.

[0129] [Odor Measurement System 900 in Second Comparative Example] Next, a description will be given of an odor measurement system 900 in a second comparative example. Fig. 12 is a diagram illustrating an example of the configuration of an odor measurement system 900 in a second comparative example.

[0130] The odor measurement system 900 includes, for example, the water treatment system 50 .

[0131] The water treatment system 50 is a sewage treatment system installed in a sewage plant, and includes, for example, a primary sedimentation tank 51, a reaction tank 52, a final sedimentation tank 53, a sterilization tank 54, and a thickening tank 55.

[0132] Specifically, the primary sedimentation tank 51 is a tank that separates, for example, organic matter and suspended solids contained in the water to be treated (sewage) by settling. The primary sedimentation tank 51 then discharges, for example, the separated organic matter and suspended solids as primary sedimentation sludge into a concentration tank 55, and discharges the water to be treated from which the organic matter and suspended solids have been separated into a reaction tank 52.

[0133] The reaction tank 52 is a tank that treats the water to be treated supplied from the primary sedimentation tank 51 by biological treatment such as the standard activated sludge method or the circulating nitrification-denitrification method.

[0134] The final settling tank 53 is a tank that separates and settles sludge contained in the water to be treated discharged from the reaction tank 52, and discharges the separated sludge as activated sludge. The final settling tank 53 supplies a portion of the activated sludge to the thickening tank 55 as excess sludge, and returns the activated sludge other than the excess sludge to the reaction tank 52 as returned sludge.

[0135] The sterilization tank 54 is, for example, a tank that sterilizes the water to be treated supplied from the final settling tank 53. Then, the sterilization tank 54 discharges the sterilized water to be treated (treated water) into, for example, a river or the like.

[0136] The thickening tank 55 thickens, for example, the primary sludge discharged from the primary sedimentation tank 51 and the excess sludge discharged from the final sedimentation tank 53, and supplies the thickened sludge to a digestion tank (not shown).

[0137] The odor measuring system 800 also includes, for example, a deodorizing device 60 and a control device 970.

[0138] The deodorizing device 60 is, for example, a device that deodorizes the water to be treated during treatment in the water treatment system 50. Specifically, the deodorizing device 60 deodorizes the water to be treated by using, for example, activated carbon or a deodorizing filter.

[0139] The control device 970 controls the deodorization device 60 based on, for example, odor information in the water treatment system 50. The control device 830 is, for example, a computer device having a CPU and memory. Below, a case where the odor information is an odor index will be described.

[0140] Specifically, as shown in Fig. 12, the operator OP acquires the water to be treated from at least one of the inlet pipe L2 that supplies the water to be treated to the primary sedimentation tank 51, the final sedimentation tank 53, the concentration tank 55, and the outlet pipe L3 through which the water to be treated (treated water) is discharged from the sterilization tank 54. The operator OP then measures the odor index of the acquired water to be treated, for example, using the operator OP's sense of smell. Furthermore, the operator OP inputs the measured odor index into the control device 970. Thereafter, the control device 970 controls the deodorizing device 60 (for example, controls the amount of activated carbon supplied and the setting of the deodorizing filter) based on the odor index input by the operator OP.

[0141] However, in this case, as in the case of the odor measurement system 800 in the first comparative example, the odor measurement system 900 imposes a heavy workload on the operator OP, and the odor index may not be measured stably due to individual differences in the measurement results between multiple operators OP, and the odor index may not be measured efficiently due to the positional relationship between the location where the water to be treated is obtained and the location where the odor index is measured. Furthermore, when an analyzer is used to analyze the odor index, costs associated with the use of the analyzer are required.

[0142] Therefore, odor prediction system 200 in this embodiment, like odor prediction system 100 in the first embodiment, generates a learning model that predicts, for example, the odor index of water to be treated in water treatment system 50. Then, odor prediction system 200 controls deodorization device 60 based on, for example, the odor index predicted by the learning model. The odor prediction system 200 in the second embodiment will be described below.

[0143] [Odor Prediction System 200 in the Second Embodiment] 13 is a diagram illustrating an example of the configuration of an odor prediction system 200 according to the second embodiment. Below, differences from the odor prediction system 200 according to the first embodiment will be described.

[0144] As shown in FIG. 13, the odor prediction system 200 includes, for example, a control device 70 and an information processing device 80.

[0145] In addition, as shown in FIG. 13, in the water treatment system 50, for example, measuring devices M11, M12, M13, M14, M15, M16, and M17 (hereinafter also referred to as measuring device M11, etc.) are installed in the inlet pipe L2, the primary sedimentation tank 51, the reaction tank 52, the final sedimentation tank 53, the sterilization tank 54, the outlet pipe L3, and the concentration tank 55, respectively.

[0146] Each of the measuring device M11 and the measuring device M12 may be, for example, at least one of a cyan meter, a water temperature meter, and an ammonia nitrogen meter. The measuring device M13 may be, for example, at least one of a dissolved oxygen concentration meter, a water temperature meter, a pH meter, an activated sludge suspended solids concentration meter, and an ammonia nitrogen meter. The measuring device M14 may be, for example, at least one of a turbidity meter, a nitrogen meter, a phosphorus meter, and a sludge interface meter. The measuring device M15 may be, for example, a pH meter. The measuring device M16 may be, for example, at least one of a cyan meter, a water temperature meter, and an ammonia nitrogen meter. The measuring device M17 may be, for example, at least one of an organic matter concentration meter and a sludge interface meter.

[0147] The control device 70 controls the deodorizing device 60 based on, for example, measurements taken in the water treatment system 10. The control device 70 is, for example, a computer device having a CPU and memory. Note that the control devices 30 and 70 may be, for example, a single computer device.

[0148] Specifically, the control device 70 acquires, for example, a measurement value measured by at least one of the measuring devices M11, etc., as shown in Fig. 13. Then, the control device 70 controls the deodorizing device 20 based on, for example, the acquired measurement value.

[0149] 3 and the like, the information processing device 80 performs, for example, a process (model generation process) for generating a learning model that predicts odor information of water to be treated in the water treatment system 50. The information processing device 80 is, for example, a computer device having a CPU and memory. Note that the control device 70 and the information processing device 80 may be, for example, a single computer device.

[0150] Then, as in the case described in Figure 4 etc., the control device 70 predicts the odor information output from the learning model MD in response to input of condition information DT11 indicating at least one of the state conditions of the water to be treated and the environmental conditions of the water to be treated as new odor information DT12 of the water to be treated.

[0151] As a result, the control device 70 in this embodiment can accurately predict the current odor information DT12 of the water to be treated in the water treatment system 50, similar to the case of the control device 30 in the first embodiment, for example.

[0152] Therefore, the odor prediction system 200 in this embodiment, similar to the odor prediction system 100 in the first embodiment, can efficiently deodorize the water to be treated in the water treatment system 50 and improve the deodorizing effect of the water to be treated. [Explanation of symbols]

[0153] 10: Water treatment system 11: Grit chamber 12: Landing well 13: Mixing pond 14: Flocculation tank 15: Sedimentation tank 16: Filtration pond 17: Purified water pond 18: Water reservoir 20: Deodorizing device 30: Control device 40: Information processing device 41: Storage device 42: Storage device 50: Water treatment system 51: Primary sedimentation tank 52: Reaction tank 53: Final settling tank 54: Sterilization tank 55: Concentration tank 60: Deodorizing device 70: Control device 80: Information processing device 100: Odor prediction system 200: Odor prediction system 800: Odor measurement system 830: Control device 900: Odor measurement system 970: Control device DT1: Odor information DT2: Condition information DT3: Training data DT3a: Teacher data DT3b: Teacher data DT11: Condition information DT12: Odor information G11: Calibration curve graph G12: Time series graph G21: Calibration curve graph G22: Time series graph G31: Calibration curve graph G32: Time series graph L1: Inflow pipe L2: Inflow pipe L3:Outflow pipe M1:Measuring device M2: Measuring device M3: Measuring device M4: Measuring device M5: Measuring device M6: Measuring device M7: Measuring device M8: Measuring device M9: Measuring device M11: Measuring device M12: Measuring device M13: Measuring device M14: Measuring device M15: Measuring device M16: Measuring device M17: Measuring device MD: Learning model OP: Operator

Claims

1. An odor prediction system comprising: a control device that inputs second condition information indicating a second state condition of new water to be treated and a second environmental condition indicating the environmental conditions of the water treatment system when the new water to be treated is treated in the water treatment system into a learning model generated by machine learning using a plurality of training data each including: first odor information of water to be treated being treated in a water treatment system; and first condition information indicating a first state condition regarding the state of the water to be treated and a first environmental condition indicating the environmental conditions of the water treatment system when the water to be treated is treated in the water treatment system; and outputs odor information output from the learning model in response to the input of the second condition information as the second odor information of the new water to be treated.

2. The water treatment system further includes a deodorizing device that deodorizes the water to be treated, The odor prediction system according to claim 1 , wherein the control device controls the deodorizing device based on the output second odor information.

3. The control device calculating the first concentration corresponding to the second odor information based on information indicating a relationship between the odor information of the water to be treated and a first concentration of a substance corresponding to each of one or more types of odor of the water to be treated; The odor prediction system according to claim 1 , wherein the calculated first concentration is output.

4. The control device calculating the second concentrations corresponding to the second odor information based on information indicating a relationship between the odor information of the water to be treated and second concentrations of one or more substances corresponding to specific types of odors in the water to be treated; The odor prediction system according to claim 1 , wherein the calculated second concentration is output.

5. A treatment system equipped with the odor prediction system according to claim 1, Further, a treatment system comprising the water treatment system.

6. An odor prediction system including an information processing device that generates a learning model that outputs second odor information for new treated water upon input of second condition information indicating second state conditions for new treated water and second environmental conditions indicating the environmental conditions of the water treatment system when the new treated water is treated in the water treatment system, through machine learning using a plurality of training data each including first odor information for the treated water being treated in a water treatment system, and first condition information indicating first state conditions for the state of the treated water and first environmental conditions indicating the environmental conditions of the water treatment system when the new treated water is treated in the water treatment system.

7. An odor prediction method comprising: inputting second condition information indicating a second state condition of new water to be treated and a second environmental condition indicating the environmental condition of the water treatment system when the new water to be treated is treated in the water treatment system into a learning model generated by machine learning using a plurality of training data each including: first odor information of water to be treated being treated in a water treatment system; and first condition information indicating a first state condition about the state of the water to be treated and a first environmental condition indicating the environmental condition of the water treatment system when the water to be treated is treated in the water treatment system; and outputting odor information output from the learning model in response to the input of the second condition information as the second odor information of the new water to be treated.

8. Acquire a plurality of training data each including first odor information of the water to be treated that is treated in the water treatment system, and first condition information indicating a first state condition regarding the state of the water to be treated and a first environmental condition indicating the environmental condition of the water treatment system when the water to be treated is treated in the water treatment system; A learning method for a learning model in which a computer performs processing to generate a learning model through machine learning using the acquired multiple pieces of training data, which outputs second odor information for the new treated water in response to input of second condition information indicating the second state condition of the new treated water and second environmental conditions indicating the environmental conditions of the water treatment system when the new treated water is treated in the water treatment system.

Images