KR101209944B1 - A method to simulate which controls activated sludge treatment process using polynomial-based radial basis function neural networks - Google Patents

Abstract

PURPOSE: A simulating method controlling an activated sludge processing method by using polynomial base radiation type basis function neuron network structure is provided to learn the polynomial base radiation type basis function neuron network structure, thereby efficiently managing the activated sludge processing method. CONSTITUTION: Polynomial base radiation type basis function neuron network structure is learned by using learning data(S100). A model is constructed based on the learned polynomial base radiation type basis function neuron network structure(S200). Legal standard value achievement is evaluated and simulated by applying data of an activated sludge processing method evaluating the legal standard value achievement of discharge water to the model(S300). The polynomial base radiation type basis function neuron network structure selects main parameters which influence on the activated sludge processing method and is learned by extracting the data. [Reference numerals] (S100) Polynomial base radiation type basis function neuron network structure is learned; (S200) Model is constructed based on the learned polynomial base radiation type basis function neuron network structure; (S300) Legal standard value achievement is evaluated and simulated by applying data of an activated sludge processing method evaluating the legal standard value achievement of discharge water to the mode

Description

A METHOD TO SIMULATE WHICH CONTROLS ACTIVATED SLUDGE TREATMENT PROCESS USING POLYNOMIAL-BASED RADIAL BASIS FUNCTION NEURAL NETWORKS} Polynomial-Based Radial Basis Function Neural Network Structure

The present invention relates to a simulation method for controlling an activated sludge treatment method, and more particularly, to a simulation method for controlling an activated sludge treatment method using a polynomial-based radial basis function neural network structure.

Activated sludge treatment in sewage treatment plants is most commonly used for the treatment of sewage and wastewater. The activated sludge treatment method is ammonium (NH ₄ ), nitrate (NO ₃ ), dissolved oxygen (Dissolved Oxygen or DO), hydrogen ion concentration (Ph), oxidation-reduction potential (ORP) due to biological characteristics These factors can be attributed to the fact that multiple layers, such as temperature, temperature, and concentration of suspended solids (MIXed Liquer Suspended Solid, or MLSS) in the aeration tank, are operated in combination with each other. When controlling, it is difficult to determine each control target.

In order to control nitrogen and phosphorus, sewage treatment plants have recently been introduced with high class treatment. At this time, the blower consumes considerable power to operate the blower to remove nitrogen. Although it is designed to control the blower and the sludge return pump of the aerobic tank by using the dissolved oxygen meter and the redox potentiometer during the advanced processing operation, the dissolved oxygen value in actual operation is the temperature, air change, inflow of the aerobic tank, and discharge section. It is difficult to predict dissolved oxygen system or redox potential by various variables such as the difference of, and actual operation is very difficult because the correlation between nitrogen and phosphorus is not accurate.

Most domestic water reproduction centers (sewage treatment plants) are designed so that the blower is operated by the dissolved oxygen meter value and the sludge return pump is operated by the concentration of suspended solids (MLSS) value in the aeration tank. There is no place where the control method is implemented among these, it is necessary to introduce a new control method.

In the existing activated sludge treatment method, the automatic operation based on the concentration of suspended solids in the dissolved oxygen system and the aeration tank is not accurate, and manual operation is performed. However, with the recent strengthening of discharge water quality standards, it is more important to manage and optimize the quality and quantity of activated sludge in sewage treatment systems. The process requires a large number of water quality measurements due to the nature of the biological mechanisms, resulting in large load changes and complex characteristics of the process. For this reason, it is difficult to select a control amount and an operation amount, and as a result, it is difficult for the operator to calculate and operate a plurality of controllers. Therefore, there is a need for a method for more efficient and systematic qualitative control.

The present invention is proposed to solve the above problems of the conventionally proposed methods, and it is possible to perform more efficient and systematic qualitative control of the existing activated sludge treatment method, which is inefficient to operate and control due to the impossible prediction of the result of the method. It is an object of the present invention to provide a simulating method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure that can help the operator to control.

In addition, the present invention further facilitates the management and optimization of the quality and quantity of activated sludge in sewage treatment systems, which can help to meet the recently enhanced effluent water quality criteria, a polynomial based radial basis function neural network structure. Another object of the present invention is to provide a simulating method for controlling the activated sludge treatment method.

Simulation method for controlling the active sludge treatment method using a polynomial based radial basis function neural network structure according to the characteristics of the present invention for achieving the above object,

(1) learning a polynomial based radial basis function neural network structure using data for neural network structure learning;

(2) constructing a model based on the polynomial based radial basis function neural network structure learned in step (1); And

(3) evaluating and simulating whether the legal standard is achieved by applying the data of the activated sludge treatment method, which should be evaluated for the achievement of the legal standard of discharged water quality, to the model established in step (2). It is characterized by the configuration.

Preferably, the step (1)

Selecting major factors affecting the activated sludge treatment method;

Extracting data using the selected main factors; And

Learning the polynomial based radial basis function neural network structure using the extracted data.

Preferably, the step (1)

By using the error backpropagation algorithm, the polynomial connection weight w = f (x) of the neural network structure can be learned to learn the polynomial-based radial basis function neural network structure.

Preferably, the step (1)

Using particle group optimization, the individual vector, which is the individual behavior of each individual, is limited to 10-20% of the total population size, and the maximum and minimum inertia loads that control the individual vector as the number of generations increase to 0.9 and 0.4, respectively. By setting, the polynomial based radial basis function neural network structure can be learned.

Preferably, the step (3)

Changing the data of the activated sludge treatment method, which should evaluate whether or not the legal standard is achieved, to conform to the model established in the step (2); And

The modified data may be applied to the model constructed in the step (2) to evaluate and simulate whether the legal standard is achieved.

According to the simulation method of controlling the activated sludge treatment method using the polynomial-based radial basis function neural network structure proposed by the present invention, the existing activated sludge treatment method is inefficient to operate and control because the result of the method cannot be predicted. It can help the operator to control more efficient and systematic qualitative control.

In addition, according to the present invention, it is possible to more easily manage and optimize the quality and quantity of activated sludge in the sewage treatment system, which can help to meet the recently strengthened discharge water quality standards.

1 is a flow chart illustrating a simulating method for controlling an active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention.
Figure 2 is a flow chart showing the detailed configuration of the step S100 in the simulating method for controlling the active sludge processing method using a polynomial based radial basis function neural network structure in accordance with an embodiment of the present invention.
Figure 3 is a view showing the configuration of the active sludge processing method, used in the simulation method for controlling the active sludge processing method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention.
4 is a diagram showing a basic neural network structure used in a simulation method for controlling an active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention.
5 illustrates a polynomial based radial basis function neural network structure as a functional module, which is used in a simulation method for controlling an active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention. Figure shown.
FIG. 6 illustrates an error backpropagation algorithm used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. FIG.
7 is a view for explaining the concept of particle population optimization, used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention.
8 is a graph evaluating the performance of the models constructed in step S200, used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. Figure shown.
9 is a flow chart showing the detailed configuration of the step S300 in the simulating method for controlling the active sludge processing method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention.
10 is a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, evaluating and simulating whether or not to achieve a legal standard value of data of an activated sludge treatment method. Figure showing an example to do.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings, in order that those skilled in the art can easily carry out the present invention. In the following detailed description of the preferred embodiments of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. In the drawings, like reference numerals are used throughout the drawings.

In addition, in the entire specification, when a part is referred to as being 'connected' to another part, it may be referred to as 'indirectly connected' not only with 'directly connected' . In addition, the term 'comprising' of an element means that the element may further include other elements, not to exclude other elements unless specifically stated otherwise.

1 is a flowchart illustrating a simulating method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. As shown in FIG. 1, the simulating method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure includes: learning a polynomial-based radial basis function neural network structure (S100), learned polynomial Constructing a model based on the base radial basis function neural network structure (S200), and evaluating and simulating whether or not to achieve the legal standard value of the data of the activated sludge treatment method (S300).

In step S100, the polynomial based radial basis function neural network structure is learned. At this time, the data for neural network structure learning can be extracted by selecting and using the main factors in the active sludge treatment method. In the simulating method for controlling an active sludge processing method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention, a polynomial based radial basis function neural network structure is constructed using data for neural network structure learning. Detailed configuration of the learning method (S100) will be described in detail later with reference to FIG. 2.

In step S200, a model is constructed based on the polynomial based radial basis function neural network structure learned in step S100. In the model established in this step, the data of the activated sludge treatment method to be evaluated in step S300 to evaluate whether or not the legal standard of discharged water quality can be input.

In step S300, it is evaluated and simulated whether the legal standard value of the data of the activated sludge treatment method is achieved. In this step, by applying the data of the activated sludge treatment method to the model established in step S200, it is evaluated whether the legal standard value is achieved. Furthermore, by changing and simulating the input data, the user can predict the result based on the changed data, and the user can find suitable conditions according to the predicted result. In a simulating method for controlling an activated sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, a method for evaluating and simulating whether or not to achieve a legal standard value of data of an activated sludge treatment method ( A detailed configuration of S300 will be described later with reference to FIG. 9.

Figure 2 is a flow chart showing the detailed configuration of the step S100 in the simulating method for controlling the active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention. As shown in FIG. 2, in the simulating method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure, step S100 includes selecting main factors (S110) and extracting data. (S120), and learning the polynomial-based radial basis function neural network structure using the data (S130).

In step S110, main factors are selected. That is, in this step, the main factors affecting the activated sludge treatment method are selected, and the selected factors are used to extract data for learning the polynomial based radial basis function neural network structure. In this case, the main factors selected may include the influence factor and the manipulation factor. The influencing factor may change the purification rate of the activated sludge in the activated sludge treatment method. The operation factor may set conditions of the activated sludge treatment method.

Based on the polynomial-based radial basis function neural network structure according to an embodiment of the present invention, the criteria based on the selection of influence factors among the main factors used in the simulation method for controlling the active sludge treatment method are shown in Table 1 below. Shown in Table 1 is a standard applied on January 1, 2011 as the discharge water quality standard of public sewage treatment facilities. As can be seen in Table 1, among the biochemical oxygen demand, chemical oxygen demand, suspended solids, total nitrogen, total phosphorus, total E. coli group, and ecotoxicity factors, depending on the embodiment biochemical oxygen demand, suspended Only material, total nitrogen, and total phosphorus factors can be selected. In particular, in the case of sewage treatment system, the process characteristics are complicated and the correlation between nitrogen and phosphorus is not accurate. According to the embodiment, the total nitrogen and total phosphorus are subdivided, and the total nitrogen is NH _3. And by the concentration of NO ₃ , the total phosphorus can be divided by the concentration of PO ₄ .

Manipulation factors may be selected in addition to the selected influence factors. Manipulation factors include Temp, the influent temperature, Airflow, which is the amount of air blown from Existing Tank 1, and Existing Tank 2, Internal Recycle, which is the amount of water that transfers the nitrate nitrogen produced in Existing Tank 1 and Existing Tank 2 to anoxic tank, and a portion of the float is the initial part of the process. Sludge Return, which is a circulating part, and Was, which is a waste volume of the suspended matter. The detailed configuration of the manipulation factors will be described in detail later with reference to FIG. 3.

In step S120, data is extracted. That is, in this step, data is extracted using the main factors selected in step S110. In addition to the numerical values of the influence factors and the manipulation factors, the influence on the activity sludge of the influence factors according to the change of the manipulation factors may be included as data. The extracted data can learn neural network structure.

In step S130, the polynomial-based radial basis function neural network structure is trained using the data extracted in step S120. That is, in this step, the polynomial-based radial basis function neural network structure is trained using the data extracted in step S120 and the algorithm proposed by the present invention. In this case, the selected algorithm proposed by the present invention may include an error back propagation algorithm and a particle group optimization algorithm. The neural network structure learned in this step may be used to build a model later in step S200.

3 is a view showing the configuration of an active sludge treatment method, which is used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. As shown, the configuration of the active sludge treatment method used in the simulating method for controlling the active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, anaerobic tank (100), Anaerobic tank 101, aerobic tank 1 (102), aerobic tank 2 (103), bell needle 104, Temp (200), Airflow (201), Internal Recycle (202), Sludge Return (203), and Was (204). It can be configured to include. In the present invention, it is possible to simulate the advanced treatment process of the sewage treatment plant, the advanced treatment process may be composed of residual suspended matter and dissolved inorganic material removal process, nitrification process, nitrogen removal process, phosphorus removal process, and nitrogen and phosphorus removal process. have. The advanced treatment process reduces the probability of eutrophication and red tide phenomena due to the decrease of dissolved oxygen in the stream and the depletion of dissolved oxygen, which can occur when sewage with high biochemical oxygen demand and nitrogen loading is drained into the stream. give. Furthermore, the advanced treatment process removes about 90% of biochemical oxygen demand and suspended solids, and removes nitrogen and phosphorus leaving only the amount of nutrients necessary for microorganisms, resulting in 10-30% nitrogen and 10-30% phosphorus. Only remove about%. Most domestic sewage treatment plants design and operate only the activated sludge process, which is a secondary treatment process. However, an advanced treatment plant, a third treatment process, may be installed to treat nitrogen and phosphorus. The basic advanced treatment process combines chemical and biological treatments. However, the chemical treatment method is dependent on the increase of the injected chemicals, and if the treatment method is maintained continuously, the treatment cost of the activated sludge is also greatly increased. Therefore, the present invention has been conducted to optimize the operation of the biological treatment method, excluding the chemical treatment method.

The anaerobic tank 100, as a phosphorus discharge tank, has no dissolved oxygen and increases the amount of organic phosphorus released. The influent to be treated with activated sludge may be introduced into the anaerobic tank 100 first to release phosphorus in the anaerobic tank 100.

The oxygen-free tank 101 is a tank without dissolved oxygen, like an anaerobic tank, and reduces the nitrate nitrogen introduced into and returned from the aerobic tank to nitrogen gas to be released into the atmosphere to remove nitrogen. The microorganisms of the anoxic tank 101 are also called denitrification tanks in terms of their functionality because they treat nitrate nitrogen with nitrogen gas using nitrate nitrogen instead of dissolved oxygen when decomposing organic matter.

The aerobic tank 1 (102) and the aerobic tank (2) 103 maintain an aerobic state in order to remove untreated organic substances in the anaerobic tank 101 and the anaerobic tank 100 and to oxidize ammonia nitrogen to nitrate nitrogen. Since the microorganism nitrifying from ammonia nitrogen to nitrate nitrogen is an aerobic microorganism, the microorganism is maintained in an aerobic state, and the nitric acid nitrogen mixture is returned to the oxygen-free tank 101 to cause denitrification. Aerobic microbes also consume phosphorus. The aerobic tank 1 102 may oxidize organic materials as much as possible. The aerobic tank 2 103 may achieve nitrification of the organic material oxidized as much as possible in the aerobic tank 1 102.

The seed needle 104 is a place for submerging the suspended matter in the water, and has a structure which can discharge only the purified water except the suspended matter. The effluent from the needle 104 is purified water free of suspended solids.

Temp 200 is an operation factor indicating the temperature of the influent water. Water temperature can affect the amount of dissolved oxygen in the water, and thus can have a great effect on the production of microorganisms. Temp 200 may be manipulated before influent enters anaerobic tank 100.

Airflow 201 is an operation factor which shows the amount of air blown by the aerobic tank 1 102 and the aerobic tank 2 103. Aerobic microorganisms can act as oxidants when oxidizing ammonia nitrogen to nitrate nitrogen while removing organic matter. Airflow 201 may be directly manipulated and applied to aerobic tank 1 (102) and aerobic tank 2 (103).

The internal recycle 202 is an operation factor indicating the amount of water to be transferred to the anoxic tank 101 to convert the nitrate nitrogen produced in the aerobic tank 1 102 and the aerobic tank 2 103 into nitrogen gas. The Internal Recycle 202 can be manipulated when water is moved from the aerobic tank 2 103 to the anaerobic tank 101.

Sludge Return 203 is an operation factor that indicates the portion where the constant portion of the submerged float of the needle 104 is purified to the initial portion of the process. The reason for manipulating the sludge return 203 is that not only the suspended matter but also the microorganisms necessary for biological treatment are concentrated in the needle 104. The sludge return 203 may be manipulated when a portion of the float is purified from the needle 104 to the anaerobic tank 100.

Was 204 is an operation factor that indicates the amount of waste of the submerged suspended matter in the seed needle 104. Since the float 104 continues to sink in the needle 104, the float needs to be removed. When removing suspended solids in this way, the amount of microorganisms should be considered and disposed of. Was 204 may be manipulated when the float is disposed of at needle 104.

Temp (200), Airflow (201), Internal Recycle (202), Sludge Return (203), and Was (204) factors directly affect the production and maintenance of microorganisms in biological treatment methods. Thus, the above factors can be selected as operational factors.

4 is a diagram illustrating a basic neural network structure used in a simulation method for controlling an active sludge processing method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. As described above, the basic neural network structure, which is used in the simulation method for controlling the active sludge processing method using the polynomial-based radial basis function neural network structure according to an embodiment of the present invention, includes an input layer 300 and a hidden layer. 301, and an output layer 302.

The present invention utilizes a polynomial based radial basis function neural network structure based on a fuzzy inference mechanism. The polynomial-based radial basis function neural network structure can use a Gaussian function based on the maximum and minimum values of the data as an active function. In addition, by using the connection weights composed of polynomial functions, it is possible to improve the typical characteristics of the existing neural network. The polynomial based radial basis function neural network structure can be represented by fuzzy rules of "If-then" in terms of linguistic interpretation, and driven by fuzzy inference mechanisms. That is, the network structure can be formed by dividing into three functional modules: a conditional part, a conclusion part, and an inference part. The input space may be divided using the maximum value and the minimum value of the conditional data, and the conclusion portion may express the divided local region as a polynomial function. Finally, the final output of the network can be assumed by the fuzzy reasoning of the inference part. Many studies have suggested several pattern recognition models based on neural networks, such as multilayer perceptrons, polynomial neural networks, and radial basis function neural networks. The basic neural network structure may be composed of an input layer 300, a hidden layer 301, and an output layer 302. 4 shows the basic structure of a neural network having one hidden layer 301. In FIG. 4, n represents the number of dimensions of the input space, and s represents the number of dimensions of the output space.

The input layer 300 receives an input vector x = [x ₁ , ..., x _n ] ^T as an input and transfers it to the hidden layer 301.

The silvering layer 301 determines the output value of the neuron by using a sigmoid function, a radial basis function, or the like as an active function. When there is one hidden layer 301 as shown in FIG. 4, the connection weight w may exist between the hidden layer 301 and the output layer 302, and the total number of hidden layers 301 and the number of output layers 302 may be determined. It can be equal to the number multiplied and can consist of a constant.

The output layer 302 determines the final output of the network using the sum of the product of the neuron's output and the connection weights at each hidden layer 301.

The polynomial based radial basis function neural network structure used in the present invention is based on a general neural network structure composed of an input layer 300 and an output of one hidden layer 301. However, for the nonlinear characteristic determination boundary, the polynomial link weight w = f (x) is used instead of the constant term link weight, and the link weight can be learned using an error backpropagation algorithm. The detailed configuration of the error backpropagation algorithm will be described in detail later with reference to FIG. 6.

5 illustrates a polynomial based radial basis function neural network structure as a functional module, used in a simulation method for controlling an active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention. 5, a polynomial as a functional module, which is used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. The underlying radial basis function neural network structure may comprise a conditional unit 400, a conclusion unit 401, and an inference unit 402.

Polynomial-based radial basis function neural network structures can be interpreted from a linguistic point of view, such as fuzzy rule expressions, and can be expressed as: If x is R _i (x) then f _ji (x). In the above expression, x is the input vector, R _i (x) is the goodness of fit generated through the radial basis at i (= 1, ..., c) by the hidden layer 301, f _ji (x) is j (= 1). , ..., represent the polynomials of the i th fuzzy rule for the s) th output.

The conditional part 400 is included before "then" in the above expression. The conditional unit 400 may perform a function of an active function in terms of network structure by using a Gaussian function that is a radial basis function. The activation function may be represented by Equation 1 below.

In Equation 1, R _i (x) is a goodness-of-fit generated through the radial basis at i (= 1, ..., c) by the hidden layer 301 as mentioned above, and represents the output value of the active function of RBFNNs. Where x represents the value of the input data (input vector), v represents the center value of the Gaussian function, and s represents the standard deviation of each input data.

The conclusion section 401 is included after "then" in the above expression. The polynomial of the conclusion unit 401 acts as a local model of fuzzy rules as network connection weights. The equation used for one output may be represented by Equation 2 below.

In Equation 2, f _i (x) represents the polynomial of the i th fuzzy rule as mentioned above.

The inference unit 402 obtains the final output of the network as the inference result of the fuzzy rule.

The polynomial based radial basis function neural network structure has a network structure based on fuzzy rules and can be divided into three functional modules such as the conditional unit 400, the conclusion unit 401, and the inference unit 402. .

FIG. 6 is a diagram illustrating an error backpropagation algorithm used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. As described above, the error backpropagation algorithm, which is used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, includes a forward propagation 500 and a backward direction. It may be configured to include a radio wave 501.

The error backpropagation algorithm is a process of deriving the relationship between a given input value and a target value as its own adaptive algorithm. In the forward propagation 500, the nodes of the hidden layer (301) (R _i (x)) value the connection weights, this value is transmitted by the (f _ji (x)), the sum of the value passed to the inference unit (402) The output value is calculated, and the error value can be measured by comparing with the original output. The backward wave propagation 501 may correct the connection weight value by reflecting the magnitude of the error value.

The amount of change in ω generated based on the error value of the original output and the model output for the connection weight value f (x) = ω may be expressed by Equation 3 below.

In Equation 3, ω represents f of Equation 2, y represents actual output data, y _out represents model output data, and α represents a learning rate, respectively.

The error backpropagation algorithm repeats the forward propagation 500 and the backward propagation 501 and can learn to minimize the error value.

7 is a view for explaining the concept of particle cluster optimization, which is used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. Optimization is not based on the evolutionary mechanisms of natural selection, but on the social behaviors of biomass such as swarms and swarms of fish. Particle population optimization is characterized by simplicity of theory, ease of implementation, and efficiency of computation. It can generate optimal solution in a short computation time and shows more stable convergence than other probabilistic methods. Particle cluster optimization is a cluster-based algorithm, and may be represented by the entity 600 and the cluster 601 as shown in FIG. 7.

The entity 600 can fly through a multidimensional space and move to the optimal location using information about their own and neighbor's experiences. To this end, the entity 600 may maintain a memory of the optimal location information previously experienced, and may adjust a parameter to balance the global search and the local search ability of the entity 600.

The group 601 means a set of individual entities 600, and the entities 600 belonging to the group 601 may operate individually under the same system. The entities 600 having the optimal position information among the entities 600 belonging to the group 601 are called Particle Best, and the entities 600 representing the most optimal position information among the entities belonging to the Particle Best are called Global Best. .

Since each object 600 operates individually, it can have a direction and a moving speed as a vector, and this feature can be called Particle Velocity. Particle Velocity can be represented by the following equation (4).

In Equation 4, v _jk (t) is a vector representing the movement speed and direction of each particle, and the product of the previous v _jk value and the inertia load w (t) which can be calculated by Equation 5 to be described later, The new values are generated by randomly reflecting the records of the optimum values. Here, pbest represents the particle best, gbest represents the global best, r ₁ and r ₂ are variables that determine the random ratio, and random change values from 0 to ₁ , and c ₁ and c ₂ represent random settings through research. As a value, most of the embodiments can be fixed to 2.0.

If the particle velocity is large, each object 600 may move at a high speed, thus passing the optimal solution. On the other hand, if the value of the particle velocity is too small, it may take a long time to find the optimal solution. Therefore, the particle velocity can be limited to 10-20% of the size of the entire population (601). Furthermore, the inertial load can adjust the particle velocity as the number of generations increases. The inertial load can be expressed by the following Equation 5.

W (t) in equation (5) represents w (t) in equation (4), and the maximum value w _max of the inertial load can be set to 0.9 and the minimum value w _min to 0.4.

8 is a graph evaluating the performance of the models constructed in step S200, used in a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention. Figure is shown. As shown in Figure 8, the performance of the models built in step S200, used in the simulation method for controlling the active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention Was evaluated by dividing the performance evaluation 700 of the primary sewage treatment system model and the performance evaluation 701 of the secondary sewage treatment system model.

The performance evaluation 700 of the primary sewage treatment system model is based on the performance of the model constructed on the basis of the neural network structure trained with data that can be extracted when the amount of water flowing into the activated sludge process of the sewage treatment plant and the water quality are fixed. Evaluated. The performance evaluation of the secondary sewage treatment system model (701) evaluates the performance of the model constructed based on the neural network structure trained with data that can be extracted when the amount of water flowing into the activated sludge process of the sewage treatment plant and the water quality change. It was. As can be seen in FIG. 8, each model can be evaluated for performance through two graphs. The graphs located on the left side of each model can represent the performance of the model constructed based on the neural network structure trained using Training Data, and the graphs on the right side show Test Data. We can evaluate the performance of each model using. Each graph can contain a total of two lines, one line representing the original output, the original output value before using the model, and the other line representing the model output. It can represent the output value from the model. The graphs can help the user evaluate the performance of the model by making it easier to compare the Model Output line to the Original Output line. Two lines representing different output values can be distinguished by different colors.

9 is a flow chart showing the detailed configuration of step S300 in the simulating method for controlling the active sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention. As shown in Figure 9, in the simulating method for controlling the activated sludge treatment method using a polynomial based radial basis function neural network structure according to an embodiment of the present invention, step S300, the data of the active sludge treatment method It may include a step (S310) to suitably change the model, and the step (S320) to evaluate and simulate whether the legal threshold is achieved by applying the changed data to the model.

In step S310, the data of the activated sludge treatment method for evaluating whether or not the legal standard value is achieved is changed to fit the model established in step S200. In this step, as in step S110, the main factors affecting the activated sludge treatment method can be selected, and after selecting the main factors, the data can be extracted using the selected main factors as in step S120. . The extracted data is data of an activated sludge treatment method to evaluate whether or not to achieve a legal standard value, and may be input to a model constructed in step S200.

In step S320, the data changed in step S310 is applied to the model constructed in step S200 to evaluate and simulate whether the legal standard is achieved. That is, in this step, the result obtained by inputting the data of the evaluation target activated sludge treatment method into the model constructed in step S200 is compared with the legal reference value to evaluate whether the reference value is achieved. The results evaluated in this step can be presented to the user visually. Furthermore, by allowing the user to change and simulate input data, the user can predict the result based on the changed data, and the user can find suitable conditions according to the predicted result. An example of evaluating and simulating whether the legal threshold is achieved by applying the data changed in step S310 to the model constructed in step S200 will be described in detail later with reference to FIG. 10.

10 is a simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, evaluating and simulating whether or not to achieve a legal standard value of data of an activated sludge treatment method. It is a figure which shows an example to do. As shown in Figure 10, in the simulating method for controlling the activated sludge treatment method using a polynomial-based radial basis function neural network structure according to an embodiment of the present invention, whether or not to achieve the legal standard value of the data of the activated sludge treatment method The diagram illustrating an example of evaluating and simulating a system includes evaluating and simulating the legal threshold of the primary sewage treatment system 800 and evaluating and simulating the legal threshold of the secondary sewage treatment system 801. It is configured by.

Evaluating and simulating the legal baseline of the primary sewage treatment system (800) and evaluating and simulating the legal baseline of the secondary sewage treatment system (801) are the models of the primary sewage treatment system and the secondary sewage treatment, respectively. The model of the system is used to assess and simulate the achievement of legal standards. As can be seen in FIG. 10, the evaluation and simulation of the achievement of the legal standard of each sewage treatment system using each model may represent detailed values of the influent, the operation amount, and the effluent. Each model may visually show whether the input data has achieved the baseline, after analyzing the data with a predetermined legal baseline after inputting the data. The visual effect of achieving the baseline can be expressed in different colors. By visually evaluating whether the legal standard is achieved, the user can more easily perceive the current state of the activated sludge treatment method. In addition, by allowing data such as influent, manipulated quantity, effluent, and legal thresholds to be changed, the user can simulate the virtual conditions of the activated sludge treatment method, compare the legal thresholds and predict the results.

The present invention may be embodied in many other specific forms without departing from the spirit or essential characteristics of the invention.

100: anaerobic tank 101: anaerobic tank
102: aerobic tank 1 103: aerobic tank 2
104: hand 200: Temp
201: Airflow 202: Internal Recycle
203: Sludge Return 204: Was
300: input layer 301: hidden layer
302: output layer 400: conditional
401: Conclusion 402: Inference
500: forward wave 501: backward wave
600: object 601: clustering
700: Performance Evaluation of the Primary Sewage Treatment System Model
701: performance evaluation of secondary sewage treatment system model
800: Evaluate and simulate the achievement of legal thresholds for primary sewage treatment systems
801: Evaluate and simulate the achievement of legal thresholds for secondary sewage treatment systems
S100: Learning polynomial based radial basis function neural network structure
S110: selecting key factors
S120: Step of Extracting Data
S130: Learning polynomial based radial basis function neural network structure using data
S200: Constructing a model based on trained polynomial based radial basis function neural network structure
S300: step of evaluating and simulating whether or not to achieve the legal standard of the data of the activated sludge treatment method
S310: changing the data of the activated sludge treatment method to the model
S320: Evaluating and simulating whether the legal standard is achieved by applying the changed data to the model

Claims (5)

In the simulation method for controlling the active sludge treatment method using a polynomial based radial basis function neural network structure,
(1) learning a polynomial based radial basis function neural network structure using data for neural network structure learning;
(2) building a model based on the polynomial based radial basis function neural network structure learned in step (1); And
(3) evaluating and simulating whether or not the legal standard is achieved by applying the data of the activated sludge treatment method, which should be evaluated to achieve the legal standard of discharged water quality, to the model constructed in step (2). A simulation method for controlling an active sludge treatment method using a polynomial-based radial basis function neural network structure.
2. The method of claim 1, wherein the step (1)
Selecting major factors affecting the activated sludge treatment method;
Extracting data using the selected main factors; And
And training a polynomial-based radial basis function neural network structure using the extracted data.
2. The method of claim 1, wherein the step (1)
The polynomial based radial basis function neural network structure is characterized by learning a polynomial based radial basis function neural network structure by learning a polynomial connection weight w = f (x) of the neural network structure using an error backpropagation algorithm. Simulating method for controlling the activated sludge treatment method.
2. The method of claim 1, wherein the step (1)
Using particle group optimization, the individual vector, which is the individual behavior of each individual, is limited to 10-20% of the total population size, and the maximum and minimum inertia loads that control the individual vector as the number of generations increase to 0.9 and 0.4, respectively. And a polynomial based radial basis function neural network structure to control the activated sludge treatment method using the polynomial based radial basis function neural network structure.
2. The method of claim 1, wherein step (3)
Changing the data of the activated sludge treatment method, which should evaluate whether or not the legal standard is achieved, to conform to the model established in the step (2); And
Applying the changed data to the model constructed in step (2) to evaluate and simulate whether or not to achieve a legal standard value. The activated sludge treatment method using a polynomial based radial basis function neural network structure Simulating method to control.

Images