JP2006002401A - Sewage inflow quantity predicting device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a sewage inflow quantity predicting device for improving accuracy in predicting a sewage inflow quantity, by accurately predicting a sewage inflow quantity and a rainwater inflow quantity by an uncomplicated sewage inflow quantity predicting model and a rainwater inflow quantity predicting model which utilize a general technology. <P>SOLUTION: This sewage inflow quantity predicting device forms sewage inflow quantity predicting data, by adding up formed sewage inflow quantity predicting data and rainwater inflow quantity predicting data, by forming the rainwater inflow quantity predicting data by inputting meteorological data to the rainwater inflow quantity predicting model, by forming the sewage inflow quantity predicting data by inputting calendar data to the sewage inflow quantity predicting model, by separately constructing the sewage inflow quantity predicting model and the rainwater inflow quantity predicting model, by using rainwater inflow quantity data and sewage inflow quantity data separately formed from sewage inflow quantity data being a sewage inflow quantity to a pump well. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a sewage inflow prediction device for predicting a sewage inflow amount of sewage flowing into a pump well of a sewage pump station or a final treatment plant.

  In order to efficiently treat sewage at the final treatment plant, it is desirable that the fluctuation of the daily sewage inflow is small. In order to reduce fluctuations in the amount of sewage inflow in this way, at present, the sewage inflow amount that flows into the pump wells that are placed in the middle or at the end of the sewage system is predicted, and the start / stop of the pump is controlled, so that The fluctuation of the amount of inflow of sewage flowing into the field is reduced.

  The RRL method (Road Research Laboratory Method) is mainly used as a method for predicting the amount of sewage inflow according to the prior art. In recent years, in addition to the RRL method, a sewage inflow prediction method has been developed. For example, Patent Documents 1 to 3 are disclosed.

  The conventional technique described in Patent Document 1 (Japanese Patent Laid-Open No. Hei 5-134715, “Neural Network Applied Rainwater Inflow Prediction Device”) uses a radar rain gauge and a neural network. It is an amount prediction, can cope with aging by the learning function of the neural network, and by using a radar rain gauge, it can use not only the rain gauge information at one point but also the area rainfall information .

Moreover, the prior art described in patent document 2 (Unexamined-Japanese-Patent No. 2000-345604, the name of an invention "inflow sewage amount prediction apparatus") is a method which can predict a combined sewage inflow amount. That is, the prediction is made without distinguishing between the sewage inflow and the rainwater inflow.
In general, the amount of inflow of sewage is due to domestic wastewater, so that the amount of inflow changes with time regardless of the weather. For example, as shown in the normal sewage inflow-time characteristic diagram of FIG. 10, the sewage inflow increased from dinner to before bedtime decreases at night, and the sewage inflow increases again in the morning. In addition, it is characterized by having an inflow peak in the morning and at night. On the other hand, the amount of rainwater inflow naturally varies depending on the amount of rainfall.
In the prior art of Patent Document 2, the method for predicting the amount of stormwater inflow and the amount of sewage inflow with these characteristics is different, and the amount of sewage inflow is calculated from the past average inflow. Calculated using a regression model.

  The prior art described in Patent Document 3 (Japanese Patent Laid-Open No. 2003-27567, “Sewage Inflow Prediction Device and Method, Server Device”) is a prediction method based on TCBM (Topological Case-Based Modeling). is there.

Japanese Patent Laid-Open No. 5-134715 JP 2000-345604 A JP 2003-27567 A

  The RRL method, which is a conventional technology, requires a huge amount of work to construct a prediction formula, such as a large amount of calculation for parameter setting in order to make a highly accurate prediction, and cannot cope with aging. was there. In other words, when the prediction formula is shifted due to changes in the population, the pavement rate, etc., there is a problem that a huge amount of work is required again to correct the prediction formula. In addition, only the rainwater inflow of the diversion sewerage can be predicted, and the inflow of sewage and the inflow of the combined sewerage cannot be predicted.

  The prior art described in Patent Document 1 has an advantage that the prediction formula is automatically corrected even if the population and the pavement rate change by the learning function of the neural network, but only the rainwater inflow can be predicted, and the sewage inflow In addition, there was a problem that the inflow of the combined sewerage system could not be predicted.

The prior art described in Patent Document 2 has an advantage that it can automatically cope with changes in population and pavement rate. However, since the sewage inflow is a simple past average value, an appropriate sewage average value cannot be calculated when there is singular data.
For example, when cleaning dirt in the sewer pipe, a method is adopted in which the dirty water in the sewage pumping station is stored, and the dirt adhering to the sewage pipe is removed at a stretch. For this reason, as shown in the sewage inflow amount-time characteristic chart in FIG. 11, the sewage inflow quantity disappears before cleaning, and the sewage inflow quantity becomes a spire shape at the time of cleaning, which is greatly different from the normal characteristic chart. Due to such a cleaning method peculiar to the sewage pipe, the sewage prediction method according to the invention of Patent Document 2 has a problem that appropriate averaging cannot be performed.
In addition, since the amount of rainwater inflow is predicted by an autoregressive equation, there is a problem that even if there is a weather forecast of rain, it cannot be reflected in the predicted value and the future weather conditions cannot be taken into account. .

  The prior art described in Patent Document 3 solves the problems of the invention described above, but since a special technique called TCBM is used, general developers cannot easily apply it. There was a request to make predictions using.

Summary,
(1) Forecast corresponding to secular change,
(2) Forecast of combined sewerage,
(3) Prediction with no adverse effect on specific data (such as when cleaning sewer pipes)
(4) Prediction using weather information such as rainfall,
(5) Prediction using well-known technology that does not use special methods,
There was a need for a sewage inflow prediction device that would satisfy all the requirements.

  The present invention has been made to solve these problems of the prior art. The purpose of the present invention is to solve the problems of sewage inflow and rainwater by using an uncomplicated sewage inflow prediction model and a stormwater inflow prediction model using general techniques. An object of the present invention is to provide a sewage inflow amount prediction device that accurately predicts the inflow amount and improves the prediction accuracy of the sewage inflow amount.

A sewage inflow prediction device according to claim 1 of the present invention is
In the sewage inflow prediction device that predicts the inflow of sewage flowing into the pump facility at the sewage pump station or the final treatment plant,
Meteorological data collection means for collecting meteorological data on weather results and weather forecasts;
Calendar data collection means for collecting calendar data about the calendar;
Water level data collection means for collecting water level data about the sewage water level in the pump well,
Pump pumping data collection means for collecting pumping volume data about pumping volume (pumped volume) of sewage of the pump well;
Data storage means for storing calendar data, meteorological data, water level data and pump pumping volume data;
Sewage inflow data calculation means for calculating sewage inflow data that is sewage inflow to the pump well based on the calendar data, the water level data, and the pumped water data;
Separation means for separating sewage inflow data and calculating sewage inflow data and rainwater inflow data;
A sewage inflow prediction model construction means for constructing a sewage inflow prediction model with calendar data as an input factor and sewage inflow data as an output factor;
A stormwater inflow prediction model construction means for constructing a stormwater inflow prediction model using weather data as an input factor and stormwater inflow data as an output factor;
A sewage inflow forecasting means for generating sewage inflow forecast data by inputting calendar data necessary for prediction into the sewage inflow forecast model,
A stormwater inflow forecasting means for generating stormwater inflow forecast data by inputting meteorological data necessary for the forecast into the stormwater inflow forecast model;
A summing means for generating the sewage inflow prediction data by adding the generated sewage inflow prediction data and the rainwater inflow prediction data;
It is characterized by having.

The sewage inflow prediction device according to claim 2 of the present invention is
In the sewage inflow prediction device according to claim 1,
The sewage inflow prediction model is
Prediction based on case-based reasoning, in which calendar data and sewage inflow data are linked and registered after singular data is removed, and the corresponding sewage inflow data is extracted when calendar data is input It is a model.

A sewage inflow prediction device according to claim 3 of the present invention is
In the sewage inflow prediction device according to claim 1 or 2,
The rainwater inflow prediction model is a prediction model using a neural network.

A sewage inflow prediction device according to claim 4 of the present invention is
In the sewage inflow prediction device according to claim 3,
The rainwater inflow prediction model is a prediction model for calculating rainwater flow rate prediction data in units of n hours,
The rainwater inflow prediction means generates rainwater flow prediction data whose time interval is shorter than n hours by interpolating the current rainwater flow data and rainwater flow prediction data or two adjacent rainwater flow prediction data. It is a means to do.

A sewage inflow prediction device according to claim 5 of the present invention is
In the sewage inflow prediction device according to any one of claims 1 to 4,
The present invention is characterized by comprising correction means for correcting the sewage inflow prediction data from an error amount calculated from the sewage inflow prediction data and the sewage inflow data at a past date and time.

A sewage inflow prediction device according to claim 6 of the present invention is
In the sewage inflow prediction device according to claim 5,
The correction unit is a unit that increases a correction amount of a near future prediction value and decreases a correction amount of a far future prediction value when correcting a future prediction value using a past prediction error. And

  According to the present invention as described above, a sewage inflow amount and a stormwater inflow amount are accurately predicted by an uncomplicated sewage inflow amount prediction model and a stormwater inflow amount prediction model utilizing general techniques, and a prediction accuracy of a sewage inflow amount is obtained. It is possible to provide a sewage inflow amount prediction device that realizes improvement of the above.

Hereinafter, the best mode sewage inflow prediction device for carrying out the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a sewage inflow prediction device according to this embodiment.
The sewage inflow prediction apparatus 1 includes an input means 10, an output means 20, a recording medium reading means 30, a data transmission means 40, a data collection means 50, a data storage means 60, a sewage inflow data calculation means 70, a separation means 80, a model. The construction means 90 and the prediction means 100 are provided.

The input means 10 is a keyboard for performing manual input.
The output unit 20 is a display, a printer, or the like, and also performs various data display / printing.
The storage medium reading means 30 is a device that reads and writes information from and to a storage medium such as an FD (Flexible Disc) or an MO (Magnet Optical Disc).
The data transmission means 40 is a means for connecting to the network 2 and can output information from the data storage means 60 via the data transmission means 40 and the network 2 to the outside.

The data collection means 50 is a means for acquiring various data from each sewage pump station, a weather information company, etc. via the network 2, and further includes a weather data collection means 51, a calendar data collection means 52, a water level data collection means 53, a pump Pumped water data collection means 54 is provided.
The meteorological data collecting unit 51 is a unit that collects meteorological data on the weather results and the weather forecast. The weather data here refers to all or any of, for example, weather (sunny, rain, cloudy, snow), rainfall, temperature, humidity, and amount of sunlight.
The meteorological results are past records, and meteorological data about the meteorological results are met by the meteorological data distributed from the weather information service company via the network 2 without going through each sewage pump station, and each sewage pump station The weather data is collected and arranged from measuring devices and sensors such as rain gauges and hygrometers, which are installed independently, and input via each sewage pump station.
The weather forecast is a future forecast, and the weather data about the weather forecast is data obtained by receiving a weather forecast from a weather information service company.
Thus, the meteorological data collection means 51 collects meteorological data related to the weather results of the sewage pump station to be predicted and meteorological data related to the weather forecast of the sewage pump station to be predicted.

The calendar data collecting means 52 is a means for collecting data related to the calendar.
The calendar data is date / time data such as year / month / day / hour / minute and the like, and feature data such as weekdays / holidays / day of the week / season (spring / summer / autumn / winter) representing the characteristics of the date / time.
The calendar data may be manually input by a device such as a keyboard of the input unit 10 or may be calculated by calculation.

The water level data collecting means 53 is means for collecting water level data.
The water level data is data representing the sewage water level in the pump well in the sewage pump station.
Water level data is input via a network 2 from a plurality of water level gauges installed in pump wells of a plurality of sewage pump stations at predetermined intervals or constantly. Accordingly, identification data for identifying the sewage pump station is also added to the water level data.

The pump yield data collecting means 54 is means for collecting pump yield data.
Pump yield data is data representing the yield from the pump well by the pumps in the sewage pump station. Pump pumping amount data is inputted via a network 2 from a flow meter provided in a pump of a plurality of sewage pump stations every predetermined period or constantly. In this case, identification data for identifying the sewage pump station is also added.
These meteorological data, calendar data, water level data, and pumped-up amount data are stored in the data storage means 60 and updated daily to store these data. Also, it can be taken out whenever necessary. In this data storage means 60, weather data, water level data, and pumped water amount data are associated and registered in a database using calendar data that can specify the measurement date, day of the week, and season as a main key.

  The sewage inflow data calculating means 70 retrieves the water level data and the pumped amount data for each hour from the calendar data by retrieving from the calendar data, and is the sewage inflow quantity to the pump well of the sewage pump station based on these data. Calculate sewage inflow data.

In the pump well of the sewage pump station, for example, as shown in the explanatory diagram of the pump well in FIG. 2, a water level gauge is usually installed, but an inflow meter is rarely installed, Calculate the amount of sewage inflow.
If the amount of sewage at a certain point in the pump well is M1, the amount of sewage before a certain time is M2, and the amount of pumped water pumped from the pump well at a certain time is P1, the amount of sewage inflow is as follows: become.

Here, the sewage amount M1 is an amount calculated from the water level h1 of the pump well at a certain time, and the sewage amount M2 is an amount calculated from the water level h2 of the pump well at a certain time before a certain time. As a calculation formula for the sewage amounts M1 and M2, a regression formula using the water level of the pump well as an input factor is often used. The simplest example is a simple proportional relationship, and the calculation formula is M1 = a × h1, M2 = a × h2.
Depending on the pump well, a complicated calculation formula may be required. In any case, the calculation formula uses the water level at a certain point in time as an input factor.

  The amount of sewage inflow includes the amount of pumped water from the upstream sewage pumping station (the amount of water delivered), which changes due to human operation. It is good also as an inflow amount except pump pumping amount (water supply amount). The upstream pump pumping amount P2 is an amount sent from the pump well on the upstream side in a certain time.

In this case, the other inflow amount M3 in FIG. 2 is calculated as the sewage inflow amount.
Of course, if there is a flow meter that measures the inflow to the pump well, the equations 1 and 2 may be omitted and the inflow data from the flow meter may be used. The inflow data calculation means 70 calculates sewage inflow data at a certain time. In the data storage means 60, weather data, water level data, pumping amount data, and sewage inflow amount data are associated and registered with calendar data as a main key.

Separating means 80 is means for calculating rainwater inflow data and sewage inflow data using sewage inflow data in order to separate the sewage inflow into stormwater inflow and sewage inflow. This is a necessary treatment in the combined sewer system, and the stormwater inflow due to rainwater and the sewage inflow due to sewage are separated from the sewage inflow. Of course, the sum of rainwater inflow and sewage inflow yields the original sewage inflow.
This is because the amount of sewage inflow during sunny hours is almost the same as the amount of sewage inflow because there is no inflow of rainwater. Hour, 2 o'clock,..., 23:00) over a number of days, and by taking the average of a number of days at the same time, the average sewage flow rate for each hour is calculated. These are used as sewage inflow data. Then, by subtracting the sewage inflow data for the corresponding time from the sewage inflow data for each hour, it becomes rainwater flow data. Although it is almost 0 on a clear day, rainwater flow rate data is calculated on a rainy day. Then, the sewage inflow data and the rainwater inflow data are registered in the data storage means 60. In the data storage means 60, weather data, water level data, pumped water amount data, sewage inflow data, sewage inflow data and rainwater inflow data are associated and registered with the calendar data as the main key.

The model construction means 90 is a means for constructing a model. FIG. 3 is a structural diagram of the model construction means. As shown in detail in FIG. 3, the model construction unit 90 includes a sewage inflow prediction model construction unit 91 and a stormwater inflow prediction model construction unit 92. Note that the learning data used in the rainwater inflow prediction model construction unit 91 and the sewage inflow prediction model construction unit 92 are past performance values registered in the data storage unit 60.
The sewage inflow prediction model construction means 91 is a means for constructing a sewage inflow prediction model using calendar data as an input factor and sewage inflow data as an output factor. The constructed sewage inflow prediction model is stored and stored by the data storage means 60.
The rainwater inflow prediction model construction means 92 is means for constructing a rainwater inflow prediction model using weather data as an input factor and rainwater inflow data as an output factor. The constructed rainwater inflow prediction model is stored and stored by the data storage means 60.
Such a prediction model construction unit 90 is activated every predetermined period (for example, every month) and updates these prediction models. In addition, as a rainwater inflow prediction model / sewage inflow prediction model, a prediction model using RRL method, neural network, case-based reasoning, or the like can be adopted. Details of the most preferable prediction model will be described later. The

Subsequently, the prediction unit 100 performs prediction using the rainwater inflow prediction model and the sewage inflow prediction model. FIG. 4 is a structural diagram of the prediction means. As shown in detail in FIG. 4, the prediction unit 100 includes a sewage inflow amount prediction unit 101, a rainwater inflow amount prediction unit 102, and a summation unit 103.
The sewage inflow amount prediction means 101 is a means for generating sewage inflow amount prediction data by inputting the calendar data at the time of prediction into the sewage inflow amount prediction model. The amount of sewage inflow depends on time, and if calendar data is input, sewage inflow prediction data can be obtained.
The stormwater inflow prediction means 102 is means for generating stormwater inflow prediction data by inputting meteorological data, which is a forecast at the time of prediction, into the stormwater inflow prediction model. The meteorological data becomes future meteorological data, and the rainwater inflow prediction data can be obtained by inputting the meteorological data of the forecast date and time.

The summing unit 102 is a unit that generates the sewage inflow prediction data by adding the generated sewage inflow prediction data and the rainwater inflow prediction data.
In addition to the generated sewage inflow prediction data, the rainwater inflow prediction data and the sewage inflow prediction data are also stored by the data storage means 60.
Such a prediction unit 100 is activated every certain period (for example, every hour) and predicts the amount of sewage inflow.

  The sewage inflow prediction data obtained in this way can be transmitted to the local control controller in the sewage pump station, for example, via the data transmission means 50 of the data storage means 60 and the network 2.

Then, the specific example of a sewage inflow amount prediction model is demonstrated. This model is preferably a model based on case-based reasoning. The output factor was sewage inflow data at the time of learning, but it was sewage inflow prediction data at the time of prediction.
It is known that the amount of sewage inflow varies little every day, and it is often the case that a past average is set in the system and used as a sewage inflow prediction. However, the method set by humans has a problem that it has to be reset every time the season changes and every time the change occurs. In the method of automating, an appropriate average value can be obtained by removing singular data as described above.
The sewage inflow data prediction model based on case-based reasoning of the present invention is a model inferred from the following three conditions.
(1) Weather (when sunny)
(2) Sewage flow shape of the day before and after (more than a certain value in correlation coefficient)
(3) Calendar information (day of week, period, etc.)

(1) About the weather (at the time of fine weather) As mentioned above, the sewage inflow data is the sum of the sewage inflow data and the rainwater inflow data. However, the amount of rainwater inflow into the sewage is a flow rate that can only be obtained in rainy weather, and the amount of rainwater inflow is 0 in fine weather. Therefore, in learning of the predicted sewage inflow prediction model based on case-based reasoning, only the sewage inflow amount during clear weather is extracted from the data within a certain past period and used as sewage inflow amount data.
(2) Concerning the shape of sewage flow before and after the day (more than a certain value in the correlation coefficient) However, these extraction conditions cannot exclude singular data. Therefore, only normal data is extracted next. The peculiar data is mainly due to cleaning in the sewer pipe, and has a characteristic that the amount of inflow of sewage temporarily disappears or increases. In addition, such data is not performed every day, but only once every few months. Therefore, the peculiar data has a feature that the shape of the sewage inflow is greatly different from the data of the previous and next days. In other words, normal data has the same shape as the sewage inflow data on the previous and next days. Using this feature, the sewage inflow data having the same shape as the previous and next days is extracted.
Here, the extraction condition can be easily extracted by setting the correlation coefficient to a certain value or more as compared with the shape of the previous and next days.
The correlation coefficient is a statistical index for measuring the degree of correlation between two data. The correlation coefficient takes a value of −1 ≦ r ≦ + 1, meaning that positive correlation is obtained when r> 0, negative correlation is given when r <0, and no correlation is given when r = 0.
When determining the shape of the amount of sewage inflow on the preceding and following days, for example, when using hourly data, the amount of sewage inflow on the target day is x1, x2, ..., x24 (mm ³ / h), and the amount of sewage inflow on the previous day. Y1, y2,. . . , Y24 (mm ³ / h), the correlation coefficient is calculated by the following equation.

  When the correlation coefficient r is close to 1 (for example, 0.8 or more), it is determined that the shape is the same, and when the correlation coefficient r is less than that, it is determined that the target date is singular data. Although only the previous day data is compared here, data of the target day and the next day of the target day may be compared.

(3) Calendar information (day of week, period, etc.)
Furthermore, the amount of sewage inflow has the property that the amount and rise time differ slightly depending on the day of the week. Therefore, the same day of the week as the prediction target day is extracted. In this way, sewage inflow data that meets the conditions is extracted for a plurality of days, but is averaged to eliminate daily variations.
If the average of spring, summer, autumn and winter is divided, the prediction accuracy will be further increased.
In this way, since it may vary depending on the day of the week and the season (spring, summer, autumn and winter), the sewage inflow data is classified by season and day of the week.

  In such a sewage inflow prediction model, singular data is automatically identified and removed by case-based reasoning, and sewage prediction is performed only with normal sewage inflow data. -Sewage inflow data corresponding to time, day of the week, and season is output, and the prediction accuracy is greatly improved over the mere average value in the prior art, and there is an advantage that the above problems can be solved.

The following describes learning of a predictive sewage inflow prediction model using case-based reasoning.
In such a sewage flow prediction model, the amount of sewage inflow is as described above, but it takes almost the same value by time except during cleaning of the pump. Create a database of sewage inflow data, which is the actual value corresponding to the past time, and input the time data (for example, April 10, Wednesday, Spring, 10:00 pm), read the corresponding sewage inflow data and read the sewage inflow Since it is a model that generates prediction data, when the date, day, season, and time are input, the corresponding sewage inflow prediction data is selected and output.

Next, the rainwater inflow prediction model described above will be described.
The rainwater inflow prediction model is specifically a prediction model using a neural network. FIG. 5 is a block diagram of a neural network that predicts the amount of rainwater inflow.
The model based on the neural network is, for example, a perceptron model having a three-layer structure of an input layer, an intermediate layer, and an output layer, and can learn to produce a desired output with respect to the input.
The neural network of FIG. 5 is a multi-input single-output neural network, but a multi-input multi-output neural network may be constructed.

  The input factor is the rain data in particular, and the output factor is the rainwater inflow data. There is usually a non-linear relationship between rainfall data and stormwater inflow data, and regression analysis, which is a linear prediction method, does not always give good prediction results, but this neural network learns data. An appropriate prediction model can be constructed simply by doing.

The input data of the neural network used in the learning stage is rainfall data, and usually several hours of data are used.
In general, these data usually use hourly rainfall data, but it is also possible to use short hourly rainfall data such as 10 minute rainfall data, or several hours of accumulated rainfall data. It may be. Furthermore, a difference value or an absolute amount of each rainfall data may be converted.

The output data is data obtained by the separating means 80 and registered in the data accumulating means 60, and is rainwater inflow data from several minutes to several hours ahead.
Normally, it is designed to output rainwater inflow data up to one hour ahead, but it is also possible to output the rainwater inflow data up to several hours in a lump by outputting multiple neural networks. Further, the rainwater inflow data may be processed to obtain a difference value from the present.

  In the present embodiment, as shown in FIG. 5, for example, as shown in FIG. 5, rainfall data (rainfall data values at a certain point in time, rainfall data one hour before a certain point in time, rainfall amounts one hour after a certain point in time) Data) as an input factor, and rainwater inflow data one hour after a certain point in time as an output factor.

  Next, learning of the above-described rainwater inflow prediction model will be described. In the learning of the neural network, it is necessary to create learning data. Rain data included in the weather data accumulated in the past by the data storage means 60 and rain water inflow data corresponding to the rain data are collected according to time, and a large number of input / output data sets of the neural network are prepared. For example, it is an input / output data set in which three points of rainfall data at a certain point in time and one hour before and after one hour are used as input factors, and rainwater inflow data one hour after a certain point in time is used as an output factor. A large number of such input / output data sets are prepared at different reference time points, and a neural network is constructed using various known learning algorithms such as a known back propagation method.

Next, the prediction of the rainwater inflow prediction model described above will be described. In the prediction stage, if the data of the same input items as in the learning are given to the neural network model created above, a desired prediction value is output. The output factor was rainwater inflow data at the time of learning, but it was rainwater inflow prediction data at the time of prediction.
For example, in the rainwater inflow prediction model shown in FIG. 5, the current rainfall data and the rainfall data one hour ago are input as the actual rainfall data distributed from a sewage pump station or a weather information service company. Is done. The rainfall data for one hour ahead is forecast rainfall data distributed from a weather information service company.

  FIG. 5 is an explanatory diagram of the amount of rainwater inflow one hour ahead. By changing input information for the same neural network, it is possible to obtain rainwater inflow prediction data further ahead of the hour. For example, if the rainfall data of 1 hour before, a certain time, and 1 hour ahead in FIG. 5 is the current, 1 hour, and 2 hours later, respectively, the rainwater inflow prediction data of 2 hours ahead can be obtained. it can. As current rainfall data, actual rainfall data distributed from a sewage pump station or a weather information service company is input. The 1-hour ahead and 2-hour ahead rainfall is forecast rain data distributed from a weather information service company. Thereby, rainwater inflow prediction data of 2 hours ahead can be obtained.

Further, the rainwater inflow prediction means 102 may be improved to obtain predicted values at shorter intervals. This improvement will be described.
The rainwater inflow prediction means 102 is a means for calculating rainwater flow rate prediction data after n hours (1 hour in this embodiment), but the current rainwater flow rate data and n hours ahead (1 hour ahead in this embodiment). As means for interpolating rainwater flow rate prediction data, or using two adjacent rainwater flow rate prediction data separated by n hours (in this embodiment, 1 hour), and converting them into rainwater flow rate prediction data having a short time interval. Also good.

  The electric pump starts immediately, but the engine pump takes about 5 minutes to start. When pump control is considered, there is a need for sewage flow prediction every 5 minutes. However, since rainfall data based on weather forecasts (rainfall forecasts) are data in units of one hour, in principle, predictions in units of 5 minutes cannot be made from rainfall data based on weather forecasts. Therefore, in the present invention, the rainwater inflow prediction data in units of n hours is simply converted into data in units of time shorter than that.

This will be specifically described below with reference to the drawings. FIG. 6 is an explanatory diagram of an interpolation principle for interpolating rainwater inflow prediction data in units of one hour into rainwater inflow prediction data in units of 5 minutes. Here, an example in which the rainwater inflow prediction data output from the neural network is in units of one hour and the value is converted into units of five minutes will be described.
The input / output data of the neural network is rainfall data in units of one hour as shown in FIG. By changing the input data, prediction is performed from 1 hour to m hours ahead. For example, when calculating rainwater inflow forecast data for two hours ahead, current rain data, one hour ahead forecast rainfall data value, and two hours ahead forecast rain data are input. When predicting rainwater inflow forecast data for 3 hours ahead, 1 hour ahead forecast rainfall data, 2 hours ahead forecast rain data, and 3 hours ahead forecast rain data are input. In this way, rainwater inflow prediction data up to m hours ahead is obtained in units of one hour. Next, the rainwater inflow prediction data in units of one hour is linearly complemented with the rainwater inflow prediction data in units of 5 minutes. As shown in FIG. 6, first, the 30-minute rainwater inflow prediction data, which is the median value, is made to coincide with the rainwater inflow prediction data in an hour unit of the rainwater inflow prediction model. Next, at other times, linear interpolation is performed with the values of preceding and following rainwater inflow prediction data. For example,

(Equation 4)
5 minutes ahead rainwater inflow prediction data = (5 x current rainwater inflow data + 1 hour ahead rainwater inflow prediction data) / 6
10-minute stormwater inflow prediction data = (4 x current stormwater inflow data + 2 x 1-hour stormwater inflow prediction data) / 6
15 minutes ahead stormwater inflow prediction data = (3 x current stormwater inflow data + 3 x 1 hour stormwater inflow prediction data) / 6
20-minute stormwater inflow prediction data = (2 x current stormwater inflow data + 4 x 1 hour stormwater inflow prediction data) / 6
25 minutes ahead stormwater inflow prediction data = (1 x current stormwater inflow data + 5 x 1 hour stormwater inflow prediction data) / 6
30-minute stormwater inflow prediction data = 1-hour stormwater inflow prediction data
35 minutes ahead stormwater inflow forecast data = (11 x 1 hour stormwater inflow forecast data + 2 hours stormwater inflow forecast data) / 12
40 minutes ahead stormwater inflow forecast data = (10 x 1 hour stormwater inflow forecast data + 2 x 2 hours stormwater inflow forecast data) / 12
45 minutes ahead rainwater inflow forecast data = (9 x 1 hour ahead rainwater inflow forecast data + 3 x 2 hours ahead rainwater inflow forecast data) / 12
50 minutes ahead rainwater inflow forecast data = (8 x 1 hour ahead rainwater inflow forecast data + 4 x 2 hours ahead rainwater inflow forecast data) / 12
55-minute rainwater inflow forecast data = (7 x 1 hour ahead rainwater inflow forecast data + 5 x 2 hours ahead rainwater inflow forecast data) / 12
60 minutes ahead stormwater inflow forecast data = (6 x 1 hour stormwater inflow forecast data + 6 x 2 hours stormwater inflow forecast data) / 12
65 minutes ahead rainwater inflow forecast data = (7 x 1 hour ahead rainwater inflow forecast data + 5 x 2 hours ahead rainwater inflow forecast data) / 12
70 minutes ahead rainwater inflow forecast data = (8 x 1 hour ahead rainwater inflow forecast data + 4 x 2 hours ahead rainwater inflow forecast data) / 12
75 minutes ahead rainwater inflow forecast data = (9 x 1 hour ahead rainwater inflow forecast data + 3 x 2 hours ahead rainwater inflow forecast data) / 12
80 minutes ahead rainwater inflow forecast data = (10 x 1 hour ahead rainwater inflow forecast data + 2 x 2 hours ahead rainwater inflow forecast data) / 12
85 minutes ahead stormwater inflow forecast data = (11 x 1 hour stormwater inflow forecast data + 2 hours stormwater inflow forecast data) / 12
90 minutes ahead rainwater inflow forecast data = (2 hours ahead rainwater inflow forecast data)

Thus, a predicted value in units of 5 minutes can be obtained.
The rainwater inflow prediction means 102 is such.
The sewage inflow amount prediction device 1 of this embodiment has been described above. According to this embodiment, it is possible to accurately predict by constructing and predicting the prediction model by dividing the inflow amount of sewage that does not vary depending on the weather and the inflow amount of rainwater that varies depending on the weather.

Next, a mode in which the prediction means described above is further improved will be described with reference to the drawings. FIG. 7 is a configuration diagram of another prediction unit 100 ′. 4 is different from the prediction unit 100 of FIG. 4 in that a correction unit 104 is further added.
The correction means 104 is means for correcting the sewage inflow prediction data from an error amount calculated from the sewage inflow prediction data and the sewage inflow data at the past date and time.
The sewage inflow prediction data obtained by adding the sewage inflow prediction data and the rainwater inflow prediction data is corrected using the past prediction error data, and corrected sewage inflow prediction data is generated. Here, the prediction error data is a collection of many errors between the sewage inflow prediction data that predicted the sewage inflow at a certain time in the past and the actual sewage inflow data at the same time, and the average value was taken as the error. It is. The correction formula is as follows.

(Equation 5)
Corrected sewage inflow prediction data = sewage inflow prediction data + prediction error data

  Thereby, the error of the sewage inflow amount prediction data can be corrected to improve the accuracy of the sewage inflow amount prediction.

When this prediction means 100 ′ is further improved and the future sewage inflow prediction data is corrected using the past prediction error data, the correction amount of the sewage inflow prediction data in the near future is increased, and the sewage inflow in the far future is increased. Means for reducing the correction amount of the amount prediction data may be used. This point will be described with reference to the drawings. FIG. 8 shows a characteristic diagram of the autocorrelation coefficient of the sewage inflow amount. The horizontal axis represents the time interval that is the time difference between the two sewage inflows taking the correlation coefficient, and the vertical axis represents the autocorrelation coefficient. The autocorrelation coefficient is a correlation coefficient with itself, and is a value obtained by creating two series of data by shifting time from the same data and calculating a correlation coefficient for each time. For example, assuming that the data for each hour {x1, x2,..., X48}, the autocorrelation coefficients one hour ahead are {x1, x2,..., X24} and {x2, x3,. -Create and calculate two series of data of x25}. The autocorrelation coefficient two hours ahead is calculated by creating two series of data {x1, x2,..., X24} and {x3, x4,.
Similarly, the autocorrelation coefficient up to 24 hours ahead can be calculated. If the autocorrelation coefficient is close to 1, it has a strong influence on itself, and in the case of sewage inflow prediction, it means that the future inflow can be predicted more easily than the current sewage inflow. If it is close to 0, it means that the future inflow cannot be predicted from the current inflow of sewage (other factors such as rainfall become stronger).
In FIG. 8, the autocorrelation coefficient of the flow rate is high in the near future, but decreases as the distant future is reached. In other words, the sewage inflow in the near future is greatly influenced by the current inflow, but the inflow in the far future is not significantly affected by the current inflow.
Therefore, in the present invention, the prediction value in the near future is corrected to a large value, and the prediction value in the far future is corrected to a small value, thereby preventing unnecessary error correction and improving the prediction accuracy.

  Next, the correction method will be described. First, the average prediction error for the past n hours is calculated by the following equation.

(Equation 6)
Average prediction error data = avg (sewage inflow prediction data [T−t] −sewage inflow data [T−t])

Alternatively, the following equation may be used.
(Equation 7)
Average prediction error data =
avg (Sewage inflow prediction data [T−t] −Sewage inflow data [T−t]) / Sewage inflow data [T−t]

Here, the sewage inflow prediction data [Tt] is the sewage inflow prediction data before time t.
Sewage inflow data [T-t] is sewage inflow data before time t

  Next, the average prediction error data is corrected by the following equation. In the following, an example of correcting the corrected sewage inflow prediction data up to the future I hour ahead is shown. When Equation 6 is used, it is as follows.

(Equation 8)
Corrected sewage inflow prediction data i =
Sewage inflow prediction data i + (I−i) × average prediction error data / I

The average prediction error data decreases as i increases, that is, in the far future.
When Expression 7 is used, the following expression is obtained.

(Equation 9)
Corrected sewage inflow prediction data i =
Sewage inflow prediction data i / (1 + average prediction error data × (I−i) / I)

  Here, 0 <i <I, and the sewage inflow prediction data i is a predicted value i hours ahead.

  The sewage inflow amount prediction device 1 of the present invention has been described above. The present invention is a computer including the above-described means, the data storage means 60 is a hard disk, and the sewage inflow data calculation means 70, the separation means 80, the model construction means 90, and the prediction means 100 are caused to function on a computer by a program. It may be a good means.

Subsequently, a prediction result obtained by performing actual prediction using the sewage inflow amount prediction device of the present invention will be described in detail with reference to the drawings. FIG. 9 is an explanatory diagram of a prediction result by the sewage inflow amount prediction device of this embodiment.
In the first embodiment, since only simulation was performed, the data stored in the data storage unit 60 was used regardless of the data collection unit 50.

The prediction conditions are as follows in detail: Predicted object: Prediction of the amount of sewage flow into the pump well of the sewage pump station where the combined sewage facilities are located: 5 minutes to 12 hours ahead (5 minute intervals),
Data for learning: Data for the past 6 months (1) Date (1 day unit)
(2) Weekdays, holidays (including Saturday) information (1 day unit)
(3) Correlation coefficient with previous day data (1 day unit)
(4) Rainfall (in units of 5 minutes)
(5) Sewage inflow (5 minutes)

Procedure 1. Calculation of sewage inflow data (sewage inflow data calculation means 70)
Using the water level data stored in the data storage means 60, the sewage inflow data was calculated according to the above Equation 1.

Procedure 2. Separation of sewage and rainwater (separation means 80)
Subsequently, the sewage inflow data is separated into sewage inflow data and rainwater inflow data. For separation, case-based reasoning was used to extract past cases for each month, weekdays, and holidays. The extraction conditions are as follows.
(1) Past 4 months not including the specified month (2) Specified day of the week (weekdays or holidays)
(3) Correlation coefficient of 0.9 or more (4) Day with a daily rainfall of 0 mm (5) All sewage inflow data are normal and no missing measurements

Under the above conditions, in particular, since the day has a daily rainfall of 0 mm, that is, it is not raining, data that can be regarded as sewage inflow data is extracted. When such sewage inflow data is extracted, a plurality of days are extracted for each month. Therefore, the average value for each hour on weekdays and holidays was taken and used as sewage inflow data. Since the extraction data this time is in units of 5 minutes, the sewage inflow data is data for 24 hours in units of 5 minutes.
Using such sewage inflow data, the daily stormwater inflow data for the daily rainfall a [mm] was obtained by subtracting the calculated sewage inflow data from the sewage inflow data.
In this way, two types of sewage inflow data are created for each month, weekdays and holidays, and a large amount of storm water inflow data is created for each day.

Procedure 3. Sewage inflow prediction model construction (sewage inflow prediction model construction means 91)
Register sewage inflow data every 24 hours a day on weekdays and sewage inflow data every 24 hours on holidays, and input as month, weekday, holiday, time (0:00, 1:00, ..., 23:00) ) Is used as a model based on case-based reasoning that outputs corresponding sewage inflow data.
Procedure 4. Rainwater inflow prediction model construction (rainwater inflow prediction model construction means 92)
The rainwater inflow prediction model is constructed by a neural network. First, learning data is created. The rainwater inflow data is calculated in the procedure 2, but the above data includes a lot of specific data such as clear weather time data and negative sewage inflow data (data in the case of reverse flow). Therefore, rainwater inflow data that satisfies the following conditions was extracted.
(1) Rainwater inflow of 0 or more,
(2) Rain 5 mm / h or more Next, a large number of learning sets are created from the extracted data according to the following format.
1 hour before rain data mm / h,
Rain data at a certain point mm / h,
1 hour ahead rainfall data mm / h,
1 hour ahead rainwater inflow data mm ³ / h
The neural network shown in FIG. 5 is learned by the back propagation method using the large number of data sets.

Subsequently, prediction is performed by the prediction unit 100 ′ (see FIG. 7).
Procedure 5. Sewage inflow prediction (sewage inflow prediction means 101)
As for sewage inflow prediction, when a month, weekday, holiday, day of the week, season, and time (0:00, 1:00,..., 23:00) are input, sewage inflow prediction data for each time is output.
Procedure 6. Rainwater inflow prediction (rainwater inflow prediction means 102)
The rainwater inflow prediction data is generated using the rainwater inflow prediction model of the neural network model constructed previously. As data to be input to the neural network, the following data is input when the inflow amount of rainwater one hour ahead is predicted.
1 hour before rain data mm / h,
Current rainfall data mm / h,
1 hour ahead rainfall data mm / h,
1 hour ahead rainwater inflow prediction data mm ³ / h
And when obtaining a predicted value for 2 hours ahead, the following data are input.
Current rainfall data mm / h,
1 hour ahead rainfall data mm / h,
2-hour ahead rainfall data mm / h
2-hour ahead rainwater inflow prediction data mm ³ / h
Similarly, 12 hours ahead of stormwater inflow prediction data is obtained. Here, as for the future rainfall, the weather forecast value should be used. However, since this example is a simulation, actual rainfall data was used.
Here, the predicted value output from the neural network is data in units of one hour. Therefore, the rainwater inflow prediction data up to 12 hours ahead by 5 minutes is calculated by converting into rainwater inflow prediction data in 5 minutes using Equation 4.

Step 7. Combined predicted sewage and rainwater (summing means 103)
The sum of sewage inflow prediction data and rainwater inflow prediction data was used as sewage inflow prediction data. Further, the sewage inflow prediction data was stored in the data storage means 60 as sewage inflow prediction data before correction.

Procedure 8. Prediction correction (correction means 104)
The average prediction error data between the sewage inflow prediction data before correction of 12 points from 5 minutes to 1 hour before read from the data storage means 60 and the actual sewage inflow data at the same time is calculated by Equation 6, The predicted value up to 2 hours ahead was corrected by Equation 8. Here, since there are 24 prediction values up to 2 hours ahead, the value of I in Equation 8 is 24,
The value of i is 1 in 5 minutes, 2 in 10 minutes, and 24 in 2 hours.

  An example of values predicted by the above method is shown in FIG. FIG. 9 shows an example of prediction from 12: 5 to 24:00 at the 12:00 stage. Although it is combined sewage, there is no significant difference between the actual value and the predicted value, and it can be predicted appropriately. In particular, at the twelve o'clock stage before the rain, it is possible to appropriately predict the tendency of the flow rate to increase using the future rain information.

  The present invention has been described above. By using the present invention, it is possible to predict in either case of a diverted sewer pipe or a combined sewer pipe. In addition, the prediction model automatically changes over time, such as population and pavement rate, and high-precision prediction results can be obtained. In addition, the present invention does not use a special method, but is based on case-based reasoning and neural network prediction that can be obtained in a commercial package, and has an advantage that a sewage inflow prediction device can be easily constructed.

It is a block diagram of the sewage inflow amount prediction apparatus of the best form for implementing this invention. It is explanatory drawing of a pump well. It is a structural diagram of a model construction means. It is a structural diagram of a prediction means. It is a block diagram of the neural network which estimates the amount of rainwater inflow. It is explanatory drawing of the interpolation principle which interpolates the rainwater inflow prediction data of a 1-hour unit into the rainwater inflow prediction data of a 5-minute unit. It is a block diagram of another prediction means. It is a characteristic view of the autocorrelation coefficient of sewage inflow. It is explanatory drawing of the prediction result by the sewage inflow amount prediction apparatus of a present Example. It is a sewage inflow amount-time characteristic figure at the normal time. It is a sewage inflow amount-time characteristic figure at the time of peculiarity.

Explanation of symbols

1: Sewage inflow prediction device 10: input means 20: output means 30: recording medium reading means 40: data transmission means 50: data collection means 51: meteorological data collection means 52: calendar data collection means 53: water level data collection means 54 : Pump pumping amount data collection means 60: data storage means 70: sewage inflow amount data calculation means 80: separation means 90: model construction means 91: rainwater inflow amount prediction model construction means 92: sewage inflow amount prediction model construction means 100, 100 ': Predictive means 101: Sewage inflow prediction means 102: Rainwater inflow prediction means 103: Summing means 104: Correction means 2: Network

Claims (6)

In the sewage inflow prediction device that predicts the inflow of sewage flowing into the pump facility at the sewage pump station or the final treatment plant,
Meteorological data collection means for collecting meteorological data on weather results and weather forecasts;
Calendar data collection means for collecting calendar data about the calendar;
Water level data collection means for collecting water level data about the sewage water level in the pump well,
Pump pumping data collection means for collecting pumping volume data about pumping volume (pumped volume) of sewage of the pump well;
Data storage means for storing calendar data, meteorological data, water level data and pump pumping volume data;
Sewage inflow data calculation means for calculating sewage inflow data that is sewage inflow to the pump well based on the calendar data, the water level data, and the pumped water data;
Separation means for separating sewage inflow data and calculating sewage inflow data and rainwater inflow data;
A sewage inflow prediction model construction means for constructing a sewage inflow prediction model with calendar data as an input factor and sewage inflow data as an output factor;
A stormwater inflow prediction model construction means for constructing a stormwater inflow prediction model using weather data as an input factor and stormwater inflow data as an output factor;
A sewage inflow forecasting means for generating sewage inflow forecast data by inputting calendar data necessary for prediction into the sewage inflow forecast model,
A stormwater inflow forecasting means for generating stormwater inflow forecast data by inputting meteorological data necessary for the forecast into the stormwater inflow forecast model;
A summing means for generating the sewage inflow prediction data by adding the generated sewage inflow prediction data and the rainwater inflow prediction data;
A sewage inflow prediction apparatus characterized by comprising: In the sewage inflow prediction device according to claim 1,
The sewage inflow prediction model is
Prediction based on case-based reasoning, in which calendar data and sewage inflow data are linked and registered after singular data is removed, and the corresponding sewage inflow data is extracted when calendar data is input A sewage inflow prediction device characterized by being a model. In the sewage inflow prediction device according to claim 1 or 2,
A sewage inflow prediction apparatus, wherein the rainwater inflow prediction model is a prediction model using a neural network. In the sewage inflow prediction device according to claim 3,
The rainwater inflow prediction model is a prediction model for calculating rainwater flow rate prediction data in units of n hours,
The rainwater inflow prediction means generates rainwater flow prediction data whose time interval is shorter than n hours by interpolating the current rainwater flow data and rainwater flow prediction data or two adjacent rainwater flow prediction data. A sewage inflow prediction device, characterized in that: In the sewage inflow prediction device according to any one of claims 1 to 4,
A sewage inflow prediction apparatus comprising correction means for correcting sewage inflow prediction data from an error amount calculated from sewage inflow prediction data and sewage inflow data at a past date and time. In the sewage inflow prediction device according to claim 5,
The correction unit is a unit that increases a correction amount of a near future prediction value and decreases a correction amount of a far future prediction value when correcting a future prediction value using a past prediction error. A sewage inflow prediction device.

Images