JPH0852459A - Sewage treatment process controller - Google Patents

Abstract

(57) [Summary] [Objective] To provide a sewage treatment process control device capable of controlling sewage in an environment in which microorganisms are most activated and sewage satisfactorily purified. [Structure] The feature quantity predicting unit 3 predicts the feature quantity representing the state of the activated sludge after a predetermined time, and the operation quantity correcting unit 4 calculates the correction quantity of the operation quantity given to the sewage treatment process 7. Repeat the correction of the operation amount by the correction amount,
The resulting manipulated variable u _n is input to the sewage treatment process 7.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a sewage treatment process control device used in, for example, a sewage treatment facility.

[0002]

2. Description of the Related Art Conventionally, in the sewage treatment process, sewage treatment by an activated sludge method utilizing the biochemical reaction of microorganisms is generally performed. This is to activate the microorganisms in the sludge by increasing the dissolved oxygen concentration in the sewage by aeration, and to repeat the growth of microorganisms that repeat the growth while oxidatively decomposing the ingested dissolved oxygen concentration and the substances in the sewage into carbon dioxide gas and water. Is a method of purifying sewage.

In order to stably maintain the activity of the microorganisms, the microorganisms live in order to control the dissolved oxygen concentration in the sewage to a constant value by manipulating the amount of aeration air and to keep the number of microorganisms almost constant. The concentration of suspended solids in the mixed liquid is controlled to keep the sludge volume constant.

However, the aeration tank and other tanks are open tanks, in which various bacteria contained in sewage and bacteria entering from the air are mixed, and sludge also becomes a mixed culture system with various microorganisms constituting the sludge. There is. Since the activated conditions differ depending on the type of microorganism, if the temperature, pH, and environmental conditions such as nutrients and toxic substances contained in the sewage change, it is suitable for that environment and The type also changes. In addition, the activity, that is, the sewage purification capacity, differs depending on the type of microorganism.

Therefore, for example, when the nutrient substance in the sewage is changed and a large number of microorganisms preferentially decompose the nutrient substance, the dissolved oxygen concentration and the suspended solid concentration in the mixed solution are controlled to be suitable for the microorganism. It is desirable to do.

[0006]

As described above, in the conventional sewage treatment process control, the dissolved oxygen concentration in the aeration tank and the mixed liquid suspended matter concentration are controlled to be constant.
The types of microorganisms contained in sludge and the environmental information that the microorganisms optimally use are not used. Therefore, even if the dissolved oxygen concentration and the mixed liquid suspended matter concentration in the aeration tank are kept constant, the sewage may not be successfully purified.

An object of the present invention is to provide a sewage treatment process control device which eliminates the above-mentioned drawbacks, controls sewage in an environment where microorganisms are most activated, and can purify sewage well. And

[0008]

In order to achieve the above object, the sewage treatment process control device of the present invention has a photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, and the photographing means. An image processing means for extracting a characteristic amount of the image; a characteristic amount predicting means for predicting the characteristic amount after a predetermined time based on the operation amount given to the sewage treatment process and the characteristic amount extracted by the image processing means;
Operation amount correction means for obtaining the correction amount of the operation amount and correction of the operation amount obtained by the operation amount correction means so that the feature amount predicted by the feature amount prediction means matches the target feature amount. Addition means for adding an amount to the manipulated variable.

According to another aspect of the present invention, there is provided a sewage treatment process control apparatus, which comprises a photographing means for photographing sewage flowing out from an aeration tank of a sewage treatment process, and an image processing means for extracting a characteristic amount of an image photographed by the photographing means. And a detection means for detecting at least one of dissolved oxygen concentration sensor information in the aeration tank, mixed liquid suspended substance concentration sensor information, inflow meter information, and excess sludge concentration sensor information, and the detection result by this detection means and sewage treatment. A feature amount prediction unit that predicts the feature amount after a predetermined time based on the operation amount given to the process and the feature amount extracted by the image processing unit, and the feature amount and target predicted by this feature amount prediction unit. The operation amount correction means for obtaining the correction amount of the operation amount and the correction amount of the operation amount obtained by the operation amount correction means are set to the operation amount so that the characteristic amount is matched. It comprises an adding means for adding.

According to still another aspect of the sewage treatment process control apparatus, an entire image of one activated sludge area is taken from an image taking means for taking an image of the sewage flowing out from the aeration tank of the sewage treatment process and an image taken by the taking means. Of the area of the activated sludge, the length of the contour line of the activated sludge, the ratio of the area of the activated sludge screen to the length of the contour line, the number of substances matching a specific shape, the color of the activated sludge, or the value of the image element, at least 1 An image processing means for extracting one value as a feature amount, a feature amount predicting means for predicting a feature amount after a predetermined time, based on the operation amount given to the sewage treatment process and the feature amount extracted by the image processing means, The operation amount correction means for obtaining the correction amount of the operation amount so that the feature amount predicted by the feature amount prediction means and the target feature amount match, and the operation amount correction means A correction amount of the order was manipulated variable comprises an adding means for adding the operation amount.

Still another aspect of the sewage treatment process control device is a photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, and an image processing means for extracting the characteristic amount of the image photographed by the photographing means. And a feature amount prediction unit configured from a neural circuit model, based on the operation amount given to the sewage treatment process and the feature amount extracted by the image processing unit, for predicting the feature amount after a predetermined time,
Realized by the error backpropagation calculation of the neural circuit model,
An operation amount correction unit that obtains a correction amount of the operation amount and a correction of the operation amount obtained by the operation amount correction unit so that the feature amount predicted by the feature amount prediction unit matches the target feature amount. Addition means for adding an amount to the manipulated variable.

According to still another aspect of the sewage treatment process control device, a photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, and an image processing means for extracting the characteristic amount of the image photographed by the photographing means, A feature amount prediction unit that predicts the feature amount after a predetermined time based on the operation amount given to the sewage treatment process and the feature amount at a plurality of past times extracted by the image processing unit, and is predicted by the feature amount prediction unit. An operation amount correction means for obtaining a correction amount of the operation amount so that the selected feature amount and the target feature amount match,
And an addition unit for adding the correction amount of the operation amount obtained by the operation amount correction unit to the operation amount.

[0013]

In the sewage treatment process control device of the present invention, on the premise that the state of the activated sludge can be estimated from the image of the activated sludge, first, the sewage flowing out from the aeration tank of the sewage treatment process is photographed, and the image of the image is taken. Extract feature quantities. Next, the feature amount after a predetermined time is predicted based on the operation amount given to the sewage treatment process and the extracted feature amount. Then, the correction amount of the operation amount is obtained so that the predicted feature amount and the target feature amount match, and the obtained correction amount of the operation amount is added to the operation amount. Therefore, by inputting the resulting manipulated variable into the sewage treatment process, the state of the activated sludge can be controlled to the target state, and the sewage can be controlled in the environment where the microorganisms are most activated.

[0014]

First, the principle of the present invention will be described prior to the description of the embodiments of the present invention.

The sewage treatment process control device of the present invention can estimate the state of the activated sludge from the image of the activated sludge, and observes the activated sludge when the microorganisms are activated and the sewage is successfully purified by a visual sensor. Based on the image signal input from the visual sensor installed in the sewage treatment process, it is given to the sewage treatment process on the assumption that the target feature amount obtained by processing the image signal is given in advance. This is to calculate a power estimation quantity.

FIG. 1 is a block diagram showing the arrangement of a sewage treatment process control apparatus according to an embodiment of the present invention. In the following description, the control unit time is Δt and the current time is nΔt. In FIG. 1, 1 is a visual sensor unit, 2 is an image processing unit, 3 is a feature amount prediction unit, 4 is an operation amount correction unit, 5 is an adder, 6 is a switch, and 7 is a sewage treatment process.

The image signal I _n obtained from the visual sensor unit 1 at time nΔt is input to the image processing unit 2. The image processing section 2, an image signal I _n and image processing, and outputs the feature quantity s _n. The subscript _n of I _n and s _n means time nΔt, and is used to mean the image signal and the feature amount observed at time nΔt, respectively. Subsequent subscripts will similarly indicate time.

The feature quantity predictor 3 receives the feature quantity s _n and the manipulated variable u _{n of} the image signal acquired at time nΔt, and inputs 1
The predicted feature amount s (hat) _{n + 1} that predicts the feature amount of the image signal acquired after a unit time is output.

The operation amount correction unit 4 provides the predicted feature amount s
(Hat) _{n + 1} and the preset target feature amount s ^d are input, and the predicted feature amount s (hat) _{n + 1} is the target feature amount s
The correction amount Δu _n of the operation amount for approaching ^d is output.

In the adder 5, the correction amount Δu _{n of} this operation amount
Is added to the manipulated variable u _n .

The corrected operation amount is fed back to the feature amount prediction unit 3 again, and the feature amount prediction unit 3 predicts the feature amount and the operation amount correction unit 4 calculates the operation amount correction amount. The switch 6 is closed when a predetermined condition is satisfied, and the corrected manipulated variable u _n is input to the sewage treatment process 7 to perform control.

In the above description, the characteristic amount and the manipulated variable at the current time are input to the characteristic amount predicting unit 3, but the dissolved oxygen concentration, the concentration of the suspended solids in the mixed liquid, the inflow amount of sewage, and the excess sludge. Various sensor information such as density may be added to the input signal, and past feature amount, various sensor information, and operation amount may be input. This makes it possible to more accurately estimate the state of microorganisms in the sludge.

Further, in the above description, the feature amount after one unit time is predicted, but the feature amount after one unit time is not limited, and the feature amount after two unit time may be predicted. . Furthermore, you may predict the feature-value of several time.

Next, the calculation procedure of the entire control device will be described with reference to the flow chart shown in FIG. Here, the feature amount predicting unit 3 causes the image feature amount at the present time and one unit time before, the manipulated variable, the dissolved oxygen concentration, the mixed liquid suspended substance concentration, the sewage inflow amount,
It is assumed that the sensor information of the excess sludge concentration is input and the image feature amount after one unit time is predicted. In the flow chart of FIG.
Although written so as to control from time 0 to time NΔt, this calculation procedure repeats the same procedure for each unit time, and therefore the calculation procedure at time nΔt is shown here.

(A) The image signal I _n by the visual sensor unit 1
Is acquired.

(B) The image processing unit 2 processes the image signal I _n acquired in (a) to extract the feature amount s _n .

(C) The value of the manipulated variable u _n-1 before the unit time is u
Substitute as the initial value of _n .

(D) The feature amount predictor 3 calculates the feature amount s _n obtained in (b), the feature amount s _n-1 one unit time ago, and the present and 1
The various sensor information t _n and t _n-1 before the unit time and the manipulated variables u _n and u _{n- 1} are input, the feature amount after the unit time is predicted, and the predicted feature amount s (hat) _{n + 1} is output. To be done.

(E) The target feature amount s ^d and the predicted feature amount s (hat) _{n + 1} obtained in (d) are input to the manipulated variable correcting unit 4, and the predicted feature amount s (hat) _{n + 1} is obtained. target feature amount s ^d to modify the amount Δu manipulated variable u _n for matching (~) _n is calculated and output.

[0030] (f) correction amount Delta] u _n of the adder 5 with the operation amount is added to the manipulated variable u _n, the operation amount u _n is corrected.

The manipulated variable u _n corrected in (g) and (f) is fed back to the feature quantity predicting section 3 again, and the calculation is repeated until a predetermined condition is satisfied.

(H) When the predetermined condition is satisfied, the switch 6 is closed and the manipulated variable u _n is input to the sewage treatment process 7.

The above procedures (a) to (h) are repeated for each unit time.

Next, another embodiment of the present invention will be described with reference to FIG. In the figure, the visual sensor unit 1 photographs the sewage flowing from the aeration tank to the final settling tank through a microscope.
The image signal processing unit 2 extracts the size of the activated sludge captured by the visual sensor unit 1 as a feature amount of the image. The feature amount prediction unit 3 is composed of a neural circuit model, and the operation amount correction unit 4 is realized by error back propagation calculation of the neural circuit model. Reference numeral 8 is a sensor for dissolved oxygen concentration, mixed liquid suspended matter concentration, sewage inflow amount, excess sludge concentration, and 9 is a comparison unit.

When sludge is added to sewage and aerated for a long period of time, a lump of muddyness gradually forms in the sewage, and when the aeration is stopped, the lump precipitates and the supernatant liquid becomes transparent.
This cloudy cloudy mass is mainly composed of gelatinous polymer organisms containing various bacteria, and protozoa inhabit it. This muddy cloudy mass is called activated sludge. Figure 4 is a photomicrograph of activated sludge (150x)
FIG. In FIG. 4, what appears to be a cloud is activated sludge. In the aeration tank, the sludge is agitated by aeration and is in the state of activated sludge.

In the visual sensor unit 1, since the sewage flowing from the aeration tank to the final settling tank is photographed through the microscope, the image signal I _n as shown in FIG. 4 is obtained.

The image processing unit 2 measures the ratio of the area of one activated sludge to the area of the entire image as the characteristic amount s _n based on the image signal I _n . In the microscope image, the length of the contour line of the activated sludge, the ratio of the area of the activated sludge screen to the length of the contour line, and the like can be considered as the characteristic amount. Furthermore, the number of substances that match a specific shape (the characteristic shape of plankton such as Vorticella and Paramecium) (representing the type and number of plankton in the image), the color of the activated sludge, and the value of the image element are considered as characteristic quantities. be able to. Further, on the screen showing the water surface, the color of sewage, the amount and size of bubbles on the water surface, and the like can be considered as the characteristic amount. Of course, these feature quantities may be combined.
The feature amount prediction unit 3, in addition to the characteristic quantity s _n, operation aeration amount as the amount u _n QAER _n and excess sludge withdrawal amount Qwas _n,
Dissolved oxygen concentration DO _n aeration tank as sensor information t _n, mixture suspended solids concentration MLSS _n, sewage inflow Qinf _n, such as excess sludge concentration CWAs _n is input. Alternatively, the past information is also input. Then, the predicted feature amount s (hat) _{n + 1} obtained by predicting the feature amount of the image signal acquired after one unit time, or the predicted feature amount in the future is output.

The feature quantity predicting section 3 is composed of a multi-layered neural circuit model, and the predictive feature quantity is calculated by forward calculation. The forward calculation of the multilayer neural network model will be described later.

The predicted feature amount s (hat) _{n + 1} and the target feature amount s ^d are input to the manipulated variable correcting unit 4, and the manipulated variable correction amount Δu _n is output. In this calculation, when the feature quantity predicting unit 3 is composed of a multi-layered neural circuit model, the correction quantity is obtained by error back propagation calculation. The error backpropagation calculation of the multilayer neural network model will also be described later.

In the adder 5, the manipulated variable modification amount Δu _n is added to the manipulated variable u _n . The corrected operation amount is fed back to the feature amount prediction unit 3 again, and the feature amount prediction unit 3 predicts the feature amount and the operation amount correction unit 4 corrects the operation amount. When the target feature amount s ^d and the predicted feature amount s (hat) _{n + 1} are input to the comparison unit 9, the comparison unit 9 compares the two feature amounts, and when the error becomes equal to or less than the determined value. , A signal is sent to the switch 6, the switch 6 is closed, and the manipulated variable u _n is input to the sewage treatment process 7.

Next, the forward calculation, the error back propagation calculation, and the learning calculation of the neural network model will be described. When the feature quantity prediction unit 3 is composed of a multilayer neural network model, the calculation of the feature quantity prediction unit 3 is performed by the forward calculation of the multilayer neural circuit model, and the calculation of the operation amount correction unit 4 is the error inverse of the multilayer neural network model. Propagation calculation is performed.

[0042] Here, the input to the feature prediction unit 3 before one unit time and the current characteristic quantity s _n-1, s _n, manipulated variable u _n-1, u
_{It is assumed} that _n and sensor information t _n-1 and t _n are input.

The forward calculation will be described first.
The forward calculation is a calculation for obtaining an output signal when an input signal is given to the neural network model. Here, of the multi-layered neural circuit models, the most commonly used three-layered neural circuit model will be used for description. FIG. 5 shows the feature quantity prediction unit 3.
FIG. 3 is a diagram of a three-layer type neural circuit model used for. A circle in the figure is a nerve cell model called a unit 10, and a line segment connecting the units is a connection 11 that transmits a signal between the units. The units are arranged in three layers, from the left, an input layer 12, an intermediate layer 13, and an output layer 14. Couplings are provided between the units of the input layer and the units of the intermediate layer, and between the units of the intermediate layer and the units of the output layer. The coupling between the units has a directivity in signal transmission, and in the forward calculation, the signal is transmitted only in one direction from the input layer unit to the intermediate layer unit and from the intermediate layer unit to the output layer unit.

The unit of the input layer has feature values s _n-1 ,
s _n, aeration amount QAER _n-1 as an operation amount, QAER _n and excess sludge withdrawal amount Qwas _n-1, Qwas _n, dissolved oxygen concentration in the aeration tank as sensor information DO _n-1, DO n, mixture suspended solids concentration MLSS _n-1 , MLSS _n , sewage inflow Qinf
_n−1 , Qinf _n , and excess sludge concentration Cwas _n−1 , Cwas _n are input to each unit one signal at a time (n + 1) Δt.
The predicted value s (hat) _{n + 1} of the feature amount is output.

The forward calculation for calculating the output signal from this input signal is shown in the formula.

When xi _i is the input value of the i-th unit of the input layer, xi ₁ = s _n-1 xi ₂ = s _n xi ₃ = Qaer _n-1 xi ₄ = Qaer _n xi ₅ = Qwas _{n- 1 xi 6 = Qwas n xi 7} = DO n-1 xi 8 = DO n xi 9 = MLSS n-1 xi 1 0 = MLSS n xi 1 1 = Qinf n-1 xi 1 2 = Qinf n xi 1 3 = Cwas _n-1 xi _{1 4} = Cwas _{n The} input / output function of the input layer unit is an identity function. The output yi _i (i = 1, ... 14) of the input layer unit is expressed as follows.

Yi _i = xi _i (i = 1, ... 14) Next, the input value xh _j to the j-th unit of the intermediate layer is the output value of the input layer unit weighted by the coupling load between the input and intermediate layers. It is a value obtained by subtracting the threshold value thh _j of the intermediate layer unit from the sum. Nh represents the number of units in the intermediate layer.

[0048]

[Equation 1] w1 _{j, i} is a coupling load between the i-th unit of the input layer and the j-th unit of the intermediate layer, th _j is a threshold value of the j-th unit of the intermediate layer, and Nh is the number of units of the intermediate layer.
I / O function of middle layer unit is sigmoid

[Equation 2] Then, the output value yh _j of the middle layer unit is the input value xh
Using _j , it can be expressed as yh _j = f (xh _j ) (j = 1, ... Nh). yh _j is the output value of the j-th unit in the hidden layer. The output value is only one unit, and the input value xo ₁ to this unit is a value obtained by subtracting the threshold value of the output layer unit from the sum of the output value of the intermediate layer unit weighted by the coupling load between the intermediate and output layers.

[0049]

(Equation 3) w2 _{k, j} is the intermediate layer j-th unit - connection weight between the k-th unit of the output layer, tho _k is the threshold of the k-th unit of the output layer. The input / output function of the output layer unit is an identity function.

Yo _k = xo _k (k = 1) The output value of this output layer unit is the predicted feature s (hat).
_{n + 1} .

S (hat) _{n + 1} = yo ₁ or more is the calculation formula for the forward calculation of the three-layer neural circuit model.

Since the number of input signals is large and the number of parameters to be set is large in the feature quantity predicting section 3, by using the three-layer type neural network model, these parameters can be set as described later. Can be easily set.

It should be noted that the unit of the input layer has s as a feature amount.
_{Although n-1} and s _n are input, the length of the contour line of the activated sludge, the ratio of the area of the activated sludge screen to the length of the contour line, When the specific shape, the color of the activated sludge, the value of the image element, the color of the sewage, the amount and size of the bubbles on the water surface, etc. are used, these signals may be input to the respective input layer units.

Further, the signal of each pixel of the screen imaged by the visual sensor unit 1 may be input as a feature amount to each separate input layer unit in the feature amount prediction unit 3.

Next, the error back propagation calculation will be described.
Feature amount s (hat) predicted by the feature amount prediction unit 3
_{When n + 1} and the target feature amount s ^d are input to the operation amount correction unit 4, the operation amount correction unit 4 calculates and outputs the operation amount correction amount Δu _n . When the feature quantity prediction unit 3 is composed of a neural network model, the correction quantity of the feature quantity can be calculated by the error backpropagation calculation. The configuration of the manipulated variable correcting unit 4 is shown in FIG.
The input and output of the three-layer type neural circuit model shown in FIG. Therefore, by providing a switch for switching between input and output in the feature quantity prediction unit 3, the operation amount correction unit 4 can share the configuration of the feature quantity prediction unit 3. A calculation formula of the operation amount correction unit 4 is shown.

First, the error function E is defined as follows from the error between the feature amount s (hat) _{n + 1} predicted by the feature amount prediction unit 3 and the target feature amount s ^d .

[0057]

[Equation 4] Next, the correction amount of the unit output value that reduces the error function E is sequentially obtained for each layer from the output layer to the input layer,
Finally, the correction amount of the input signal is obtained. First, the correction amount of the output value of the output layer unit is calculated by using the positive constant η.

(Equation 5) Is calculated. Next, the correction amount of the output value of the intermediate layer unit is expressed as follows using the correction amount of the output layer unit.

[0058]

(Equation 6) Similarly, the correction amount of the output value of the input layer unit, using the correction amount of the middle layer unit,

(Equation 7) It is expressed as f ′ (x) represents the derivative of the sigmoid function f (x), and specifically,

[Equation 8] Can be expressed as Since the input / output function of the input layer unit is an identity function, the correction amount of the output value becomes the correction amount of the input value as it is, and the aeration air flow Qaer _n and the excess sludge removal amount Qwas _n that are the operation amounts at the current time are corrected. Quantity ΔQaer _n , ΔQwas
_n is expressed as ΔQaer _n = Δyi ₄ ΔQ was _n = Δyi ₆ . The above is the error backpropagation calculation and the correction amount of the operation amount calculated by the calculation.

Therefore, since the operation amount correction unit 4 uses the error backpropagation calculation as described above, the operation amount correction amount Δ
u _n can be easily calculated.

Next, learning of the multilayer neural network model will be described. Learning is performed in advance based on past data before performing control by this control system, but even if conditions such as the environment of the sewage treatment process change, the feature amount can be predicted accurately, so control is in progress. It will be held at any time. In the present embodiment, the learning data is generated from the characteristic amount and the operation amount obtained by processing the image signal obtained from the past visual sensor and the sensor information.

In FIG. 6, reference numeral 15 is a database holding past feature quantities and operation quantities, 16 is a subtracter, and 17 is an error backpropagation learning device. Further, 18 is an arrow indicating that the output of the input layer unit of the neural network model is input to the error backpropagation learning device, and 19 is the output of the intermediate layer unit of the neural circuit model input to the error backpropagation learning device. , 20 is an arrow indicating that the coupling weight between the intermediate layer and the output layer of the neural network model is input to the error backpropagation learning device,
Reference numeral 21 is an arrow indicating that the correction amount of the coupling load between the input layer and the intermediate layer calculated by the error back-propagation learning device is output to each coupling load, and each coupling weight is corrected, and 22 is an error back-propagation learning device. An arrow indicating that the calculated correction amount of the coupling load between the intermediate layer and the output layer is output to each coupling load and each coupling load is corrected is shown.

Characteristic amounts s _m-1 and s _{m of} past time (m-1) Δt and mΔt, manipulated variables Qaer _m-1 , Qaer _m and Qwas
_m-1 , Qwas _m , sensor information DO _m-1 , DO _m , MLS
_{S m- 1, MLSS m, Qinf} m-1, Qinf m, Cwas
_m-1 and Cwas _m are input signals, and the time (m +
1) The feature amount s _{m + 1 of} Δt is output from the database 15 to the neural circuit model as a teacher signal. In the neural network model, forward calculation is performed on the input signal, and the output signal s
(Hat) Output _{m + 1} . In the subtractor 16, the teacher signal s
The error signal between _{m + 1} and the output signal s of the neural circuit model (hat) _{m + 1} is calculated and inputted to the error back propagation learning device 17.
In addition to the error signal, the output of each input layer unit of the neural network model, the output of each intermediate layer unit, and the connection weight between each intermediate layer and the output layer are input to the error back-propagation learning device 17, and the correction amount of the connection weight is input. Is calculated.

Assuming that the learning data is created from the past data from time M1Δt to time M2Δt, the formula for calculating the correction amount will be described.

When the output of the neural network model is s (hat) _{m + 1} with respect to the input of the data of the times mΔt and (m-1) Δt, the error function E is

[Equation 9] The connection weight of the neural network model and the correction amount of the threshold value are calculated as follows according to the error back-propagation learning method.

[0065]

[Equation 10] Δw2 _{k, j} is the j-th unit in the middle layer-the k-th unit in the output layer
Correction amount of coupling load w 2 _{k, j} between the th unit, Δth
o _k is the correction amount of the threshold tho _k of the k-th unit output layer, .DELTA.W1 _{j, i} is the input layer i-th unit - coupling weight w1 _j between the intermediate layer j-th surface _unit, the correction amount of _i , Δt
hh _j is the correction amount of the threshold value thh _j of the j-th unit in the middle layer, and ε is a learning constant.

[0066]

[Equation 11] The coupling load and the threshold value are corrected based on these correction amounts.

The forward calculation, the error backpropagation calculation, and the error backpropagation learning method have been described for the three-layer neural circuit model, but the same calculation can be performed for the neural circuit model composed of two or more layers. It can be used for feature amount prediction and operation amount correction.

[0068]

As described above, according to the present invention,
Since the state of the activated sludge can be controlled to the target state, the sewage can be controlled to an environment in which the microorganisms are most activated and can be satisfactorily purified.

[Brief description of drawings]

FIG. 1 is a schematic block diagram of an entire sewage treatment process control device according to an embodiment of the present invention.

FIG. 2 is a flow chart showing a calculation procedure of the entire sewage treatment process control device shown in FIG.

FIG. 3 is a schematic block diagram of the entire sewage treatment process control device according to another embodiment of the present invention.

FIG. 4 is a diagram schematically showing a micrograph of activated sludge.

FIG. 5 is a diagram of a three-layer neural circuit model used in the feature amount prediction unit shown in FIG.

FIG. 6 is a diagram illustrating learning of the neural circuit model shown in FIG.

[Explanation of Codes] 1 ... Visual sensor unit 2 ... Image processing unit 3 ... Feature amount prediction unit 4 ... Manipulation amount correction unit 5 ... Adder 6 ... Switch 7 ... Sewage treatment process 8 ... Various sensors 9 ... Comparison unit

Claims (5)

[Claims] 1. A photographing means for photographing sewage flowing out from an aeration tank of a sewage treatment process, an image processing means for extracting a characteristic amount of an image photographed by the photographing means, an operation amount given to the sewage treatment process, and Based on the feature quantity extracted by the image processing means, the feature quantity predicting means for predicting the feature quantity after a predetermined time and the feature quantity predicted by the feature quantity predicting means match the target feature quantity, Sewage treatment process control, comprising: an operation amount correction unit for obtaining the operation amount correction amount; and an addition unit for adding the operation amount correction amount obtained by the operation amount correction unit to the operation amount. apparatus. 2. A photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, an image processing means for extracting the characteristic amount of the image photographed by the photographing means, dissolved oxygen concentration sensor information of the aeration tank, and mixing. A detection unit that detects at least one of the liquid suspended solid concentration sensor information, the inflow meter information, and the surplus sludge concentration sensor information, the detection result by the detection unit, the operation amount given to the sewage treatment process, and the image processing unit. Based on the extracted feature amount, a feature amount predicting unit that predicts a feature amount after a predetermined time, and the operation amount of the operation amount so that the feature amount predicted by the feature amount predicting unit and the target feature amount match. It is provided with an operation amount correction unit for obtaining a correction amount, and an addition unit for adding the correction amount of the operation amount obtained by the operation amount correction unit to the operation amount. Sewage treatment process control apparatus that. 3. A photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, and the ratio of the area of one activated sludge to the whole area of the image, the contour line of the activated sludge, from the image photographed by this photographing means. Length, the ratio of the area of the activated sludge screen to the length of the contour line, the number of substances that match a specific shape, the color of the activated sludge, or the image processing for extracting at least one of the values of the image elements as a feature amount. Means, a feature quantity predicting means for predicting the feature quantity after a predetermined time based on the operation quantity given to the sewage treatment process and the feature quantity extracted by the image processing means, and the feature predicted by the feature quantity predicting means. And a target feature amount such that the amount and the target feature amount match, the operation amount correction means for obtaining the operation amount correction amount, and the operation amount correction amount obtained by the operation amount correction means are added to the operation amount. Sewage treatment process control apparatus characterized by comprising adding means for. 4. A sewage treatment process comprising a photographing means for photographing the sewage flowing out of the aeration tank of the sewage treatment process, an image processing means for extracting the characteristic amount of the image photographed by the photographing means, and a neural circuit model. Based on the operation amount given to the and the feature amount extracted by the image processing unit, a feature amount prediction unit that predicts a feature amount after a predetermined time, and is realized by error backpropagation calculation of the neural circuit model,
A manipulated variable correcting unit that obtains a correction amount of the manipulated variable so that the feature amount predicted by the feature amount predicting unit and the target feature amount match, and the operation amount corrected by the operation amount correcting unit A sewage treatment process control device, comprising: an addition unit that adds an amount to the operation amount. 5. A photographing means for photographing the sewage flowing out from the aeration tank of the sewage treatment process, an image processing means for extracting the characteristic amount of the image photographed by the photographing means, an operation amount given to the sewage treatment process, and Based on the feature values at multiple past times extracted by the image processing means,
Feature amount predicting means for predicting the feature amount after a predetermined time, and operation amount correcting means for obtaining the correction amount of the operation amount so that the feature amount predicted by the feature amount predicting means and the target feature amount match. A sewage treatment process control apparatus comprising: a sewage treatment process control unit; and an addition unit that adds the correction amount of the operation amount obtained by the operation amount correction unit to the operation amount.