KR100331708B1 - Method and apparatus for automatically calculating coagulant injection rate in portable water purification system - Google Patents

Abstract

PURPOSE: A method and an apparatus for automatically calculating coagulant injection rate in portable water purification system are provided to efficiently cope with water quality fluctuation, thus suitably controlling turbidity of portable water. CONSTITUTION: The apparatus includes a water quality analysis instrument(211) which measures turbidity, water temperature, pH and alkalinity of raw water introduced through a water intake; a water quality analysis instrument(217) which measures turbidity of influent water flowing into a settling tank; a calculation part(212) which calculates the amount of coagulant to be injected; a PID controller(215) which injects a certain amount of coagulant into the water intake according to coagulant injection rate calculated from the calculation part; a data collection part(213) which periodically collects JAR test data; a model formula modification part(214) which corrects coagulant injection rate by receiving JAR test date transmitted from the data collection part.

Description

Automatic Computing Device and Method for Injecting Coagulant in Water Treatment System

The present invention relates to a water treatment system, and more particularly, to an apparatus and method for automatically calculating a coagulant injection rate of a water treatment system for automatically calculating and controlling an optimal drug injection rate in a coagulant injection process for efficient and systematic water quality control.

Today, due to the rapid development of industry, the quantitative increase of untreated sewage and waste water causes the increase of the cause of river water pollution, and large reservoirs such as multi-purpose dams are being built, so that the pollutants with the rainfall and the contaminated water for a long time As various nutrients are mixed, water pollution phenomenon is eutrophicated.

As a result of the serious social problems caused by the severe water quality degradation, more effective and systematic water treatment is urgently needed.

A general water treatment system is shown in the schematic diagram of FIG.

These water treatment plants are equipped with a remote station (R) in the intake station for the intake pump control, the chemical input chamber for the chemical injection rate, the filter paper for the automatic control of the filter paper, the chlorine chamber for the sterilization and the water supply chamber for the water supply and the water pump control. / S # 1 to R / S # 7) installed in the central control room, operator operator for data collection and operator console from each remote station (R / S # 1 to R / S # 7), full control and operation A central computer is installed for the distributed control.

The main purpose of a water treatment plant is to provide safe and clean drinking water. In general, raw water contains a large amount of turbidity, so that raw water is purified through filtration, flocculation and separation to remove it.

The general water treatment process is shown in the schematic diagram of Figure 2, the raw water is drawn from the intake station, and after the pre-chlorination for sterilization and disinfection, the chlorine to remove the colloidal suspension in the water ) Is mixed with coarse particles to settle economically and sedimentation basin, filter paper to remove contaminated water through the porous layer through the porous layer, filter water sterilized through the above process to control the amount of water from the drainage Supply to the consumer.

In general water treatment system, chemical injection control is divided into flocculation injection process for flocculation and sedimentation and chlorine injection process for sterilization and disinfection.

Agglomeration and precipitation in the flocculant injection process are the most basic processes and account for the largest share of the water purification plant.

The flocculant injection process of a general water treatment system is as shown in FIG.

The flocculant is injected from the impingement well and causes the impurities in the water to aggregate and become heavy so that the impurities do not quickly disappear from the sedimentation basin.

Considering the flocculation and precipitation reactions in this water treatment, water quality items such as raw water turbidity, water temperature, pH, alkalinity, and electrical conductivity are known to be relatively related to the flocculation process.

The measurement position of the water quality item in the general flocculant injection process is as shown in FIG. For example, the turbidity of raw water is generally 10-odd mg / l and can be increased to several hundred mg / l, but most of the impurities are removed from the settling basins by flocculants and the turbidity at the outlet is only a few mg / l.

Thereafter, residual impurities are removed by sand filtration to reduce the turbidity to 1 mg / l or less.

It can be seen from this that the flocculant plays a very important role.

This flocculant injection process has the following characteristics.

1) The quality of raw water varies according to the four seasons of environmental changes and rivers.

2) Reaction of flocculant The processor is involved in water temperature, alkali, etc. in addition to turbidity, and the reaction itself is not well known.

3) The process residence time of raw water is long and there is no feedback.

4) Control failure is not allowed because the output of the process is a beverage.

Therefore, in the flocculant injection process, the flocculant input amount to minimize the turbidity of the sedimentation basin should be determined and injected.

Generally, the coagulation agent injection rate can be set by 1) a JAR test or 2) using a statistical method.

In practice, two methods are used in parallel.

Coagulant injection in the prior art is as shown in the flowchart of FIG.

The water quality analyzer 111 installed in the water well 110 analyzes the water quality items (turbidity, water temperature, pH, alkalinity) of raw water and uses a statistical software package in the statistical model 112 that receives the water quality items of the raw water. Calculate the injection rate.

That is, the statistical model 112 creates a linear model that calculates the coagulant injection rate statistically from past performance data, and calculates the coagulant injection rate by substituting water quality items.

In other words, as shown in Fig. 6, the output (Y) is the coagulant injection rate and the input is made the following linear equation with raw water turbidity, water temperature, pH, and alkalinity, and the coefficients (a, b, c, d, e)

Coagulant injection rate (Y) = a * raw water turbidity + b * water temperature + c * pH + d * alkalinity + e

The coagulant injection rate is calculated by substituting the water quality items (turbidity, water temperature, pH, alkalinity) of the raw water flowing into the impingement 110 in the formula implemented in this way.

In addition, in the JAR tester 113, the operator manually inserts raw water into 5 or 6 beakers and injects the flocculant injection rate differently to observe the precipitation state of the turbidity, and among them, manually selects the injection rate having the best suspension state. do.

In this case, when the driver selects an optimal injection rate among the injection rates calculated by the statistical model 112 wave JAR tester 113, the driver is input to the PID controller 115 by the switch 114.

Accordingly, the drug input pump 116 injects the appropriate amount of flocculant into the landing well 110 by the PID controller 115.

However, in the conventional method of injecting a flocculant by JAR test, the test takes a long time, and thus it is difficult to cope with a sudden change in the quality of raw water, and the selection of the injection rate is highly dependent on the skill and experience of a skilled expert. There is a disadvantage of not operating effectively.

In addition, in the case of the flocculant injection method using the conventional statistical software package, the flocculation reaction process is limited to be expressed in a linear form such as a statistical model. Since it does not reflect, there is a disadvantage that the error of the calculated flocculant injection rate becomes large.

Therefore, the most important problem in setting the flocculant injection rate is that the flocculant injection rate must correspond to various changes in the water quality of the raw water.

In order to improve the disadvantages of the prior art, the coagulant injection process model formula for minimizing the turbidity of the sedimentation basin in consideration of the water quality of the raw water and the turbidity of the sedimentation basin from the water quality analyzer of the impingement well It is an object of the present invention to provide an apparatus and method for automatically calculating the coagulant injection rate of the water treatment system, which is designed to maximize the effect of coagulation sedimentation and to supply the optimal beverage by automatically recalculating the process and periodically re-learning the process.

That is, the present invention uses a neural network for modeling the coagulant injection process of the water treatment system, and learns the coagulant injection process from the JAR test data using the neural network to automatically calculate and control the optimal coagulant injection rate.

In other words, the present invention of the above object is that the general plant operation is dependent on the experience of the skilled operator, so that the database of expert knowledge can be used on a computer, and the neural network modeling of the data including the knowledge of the driver To provide driving knowledge.

1 is a block diagram of a general water treatment system.

Figure 2 is a schematic diagram showing a general water treatment process.

Figure 3 is an exemplary view showing a flocculant injection process in FIG.

4 is an exemplary view showing a water quality item measurement position in FIG.

Figure 5 is a flow chart showing the flocculant injection rate in the prior art.

6 is an exemplary view showing a statistical model in FIG.

7 is a block diagram of an apparatus according to the present invention.

8 is a signal flow diagram for calculating the flocculant injection rate in the present invention.

9 is a structural diagram of a neural network applied to a model-correction unit in FIG. 7;

10 is a table showing the mean error of the prior art and the present invention.

11 to 13 are waveform diagrams comparing the actual injection rate and the model injection rate.

*** real name of sign for main part of drawing ***

210: Onset Crystal 211,217: Water Quality Analyzer

212: coagulant injection rate calculation unit 213: data collection unit

214: model correction unit 215: PID controller

216: chemical injection pump 220: immersion paper

As shown in FIG. 7, the water quality analyzer 211 and the sedimentation basin 220 which detect raw water turbidity, water temperature, pH, and alkalinity, which are the water quality items of the raw water flowing into the landing well 210, are illustrated in FIG. 7. Coagulant injection rate calculation unit for calculating the coagulant input rate by calculating the water quality analysis unit 217 for detecting the turbidity of the treated water and the neural network model formula of the water quality analysis unit (211, 217) ( 212) and the PID controller 215 for controlling the chemical injection pump 216 according to the coagulant input rate of the coagulant injection rate calculation unit 212 to inject an appropriate amount of coagulant into the landing well 210, and periodically Model number for modifying the neural network model of the coagulant injection rate calculation unit 212 by learning the data collection unit 213 for collecting JAR test data and the JAR test data collected by the data collection unit 213. It consists of the government (214).

The model correction unit 214 is composed of a neural network as shown in Figure 9 for modeling the flocculant injection process.

If described in detail the operation and effect of the embodiment according to the present invention as follows.

First, the water quality analyzer 211 detects turbidity, alkalinity, water temperature, and pH values of the water quality items of the raw water flowing into the impingement 210 and inputs them to the coagulant injection rate calculator 212.

Then, the water quality analyzer 217 detects the turbidity of the treated water flowing into the sedimentation basin 220 and inputs it to the flocculant injection rate calculation unit 212.

At this time, the flocculant injection rate calculation unit 212 determines the flocculant injection rate by substituting the input values from the water quality analyzers 211 and 217 into the coagulant injection process neural network model formula constructed by the JAR test data. The coagulant injection rate is stored in the internal memory and output to the PID controller 215.

Accordingly, the PID controller 215 injects the appropriate amount of flocculant into the landing well 210 by controlling the chemical injection pump 216 using the injection rate set value determined by the flocculant injection rate calculation unit 212 as a new set value. .

Meanwhile, the data collector 213 periodically collects JAR test data and inputs it to the model formula corrector 214.

Accordingly, the model correction unit 214 re-learns the process with JAR test data to modify the coagulant injection process neural network model equation of the coagulant injection rate calculation unit 212.

The flocculant injection process is one of several processes executed by the host computer, and the host computer executes each process at regular intervals, and when the execution of one process is finished, the next process proceeds.

The flocculant injection process is performed by the signal flow as shown in FIG. 8, and the operation thereof is as follows.

First, the flocculant injection rate calculation unit 212 reads turbidity, alkalinity, pH, and water temperature values of the water quality items of the raw water flowing into the landing well 210 from the water quality analyzer 211 and flows into the sedimentation basin 220. Is turbidity is read from the water quality analyzer (217).

Accordingly, the coagulant injection rate calculation unit 212 calculates the coagulant injection rate by calculating the values read from the water quality analyzers 211 and 217 in a coagulant injection process neural network model, and determines the coagulant injection rate. Set to the set value of the PID controller 215 and store the operation result data.

In addition, when the JAR test is periodically collected by the data collection unit 213, the model formula correction unit 214 modifies the model formula of the coagulant injection rate calculation unit 212 by relearning the process using the JAR data. Improve the efficiency of the process

In the above, the water quality analyzer 211 and the water quality analyzer 217 mounted in the sedimentation basin 210 are used to determine the water quality of raw water (raw turbidity, water temperature, pH, alkalinity) and turbidity of the treated water. Send it to the host computer periodically.

JAR test data collected in the data collection unit 213 is made by a skilled expert.

The JAR test data is largely divided into a normal state and an abnormal state based on the turbidity of the raw water flowing into the landing well 210.

Generally, turbidity of raw water is considered abnormal when turbidity is above 30 NTU and steady state when turbidity is below 30 NTU.

Therefore, in order to effectively perform neural network modeling of flocculant injection process, it is necessary to make data by JAR test while increasing the variation of water quality.

In addition, the model correction unit 214 is implemented as a neural network to model the flocculant injection process using JAR test data.

At this time, the neural network model is divided into normal and abnormal cases according to the water quality of the raw water, and the five independent variables, turbidity, water temperature, pH, alkalinity, and turbidity of the treated water are assigned as input neurons and coagulants. Assign injection rates to output neurons.

Therefore, the structure of each neural network model is shown in FIG.

If the water quality of the raw water changes suddenly, the current flocculant injection rate should be changed. Therefore, the neural network model learns the current coagulation state to cope with the rapid change in water quality of the raw water.

At this time, the learning of the neural network model of the flocculant injection process uses a back propagation algorithm (BP) proposed by Rumelhart et al.

Accordingly, by repeatedly learning the coagulant injection process neural network model using the backpropagation algorithm, the connection loads (wij, Vj, Wj) are determined to modify the model equation of the coagulant injection rate calculation unit 212.

The learning process in the above is performed using a set of input and output data obtained in the process. That is, the learning operation of the coagulant injection process using the neural network model as shown in FIG. 9 is as follows.

First, the data collection unit 213 inputs the input data obtained through the experiment of the expert to the model formula correction unit 214, wherein the output value of the input layer 310 is a value normalized the value of the input data.

Accordingly, the sum of the products of the normalized inputs and the connection load value W _ij is input to the hidden layer 320 (or the middle layer) 320 as shown in Equation (1) below.

At this time, the wing layer 320 outputs the sigmoid function value of Equation (1) and the output value (Z _i ) is expressed as Equation (2) below.

Therefore, the output layer 330 calculates the total of the output value Z _i from the vane layer 320 by the connection load W _j as the coagulant injection rate Y, as shown in Equation (3) below.

Accordingly, when the output value Y obtained by the above equation (3) and the output pattern obtained by the experiment coincide, the model equation correction unit 214 ends the learning.

If the value (Y) according to equation (3) does not coincide with the experimental output pattern, the model equation corrector 214 connects using the gradient descent in the downward direction to reduce the error. Correct the load value W _j (W _ij ).

At this time, in order to correct the connection load W _j (W _ij ), _{ΔW j} and _{ΔW ij} are obtained by the following equation (4) (5).

Here, Y is an actual value, Y 'is a calculated value, and (eta) has a value of "0 <(eta) <1" as a learning rate.

Thereby, the connection load W _j (W _ij ) is corrected as in the following equations (6) and (7).

Here, k means the number of learning.

Therefore, the output value (Y ') extracted by the repetitive learning in the repetitive learning of the flocculant injection process by correcting the connection load (W _j ) (W _ij ) of the neural network having the structure as shown in FIG. The error of the output data Y is calculated as shown in Equation (8) below, and if the error Err is within a predetermined range, the learning ends and the modeling operation ends.

Thereafter, when the coagulant injection process modeling is completed in the same manner as described above, the coagulant of the coagulant injection rate calculation unit 212 is corrected by correcting the connection load W _j (W _ij ) of Equation (1) (2) (3). The injection rate model is modified.

Accordingly, the flocculant injection rate calculation unit 212 calculates the optimal flocculant injection rate using the modified model equation.

On the other hand, the present invention was tested using, for example, a water treatment plant equipped with an average processing capacity of 1.32 million tons per day and a distributed control system.

The chemical injection process of the plant uses a combination of JAR testing and statistical techniques, and the flocculant uses PAC.

At this time, the neural network model of the flocculant injection process was implemented and modeled with the one-year JAR test data of the target plant and compared with the actual injection rate for one year.

Accordingly, the model error of the neural network is shown in the table of FIG. 10 by comparing the average error between the conventional statistical model and the coagulant injection rate. Here, the error value shown in the table of Fig. 10 is the least square error.

The injection rate calculated from the actual injection rate and the neural network model is as shown in the waveform diagram of FIG.

In addition, the coagulant injection process model formula constructed above was applied to the actual plant, and the coagulant injection rate control was performed on the site data for one year at the four series of the target water purification plant. The injection rate calculated from the actual injection rate and the neural network model was used. Comparing to the same as the waveform of Figure 12 in the normal state and the waveform of Figure 13 in the abnormal state.

Here, the portions where the actual injection rate and the calculated injection rate differ slightly in FIGS. 12 and 13 are interpreted as cases where the actual injection rate itself is not corrected despite the actual water quality change.

As described in detail above, the present invention implements a coagulant injection process model as a neural network and learns a process using a back propagation technique to determine an optimal coagulant injection rate to minimize turbidity of water quality. Using JAR test data with knowledge has the effect of improving the reliability of the system for determining flocculant injection rate.

The present invention can be utilized for the water quality always by the optimal flocculant injection, the reduction of stabilization and maintenance costs, and the rational use of water resources by water quality control.

Claims (5)

A water quality analyzer 211 for detecting turbidity, mercury, pH, and alkalinity of raw water, which are the water quality items of the raw water flowing into the water well, a water quality analyzer 217 for detecting the turbidity of the treated water flowing into the sedimentation basin, and the water quality. Coagulant injection rate calculation unit 212 that calculates a coagulant injection rate by substituting the water quality items of the analyzers 211 and 217 into the neural network model equation, and according to the coagulant injection rate of the coagulant injection rate calculation unit 212. PID controller 215 for controlling the chemical injection pump 216 to inject an appropriate amount of flocculant into the landing well, a data collection unit 213 for periodically collecting JAR test data, and the data collection unit 213 A coagulant injection rate automatic calculation device for a water treatment system, comprising a model formula correction unit (214) for learning a JAR test data to modify the neural network model equation of the coagulant injection rate calculation unit (212). The coagulant injection rate automatic calculation device according to claim 1, wherein the model correction unit (214) comprises neural networks in a steady state and an abnormal state according to turbidity of the water quality. The apparatus of claim 1 or 2, wherein the model correction unit (214) learns the JAR test data periodically using a backpropagation algorithm. A first step of reading the water quality items of the impingement well and the water turbidity of the immersion paper; a second step of calculating the coagulant injection rate by calculating the parameters read above by a neural network model; A third step of executing the flocculating agent, a fourth step of periodically reading the JAR test data, and a fifth step of modifying the neural network model equation by learning the JAR test data read above. Automatic calculation method of the flocculant input rate of the water treatment system characterized by the above-mentioned. 5. The method of claim 4, wherein the fifth step is a first process of calculating the coagulant injection rate (Y ') by calculating JAR test data by an expert, and the calculated value (Y') is the output pattern of the experiment. A second process of comparing the same, a third process of correcting the connection load W _j (W _ij ) if the comparison result does not match, and a connection load W _i (W _ij ) corrected above. A fourth process of repeating the first process and comparing whether the calculated value and the actual data (Y) are within a predetermined range; and if the error is not within the predetermined range, a fifth process of repeating from the third process; And a sixth process of terminating learning and modifying a neural network model equation if the error is within a predetermined range.