Turbidity Prediction System

Abstract

To provide a turbidity prediction system capable of easily acquiring a timing as a predicted value when a first threshold value and a second threshold value are exceeded without monitoring by a monitor or an operator whether or not the turbidity always exceeds the first threshold value and the second threshold value.SOLUTION: The turbidity prediction system of the present invention includes a prediction unit that acquires predicted turbidity which is a predicted value of the turbidity of a river at a predetermined timing by using a machine learning model that predicts the turbidity from input data of the turbidity in a predetermined period in the past with respect to the timing and data relating to a flow rate in the period.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a turbidity prediction system. To Regarding. [Background technology]

[0002] Conventionally, water taken from rivers for use as tap water has been sent to a water purification plant, where it is purified to meet the necessary water quality standards before being supplied to users' taps. Water taken from rivers (hereinafter referred to as "taken water") contains various components that cause turbidity, such as turbidity caused by organic matter such as sediment, bacteria such as E. coli, and pH level, and purification is carried out at water purification plants according to each component (see, for example, Patent Document 1). In the above purification process, the purification process for turbidity is a particularly important item because the evaluation is easy to visually confirm, that is, transparency.

[0003] Turbidity in water taken from a river is greatly affected by rainfall. The more rain that falls, the higher the river's water level becomes. This large amount of flowing water erodes the sediment in the riverbed and riverbank as it moves. Turbidity is managed by measuring it every hour using a turbidity meter installed at a designated location upstream of the water intake for the water purification plant. Generally, turbidity is measured using a turbidity meter by a manager who visually reads the values.

[0004] If the turbidity exceeds the specified first threshold and the water purification plant enters the high turbidity response preparation zone, the plant will gather manpower and begin preparations for alert deployment. In addition, if the turbidity exceeds the second threshold and the area becomes a high turbidity alert zone, preparations will be made for further measures, such as managing the intake pumps that take water from the river to prevent them from becoming clogged due to dirt, increasing the amount of chemicals that cause the dirt to settle depending on the turbidity, and issuing an alert to users in the worst-case scenario where the water quality is not satisfactory.

[0005] In other words, in the event of flooding caused by heavy rain or typhoon-induced rainfall, each water purification plant will prepare for and process alert deployment according to a timeline corresponding to each of the above-mentioned first and second thresholds that have been predetermined. At this time, if the turbidity exceeds the first threshold, the manager will assemble manpower on alert, and if the turbidity exceeds the second threshold, the assembled manpower will monitor for blockages in the intake pump and begin preparations for adjusting chemicals to settle the dirt. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2007-146422 A Summary of the Invention [Problem to be solved by the invention]

[0007] However, when there is heavy rain or rainfall caused by a typhoon, it is necessary to constantly monitor the turbidity meter until the first threshold is exceeded, and furthermore, the burden of considering when to gather manpower falls on the manager. In addition, after manpower is gathered, workers must constantly monitor for blockages in the intake pump and wait while preparing to adjust the chemicals that settle the dirt until the second threshold is actually exceeded, which places a burden on the workers.

[0008] The present invention has been made in consideration of the above circumstances, and provides a turbidity prediction system that enables a supervisor or operator to easily obtain a predicted value for the timing at which the turbidity will exceed a first threshold value and a second threshold value, without constantly monitoring whether the turbidity exceeds a first threshold value and a second threshold value. M The purpose is to provide. [Means for solving the problem]

[0009] In order to solve the above problem, the turbidity prediction system of the present invention includes a prediction unit that obtains a predicted turbidity, which is a predicted value of the turbidity of a river at a predetermined timing, from input data of the turbidity for a predetermined period in the past relative to the timing and related data, which is data related to the flow rate for the period, using a machine learning model. The machine learning model is configured to predict the predicted turbidity in hourly units, and the timing is the predicted turbidity n hours from the current time, and the predicted turbidity is output as an hourly predicted turbidity based on the turbidity measured at each time from the current time to m hours ago and the related data measured at each time from the current time to m hours ago; and the machine learning model is configured to predict the predicted turbidity on a daily basis, and the timing is the predicted turbidity p days from the current day, and the predicted turbidity is output as the maximum turbidity value p days from the current day using as input data the maximum value of the turbidity measured hourly on each day from the current day to q days ago and the related data, which are weather data for each day from the day after the current day to the day before the p day, and if the predicted daily maximum turbidity value exceeds the daily pollution threshold, the prediction result output unit notifies that the zone will become a high turbidity response preparation zone in p days from now, and if the predicted predicted turbidity data exceeds the hourly pollution threshold, the prediction result output unit notifies that the zone will become a high turbidity alert deployment zone in n hours from now. It is characterized by:

[0012] The turbidity prediction system of the present invention is characterized in that the input data to the machine learning model includes the maximum values ​​in the associated data measured every hour on each day from the current day to q days ago, which is the associated data. Effect of the Invention

[0013] According to the present invention, there is provided a turbidity prediction system that allows a supervisor or an operator to easily obtain a predicted value of the timing at which the turbidity will exceed the first threshold value and the second threshold value without constantly monitoring whether the turbidity exceeds the first threshold value and the second threshold value. M can be provided. [Brief description of the drawings]

[0014] [Figure 1] 1 is a block diagram showing an example of the configuration of a turbidity prediction system according to a first embodiment of the present invention. [Diagram 2] FIG. 2 is a conceptual diagram showing the relationship between input data and output data of the time unit turbidity prediction model in the first embodiment. [Diagram 3] FIG. 2 is a conceptual diagram showing a correspondence relationship between a time unit turbidity prediction model and a time to be predicted in the first embodiment. [Figure 4] 11 is a flowchart showing an example of the operation of a process for generating a turbidity prediction model for a predetermined time ahead and predicting turbidity for a predetermined time ahead in the turbidity prediction system according to this embodiment. [Diagram 5] FIG. 11 is a block diagram showing an example of the configuration of a turbidity prediction system according to a second embodiment of the present invention. [Figure 6]FIG. 11 is a conceptual diagram showing an example of a relationship between input data and output data of a time unit turbidity prediction model in the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Fig. 1 is a block diagram showing an example of the configuration of a turbidity prediction system according to a first embodiment of the present invention. In Fig. 1, a turbidity prediction system 10 includes a data input / output unit 11, a time unit turbidity prediction model generation unit 12, a time unit turbidity prediction unit 13, a prediction result output unit 14, a past data storage unit 15, a machine learning model storage unit 16, and a prediction result storage unit 17. The data input / output unit 11 reads teaching data used for learning the time unit turbidity prediction model, which is a machine learning model, from an external device, and writes and stores the data in the past data storage unit 15.

[0016] Here, the teacher data is turbidity data, which is data on turbidity (mg / L) over the past few years, for example, 3.5 years, and water level data, which is data on water level (m) over the past 3.5 years. The turbidity data is data acquired by a turbidity meter placed near the water intake of the water purification plant in an hourly cycle from 0:00 to 23:00. The water level data is data acquired by a water level meter placed upstream of the river where the water intake of the water purification plant is located, preferably the water level meter closest to the water intake, in an hourly cycle from 0:00 to 23:00.

[0017] In this embodiment, the water level is used, but related data that is correlated with the water volume in the vicinity of the intake of the river, such as flow velocity, may also be used. That is, the flow rate is represented by the cross-sectional area perpendicular to the flow direction of the water flowing in the river and the flow velocity of the water. The water level corresponds to a height that represents the cross-sectional area of ​​the water flowing in the river. The data input / output unit 11 writes and stores the turbidity data and water level data as teaching data in association with a time stamp (date and time).

[0018] In a mode for learning a machine learning model, the hourly turbidity prediction model generating unit 12 generates an hourly turbidity prediction model by learning a predetermined machine learning model using each of the past turbidity data and water level data stored in the past data storage unit 15 as teacher data. In the embodiment described below, the machine learning model in this embodiment is configured to input a predetermined number of input data, for example, six input data, and output one predicted value as output data, and deep learning, a neural network, or the like is used.

[0019] The hourly turbidity prediction model in this embodiment will be described below. Fig. 2 is a conceptual diagram showing an example of the relationship between input data and output data of the hourly turbidity prediction model in this embodiment. Fig. 2 shows the relationship between the turbidity data and water level data, which are input data of the hourly turbidity prediction model that outputs turbidity data five hours ahead from the current time as a predicted value, and the turbidity data five hours ahead (n hours ahead) from the current time (current time), which is output data. This time unit of five hours ahead is set because the time it takes for the turbidity data value to transition from the high turbidity response preparation zone to the high turbidity alert deployment zone is five hours in past data statistics.

[0020] Here, when the time to be predicted is five hours from the current time, the turbidity data used is the measurement values ​​(actual values) at the current time, one hour ago, and two hours ago (m hours ago) for the time to be predicted. Similarly, when the time to be predicted is five hours from the current time, the water level data used is the measurement values ​​(actual values) at the current time, one hour ago, and two hours ago for the time to be predicted. The hourly turbidity prediction model inputs the turbidity data and water level data, which are the above-mentioned past measurement values, as input data, and outputs a predicted value of the turbidity data for the time to be predicted.

[0021] Returning to FIG. 1, when generating the above-mentioned hourly turbidity prediction model, the hourly turbidity prediction model generating unit 12 extracts teacher data for training the machine learning model from past data to generate it. The hourly turbidity prediction model generating unit 12 extracts turbidity data and water level data in the order of oldest time stamp from the past data as turbidity data and water level data from two hours ago, one hour ago, and the current time, and similarly extracts turbidity data from the past data five hours ahead of the current time as a label, to generate teacher data consisting of six input data and one output data label. Then, the hourly turbidity prediction model generating unit 12 shifts the process of extracting the hourly turbidity data and water level data as the above-mentioned input data and the turbidity data as the label in the past data in the past data storage unit 15 in the time progression direction by one hour at a time, to sequentially generate teacher data for training the machine learning model.

[0022] The hourly turbidity prediction model generating unit 12 inputs, for example, six pieces of input data, namely, turbidity data and water level data for two hours before, one hour before, and the current time of the prediction target, and causes the machine learning model to learn so that turbidity data is output as a label of the prediction target time. At this time, the hourly turbidity prediction model generating unit 12 repeats learning using all teacher data extracted from past data a preset number of times to generate the hourly turbidity prediction model.

[0023] The time-unit turbidity prediction model generating unit 12 repeats the process until the root mean square of the difference between the turbidity data of the labels in the teacher data and the predicted turbidity data falls within a preset range. The time-unit turbidity prediction model generating unit 12 then writes and stores the generated time-unit turbidity prediction model in the machine learning model storage unit 16.

[0024] In a mode for outputting a predicted value, when measurement values ​​of turbidity data and water level data for the current time, one hour ago, and two hours ago for the time to be predicted are input via the data input / output unit 11, the hourly turbidity prediction unit 13 reads out an hourly turbidity prediction model from the machine learning model storage unit 16. Then, the hourly turbidity prediction unit 13 inputs six pieces of data, namely, turbidity data and water level data for the current time, one hour ago, and two hours ago, as input data to the read hourly turbidity prediction model.

[0025] As a result, the hourly turbidity prediction model outputs a predicted value of turbidity data (predicted turbidity data) for the prediction target time in response to the input data. The time unit turbidity prediction unit 13 outputs the predicted turbidity data output by the time unit turbidity prediction model to the prediction result output unit 14, and also writes the data in the prediction result storage unit 17 for storage.

[0026] The prediction result output unit 14 supplies the predicted turbidity data supplied from the hourly turbidity prediction unit 13, together with the date and time indicating the time to be predicted, to a mobile terminal (smartphone) carried by a manager and an operator at the water purification plant. In the above-described embodiment, the time to be predicted for which the hourly turbidity prediction model obtains predicted turbidity data is set to 5 hours from the current time, but the time to be predicted for the predicted pollution may be set to any number of hours in the future, such as 2 hours, 3 hours, 4 hours, 6 hours, etc., as long as it is 1 hour or more in the future.

[0027] For example, if the time to be predicted for which predicted turbidity data is to be obtained is one hour ahead of the current time, the current time, one hour ago, and two hours ago are input data for the time to be predicted into the hourly turbidity prediction model (one-hour-ahead turbidity prediction model described below). In this case, past data to be used as turbidity data and water level data for two hours ago, one hour ago, and the current time in the training data are extracted in order of oldest timestamp from the past data, and training data consisting of six input data and one output data label is generated using the turbidity data one hour ahead of the current time from the past data as a label, and a machine learning model is trained using each of these training data to generate an hourly turbidity prediction model.

[0028] Fig. 3 is a conceptual diagram showing the correspondence relationship between the hourly turbidity prediction model and the time of the prediction target in the first embodiment. In Fig. 3, the 5-hour turbidity prediction model is an hourly turbidity prediction model generated with, for example, 5 hours from the current time as the time of the prediction target for outputting the predicted pollution. Similarly, each of the other 1-hour, 2-hour, 3-hour, and 4-hour turbidity prediction models is an hourly turbidity prediction model generated with, for example, 1 hour, 2 hours, 3 hours, and 4 hours from the current time as the time of the prediction target for outputting the predicted pollution.

[0029] For example, the 1-hour-ahead turbidity prediction model and the 3-hour-ahead turbidity prediction model each use the same turbidity data and water level data for the current time, and 1 hour, 2 hours, 3 hours, and 4 hours ago as input data, but are hourly turbidity prediction models that predict pollution for different times, 1 hour and 3 hours ahead.

[0030] When extracting teacher data to be used in generating a one-hour-ahead turbidity prediction model, turbidity data and water level data from past data are extracted in order of oldest time stamp as turbidity data and water level data from 4 hours ago, 3 hours ago, 2 hours ago, 1 hour ago, and the current time, and turbidity data from the current time one hour ahead is extracted as a label (output data), generating teacher data consisting of 10 input data and 1 output data label. The hourly turbidity prediction model generation unit 12 trains a machine learning model using the above-mentioned teacher data to generate a one-hour-ahead turbidity prediction model.

[0031] Furthermore, when extracting teacher data used to generate a three-hour-ahead turbidity prediction model, turbidity data and water level data from the past data are extracted in order of oldest time stamp as turbidity data and water level data from four hours ago, three hours ago, two hours ago, one hour ago, and the current time, and turbidity data from three hours ahead from the current time is extracted as a label (output data), generating teacher data consisting of ten input data and one output data label. The hourly turbidity prediction model generation unit 12 trains a machine learning model using the above-mentioned teacher data to generate a three-hour-ahead turbidity prediction model.

[0032] 4 is a flowchart showing an example of the operation of a process for generating a turbidity prediction model for a predetermined time ahead and predicting turbidity for a predetermined time ahead in the turbidity prediction system according to this embodiment. As an example, a case will be described in which the hourly turbidity prediction model is a 5-hour turbidity prediction model that predicts turbidity data (predicted turbidity data) for 5 hours ahead based on turbidity data and water level data for the current time, 1 hour ago, and 2 hours ago.

[0033] Step S11: The data input / output unit 11 reads each of the turbidity data and water level data acquired in hourly increments over multiple years from an external device as past data, and writes and stores it in the past data storage unit 15. Here, the data input / output unit 11 writes and stores each of the turbidity data and water level data at the current time acquired every hour in the past data storage unit 15, and performs an editing process of adding it to the above-mentioned past data.

[0034] Step S12: The hourly turbidity prediction model generating unit 12 uses each of the past turbidity data and water level data stored in the past data storage unit 15 as teacher data to train a predetermined machine learning model, and generates an hourly turbidity prediction model (such as the above-mentioned 1-hour ahead turbidity prediction model to 5-hour ahead turbidity prediction model) that outputs a predicted value of turbidity data a predetermined time ahead. At this time, the hourly turbidity prediction model generating unit 12 reads, from the machine learning model storage unit 16, a machine learning model having a configuration corresponding to the number of input data and the number of output data of the hourly turbidity prediction model to be generated, as the machine learning model to be trained.

[0035] Furthermore, when an hourly turbidity prediction model predicting pollution at a prediction target time has already been generated, the hourly turbidity prediction model generation unit 12 performs additional learning of the hourly turbidity prediction model by including new turbidity data and water level data in the past data only when new turbidity data and water level data are added to the past data.The hourly turbidity prediction model generation unit 12 then writes the generated hourly turbidity prediction model in the machine learning model storage unit 16 for storage.

[0036] Step S13: The time-unit turbidity prediction unit 13 judges whether or not a request to predict predicted pollution has been input from an input means (not shown). At this time, if a request to predict predicted pollution has been input from the input means, the time-unit turbidity prediction unit 13 advances the process to step S13. On the other hand, if a request to predict predicted pollution has been input from the input means, the time-unit turbidity prediction unit 13 advances the process to step S11.

[0037] Step S14: When a request to predict pollution is input from an input means (not shown), the hourly turbidity prediction unit 13 reads out an hourly turbidity prediction model from the machine learning model storage unit 16. Then, for the hourly turbidity prediction model that has been read out, the hourly turbidity prediction unit 13 extracts turbidity data and water level data for the current time, one hour ago, and two hours ago from the past data stored in the past data storage unit 15 (acquisition of input data). Here, when turbidity data and water level data for the current time are added to the request, this data is used.

[0038] Step S15: The hourly turbidity prediction unit 13 inputs the turbidity data and water level data of the current time, one hour ago, and two hours ago as input data to the hourly turbidity prediction model. Here, the hourly turbidity prediction unit 13 obtains output data that the hourly turbidity prediction model predicts and outputs based on input data as predicted turbidity data for five hours ahead.

[0039] Step S16: The prediction result output unit 14 compares the predicted turbidity data supplied from the time-unit turbidity prediction unit 13 with a preset time-unit pollution threshold value (second threshold value). Then, when the predicted turbidity data is equal to or greater than the hourly pollution threshold, the prediction result output unit 14 supplies the information that the pollution will exceed the hourly pollution threshold in 5 hours and there is a possibility that the zone will become a high turbidity alert zone, together with the date and time indicating the time to be predicted, to mobile terminals carried by managers and workers at the water purification plant.Then, the prediction result output unit 14 advances the process to step S13.

[0040] On the other hand, if the predicted turbidity data is less than the hourly pollution threshold, the prediction result output unit 14 supplies the information that the pollution will not exceed the hourly pollution threshold in 5 hours and the water will not become a high turbidity alert deployment zone, together with the date and time indicating the time to be predicted, to mobile terminals carried by managers and workers at the water purification plant.Then, the prediction result output unit 14 advances the process to step S13.

[0041] According to the present embodiment, the above-mentioned configuration uses the hourly turbidity prediction model, and the turbidity data and water level data at the current time and a predetermined time before in the past data are used as input data, and the turbidity at a future time to be predicted relative to the current time can be obtained as predicted turbidity data. Therefore, according to the present embodiment, the time when the pollution becomes a high turbidity alert deployment zone is detected as the predicted time by comparing every hour with the hourly pollution threshold value for determining that the pollution is a high turbidity alert deployment zone, thereby shortening the time required by workers to check for blockage of the water intake at the water purification plant and to prepare for the adjustment of the chemicals that settle the pollution, thereby reducing the burden on managers and workers.

[0042] In this embodiment, there is one river from an upstream water source such as a dam to the water intake of the water purification plant. On the other hand, when rivers from multiple water sources merge into one river upstream of the water purification plant, the input data are the turbidity data and water level data of the past data of the multiple rivers before the merger and the river after the merger, and an hourly turbidity prediction model is generated and used to estimate the turbidity at the water intake of the water purification plant. For example, when taking water from a river where three rivers merge, the input data are 24 pieces of turbidity data and water level data of the three rivers before the merger and the river after the merger, including the current time, 1 hour ago, and 2 hours ago.

[0043] <Second embodiment> Fig. 5 is a block diagram showing an example of the configuration of a turbidity prediction system according to a second embodiment of the present invention. In Fig. 5, a turbidity prediction system 20 includes a data input / output unit 21, a daily turbidity prediction model generation unit 22, a daily turbidity prediction unit 23, a prediction result output unit 24, a past data storage unit 25, a machine learning model storage unit 26, and a prediction result storage unit 27. The data input / output unit 21 has the same configuration as the data input / output unit 21 in the first embodiment.

[0044] In a mode for learning a machine learning model, the daily turbidity prediction model generating unit 22 generates a daily turbidity prediction model by learning a predetermined machine learning model using each of the past turbidity data and water level data stored in the past data storage unit 25 as teacher data. In the embodiment described below, the machine learning model is configured to input, for example, six pieces of input data and output one predicted value as output data, and deep learning, a neural network, or the like is used.

[0045] The daily turbidity prediction model in this embodiment will be described below. Fig. 6 is a conceptual diagram showing an example of the relationship between input data and output data of the hourly turbidity prediction model in this embodiment. Fig. 6 shows the relationship between the turbidity data and water level data, which are input data of the daily turbidity prediction model that outputs turbidity data for 7 days (one week) from the current day (the current day as the day for making the prediction), and the turbidity data for 7 days ahead, which is output data. This 7-day unit is set because when the turbidity reaches the high turbidity response preparation zone, the water purification plant manager needs to gather workers, but preparations to gather workers need to be made 7 days in advance.

[0046] Figure 6(a) shows a daily turbidity prediction model that is configured to predict turbidity (daily maximum value, which is the maximum value for one day) on a target day by inputting as input data the daily maximum value of turbidity (daily maximum value of turbidity) for the current day and past data, as well as forecast data from the day after the current day to the day before the target day. Here, the daily maximum turbidity value refers to the maximum turbidity value among 24 turbidity values ​​acquired every hour from 0:00 to 23:00 in a day. The forecast data also indicates, for example, the total amount of rainfall predicted in a day by the weather forecast, or the name of the weather, such as rain, cloudy, sunny, or snow.

[0047] 6(a), when generating the above-mentioned daily turbidity prediction model, the daily turbidity prediction model generating unit 22 extracts teacher data for training the machine learning model from past data to generate it. The hourly turbidity prediction model generating unit 12 extracts the daily maximum turbidity values ​​for each day in the past data in order of the oldest day in the timestamp, and extracts each of the daily maximum turbidity values ​​in the turbidity data for one day as the daily maximum turbidity values ​​for two days ago, one day ago, and the current day, and similarly extracts the daily maximum turbidity value for the day 7 days ahead of the current day from the past data as a label.

[0048] The hourly turbidity prediction model generating unit 12 also extracts weather data for six days ahead (the day before seven days ahead) from the day after the current day. Here, the weather data is included in one of the related data described in the first embodiment, since the state and amount of rainfall are related to the river flow rate. The daily turbidity prediction model generating unit 22 then generates training data consisting of 10 input data and a label for one output data. The daily turbidity prediction model generating unit 22 then shifts the process of extracting the daily maximum turbidity value and forecast data on a daily basis as input data and the daily maximum turbidity value as a label from the past data in the past data storage unit 15 by one day in the time direction, thereby sequentially generating training data for training the machine learning model.

[0049] Fig. 6(b) shows a daily turbidity prediction model configured to predict turbidity (daily maximum value, which is the maximum value for one day) on a day to be predicted by inputting the following data: the daily maximum turbidity values ​​for the current day and past data, the daily maximum water level values ​​(daily maximum water level values) for the current day and past data as external data, and forecast data from the day after the current day to the day before the day to be predicted. Here, the daily maximum water level value indicates the maximum water level value among 24 water levels acquired every hour from 0:00 to 23:00 in a day.

[0050] In Figure 6(b), the hourly turbidity prediction model generation unit 12 extracts the daily maximum turbidity values ​​for each day in the past data in order of the oldest day's timestamp, and extracts each of the daily maximum values ​​in the turbidity data for one day as the maximum turbidity values ​​for two days ago, one day ago, and the current day.Similarly, it extracts the daily maximum turbidity value for the day 7 days ahead of the current day from the past data as a label. In addition, the hourly turbidity prediction model generation unit 12 extracts the daily maximum turbidity values ​​for each day in the past data in order of the oldest timestamp, and extracts each of the daily maximum values ​​in the water level data for one day as the daily maximum water level values ​​for two days ago, one day ago, and the current day.

[0051] The hourly turbidity prediction model generation unit 12 also extracts weather data for six days from the day after the current day (the day before the day seven days from the current day). The daily turbidity prediction model generation unit 22 then generates teacher data consisting of 13 input data and a label for one output data. The daily turbidity prediction model generation unit 22 then moves the process of extracting the daily maximum turbidity value, daily maximum water level value, and forecast data on a daily basis as the above-mentioned input data, and the daily maximum turbidity value as a label, from the past data in the past data storage unit 15, one day at a time in the time direction, to sequentially generate teacher data for training the machine learning model.

[0052] In the case of the daily turbidity prediction model shown in Fig. 6(a), when the daily maximum turbidity values ​​for the current time, one day ago, and two days ago and weather data for one day to six days ahead are input via the data input / output unit 21 in a mode for outputting predicted values, the daily turbidity prediction unit 23 inputs these as input data to the daily turbidity prediction model (Fig. 6(a)) read out by the machine learning model storage unit 16. As a result, the daily turbidity prediction model shown in Fig. 6(a) outputs a predicted value of the daily maximum turbidity value for the day to be predicted (predicted daily maximum turbidity value) in response to the input data.

[0053] On the other hand, in the case of the daily turbidity prediction model shown in Fig. 6(b), when the daily maximum turbidity and daily maximum water level values ​​for the current time, one day before, and two days before, and weather data for one day to six days before are input via the data input / output unit 21 in a mode for outputting predicted values, the daily turbidity prediction unit 23 inputs these as input data to the daily turbidity prediction model (Fig. 6(b)) read out by the machine learning model storage unit 16. As a result, the daily turbidity prediction model shown in Fig. 6(b) outputs a predicted value of the daily maximum turbidity for the day to be predicted (predicted daily maximum turbidity) in response to the input data.

[0054] The prediction result output unit 24 supplies the predicted daily maximum turbidity value supplied from the daily turbidity prediction unit 23 together with the date and time indicating the day to be predicted to mobile terminals carried by managers and workers at the water purification plant. The prediction result output unit 24 also writes and stores the predicted daily maximum turbidity value together with the date and time indicating the day to be predicted in the prediction result storage unit 27. In the embodiment described above, the day for which the daily turbidity prediction model obtains the predicted daily maximum turbidity is set to seven days from the current day, but the day for which the daily maximum turbidity is predicted may be set to any number of days in the future, such as one day or more from the current day, two days to six days, or eight days or more.

[0055] According to the present embodiment, the above-mentioned configuration uses a daily turbidity prediction model, and the daily maximum values ​​of the turbidity data (and water level data) for the current day and a predetermined number of days ago in the past data, and the weather data from the day after the current day to the day before the prediction are used as input data, and the daily maximum value of turbidity at the future prediction target time relative to the current day can be acquired as the predicted daily maximum value of turbidity. Therefore, according to the present embodiment, the daily maximum value of turbidity is compared with the daily pollution threshold value (first threshold value) for determining the high turbidity response preparation zone on a daily basis, and the time when the daily maximum value of turbidity becomes the high turbidity response preparation zone is detected as the prediction target day, so that the manager can prepare in advance to check the blockage of the water intake at the water purification plant and to arrange workers to adjust the chemicals that settle the dirt, and since there is no need to gather workers early, the waiting time at the water purification plant can be shortened and the burden on the manager and workers can be reduced.

[0056] In the above-described embodiment, as an example, the maximum value of the turbidity data in one day, that is, the daily maximum value of the turbidity data, is used as the value to be compared with the daily pollution threshold value, which indicates that the zone has become a preparation zone for high turbidity and is empirically set based on daily changes in past turbidity data. However, as another example of the configuration of this embodiment, the daily pollution threshold value may be, for example, a daily average turbidity, which is the average value of the turbidity data on days when the daily maximum turbidity value is equal to or greater than a predetermined value, and the minimum value of each of the calculated daily average turbidities may be extracted and used, as shown below. In this configuration, when generating a daily turbidity prediction model, rather than using the above-mentioned daily maximum turbidity value as the turbidity data in the training data, a daily average turbidity, which is the average value of the turbidity data measured in a day, is used, and a daily turbidity prediction model is generated that outputs the predicted daily average turbidity as the output predicted value.

[0057] Then, in the past turbidity data used as the training data, the days on which the daily maximum turbidity value exceeds the threshold value previously used to set up the high turbidity preparation zone (for example, 100 mg / L, i.e., the above-mentioned predetermined value) are extracted as candidate days. Then, the daily average turbidity is calculated from the turbidity data on each of the extracted candidate days, and the minimum daily average turbidity is extracted as the daily pollution threshold value. That is, the daily average turbidity of each of the candidate days is compared, and the minimum daily average turbidity (for example, 30 mg / L) is extracted from the daily average turbidity of each candidate day, and this extracted daily average turbidity is set as the daily pollution threshold value.

[0058] That is, in the case of this embodiment, since the daily change in turbidity is usually the daily average of turbidity data, the prediction accuracy of the daily turbidity prediction model can be improved by training the model to predict the predicted daily average turbidity as the daily average change in turbidity, compared to when the daily turbidity prediction model is trained to predict the daily maximum value of predicted turbidity, which is unknown to occur at any time during the day.

[0059] Furthermore, by setting the minimum daily average turbidity among the daily average turbidity for each of the candidate days as the daily pollution threshold, as described above, rather than the threshold used as the high turbidity response preparation zone, it is expected that the day on which the predicted daily average turbidity predicted by the daily turbidity prediction model for each of the candidate days exceeds the daily pollution threshold will relatively correspond to the day on which the daily maximum turbidity value exceeds the threshold used as the high turbidity response preparation zone.

[0060] In addition, the above-mentioned specified numerical value was previously used as the threshold value for determining whether a zone was a high turbidity response preparation zone, but when generating a daily turbidity prediction model for calculating the predicted daily average turbidity seven days into the future, it is also possible to use, in the past turbidity data used as training data, the minimum daily average turbidity among the daily average turbidity values ​​from the day on which the daily maximum turbidity value seven days into the future becomes the high turbidity alert deployment zone to two days prior to this day, or the average daily average turbidity value for each day.

[0061] In this embodiment, as in the first embodiment, there is one river from an upstream water source such as a dam to the water intake of the water purification plant. On the other hand, when rivers from multiple water sources merge into one river upstream of the water purification plant, the input data are the daily maximum turbidity (and daily maximum water level) and weather data for each day of past data for the multiple rivers before the merger and the river after the merger, and a daily turbidity prediction model is generated and used to estimate the turbidity at the water intake of the water purification plant. For example, when water is taken from a river where three rivers merge, the input data are the daily maximum turbidity (and daily maximum water level) and weather data for the current day, one day before, and two days before for the three rivers before the merger and the river after the merger.

[0062] <Third embodiment> The third embodiment is a turbidity prediction system 1 configured to use a combination of the functions of the turbidity prediction system 10 of the first embodiment and the functions of the turbidity prediction system 20 of the second embodiment. Here, in the turbidity prediction system 1, for example, the turbidity prediction system 10 predicts turbidity data five hours ahead as predicted turbidity data, and the turbidity prediction system 20 predicts the daily maximum turbidity value seven days ahead as the predicted daily maximum turbidity value.

[0063] Then, when the daily maximum turbidity value predicted by the turbidity prediction system 20 exceeds the daily pollution threshold, the prediction result output unit 24 notifies the manager's mobile device that the area will become a high turbidity response preparation zone in 7 days from now. This allows the manager to start preparing to arrange for workers on the current day when the daily maximum turbidity value exceeds the daily pollution threshold. The manager also arranges for workers within 7 days after the day the daily pollution threshold is exceeded, and starts predicting turbidity data by hour using the turbidity prediction system 10 from 7 days after the day the daily pollution threshold is exceeded.

[0064] When the predicted turbidity data predicted by the turbidity prediction system 10 exceeds the hourly pollution threshold, the prediction result output unit 14 notifies the manager's mobile terminal that the area will become a high turbidity alert deployment zone in five hours. Upon receiving this notification, the manager performs processing to deploy workers to the water intake and settling tank of the water purification plant within five hours from the current time.

[0065] According to this embodiment, it is possible to predict the day when the daily maximum turbidity value is likely to become a high turbidity response preparation zone, and also to predict the time when the turbidity data may become a high turbidity alert deployment zone, so that the manager can make preparations in advance to check for blockages in the water intake at the water purification plant and arrange for workers to adjust the chemicals to settle the dirt. It is also possible to gather workers earlier and shorten the long period of time that workers are kept waiting at the water purification plant, and further to shorten the time that workers spend observing the water intake, settling tank, etc., thereby reducing the burden on the manager and workers.

[0066] In addition, a program for realizing the functions of the turbidity prediction system 10 of FIG. 1 and the turbidity prediction system 20 of FIG. 4 in the present invention may be recorded in a computer-readable recording medium, and the program recorded in the recording medium may be read into a computer system and executed to predict the pollution at a specific prediction time or the maximum daily turbidity value on a specific prediction day. The term "computer system" as used herein includes hardware such as an OS (Operating System) and peripheral devices. The term "computer system" also includes a WWW (World Wide Web) system equipped with a homepage providing environment (or display environment). The term "computer-readable recording medium" refers to portable media such as flexible disks, optical magnetic disks, ROMs (Read Only Memory), and CD-ROMs (Compact Disc - Read Only Memory), and storage devices such as hard disks built into a computer system. The term "computer-readable recording medium" also includes those that hold a program for a certain period of time, such as volatile memory (RAM (Random Access Memory)) inside a computer system that becomes a server or client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line.

[0067] The above program may also be transmitted from a computer system in which the program is stored in a storage device or the like to another computer system via a transmission medium, or by transmission waves in the transmission medium. Here, the "transmission medium" that transmits the program refers to a medium that has the function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The above program may also be for realizing part of the above-mentioned functions. Furthermore, it may be a so-called difference file (difference program) that can realize the above-mentioned functions in combination with a program already recorded in the computer system. [Explanation of symbols]

[0068] 10, 20... Turbidity prediction system 11, 21... Data input / output unit 12... Hourly turbidity prediction model generation unit 13... Hourly turbidity prediction unit 14, 24... Prediction result output unit 15, 25... Past data storage unit 16, 26... Machine learning model storage unit 17, 27... Prediction result storage unit 22... Daily turbidity prediction model generation unit 23... Daily turbidity prediction unit

Claims

1. A prediction unit that obtains a predicted turbidity, which is a predicted value of the turbidity of the river at a specified timing, using a machine learning model that predicts the predicted turbidity from input data of the turbidity for a specified period in the past relative to the specified timing and related data, which is data related to the flow rate for the specified period. Equipped with The machine learning model is configured to predict the predicted turbidity in one-hour units, and outputs the predicted turbidity n hours from the current time as the timing as an hourly predicted turbidity based on the turbidity measured at each time from the current time to m hours ago and the related data measured at each time from the current time to m hours ago, and, The machine learning model is configured to predict the predicted turbidity on a daily basis, and outputs the predicted turbidity p days from the current day as the timing, using as input data the maximum value of the turbidity measured hourly on each day from the current day to q days ago, and weather data on each day from the day after the current day to the day before the pth day, which is the related data, as the predicted turbidity, as the maximum value of the turbidity on a day p days away; When the predicted daily maximum turbidity value exceeds the daily pollution threshold value, the prediction result output unit notifies that the zone will be a high turbidity response preparation zone in p days ahead, A turbidity prediction system characterized in that, when the predicted turbidity data exceeds the hourly pollution threshold, the prediction result output unit notifies that a high turbidity alert deployment zone will occur n hours in the future.

2. The input data to the machine learning model includes a maximum value in the relevant data measured every hour on each day from the current day to q days ago, which is the relevant data; The turbidity prediction system according to claim 1 .

Images