JPH03135497A - Chlorine injection control method - Google Patents

Abstract

PURPOSE:To easily perform formation, addition, delation and improvement of a membership function and the rule of inference and to perform the operation equal to an operator by performing forecast and control of injected amount of chlorine based on experience of the past by a neural net. CONSTITUTION:A neural circuit model 701 having a hierarchical structure is equipped which is constituted of an input layer 710, at least one layer of intermediate layer 720 and an out layer 730. The meteorological conditions in a plurality of points of time of the past and the conditions of water quality are inputted. Further the injected amount or the concn. of chlorine correspondent to the input is learned in the model 701 as an educator pattern. The meteorological conditions forecast in a point of the present time or in the future and the conditions of water quality are inputted to the learned model 701. The injected amount or the concn. of chlorine correspondent to the input is forecast. An injecting means of chlorine is operated on the basis of this forecast value. Thereby operation conformed to experience and a precedent is made easy and rapidly followed to change in situation and forecast is performed.

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial Application Field] The present invention relates to a method for controlling the amount or concentration of pre-chlorine injection in order to properly maintain the chlorine injection effect in pre-chlorine injection control in a water purification plant.

[Conventional technology]

At water treatment plants, chlorine is injected into raw water to (1) remove iron, manganese, ammonia nitrogen, nitrite nitrogen, and organic matter;
(2) Preventing the growth of bacteria, bacteria, protozoa, and algae. Chlorine is consumed by reactions with these substances or by reactions due to solar radiation, but residence time (dead time) also affects the injection, so injection depends on the operator's intuition and experience.

For this reason, in addition to the method of constructing a physicochemical mathematical model and determining the appropriate amount of chlorine injection through simulation, a fuzzy control method that simulates an operator's operation has been proposed.

[The problem that the invention attempts to solve]

In fuzzy control, creating membership functions, which are characteristic functions that quantify ambiguity, and creating rules for inference require a great deal of effort, including data analysis and operator interviews.

You can also add, delete, and add membership functions and inference rules.
Software maintenance such as improvements requires a lot of effort.

[Means to solve the problem]

Neural networks (
neural circuit model).

The present invention is characterized in that the pre-chlorine injection amount is predicted and controlled based on past performance using a neural network. The operator determines the expected weather, the season, time and measured solar radiation, measured water volume, water quality, water temperature, and other factors that affect chlorine consumption, and the amount or concentration of chlorine injected under the relevant conditions. He mentally predicts the amount or concentration of chlorine to be injected based on the past history of chlorine. Here, the concentration refers to the average chlorine concentration of purified water after chlorine injection.

In particular, since the chlorine consumption reaction is complex, operators must consider past injection results (injection history) when injecting chlorine. Applying a neural network to artificially perform this action 6 First, the neural network is given weather information at multiple points in the past,
■Season, time, and measured solar radiation, ■Measured water volume,
Have students learn ■ Chlorine injection amount or concentration under the conditions of water quality, water temperature, and ■ ~ ■. . The specific configuration of the neural network that executes this is a hierarchical structure consisting of an input layer based on a neuron element model, at least one intermediate layer, an output layer, a comparison layer, and a teacher layer. The method is performed by a known error backpropagation method. The trained neural network automatically builds a way for the operator to drive based on past performance. Next, the input layer of the trained neural network is inputted with the unlearned predicted weather, ■season, time, and measured solar radiation, ■measured water volume, water quality, and temperature, and ■chlorine injection amount or Past actual concentration values are input, and the output layer outputs (predicts) the previous chlorine injection amount or concentration. The amount of pre-chlorine injection is controlled based on this predicted value.

[Effect]

In the present invention, the learning and prediction ability of a neural network is applied to a pre-chlorination injection control method and system. In other words, first, the neural network is given ■weather at multiple points in the past,
■Season, time, and measured amount of solar radiation; ■Measured water volume, water quality, and water temperature; and ■Chlorine injection amount or concentration history.

At this time, the distribution of the weighting coefficients of the neural network changes.

This corresponds to the operator memorizing various past achievements. In this trained neural network, unlearned information such as ■expected weather, ■season, time, and measured solar radiation.

By inputting (1) the measured water quantity, water quality, and water temperature, and (2) past actual values of the chlorine injection amount or concentration, it is possible to predict the target value of the chlorine injection amount or concentration to be injected. This simulates the method by which an operator predicts the amount of chlorine to be injected by taking various past results into account.

〔Example〕

The present invention uses a neural network to record multiple past history patterns of ■ weather, ■ season and time, measured solar radiation, ■ measured water volume, water quality, water temperature, ■ chlorine injection amount or concentration, etc. Example: After learning using a neural network, a trained neural network (neural network for prediction)
The system inputs ■ to ■ on the current day and predicts unknown ■ to control the pre-chlorine injection amount or concentration.

The execution process includes (1) a learning process using a learning neural network model (learning neural network), (2) a prediction process for the amount or concentration of chlorine injection by association with a trained neural network (prediction neural network), (3) It consists of a process of controlling the amount or concentration of chlorine injection based on the predicted results.

The configuration and operation will be explained using the embodiment shown in FIG. First, I will explain the flow of a water treatment plant. In FIG. 1, raw water from a river or lake (not shown) is introduced into the landing well 9. In the rapid mixing tank 10, the water from the landing well 9 is introduced into the flocculant tank I.
A flocculant is injected from the IA and stirred by a stirrer 13 to flocculate suspended fine particles in the raw water. Although not shown in the diagram,
Alkaline agent is injected as needed. In the floc formation pond 15 (there are a plurality of ponds, but only one pond is omitted in the figure), flocs (agglomerates) grow due to stirring by the stirring paddle 17. The grown flocs settle in a sedimentation basin 16, and the supernatant liquid is filtered in a filtration basin 17. Filtered water is from water purification pond 20
The water is temporarily stored in the water distribution reservoir 21, and is sent to consumers via the piping network 24 by the pump 22. Chlorine storage tank 25 for sterilization (stores liquid chlorine, sodium hypochlorite, etc.)
An appropriate amount of chlorine is injected into the receiving well 9 and the water purification pond 20 from the chlorine injection pumps 26A and 26B. The chlorine injected into the receiving well 9 is called "pre-chlorine," and the chlorine injected into the water treatment pond 20 is called "post-chlorine."

Next, the measuring device will be explained. Water landing well 9, floc formation pond 15, settling basin 16, filter suitable area 17 outlet, water purification pond 20,
Each water distribution reservoir 21 has 9M of measuring equipment.

15M, 16M, 17M, 20M, and 21M will be installed. Measurement items are water temperature, turbidity, alkalinity, PH1 conductivity, residual chlorine concentration, chlorine demand, organic matter, iron, manganese, ammonia nitrogen, organic nitrogen, water volume (process flow rate)
, water level, head loss, etc.

Additionally, a pyranometer 8M is set up to measure the amount of solar radiation (solar radiation intensity, ultraviolet intensity, etc.).

Next, the configuration of the predictive operation control device 80 will be explained. The predictive operation control device 80 is a computer system, and a flowchart of the processing is shown in FIG. 1 to facilitate explanation of the present invention. In addition, the part described as network in the figure indicates a large number of wirings (or information communication routes) corresponding to a solid line 702 in FIG. 4, which will be described later.

First, the history pattern data file 71F in the predictive operation control device 80 is
The weather entered in , ■ the season and time, and the amount of solar radiation measured by the pyranometer 8M, ■ the measuring device 9M.

The water volume, water quality, and temperature measured at 15M, 16M, 17M, 20M, and 21M, ■The amount or concentration of chlorine injected with the chlorine injection pump 26A (average chlorine concentration after injection), and the above ~■ time series data are stored. Ru. Automatically from the historical pattern data file 71F or by an instruction 71S from the communication means 46, ■a selected data string T of the weather;
Selected data string N of season, time, and amount of solar radiation, ■water amount, and water quality.

A data string C in which the water temperature is selected and a data string Q in which the chlorine injection amount or concentration is selected are each output to the learning neural network 71. These data T, N, C, Q
Learning is executed using the execution results 71S1 and 71S2.
is output to the prediction neural network 72. The prediction neural network 72 selects data T, N, C, and Q necessary for prediction from the history pattern data file 71F and uses them for prediction. The prediction result is displayed on the communication means 46 as a prediction signal 72S, and is also displayed in the operation control process 75.
Output to. The operation control process 75 receives the prediction signal 72S and transmits the target value signal 75S of the chlorine injection amount to the chlorine injection pump 26.
A is output to control the injection amount. At the same time, the injection amount target value signal 75S of the target value is displayed on the communication means 46, and the actual injection amount is corrected according to selection by the operator 101 as necessary.

The correction value is again output to the chlorine injection pump 26A as the injection amount target value signal 75S.

Next, the present invention will be explained in detail using FIG.

FIG. 2 and subsequent figures will be used to explain the detailed configuration and operation.

In the description, an example will be described in which the amount of chlorine injection is controlled, but the same method can be applied to the case of controlling the chlorine injection concentration.

First, a method for storing data in the history pattern data file 71F will be explained. The historical pattern data at time t=0 is: ■ Weather T (0, 1), ■ Season 2 time 2 solar radiation N (0, 1), N (0, 2), N (0, 3),
■Water quantity, water quality, water temperature C(0,1), C(0,2), C
(0,3) and (2) chlorine injection amount Q(0,1). The water quality C(0,2) is each of the aforementioned multiple measured quantities at time 1=0, and it goes without saying that accuracy will be improved by using these quantities, so a detailed explanation of the case of multiple quantities will be omitted.This is one example. Here, they are collectively expressed as C(0,2). These T
,N.

The data of C and Q are stored in the history pattern data file 71F. If you repeat this, 1=0°-1,-2,・
... data is stored sequentially. In this embodiment, a case where this time interval is one hour will be explained, but the implementation of the present invention is not restricted even if it is shorter or longer than one hour.

Next, the learning neural network 71 in the learning process
The operation will be explained below. Learning neural network 71
Now, learning is executed by receiving data selected from the history pattern data file 71F. This data selection and learning method will be explained below using FIGS. 1 and 2.

As shown in Figure 2, ■Weather T(0,1) and ■Season.

Time 9 solar radiation N(0,1), N(0,2), N(0,
3) and ■Water volume.

Water quality, water temperature C(0,1), C(0,2), C(0,3
), and ■ Chlorine injection amount Q (0, 1) at any time 1
=1□ as a reference and go back to the past and calculate tl-1, t knee 2. First learn... Similarly, 1=12 (12≠t
i) as a reference, t2-1. The pattern data of t, -2, . . . are learned, and a total of q pattern data are learned. Learning is performed by dividing these data into input data and teacher data. As shown in FIG.
0 is (1) any time 1 (1=1, tl,...)
T(t, 1) at.

N(t, 1), N(tl2)IN(tl3)? C(tt
l) Ic(tl2) Ic(tl3) and T(t-1, 1) retroactively from time t.

N(t-1,1), N(t-1,2), N(t-1,3
), C(t-1,1), C(t-1,2), C(t-1
, 3), and Q(t-1, 1) are input sequentially. Here, Q
Note that (t-1,1) has been added.

On the other hand, Q(t, 1) is input to the teacher layer 750.

Learning is performed on the input layer 71o, the intermediate layer 720 . Output layer 730° Comparison layer 740. This is performed using a learning neural network 71 consisting of a teacher layer 750.

The signal processing method in the learning method will be explained below using FIG. The configuration and signal processing method shown in FIG. 3 are known, except for the input data setting method described above. That is, the configuration and signal processing method shown in FIG. 3 are publicly known techniques devised by Rumelhart et al.
allell DistributedProcess
ing, MIT Press, Vol. 1. (198
Please refer to 6)).

The configuration and operation of FIG. 3 will be outlined. First, the symbols will be explained. In FIG. 3, O is a neuron element model 701 having a product-sum operation and a sigmoid transformation function, and a solid line 702 connecting O and O indicates that information is exchanged between the neuron element models 701. Here, each layer consists of a finite number of neuron element models, and all neuron element models in adjacent layers are connected. middle layer 72
0 may have a plurality of layers, but in this embodiment, for simplicity of explanation, an example in which the number of intermediate layers is one is shown. The configuration shown in FIG. 3 is called a neural network (neural circuit model).

Next, the basic operations of the neuron element model 701 will be explained using FIG. 4. Data T, N, C, and Q input to the input layer 710 are collectively input (assuming there are n pieces in total).
Then, as shown in Figure 4, there are n variables and the variable values are Y□~Yn.
It is written as An operation (sum-of-products operation) in which each of the input signal values Y-Yn is multiplied by a weighting coefficient Wji and then added is calculated using equation (1).

Here, Yi (1): Yi value of input layer (first layer), Wj
i (2←1) Weighting coefficient from the i-th variable of the two-person rower (first layer) to the j-th neuron element model of the middle layer (second layer), Zj (2): Weight coefficient of the middle layer (second calendar) ) is the total input value to the j-th neuron element model.

In the neuron element model 701, the output value here is expressed as Equation (2) (sigmoid transformation) depending on the size of Zj(2).
It is calculated by

Although the calculation content of equation (2) is nonlinear transformation as shown in FIG. 5, the same effect can be obtained even if linear transformation is applied. The calculated value Yj(2) is further sent to the output layer, and the same calculation is performed in the output layer.

Next, an overview of the calculation method using a neural network will be explained. The aforementioned variable value yi(1) is input to the input layer 7 in FIG.
10, and this signal value is output to the neuron element model of the intermediate layer 720.6 In the neuron element model of the intermediate layer 720, these output values Yi(1) and the weighting coefficient Wi
Calculate the sum of products Zj (2) with j (2←1) using formula (1),
Depending on the magnitude, the output value Yj(2) to be output to the output layer 730 is determined using equation (2). Similarly, the output value Yj(2) of the intermediate layer 720 is further calculated by the weighting coefficient Wij(3←2) between the intermediate layer (second layer) 720 and the output layer (third layer) 73o.
7) Calculate the sum of products Zj (3) using equation (3).

Kokote, Yi (2): value of the middle layer (second layer), Wji (
3←2): Output M from the first variable of the middle layer (second layer)
Weighting coefficient for the j-th neuron element model of the (third layer), Zj (3): 2, which is the total input value of the output N (third layer) for the j-th neuron element model, Zj (3) The output value Yj(3) to the output layer 730 is calculated according to the magnitude of Yj(3) using equation (4).

In this way, the calculated value Yj(3) of the output layer is obtained. Yj(3) is Q(t,l)* in this example.

To perform learning with a neural network, the output layer 73
After 0, a comparison 5740 and a teacher signal layer 750 are further provided, and the output W! J73o signal 730S and teacher signal layer 7
The teacher signal 750S of 50 enters the comparison layer 740, where the output signal 730S compares the magnitude of the teacher signal 750s1' and the teacher signal 750s1' and Q(t,1), respectively. In order to reduce this deviation, the weighting coefficient Wji (3
←2) and Wji(2←1) (7) Correct the size.

When the calculations of equations (1) to (4) are executed again using this corrected value and the magnitudes of the output signal 730S and the teacher signal 750S are compared, a deviation similarly appears. Based on this deviation, the weighting coefficients W j i (3←2) and Wji (2
In this way, the weighting coefficient wji is repeatedly corrected until the deviation becomes equal to or less than a predetermined value. Initially, the weighting coefficients are randomly given by generating random numbers, so the deviation is naturally large, but the output signal value gradually approaches the teacher signal value. At this time, in equations (1)-(4), only the weighting coefficient value wjj is changed, so the learning result is W
This will be reflected in the distribution of ji values. Note that this Wji value is used in the prediction process.

The method of correcting the deviation in this manner is called the error backpropagation method, and uses a known technique devised by Rumelhart et al. For details, refer to the literature (ParallelDis
Tributed Processing, HIT P
(1986)). Although such a learning method itself is publicly known, the present invention particularly focuses on learning at different times (1=1 tt2 tt3
1... The history pattern data group of tj is repeatedly learned. As a result, a pattern grasping ability comparable to the operator's past experience is accumulated and stored in the distribution of the weighting coefficients Wji of the neural network. This operation was made to have the same effect as the operator's past experience.

A method of learning historical pattern data groups at multiple times t"tt ttz +ta,...t will be explained below. Multiple historical pattern data groups are learned by arbitrarily changing time t. On this day t1rttt"-3+ ...t may be selected by the operator or automatically, and both will be explained below.

The operator selects a typical pattern of the variable values Y1 to Yn that the operator would like to reflect on the operation later, or an abnormal pattern that the operator would like to refer to later. As a result, the neural network outputs results according to what it has learned, so this selection is important. Leaving the selection to the operator relies on the operator's empirical and comprehensive ability to judge data. In this case, the patterns to be learned are patterns at different times in the past, and a plurality of patterns are repeatedly learned. This gives the neural network the ability to grasp patterns comparable to the operator's past experience. The operator sets the time by man-machine conversation via the communication means 46. For example, 12 patterns are learned from a typical pattern for January to a typical pattern for December.

On the other hand, when performing automatically, statistical analysis of the data string is performed in advance. In other words, the system uses statistical analysis to find the most frequently occurring pattern and considers it as a typical example of normal conditions and learns it.On the other hand, it learns patterns that occur less frequently as typical examples of abnormal conditions. . In the embodiment shown in FIG. 1, there is only one neural network, but it would be more effective if the neural network was divided into a neural network for normal times and a neural network for abnormal times (unsteady times).

Next, the prediction process by the prediction neural network 72 will be explained. The configuration of the prediction neural network 72 is shown in FIG. As shown in FIG. 1, the prediction neural network 72 predicts the amount of chlorine injection based on the learning results of the learning neural network 71, that is, the distribution of the weighting coefficient values Wji. Weighting coefficient W obtained as a result of learning
The values of ji (3←2) and Wji (2←1) are the signal 71S
1 and 71s2 to the prediction neural network 72.

Prediction is performed using these. Although the learning neural network 71 and the prediction neural network 72 are shown separately in FIG. 1 for convenience, it is of course possible to use the same neural network. .

Next, the operation of the prediction neural network 72 will be explained using FIGS. 6 and 7. As shown in FIG. 6, the prediction neural network 72 has a configuration in which the comparison M 740 and the teacher layer 750 are removed from the learning neural network 71. What the prediction neural network 72 performs is based on the currently predictable or known weather T, season 2 time 2 solar radiation N, water amount, water quality, and water temperature C1, and the unknown
This is to predict the amount Q of chlorine injection. For this purpose, first, a variable value Yi set on the basis of 720 is input to the input layer 710 as an input layer pattern. As shown in Figure 1 or Figure 7, this variable value Yi is determined by (1) the currently known weather T (0,1), and ■season 2 time 2 solar radiation (7) N (0,1). , N(0,2), N(0,3)
and ■C(0,1), C(0,2) of water quantity, water quality, and water temperature.
, C(0,3).

(2) Data of ■ to ■ in the past (t=-1) and ■ Chlorine injection amount Q(-1,1), in addition to this, T of t knee 2
,N.

Values for C and Q may also be included; these are all actual values or known data. How much chlorine should be injected now based on these values, i.e. the unknown chlorine injection target value Q
Predict the (0,1) book. The results are output on output 1730. The calculation method up to the output is the same calculation as described above using equations (1) to (4).

The chlorine injection target value Q(0,1)* is the signal 72S.

The operation control process 75 will be explained below. Operation control process 75
Then, in response to this Q(0,1) quintillion signal 72S, a target value signal 75S for the chlorine injection flow rate is output to the chlorine injection pump 26A to control the flow rate. The calculation frequency of the signal Q(0,i)* is usually every hour, but this time unit can be set arbitrarily. Of course, the smaller the time interval, the better the prediction accuracy. If the set time interval (one hour in this embodiment) is too long to predict the amount of chlorine to be injected over a short period of time (for example, one minute), the prediction is made by mathematical interpolation. at the same time,
The injection amount target value signal 75S of the target value (prediction result) is displayed on the communication means 46, and the actual flow rate is corrected by selection by the operator 101 as necessary. For example, the predicted flow rate is corrected by multiplying the proportional coefficient by 4 and adding 2 to the constant, and this correction value is again applied to the injection amount target value signal 75S.
The chlorine injection pump 26A injects chlorine into the landing well 9 by this flow rate.

In addition, in the explanation of this example, the amount of chlorine injection that is actually operated is explained as a target value, but since the first purpose is to control the chlorine concentration in water in the water treatment plant, the chlorine concentration after injection is set as the target value. The method is practical. In this case, the average chlorine concentration after injection, which can be calculated using equation (5), is used instead of the chlorine injection amount Q(t, 1).

Here, Qo(0) is the processing flow rate. What is the specific method?

Qc(0) is used instead of the explanation using Q(t, 1).

〔Effect of the invention〕

The effects of the present invention will be described below. The traditional method of predicting and controlling the amount of chlorine injection (or concentration) using fuzzy control requires a lot of effort, including analyzing large amounts of data, interviewing operators, and creating, adding, modifying, deleting, and improving inference rules. By applying the present invention, the operator can easily carry out "operation based on actual results and precedents" with less effort. Furthermore, since learning can be performed at any time, it is possible to quickly follow changes in the situation, learn and predict.

[Brief explanation of the drawing]

Fig. 1 is a block diagram showing an embodiment of the present invention, Fig. 2 is an explanatory diagram of a learning pattern, Fig. 3 is a block diagram of a neural network,
Fig. 4 is an explanatory diagram showing the neuron element model, Fig. 5 is a characteristic diagram showing signal conversion in the neuron element model, and Fig. 6 is an explanatory diagram showing the neuron element model.
The figure and FIG. 7 are explanatory diagrams for explaining the prediction process. 9...Water landing well, 25...Chlorine storage tank, 26A...
Chlorine injection pump, 9M, 15M, 16M, 17M, 20
M. 21M...Measuring instrument, 46...Communication means, 101...
- Operator, 71F... History pattern data file, 71... Neural network for learning, 72... Neural network for prediction, 75... Operation control process, 71
0...Input layer, 720...Middle layer, 730...Output layer. 74, 0... Comparison layer, 750... Teacher layer, 701.
・New Figure 3 Input layer Figure 4 Figure 5 Figure 6 Prisoner input layer Figure 7 'RO?

Claims (1)

[Claims] 1. A method of manipulating the amount or concentration of chlorine injection in accordance with weather conditions, which are weather and solar radiation, and water quality conditions, which include at least one of water volume, water temperature, and water quality, which change over time. The neural circuit model has a hierarchical structure consisting of an input layer, at least one intermediate layer, and an output layer, and inputs the weather conditions and water quality conditions at multiple points in the past. The neural circuit model is made to learn the chlorine injection amount or concentration corresponding to the above as a teacher pattern, and the learned neural circuit model is inputted with the weather conditions and water quality conditions expected at present or in the future. A chlorine injection control method characterized by predicting a corresponding chlorine injection amount or concentration and operating a chlorine injection means based on this predicted value.