JP2000218263A - Water quality controlling method and device therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a new water quality controlling method having excellent learning ability and self-performance diagnostic ability against the characteristics of water to be treated. SOLUTION: After calculating the deviation between a target water quality and a treated water quality predicted by neural network from the water quality of the water to be treated and the injection ratio of a flocculating agent, review of the neural network is executed so that the deviation may be fixed within a prescribed range. The flocculating agent injection ratio is obtained by adding the flocculating agent supplementary injection ratio obtained by fuzzy inference from the membership function between a chlorine agent injection ratio and an alkali agent injection ratio to a basic design injection ratio, or the like, as a parameter. The membership function is adjusted for improving the review of the neural network.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and an apparatus for controlling water quality in a water purification plant.

[0002]

2. Description of the Related Art Many common water purification facilities in Japan use coagulation sedimentation + sand filtration + chlorination as main processes.

[0003] Depending on the quality of raw water, there is a place where a powdered activated carbon treatment is installed in a water pipe from a river, and an ozone + activated carbon treatment is additionally provided for the treatment of off-flavor substances and organic substances. In addition, according to these facilities, mainly coagulant,
Chlorine and alkaline agents are injected as chemicals.

[0004] The problems of the water purification equipment and their countermeasures will be described with reference to FIG.

[0005] Raw water flowing into the water purification facility is supplied from a lake 131, a river 133, and a raw water regulating pond 135. The raw water introduced into the landing well 137 is uniformly mixed with the coagulant added in the rapid mixing pond 139, and is gently stirred in the floc forming pond 141 in order to obtain a settling floc. The flocs formed by flocculation of the pollutants in the raw water in the floc formation pond 141 are separated into solid and liquid in the sedimentation pond 143. The solid-liquid separation treatment water is supplied to the rapid filtration pond 151, and is further transferred to the advanced water purification treatment facility 145 in the case of further processing off-flavor substances and dissolved organic substances. Offensive odor components and dissolved organic components in the solid-liquid separation treatment water introduced into the advanced water purification treatment equipment 145 are oxidized by the ozone supplied from the ozone generator 144 in the ozone contact pond 147. The ozone-treated water for solid-liquid separation is treated in the activated carbon adsorption pond 149, and then supplied to the rapid filtration pond 151. The filtered water is used as clean water or middle water through the water purification reservoir 153 and the distribution reservoir 155, and a part of the water is discharged.

[0006] With the recent deterioration of raw water quality and the diversification of substances to be removed, the above-described chemical injection process has become complicated. One specific example of the deterioration of raw water quality is the eutrophication phenomenon in lakes and marshes 131, which is caused by domestic wastewater flowing from residential and arable lands around lakes and marshes, and groundwater containing nitrogen and phosphorus from fertilizers. Is due. When the water of the lake 131 flows into the river 133, it causes river pollution.

As a result, in the floc formation pond in the water purification facility, solid-liquid separation occurs due to an increase in the amount of chlorine required due to the production of trihalomethane by iron manganese treatment and pre-chlorination, an increase in disinfection by-products, and coagulation and sedimentation obstacles. As a result, the turbidity of the treated water becomes worse. As a result, secondary obstructions such as residual ozone by-products in the ozone contact pond and biological leakage in the activated carbon adsorption pond in advanced water treatment plants, and leakage of cryptosporidium due to reduced filtration function in the rapid filtration pond are caused. It has become. In addition, aging of facilities, for example, aging of buried pipes, and the like, may cause deterioration of treated water quality.

[0008] In order to solve these problems, in the river 133 serving as a raw water supply source of the water purification equipment, water quality monitoring devices 132 and 134 are installed near the upstream and downstream water purification equipment so that
We monitor the quality of incoming raw water.

In the water purification equipment, not only the conventional flocculant + sand filtration + chlorination treatment but also the removal and sterilization of pollutants, as well as a countermeasure against ammonia, a low concentration UV meter 136 and a three-phase nitrogen meter in front of the rapid mixing pond 139. 138 is installed, and chlorination and alkali treatment are performed according to the fluctuation of the water quality while constantly monitoring the water quality. Further, chlorination for reducing trihalomethane production is performed as an intermediate treatment, and alkali treatment for protecting a water pipe is performed as a post-treatment. In advanced water treatment facilities, low-concentration UV meters 142 and 148 are installed on the inflow side and the discharge side.
At 147, a dissolved ozone meter is installed to evaluate the function of the equipment, and to examine the ozone contact amount and the cleaning and replacement of activated carbon. In the rapid filtration pond 151, turbidity monitoring of the sand filtration process effluent has become important as a measure against Cryptosporidium, which has recently become a problem, and adjustment of coagulant injection has been highlighted again. Therefore, a turbidity meter 140 and high-sensitivity turbidity meters 150 and 152 were installed to judge the performance of the filtration basin and replace the filter media, as well as to examine the adjustment of coagulant injection in the rapid mixing basin 139 and floc formation basin 141. Are doing.

[0010] Regarding the problem of aging of buried pipes, a water distribution water quality monitor 154 is placed at an appropriate place of the drainage pipe network 157 in the equipment, and while monitoring the water quality of the distribution water, a search is made for a contamination source and an aging part in the equipment. If so, it has been repaired.

[0011] As described above, in the water purification equipment where good water quality is obtained, operation management such as water quality control by chemical injection is performed based on the abundant experience of skilled operators. It is. In addition, since the water quality is different in each place, it is the present condition that each water treatment plant has a water quality control method.

[0012]

The adjustment of the operation in today's water purification equipment, for example, the adjustment of the chemical injection rate, is almost always performed manually. In general, the adjustment of the chemical injection rate is performed by an operator performing a beaker test visually on site based on the basic injection rate calculated from the design calculation.

However, the final criterion for determining the chemical injection rate by the beaker test is based on the operator's abundant practical experience and intuition cultivated in most cases. As described above, since the determination of the chemical injection rate is subjective, there are not few errors in the determination, and at present, the optimum chemical injection rate for the characteristics of the water to be treated has not yet been determined.

[0014] In addition, there are many situations in which the chemical injection rate is different even with the same water quality due to fluctuations in the quality of the water to be treated and changes in seasons.
For this reason, it is difficult to determine the chemical injection rate and to maintain the performance of the equipment without a skilled operator having a situation analysis capability and an appropriate response capability, and the operator is always stationed. Furthermore, if the dynamic characteristics of the equipment change due to modification or the like, maintenance cannot be performed by anyone other than the developer.

SUMMARY OF THE INVENTION The present invention has been made in view of the above circumstances, and provides a new water quality control method having excellent learning ability and self-performance diagnosing ability with respect to the characteristics of the water to be treated which tends to fluctuate. As an issue.

[0016]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides a first invention which comprises a chemical injection rate, a raw water temperature,
Raw water pH, raw water alkalinity, raw water turbidity, etc. are supplied to the neural network to predict the treated water quality, and the deviation between the target water quality and the predicted treated water quality is calculated, and the deviation is determined within a predetermined range. As described above, the operation is performed by the neural network, and the processed water turbidity prediction signal is sent to the output.

The second invention is characterized in that a chemical injection rate of the water quality data is determined by a membership function, calculated by fuzzy inference, and then supplied to the neural network.

According to a third aspect of the present invention, the neural network is formed in a plant model, and when the deviation between the target water quality and the predicted water quality is out of a predetermined range, the plant model is corrected by re-learning the neural network. It is characterized by doing.

According to a fourth aspect of the present invention, the plant model is corrected by comparing the difference between the target water quality and the predicted water quality before and after the correction, and when the corrected difference is small, updating the plant model. When the difference is large, the membership function is adjusted.

According to a fifth aspect of the present invention, there is provided a water quality data storage unit storing water quality data including a chemical injection rate, raw water temperature, raw water pH, raw water alkalinity, raw water turbidity, and the like, and water quality data output from the water quality data storage unit. And a neural network that predicts treated water quality from these data, calculates a deviation of turbidity from a target water quality, and sends out an appropriate treated water turbidity prediction signal to an output.

According to a sixth aspect of the present invention, the chemical injection rate stored in the water quality data storage section is determined by a membership function, fuzzy inference operation is performed, and an appropriate chemical injection rate obtained at the output is supplied to the neural network. It features a fuzzy inference unit. In a seventh aspect of the present invention, the water quality data accumulating section stores and retrieves the chemical injection rate calculated by the fuzzy inference section, and the processed water turbidity prediction signal predicted from the water quality data and the chemical injection rate by the neural network. It is also characterized in that it can be stored and withdrawn.

The coagulant injection rate is adjusted according to the characteristics of the raw water. When the raw water turbidity is stable, when the deviation between the actual value and the target value of the treated water turbidity is 0.2 degrees or more for a certain period of time,
It is characterized in that the basic injection rate, which is one of the parameters of the coagulant injection rate, is examined. That is, when the value of the basic injection rate is increased or decreased by one, the treatment water turbidity prediction according to the first invention is performed, and the deviation between the actual value and the target value of the treatment water turbidity is within 0.2 degrees for a certain period. If so, the value of the basic injection rate is adopted as a parameter.

When the raw water turbidity rises and the deviation between the actual value and the target value of the treated water turbidity is 0.2 degrees or more for a certain period of time, a power index of the raw water turbidity, which is one of the parameters of the coagulant injection rate, It is characterized by examining the coefficient of the power value. That is, when the coefficient value is increased or decreased by 0.1 at a time, the treatment water turbidity prediction according to the first invention is performed, and the deviation between the actual value and the target value of the treatment water turbidity is within 0.2 degrees for a certain period. If so, the value of the coefficient is adopted as a parameter.
If the deviation does not fall within 0.2 degrees, the power index is examined. That is, when the value of the power index is increased or decreased by 0.01, the treatment water turbidity prediction according to the first invention is performed, and if the difference between the actual value and the target value of the treatment water turbidity is within 0.2 degrees of a certain period. In this case, the value of the exponent is used as a parameter.

At the time of low turbidity after the rise of raw water turbidity, if the deviation between the actual value and the target value of the treated water turbidity is 0.2 degrees or more for a certain time, it is one of the parameters of the coagulant injection rate. It is characterized by studying a certain coagulant correction injection rate. That is, the vertex of each output membership function is set to 0.1
When the difference between the actual value and the target value of the treated water turbidity is within 0.2 degrees for a certain period, the coagulant correction injection rate is changed. Adopt as parameters.

[0025]

Embodiments of the present invention will be described below with reference to the drawings.

(First Embodiment) A plant model for predicting treated water quality using a neural network.

A treatment water turbidity prediction system using the neural network constructed in this embodiment will be described. Neural networks attempt to simulate the information processing system of the brain in an engineering manner, and differ in principle from conventional signal processing systems, and generally have features such as nonlinearity of input / output relations and adaptability through learning. ing. Therefore, a neural network imitating a skilled operator having excellent learning ability and performance diagnosis ability can be applied to operation control of a water purification facility to which a non-linear control factor such as raw water is supplied. Here, a schematic diagram of the neural network is shown in FIG. In FIG. 1, 1 is an input layer, 2 is a first intermediate layer, 3 is a second intermediate layer, 4 is an output layer,
5 is a teacher number.

FIG. 2 is a block diagram showing a first embodiment of the present invention. In FIG. 2, reference numeral 21 denotes a coagulant injection rate, a water temperature,
The water quality data storage unit stores data of pH, alkalinity, raw water turbidity, and raw water flow rate. Each data from the water quality data storage unit 21 is supplied to the neural network 22. Reference numeral 23 denotes a chemical injection fuzzy control inference unit to which a coagulant injection rate is supplied from the accumulation unit.
Is supplied to the neural network 22. The neural network 22 calculates each supplied input data, and outputs a treated water quality prediction signal (the treated water is the supernatant water of the sedimentation basin, and outputs the supernatant water of the sedimentation pond to the treated water in the following embodiments as well). ) And auto-tuning (automatic control) of the membership function of the fuzzy control inference unit 23 described in detail later.

The above-mentioned raw water flow rate is an important factor for maintaining the performance of the water purification equipment. In this embodiment, a simulation system was constructed in which the raw water flow rate was incorporated as one of the input factors to the neural network. In the present embodiment shown in FIG. 2, past time series data was learned under the following conditions, and a prediction model was examined. Back propagation was used as the learning algorithm.

1) Input / output items Input items: coagulant injection rate, raw water temperature, raw water pH, raw water alkalinity, raw water turbidity and raw water flow rate Output items: treated water turbidity 2) Teacher data (output), input data input Data P: 24 × 30 days−11 hours = 709 points (6 rows, 709 columns) Teacher data T: 24 × 30 days−11 hours = 709 points (1 row, 709 columns) A time delay (2 hours) is considered.

3) Selection of neuron input / output function Hidden layer (first and second): logarithmic sigmoid transfer function (tans
ig) Output layer: linear transfer function (purelin) 4) Selection of network structure 4-layer model: first hidden layer / second hidden layer / output layer = tansig
/ Tansig / purelin Number of inputs: input layer / first intermediate layer / second intermediate layer / output layer = 6
5/3/15) Training (learning and speeding up) In the back propagation method, it is preferable to use the momentum method or the Levenbert-Marquardt optimization method since a large amount of calculation time is required. As shown in FIG.
When the neural network of the first embodiment shown in FIG. 1 is used, the turbidity of the treated water can be predicted almost accurately.
As shown in FIG. 3, the predicted value of the turbidity was able to obtain a result that followed a temporal change that almost coincided with the actually measured value.

In the conventional operation and maintenance of a water purification plant, after the setting of the raw water flow rate and the coagulant injection rate is manually changed, the performance is determined by turbidity observation or the like, and the setting is examined. Therefore, if setting is wrong, much time is required to obtain good water quality. Therefore, by introducing the neural network according to the present embodiment, it is possible to quickly and accurately predict the treated water turbidity after a certain period of time, as a result, and to avoid such a situation. Become.

As described above, the plant model can be constructed by the simulation obtained from the actually measured data. Also, since the neural network is capable of self-learning, it has become possible to introduce it into the automatic driving control system.

(Second Embodiment) Water Quality Prediction Plant Control System Using Neural Network The system using fuzzy inference and the neural network shown in the first embodiment needs to adjust the fuzzy membership function and the network as time passes. Becomes
Neural nets are not always perfect, as water quality can deteriorate over time. Therefore, in this embodiment, a water quality prediction plant control system was constructed in which the plant model was sequentially reviewed and the membership function was adjusted if necessary.

FIG. 4 is a flowchart showing a method for operating a water quality control system using neuro-fuzzy based on the second embodiment. In FIG. 4, water quality / operation data is input to a plant model using a neural network for water quality prediction in step S2 in step S1,
It is input to the drug injection fuzzy control inference unit in step S3, and the drug injection rate is determined by a fuzzy membership function. The chemical injection rate obtained in step S3 is
Is input to

In the normal operation, the chemical injection rate is determined based on fuzzy inference. However, after a lapse of time, it becomes necessary to adjust the fuzzy membership function and the neural network plant model in step S2. Although the neural network is adopted as the plant model in step S2, the water quality may deteriorate over time, and is not always perfect. Therefore, the plant model is reviewed, and if the model is determined to be valid, the membership function is reviewed.

After the data when the treated water turbidity is reduced by the chemical injection in step S2 is accumulated in step S4 as the operation result data, the difference between the actually measured treated water turbidity and the target treated water turbidity is determined in step S5. calculate. As a result of the calculation, if the turbidity deviation is within the predetermined range, the operation is continued as it is via step S6. If the turbidity deviation is within the predetermined range, the plant model is corrected by re-learning the neural network in step S8 via step S7. Thereafter, the difference between the actual measured water turbidity before and after the correction and the target treated water turbidity is compared in step S9, and step S10
If the difference after the correction is small, the existing net file is updated in step S11 and the operation is continued, and if the difference after the correction is large in step 12, the fuzzy membership function is adjusted in step 13.

The method of adjusting the fuzzy membership function in this embodiment is as follows.

1) Calculate the actual measured water turbidity and the target water turbidity (turbidity deviation) 2) Taking into account the time delay, add or subtract the turbidity deviation to the injection results, and use this as a chemical (coagulant) It is used as teacher data for the injection rate output value.

3) Automatically adjust the fuzzy membership function to match this value.

As described above, it is possible to update the plant model of the neural network and automate the adjustment timing of the fuzzy membership function.

Next, a modification of the teacher data (output) and the input data in the first embodiment will be described. 12 items of raw water temperature, raw water pH, raw water turbidity, raw water alkalinity, raw water conductivity, coagulant injection rate as input data and one hour before, and one item of treated water turbidity after 2 hours as output data did. The output was made after 2 hours because the residence time in the floc formation pond and sedimentation pond was considered.

A prediction model was studied by learning input data under the following conditions.

1) Input items Input items: raw water temperature, raw water pH at present and one hour before,
Raw water turbidity, raw water alkalinity, raw water conductivity, coagulant injection rate Output item: treated water turbidity 2) Input data, teacher data (output) Input data P: 12 x 24 x 3 days-24 hours-12 hours = 828 points (12 rows, 828 columns) Teacher data T: 12 × 24 × 3 days−24 hours−12 hours = 828 points (1 row, 828 columns) 3) Selection of neuron input / output function Hidden layer (first, second) : Log sigmoid transfer function (tans
ig) Output layer: linear transfer function (purelin) 4) Selection of network structure 4-layer model: first hidden layer / second hidden layer / output layer = tansig
/ Tansig / purelin Number of inputs: input layer / first intermediate layer / second intermediate layer / output layer = 12
5/10/15 5) Training (learning and speeding up) Learning algorithm: Back propagation method Learning speeding up means: Momentum method or Levenbert-Marq
uardt optimization method 6) Error Treatment water turbidity within 0.1 degree FIG. 5 shows the prediction result of treatment water turbidity using a neural network in the above-mentioned modification.

In the conventional back propagation method, much time was required for calculation. Therefore, by introducing the current value of the water quality data and the value one hour ago into the neural network, the treated water turbidity two hours later can be predicted almost accurately. For this reason, as shown in FIG. 5, it is possible to obtain a result in which the predicted value of the turbidity follows a temporal change that substantially coincides with the actually measured value.

As described above, with the water quality control system according to the present embodiment in which fuzzy inference and neural network are combined, the optimum coagulant injection amount is set efficiently in accordance with the characteristics of raw water that is likely to fluctuate over time. It is possible to provide good and stable water quality.

(Third Embodiment) Control of Chemical Injection Rate by Fuzzy Inference As described above, most of the operation management know-how of water purification equipment is based on the practical experience of operators. The fuzzy control has an advantage that the operator's rule of thumb can be embedded in the injection control. In some water purification plants,
A method using fuzzy inference for determining the chemical injection rate has also been put to practical use.

The determination of the chemical injection rate using fuzzy inference is performed by using if-
The feature is that the rules can be created in the then format.
Even if the control target is a non-linear one such as raw water in a water purification plant, an appropriate output can be obtained.

In this embodiment, a method for controlling the injection of a flocculant as a chemical using fuzzy inference was constructed. FIG. 6 shows a block diagram of the coagulant injection control. In FIG. 6, 61 is a water quality data storage unit, 62 is a manual input unit, 63 is a fuzzy control inference unit, and 64 is an addition unit.

The water quality data storage section 61 stores values relating to raw water turbidity, pre-chlorine injection rate, pre-caustic injection rate, raw water organic matter (UV-VIS), and feedback.

From the manual input unit 62, the basic injection rate k and k of the power n are used as parameters other than the fuzzy control inference.
₁ times and k ₂ times are input. The basic injection rate k is the basic injection rate D
Is supplied to the adder 64 as _1, the raw water turbidity is supplied to the adder 64 to ₁ × k power of n as D _2, also raw water organics (UV
-Vis) is supplied to the addition section 64 as D ₄ is _doubled k.

[0052] In fuzzy controller inference unit 63, and supplies the pre chlorine injection rate and the previous caustic injection water sources coagulant calculated by fuzzy inference from factor correction infusion rate to the adding unit 64 as D _3.

Finally, the term D ₅ relating to the feedback is supplied to the adding unit 64 to obtain the coagulant injection rate D. Regarding the feedback system, turbidity removal rate, treated water turbidity, and filtered water turbidity constant control can be considered, but they are not actually added because they have not been verified this time.

As shown in FIG. 7, the fuzzy inference unit 43 has two inputs and one output. The input is the pre-chlorine injection rate 41 and the pre-caustic injection rate 42, and the output is the coagulant correction injection rate 44. Table 1 below shows the fuzzy control rules for the coagulant correction injection rate, and FIGS. 8, 9 and 10 show the membership functions of the pre-chlorine injection rate, pre-caustic injection rate, and coagulant correction injection rate.

[0055]

[Table 1]

In Table 1, FIGS. 8 and 9, PB means high injection rate, ZE means proper, and NB means low injection rate. In Table 1 and FIG. 10, PB means a large correction injection rate, PS means a slightly high correction injection rate, ZE means no correction is needed, NS means a small correction injection rate, and NB means a small correction injection rate.

The fuzzy inference unit 43 has a function of performing fuzzy inference based on the fuzzy control rules in Table 1 and calculating by the membership functions shown in FIGS. 8 and 9 to correct the coagulant injection rate. For example, it functions as follows. As the pre-chlorine injection rate and the pre-caustic injection rate increase, the coagulant correction injection rate increases. If the pre-chlorine injection rate is appropriate and the pre-caustic injection rate is low, the coagulant correction injection rate will be slightly lower. As the pre-chlorine injection rate increases and the pre-caustic injection rate decreases, the rate at which the coagulant injection rate does not need to be corrected increases.

FIG. 11 shows the results of the simulation. In the graph notation, one step is equivalent to 5 minutes, and 288 pieces per day (2
4 × 12) data. The dotted line indicates the actual value of the coagulant injection rate, and the solid line indicates the calculated value. As shown in FIG. 11, 0 to 700
Until the step, most of the simulation results could be matched with the actual values.

Conventionally, the operator of the water treatment facility has set the amount of coagulant to be injected based on basic knowledge in water treatment, abundant practical experience, and insights supported by these. Although the coagulant injection amounts determined by them are almost similar, the optimal injection amount has not yet been determined because of subjective judgment.

The introduction of the fuzzy inference according to the present embodiment makes it possible to determine an optimum coagulant injection rate according to the past coagulant injection results.

(Fourth Embodiment) Treatment Water Turbidity Prediction System Diagram
Fig. 12 shows a block diagram of the treatment water turbidity prediction system. The system according to this embodiment includes a data storage unit 122, a coagulant injection rate inference unit 123, and a treated water turbidity prediction unit 124.

The data storage unit 122 registers the water quality data 121 once. The reason why the data is registered in the data storage unit 122 is that the data is used as past data in the coagulant injection rate inference 123 and the treated water turbidity prediction unit 124.

The coagulant injection rate inference unit 123 includes a data storage unit 1
It has a function of calculating accumulated flocculant injection rate by extracting accumulated data from 22 and performing calculation by fuzzy inference according to the third embodiment. The injection rate obtained by the calculation is registered in the data storage unit 122.

On the other hand, the treated water turbidity predicting section 124 stores the coagulant injection rate obtained by the calculation and the registered water quality data 121.
Is extracted from the data storage unit 122 and a function of predicting the treated water turbidity using the neural network according to the first embodiment. The predicted result is registered in the data storage unit 122 again.

As described above, it is possible to construct a treatment water turbidity prediction system that combines fuzzy inference for an arbitrary water quality and a neural network.

By using the prediction system according to the present embodiment, it is possible to newly provide a basic design guideline and an operation method of a chemical injection facility in a water purification facility utilizing empirical rules for arbitrary water quality.

(Fifth Embodiment) Process Water Turbidity Prediction Adjustment Method Process water turbidity prediction is not always accurate. Therefore, in the present embodiment, a process based on the fourth embodiment is characterized in that it is determined whether or not the prediction is accurate based on the predicted data and the actual data of the treated water turbidity, and the adjustment is performed. A method for adjusting water turbidity prediction is provided.

In FIG. 12, the treated water turbidity predicting unit 124
The actual data of the treated water turbidity stored in the data storage unit 122 is compared with the actual data of the treated water turbidity predicted from the coagulant injection rate and the water quality data actually injected during the same period.

If the error between the two is within 0.2 degrees for a predetermined time, the current model is used continuously. Conversely, if the error is 0.2 degrees or more for a certain period of time, the learning is performed again using the data stored in the data storage unit 122 to update the network.

Learning based on the stored water quality data makes it possible to always accurately predict the treated water turbidity.

(Sixth Embodiment) Method of Adjusting Coagulant Injection Control Parameters The performance of water treatment is easily affected by seasonal fluctuations. For example, in winter, the viscosity of water is higher than in summer, so that large flocs are formed, and as a result, the pollutant removal function is improved. In the fuzzy control, since there is no parameter of the water temperature, it is difficult to cope with a minute fluctuation of the coagulant injection rate due to the difference of the water temperature.

The present embodiment provides a control method based on the fourth embodiment, characterized in that the actual value of the treated water turbidity in the data storage unit corresponding to this problem is evaluated.

If the error between the actual value and the target value of the treated water turbidity is 0.2 degrees or more for a certain period, the coagulant injection control parameters shown in FIG. 6 are adjusted. However, the treated water turbidity has an error of 0.
The parameters to be adjusted are determined according to the period exceeding two times. The parameters to be adjusted are as follows.

1) Clear weather, when raw water turbidity is stable → basic injection rate k
Adjustment of.

[0075] 2) the raw water turbidity increases during the → power n or adjustment of the multiplier k ₁ of the raw water turbidity.

3) At the time of low turbidity after the increase of raw water turbidity, at the time of decrease of conductivity or at the time of injection of alkali → adjustment of fuzzy membership function.

A simulation is performed with the coagulant injection rate after the parameter adjustment. If the error between the actual value and the target value of the treated water turbidity is within 0.2 degrees for a certain period, the parameter is used as it is as a parameter.

By storing the past data and calculating as described above, it is possible to know which parameter should be adjusted, and to always carry out the optimal coagulant injection in response to the fluctuation of the raw water turbidity depending on the change of the water temperature. It becomes possible.

In the coagulant injection control means according to the characteristics of the raw water, the coagulant injection control means when the raw water turbidity is stable, when the raw water turbidity increases, and when the raw water turbidity is low, is as follows. A description will be given of an embodiment.

(Seventh Embodiment) Method of Adjusting Coagulant Injection Control Parameter When Raw Water Turbidity is Stabilized A parameter adjusting method for coagulant injection control when raw water turbidity is stable will be described.

When the error between the actual value of the treated water turbidity and the target value is 0.2 degrees or more for a certain time, the basic injection rate k, which is a parameter of the coagulant injection control system in FIG. 6, is examined.

First, if the error is always positive, the k value is increased by one because the coagulant is insufficient. On the other hand,
If the error is always negative, then k
Decrement the value by one.

When the value of the basic injection rate k is increased or decreased by one, the treatment water turbidity prediction according to the fourth embodiment is performed, and the error between the actual value of the treatment water turbidity and the target value is constant. If it is within 0.2 degrees during the period, the k value is adopted as a parameter.

By such means, it becomes possible to determine the optimum coagulant injection rate when the raw water turbidity is stable, based on the prediction of the treated water turbidity.

(Eighth Embodiment) Parameter Adjustment Method of Coagulant Injection Control When Raw Water Turbidity Increases A parameter adjustment method for coagulant injection control when raw water turbidity increases will be described.

When the error between the actual value of the treated water turbidity and the target value is 0.2 or more for a certain time, the multiplier k ₁ of the power of the raw water turbidity, which is a parameter of the coagulant injection control system in FIG. 6, is examined. .

If the error is always positive, the coagulant is insufficient, and the k ₁ value is increased by 0.1. On the other hand, if the error is always negative, the k ₁ value is reduced by 0.1 because the coagulant is excessive.

A simulation is performed when the value of the multiplier k ₁ is increased or decreased by 0.1 in this way. If the error between the actual value of the treated water turbidity and the target value is within 0.2 degrees for a certain period, the k ₁ value Is adopted as a parameter.

When the angle does not fall within 0.2 degrees, the power exponent n which is another parameter in the system is examined. When the error between the actual value and the target value of the treated water turbidity is 0.2 degrees or more for a certain period, if the value is always positive, the n value is increased by 0.01 because the coagulant is insufficient. Conversely, if the error is always negative, the coagulant is in excess and the n value is reduced by 0.01.

As described above, a simulation is performed when the power exponent n is increased or decreased by 0.01, and if the difference between the actual value and the target value of the treated water turbidity is within 0.2 degrees of a certain period, the n value is not changed. Is adopted as a parameter.

By such means, it becomes possible to determine the optimum coagulant injection rate when the raw water turbidity increases based on the prediction of the treated water turbidity.

(Ninth Embodiment) Method of Adjusting Coagulant Injection Control Parameter at Low Turbidity After Raw Water Turbidity Increases A method of adjusting the coagulant injection control parameter at low turbidity after raw water turbidity increase will be described.

[0093] When the error between the actual value and the target value of the treated water turbidity is not less than 0.2 degrees for a predetermined time, the study of coagulant correction injection rate D ₄ which is a parameter of the coagulant injection control system of FIG. 6 Do.

If the error is always positive, since the coagulant is insufficient, the vertex of each output membership function in the water quality factor coagulant fuzzy inference is increased by 0.1.
On the other hand, if the error is always negative, the vertex of each output membership function is reduced by 0.1 because the coagulant is excessive.

As described above, a simulation is performed when the vertices of each output membership function are increased or decreased by 0.1, and if the difference between the actual value and the target value of the treated water turbidity is within 0.2 degrees for a certain period, the aggregation is performed. The agent correction injection rate is adopted as a parameter.

By such means, it is possible to determine the optimum coagulant injection rate at the time of low turbidity after the increase of raw water turbidity based on the prediction of treated water turbidity.

[0097]

As described above in detail, according to the method for controlling water quality according to the present invention, a neural network having a special judgment power of a skilled operator for raw water whose characteristics are liable to change with time.・ By introducing fuzzy,
The current system can be evaluated and learned from the past water quality and results, and the chemical injection rate can be automatically adjusted, and a stable water quality control function can always be obtained.

Various conditions can be input to the neural network fuzzy according to the present invention.
This makes it possible to predict and evaluate the performance of the water purification plant under various conditions, and presents a new method to the basic design of the water purification plant.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a neural network.

FIG. 2 is a block diagram showing a first embodiment.

FIG. 3 is a characteristic diagram showing a temporal change in treated water turbidity according to the first embodiment.

FIG. 4 is a schematic diagram of a water quality control system according to the present invention.

FIG. 5 is a characteristic diagram showing a change over time in turbidity of treated water according to a second embodiment.

FIG. 6 is a block diagram of controlling a coagulant injection rate according to the present invention.

FIG. 7 is a schematic diagram of a fuzzy inference system according to the present invention.

FIG. 8 is a membership function of a pre-chlorination agent injection rate.

FIG. 9 is a membership function of a pre-alkali agent injection rate.

FIG. 10 is a membership function of the coagulant correction injection rate.

FIG. 11 is a characteristic diagram showing a change in a coagulant injection rate according to the third embodiment.

FIG. 12 is a schematic diagram of a treatment water turbidity prediction system according to the present invention.

FIG. 13 shows current issues of water purification equipment and countermeasures.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input layer 2 ... 1st intermediate layer 3 ... 2nd intermediate layer 4 ... Output layer 5 ... Teacher signal 21, 61 ... Water quality data storage part 22 ... Neural network 23 ・ ・ ・ Fuzzy control inference 62 ・ ・ ・ Manual input unit 63 ・ ・ ・ Fuzzy control inference unit 64 ・ ・ ・ Addition unit 71 ・ ・ ・ Pre-chlorine injection rate membership function 72 ・ ・ ・ Pre-caustic injection rate member Ship function 73 ・ ・ ・ Fuzzy inference unit 74 ・ ・ ・ Coagulant correction injection rate membership function 121 ・ ・ ・ Water quality data 122 ・ ・ ・ Data storage unit 123 ・ ・ ・ Coagulant injection ratio inference unit 124 ・ ・ ・ Treatment water Turbidity prediction unit

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Kazuji Ikeda 2-1-117 Osaki, Shinagawa-ku, Tokyo Inside the company Meidensha Co., Ltd. (72) Kaoru Hatano 2-1-17-1 Osaki, Shinagawa-ku, Tokyo Shareholder F-term (reference) in the company Meidensha 5H309 AA07 BB16 CC07 DD02 DD12 DD16 DD20 DD38 GG05 HH12 HH25 HH28 KK04

Claims (7)

[Claims] 1. A water quality data comprising a chemical injection rate, a raw water temperature, a raw water pH, a raw water alkalinity, a raw water turbidity, etc., is supplied to a neural network to predict a treated water quality, and a deviation between a target water quality and the predicted treated water quality. A water quality control method, wherein a calculation is performed by the neural network so that the deviation is determined within a predetermined range, and a processed water turbidity prediction signal is sent to an output. 2. The water quality control method according to claim 1, wherein a chemical injection rate of the water quality data is determined by a membership function, calculated by fuzzy inference, and supplied to the neural network. 3. The method according to claim 2, wherein the neural network is formed in a plant model, and when a deviation between the target water quality and the predicted water quality is out of a predetermined range, the plant model is corrected by re-learning the neural network. The water quality control method according to claim 1 or 2, wherein 4. The plant model is modified by comparing the difference between the target water quality and the predicted water quality before and after the modification, and when the difference after the modification is small, updating the plant model;
4. The water quality control method according to claim 3, wherein when the difference after the correction is large, the membership function is adjusted. 5. A water quality data storage unit storing water quality data including a chemical injection rate, raw water temperature, raw water pH, raw water alkalinity, raw water turbidity, and the like, and water quality data output from the water quality data storage unit. And a neural network for predicting a treated water quality from these data to calculate a turbidity deviation from a target water quality, and transmitting an appropriate treated water turbidity prediction signal to an output. 6. A fuzzy inference unit that determines a chemical injection rate stored in the water quality data storage unit by a membership function, performs fuzzy inference operation, and supplies an appropriate chemical injection rate obtained as an output to the neural network. The water quality control device according to claim 5, further comprising: 7. The water quality data accumulating section is capable of storing and retrieving a chemical injection rate calculated by a fuzzy inference section, and a processing water turbidity prediction signal predicted from the water quality data and the chemical injection rate by the neural network. 7. The water quality control device according to claim 5, wherein storage and withdrawal of the water quality are also possible.