KR102097552B1 - Method for determination of the amount of a model input data for predicting membrane fouling in reverse osmosis process and device using the same - Google Patents

Abstract

One embodiment of the present invention comprises the steps of building a database of monitoring data collected from sensors of a reverse osmosis process; Constructing a pollution prediction model of a reverse osmosis membrane using the database; Calculating a prediction accuracy by changing the amount of input data according to a future prediction time point and a type of input variable using the reverse osmosis membrane contamination prediction model; And Determining the amount of input data when having the highest prediction accuracy of the prediction accuracy; Provides a method for determining the amount of sample input data of the reverse osmosis membrane contamination prediction model to predict the exact degree of membrane contamination reverse osmosis membrane cleaning and replacement timing Can contribute to the decision.

Description

Method for determination of the amount of a model input data for predicting membrane fouling in reverse osmosis process and device using the same}

The present invention relates to a method for determining the amount of sample input data of a reverse osmosis membrane contamination prediction model and an apparatus using the same, and more specifically, to use a machine learning-based reverse osmosis membrane pollution prediction model for predicting membrane contamination required for maintenance of the reverse osmosis process. In order to improve the accuracy of the reverse osmosis membrane contamination prediction model, the present invention relates to a method for determining the amount of sample input data and a reverse osmosis membrane pollution prediction apparatus using the same.

The seawater desalination technology is a technology that can solve the water shortage problem that has recently emerged, and refers to a technology for desalination of seawater that can secure a large amount of water resources regardless of seasons or weather conditions. The seawater desalination technology is a technology that seeks low energy, large-scale, stability, and eco-friendliness as it has a shorter construction period and a smaller initial investment cost compared to dam construction.

The seawater desalination plant collectively refers to processes and equipment that produce fresh water that can be used by humans, such as industrial water and drinking water, by evaporating seawater or passing through a membrane to remove salts as well as a large number of inorganic salts.

The desalination method can be largely divided into a distillation / evaporation method using thermal energy and evaporation of water, and a reverse osmosis (RO) method using differential and selective permeability of membranes. Among the main methods of seawater desalination plants, the reverse osmosis (RO) method, that is, the Seawater Reverse Osmosis (SWRO) method, is a conventional multi-stage flash distillation (MSF) or multiple-effect evaporation method. Energy efficiency is higher than Evaporation (ME), and commercial performance has recently increased. The reverse osmosis method is a method of separating fresh water by adding seawater to a reverse osmosis membrane through which water passes and solute does not permeate using pressure.

In the seawater desalination process using a reverse osmosis membrane, contamination of the membrane occurs as the reverse osmosis progresses. Membrane contamination means that substances present in the influent adhere to the membrane and degrade the performance of the separation membrane. In particular, in seawater desalination, substances present in the influent due to high pressure conditions (50 to 70 bar) are strongly accumulated in the membrane. When the membrane is contaminated, the efficiency of the entire process is drastically reduced, so controlling the membrane contamination is very important in the reverse osmosis process. When membrane contamination progresses, the membrane should be periodically cleaned, or the membrane should be replaced when cleaning cannot be resolved.

The problem is that it is necessary to determine the degree of membrane contamination and determine when to clean or replace the membrane, but there is a limit to monitoring or predicting the degree of membrane contamination in real time. The progress of membrane contamination can be indirectly determined by measuring the degree of decrease in pressure or the flow of influent or production water, but it is not determined the degree of contamination of the membrane itself. However, if the replacement time is decided, cleaning or replacement at an inappropriate time may occur, resulting in unnecessary maintenance costs. In order to solve this, an approach using an algorithm has been sought to determine the degree of membrane contamination.

In the prior art, a database is constructed by measuring the water quality of the incoming source water flowing into the reverse osmosis process, the operating conditions of the process, and the water quality of the production water. It provides a technology that predicts changes in permeation rate and film resistance, and predicts performance changes in the process.

However, because it uses a fixed amount of input data regardless of the future forecasting time and the type of input data, the best accuracy is not maintained, and in particular, the real-time changes in data generated in the prediction process, such as the occurrence of variables missing from the input data, can be accurately performed. There is a problem in that the prediction accuracy is lowered because it cannot be reflected.

Republic of Korea Registered Patent No. 10-1187416

The technical problem to be achieved by the present invention is to solve the above-mentioned problems, and improves prediction accuracy by reflecting a tendency to change data, and a predictive model of reverse osmosis membrane contamination that can maintain prediction accuracy above a certain level even if there are missing variables. Is to provide a way to determine the amount of sample input data.

The technical problems to be achieved by the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by those skilled in the art from the following description. There will be.

In order to achieve the above technical problem, an embodiment of the present invention comprises the steps of building a database of monitoring data collected from sensors of a reverse osmosis process; Constructing a pollution prediction model of a reverse osmosis membrane using the database; Calculating a prediction accuracy by changing the amount of input data according to a future prediction time point and a type of input variable using the prediction model; And determining the amount of input data when having the highest prediction accuracy of the prediction accuracy; provides a method for determining the amount of sample input data in a reverse osmosis membrane contamination prediction model.

In one embodiment of the present invention, the monitoring data collected from the sensors of the reverse osmosis process is influent temperature, influent TDS concentration, operating pressure, influent flow, production water flow, production water TDS concentration, production water pressure and concentrated water pressure It may be any one or more of.

In one embodiment of the present invention, there may be one or more missing monitoring data collected from the sensors of the reverse osmosis process.

In one embodiment of the present invention, the step of constructing a pollution prediction model of the reverse osmosis membrane using the database may be performed through machine learning.

In one embodiment of the present invention, the machine learning may be performed through a time delay neural network.

In one embodiment of the present invention, the time delay neural network may have a hidden layer from 1 to 3 layers.

In one embodiment of the present invention, the amount of input data may increase as the future prediction time is farther from the present.

In one embodiment of the present invention, the input data includes missing data, and as the number of missing data increases, the amount of input data may increase.

In one embodiment of the present invention, the prediction accuracy may be calculated according to the following equation (1).

(1)

(One)

는 t에서 모의값,

는 t에서 관측값,

은 관측값의 평균값임)(here,

Is a simulated value at t,

Is the observation at t,

Is the mean of the observations)

In one embodiment of the present invention, in the step of determining the amount of input data when the highest prediction accuracy is obtained, the amount of data may be determined in units of days.

In one embodiment of the present invention, providing the determined amount of data in real time to reverse osmosis membrane contamination prediction; may further include a.

In one embodiment of the present invention, constructing a database of monitoring data collected from sensors of the reverse osmosis process; and constructing a prediction model of contamination of the reverse osmosis membrane using the database; It may further include; normalizing the input value between (normalizing).

In order to achieve the above technical problem, another embodiment of the present invention is a data collection unit for collecting influent and production water factors in a reverse osmosis process; A driving unit receiving the influent factor and the production water factor and driving a reverse osmosis membrane contamination prediction model; A calculating unit that calculates prediction accuracy by changing the amount of input data according to the future prediction time point and the type of input factor, and determines the amount of input data when having the best prediction accuracy; It provides a reverse osmosis membrane contamination prediction apparatus comprising a; feedback unit for transmitting the amount of data determined in the calculation unit in real time to the driving unit.

In another embodiment of the present invention, the reverse osmosis membrane contamination prediction model may use a time delay neural network.

In another embodiment of the present invention, when determining the amount of input data when the calculator has the highest prediction accuracy, the amount of data may be determined in units of days.

According to an embodiment of the present invention, when predicting the degree of membrane contamination of a process desired by a decision maker through machine learning during operation management of the reverse osmosis process, the amount of sample data required for predicting the degree of membrane contamination and improving prediction accuracy is proposed. Thus, it is possible to contribute to determining the timing of cleaning and replacing the reverse osmosis membrane by predicting the exact degree of membrane contamination.

It should be understood that the effects of the present invention are not limited to the above-described effects, and include all effects that can be deduced from the configuration of the invention described in the detailed description or claims of the present invention.

1 is a flowchart briefly showing the structure of an artificial neural network according to the prior art.
2 is a flowchart briefly showing the structure of a time delay neural network according to the present invention.
3 is a graph showing prediction accuracy according to an embodiment of the present invention.
4 is a graph showing prediction accuracy according to a comparative example of the present invention.

Hereinafter, the present invention will be described with reference to the accompanying drawings. However, the present invention may be implemented in various different forms, and thus is not limited to the embodiments described herein. In addition, in order to clearly describe the present invention in the drawings, parts irrelevant to the description are omitted, and like reference numerals are assigned to similar parts throughout the specification.

Throughout the specification, when a part is "connected (connected, contacted, coupled)" to another part, this is not only when it is "directly connected", but also "indirectly" with another member in between. "It also includes the case where it is. Also, when a part is said to “include” a certain component, this means that other components may be further provided instead of excluding the other component unless otherwise stated.

The terms used in this specification are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this specification, terms such as “include” or “have” are intended to indicate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, and that one or more other features are present. It should be understood that the existence or addition possibilities of fields or numbers, steps, operations, components, parts or combinations thereof are not excluded in advance.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Hereinafter, a method of determining the amount of sample input data of the reverse osmosis membrane contamination prediction model will be described.

The method for determining the amount of sample input data of the reverse osmosis membrane contamination prediction model according to the present invention comprises: building a database of monitoring data collected from sensors of a reverse osmosis process; Constructing a pollution prediction model of a reverse osmosis membrane using the database; Calculating a prediction accuracy by changing the amount of input data according to a future prediction time point and a type of input variable using the prediction model; And determining the amount of input data when having the highest prediction accuracy among the prediction accuracy.

The monitoring data collected from the sensors of the reverse osmosis process may be any one or more of influent temperature, influent TDS concentration, operating pressure, influent flow rate, production water flow rate, production water TDS concentration, production water pressure and concentrated water pressure.

One or more of the monitoring data collected from the sensors of the reverse osmosis process may be missing.

In a seawater desalination plant using reverse osmosis, the influent source water becomes seawater, and the production water becomes fresh water. By adjusting the temperature, pressure or flow rate of the influent water, the flow rate or water quality of the production water can be improved. Water quality is usually determined in units of Total Dissolved Solids (TDS).

The flow rate and water quality data of the production water according to the control of the temperature, pressure, or flow rate of the influent water may be accumulated over several days to several years, and this is built into a database and used in the present invention. The type of monitoring data is not limited to the above, and can be extended or added to the extent necessary for predicting membrane contamination.

The step of constructing a model for predicting contamination of a reverse osmosis membrane using the database may be performed through machine learning, and machine learning may be performed through a time delay neural network.

Machine learning is a field of artificial intelligence that refers to the field of developing algorithms and technologies that enable computers to learn. It gives computers the ability to learn without explicit programming, and artificial neural network models are an area of machine learning.

1 is a flowchart schematically showing the structure of an artificial neural network used in the prior art. Artificial neural networks are algorithms that mimic the way the human brain recognizes patterns. The circle seen in FIG. 1 simulates neurons constituting a human neural network, and is called a node. Nodes are connected by synapses to form a network. A plurality of nodes are arranged to form a layer, and the layer is formed of an input layer, a hidden layer, and an output layer.

Nodes actually perform computations, which are designed to simulate the processes in neurons that make up the human neural network. When a node receives a stimulus of a certain size or more, it responds, and the magnitude of the reaction is approximately proportional to the value multiplied by the input value and the node's coefficient (or weight). In general, a node receives multiple inputs and has coefficients as many as the number of inputs. Therefore, different weights can be assigned to different inputs by adjusting this coefficient. Finally, all the multiplied values are added and the sum goes into the input of the active function. The result of the active function corresponds to the output of the node, and this output value is ultimately used for classification or regression analysis.

The purpose of artificial neural network learning is to minimize errors in output. Initialize all coefficients on the network before learning begins. And by sending data repeatedly, we learn. If learning is successful, the coefficients will be updated with appropriate values, and the artificial neural network can be used for various classifications and predictions. In the learning process, the coefficient update of the same principle occurs repeatedly. The principle of coefficient update is to first estimate the coefficient, measure the error that occurs when using the coefficient, and then update the coefficient slightly based on the error. First, when the input enters the neural network, the result is output using the coefficients of the current state. Then, the estimated value is compared with the actual correct answer. The difference between the correct answer and the estimated value is an error, and the neural network measures the error and corrects the coefficient to reflect this error. Repeating this process is a learning process.

In the prior art, an artificial neural network was trained to build a model, and the model was used to predict changes in inter-membrane pressure, water permeation, and membrane resistance, which are indicators of the degree of membrane contamination, and predicted the process performance change.

However, since the artificial neural network predicted the result while the size of the sample data was fixed, there is a problem that the accuracy of prediction decreases when the size of the sample data changes. It is a time-delayed neural network that performs self-learning with the function of considering the size of the sample input data added to the artificial neural network, and performs prediction considering the past data according to the size of the sample data.

2 is a flowchart schematically showing the structure of the time delay neural network of the present invention. Compared to artificial neural networks, time-delayed neural networks differ in the input layer. In the artificial neural network, one row of nodes is arranged on the input layer, and each node delivers only one data to the nodes of the hidden layer at the same time. However, the time delay neural network considers the case of missing input data and the data of t _n time and t _n-1 , t _n-2 ,… … Enter the data of the group together. In the case of artificial neural networks, if there is a missing data in the input data, the corresponding node has no signal to transmit to the next layer, but in the case of a time-delayed neural network, data from the reference time point to any past time point is bundled together, so that some of the input data is missing. However, the probability that a node of the input layer cannot transmit a signal to the next layer is reduced.

In a time-delayed neural network, data from a reference time point to any past time point are bundled together, and there is a limit in processing data that is continuously input over time. Therefore, in order to process data that is continuously input according to the passage of time, it is necessary to determine the time interval to be bundled together in advance and to bundle them. However, in order to maintain the continuity of continuous data, it is common to pre-process data to be overlapped at regular time intervals, rather than to group data divided according to time intervals. For example, t _n , t _n-1 , t _n-2 ,… … If there is data of t, and t _n to t _n-3 and t _n-4 to t _n-7 are divided by time intervals, the first data bundle is t _n to t _n-3 , the second data The bundle contains data from t _n-2 to t _n-5 , and the third bundle contains data from t _n-4 to t _n-7 . That is, three bundles of data must be transmitted during two time intervals from t _n to t _n-3 and t _n-4 to t _n-7 , and a time delay occurs.

In the present invention, the time-delayed neural network may have a hidden layer of 1 to 3 layers. The hidden layer of the neural network can have multiple layers, and each layer can learn different features. By combining these features, more complex data can be classified and clustered. In the present invention, the purpose of the reverse osmosis membrane performance prediction is to predict the performance of the reverse osmosis membrane in consideration of factors such as inter-membrane pressure difference, water permeation change, and membrane resistance, so that each layer can be characterized by placing multiple layers in the hidden layer, It is also possible to have one layer of learning. However, at least one hidden layer is necessary, and if there are more than three hidden layers, it is not preferable because learning takes too long and requires a lot of system resources.

In the present invention, the step of calculating the prediction accuracy by changing the amount of input data according to the future prediction time point and the type of the input variable using the prediction model is how far from the current time point to the future prediction time point or the type of the input variable. The prediction accuracy is calculated by inputting and processing the reverse osmosis membrane contamination prediction model constructed by varying the amount of input data according to how much. The amount of input data increases as the future prediction time increases from the present.

The steps of constructing the reverse osmosis membrane contamination prediction model or calculating the prediction accuracy through the time-delayed neural network are the same as inputting the data and obtaining the output value, but the steps of constructing the reverse osmosis membrane contamination prediction model are as follows. If the objective is to update the coefficients of each node while initializing and repeating data input and output, the step of calculating the prediction accuracy is aimed at determining the amount of input data that will represent the best accuracy to be described later.

The input data from the current time point to the past time point is input to the reverse osmosis membrane contamination prediction model by the time difference from the current time point to the future prediction time point, and the prediction accuracy according to the size of the input data is calculated. A time difference from a time point to a future prediction time point or a corresponding past time point, that is, it may be determined corresponding to a length of time. However, even if the length of time is doubled as described above, the data to be actually bundled and pre-processed by overlapping a certain portion is pre-processed, so the size of the actual input data will be twice or more, and is not directly proportional to the length of time.

The input data may include missing data, and as the number of missing data increases, the amount of input data increases. When the input data increases, the missing value can be supplemented by referring to the past data as described above.

The prediction accuracy can be calculated according to the following equation (1).

(1)

(One)

는 t에서 모의값,

는 t에서 관측값,

은 관측값의 평균값임)(here,

Is a simulated value at t,

Is the observation at t,

Is the mean of the observations)

In the above equation, the simulated value is a value calculated by the reverse osmosis membrane contamination prediction model, and the observed value is an actual value based on monitoring data. The closer the NSE value is to 1, the higher the prediction accuracy, and the farther from 1 the lower the prediction accuracy.

The amount of input data when having the highest prediction accuracy among the prediction accuracy can be finally determined, and then the data can be input in units of the determined amount to drive the reverse osmosis membrane prediction model. Preferably, the amount of data is determined in units of days. If it is determined by the time unit, it may take too much time to calculate the prediction accuracy and determine the amount of data, and if it is determined by the weekly unit over one day, the effectiveness in determining the cleaning and replacement cycle according to the membrane contamination prediction may be ineffective. This is not preferred.

The present invention may further include providing the determined amount of data in real time to reverse osmosis membrane contamination prediction. If the monitoring data is collected in real time from the reverse osmosis process, and the coefficient of the reverse osmosis membrane pollution prediction model is used in real time, the accuracy of the membrane pollution prediction is continuously maintained if the amount of data representing the highest accuracy is provided to the pollution prediction model in real time. Can be improved.

In addition, the present invention comprises the steps of constructing a database of monitoring data collected from sensors of the reverse osmosis process; and constructing a prediction model of contamination of the reverse osmosis membrane using the database; It may further include; normalizing the input value between (normalizing).

Normalizing is a procedure that transforms data according to certain rules and makes it easy to use. It can be said to be a procedure for unifying data standards. The monitoring data collected from the sensors of the reverse osmosis process has different ranges depending on the type, and when machine learning by inputting them into the time-delayed neural network as it is, it takes a long time to understand and learn the correlation between the data due to the too wide data range. It may take time. In addition, unnecessary load may be applied to the neural network through the input of redundant data. To solve this, all data can be converted to come within a certain range, or values that deviate significantly from the average value due to sensor errors can be excluded. If the values calculated through these processes are used for machine learning, it is possible to increase the construction efficiency of the reverse osmosis membrane prediction model.

Hereinafter, a reverse osmosis membrane contamination prediction apparatus will be described.

The present invention also provides a reverse osmosis membrane contamination prediction apparatus using the above method. The reverse osmosis membrane contamination prediction apparatus according to the present invention includes a data collection unit for collecting influent and production water factors in a reverse osmosis process; A driving unit receiving the influent factor and the production water factor and driving a reverse osmosis membrane contamination prediction model; A calculating unit that calculates prediction accuracy by changing the amount of input data according to the future prediction time point and the type of input factor, and determines the amount of input data when having the best prediction accuracy; And a feedback unit that delivers the amount of data determined by the calculation unit to the driving unit in real time.

The reverse osmosis membrane contamination prediction model uses a time-delayed neural network, and when determining the amount of input data when the operation unit has the highest prediction accuracy, the amount of data may be determined in units of days.

The data collected by the collection unit may be accumulated in units of days to years to be constructed as a database. The built-in database is connected to the driving unit to drive the reverse osmosis membrane contamination prediction model to update the coefficients, change the amount of data in the calculation unit, and calculate the prediction accuracy to determine the amount of data at the highest accuracy. The determined amount of data may be configured to be fed back to the driving unit and input to the reverse osmosis membrane contamination prediction model as much as the determined amount of data.

Hereinafter, the present invention will be described with reference to Examples.

Example 1

In the reverse osmosis membrane desalination process, influent temperature, influent TDS concentration, operating pressure, influent flow rate, production water flow rate, and production water TDS concentration data were monitored to database the data for 2 years. A time-delay neural network model was driven to input the values of the database at arbitrary constant time intervals, and the coefficient was corrected to construct a reverse osmosis membrane prediction model. The amount of data input to the constructed model at intervals of 2 days, 4 days, and 6 days is calculated and the result prediction accuracy after 1 day is shown in Table 1 below. The amount of data was determined by setting the time interval with the highest prediction accuracy.

Example 2

Except that the production water pressure and concentrated water pressure data was added, the same results as in Example 1 were calculated and the prediction accuracy of the results after 1 day is shown in Table 1 below. The amount of data was determined by setting the time interval with the highest prediction accuracy. Prediction accuracy over time for the determined amount of data is shown in FIG. 3.

Comparative Example 1

In the reverse osmosis membrane desalination process, influent temperature, influent TDS concentration, operating pressure, influent flow rate, production water flow rate, and production water TDS concentration data were monitored to database the data for 2 years. The artificial neural network model was driven to input the values from the database and correct the coefficients to construct a reverse osmosis membrane contamination prediction model. Table 1 below shows the accuracy of predicting the result after 1 day by inputting data into the constructed model.

Comparative Example 2

Except for adding the product water pressure and concentrated water pressure data, the same results as in Comparative Example 1 were calculated and the result prediction accuracy is shown in Table 1 below. The prediction accuracy over time is shown in FIG. 4.

division Number of input variables Time delay
(Time Delay) Train
(NSE) Test
(NSE) Comparative Example 1 6 0 0.93 0.68 Example 1 6 2 0.97 0.75 6 4 0.98 0.85 6 6 0.97 0.63 Comparative Example 2 8 0 0.98 0.53 Example 2 8 2 0.99 0.66 8 4 0.99 0.86 8 6 0.99 0.51

In Comparative Example 1 and Comparative Example 2, since the time delay does not occur using an artificial neural network, the time delay value is 0.

When predicting with 6 data, when the time delay was 4, the highest accuracy was 0.85.

When predicting with 8 data, when the time delay was 4, the highest accuracy was 0.86.

It can be seen that the accuracy of the embodiments is higher than that of Comparative Example 1 and Comparative Example 2 without time delay.

Example 3

The amount of data was determined in the same manner as in Example 1 except that the amount of data was input at intervals of 4 days, 6 days, and 8 days to calculate the result prediction accuracy after 7 days.

Example 4

The amount of data was determined in the same manner as in Example 2 except that the amount of data was input at intervals of 4 days, 6 days, and 8 days to calculate the result prediction accuracy after 7 days.

Comparative Example 3

It was performed in the same manner as in Comparative Example 1, except that the prediction accuracy of the results after 7 days was calculated.

Comparative Example 4

It was performed in the same manner as in Comparative Example 2, except that the prediction accuracy of the results after 7 days was calculated.

The prediction accuracy calculated in Examples 3 and 4 and Comparative Examples 3 and 4 is shown in Table 2 below.

division Number of input variables Time delay
(Time Delay) Train
(NSE) Test
(NSE) Comparative Example 3 6 0 0.92 0.40 Example 3 6 4 0.94 0.48 6 6 0.87 0.62 6 8 0.94 0.45 Comparative Example 4 8 0 0.90 0.45 Example 4 8 4 0.90 0.53 8 6 0.92 0.32 8 8 0.98 0.55

In Comparative Example 3 and Comparative Example 4, since the time delay does not occur using an artificial neural network, the time delay value is 0.

When predicting with 6 data, when the time delay is 6, the highest accuracy is 0.62.

When predicting with 8 data, when the time delay was 8, the highest accuracy was 0.55.

The above description of the present invention is for illustration only, and those skilled in the art to which the present invention pertains can understand that it can be easily modified to other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and similarly, components described as distributed may be implemented in a combined form.

The scope of the present invention is indicated by the following claims, and all changes or modifications derived from the meaning and scope of the claims and equivalent concepts should be interpreted to be included in the scope of the present invention.

Claims (15)

Constructing a database of monitoring data collected from sensors of the reverse osmosis process;
Constructing a reverse osmosis membrane contamination prediction model through time-delayed neural network machine learning using the database;
In order to prevent a signal from being transmitted from an input layer node to a next layer node due to data missing, to prevent prediction accuracy deterioration, the reverse osmosis membrane contamination prediction model is grouped with data from a reference point to any past point in time, and Calculating a prediction accuracy by changing and inputting an amount of input data according to the type; And
And determining the amount of input data when having the highest prediction accuracy among the prediction accuracy. 2. A method for determining the amount of sample input data in a reverse osmosis membrane contamination prediction model. According to claim 1,
The monitoring data collected from the sensors of the reverse osmosis process is a reverse osmosis membrane, characterized in that at least one of influent temperature, influent TDS concentration, operating pressure, influent flow rate, production water flow rate, production water TDS concentration, production water pressure and concentrated water pressure. How to determine the amount of sample input data in a pollution prediction model. According to claim 2,
Method for determining the amount of sample input data in a reverse osmosis membrane prediction model, characterized in that there is one or more missing monitoring data collected from the sensors of the reverse osmosis process. delete delete According to claim 1,
The time-delayed neural network method of determining the amount of sample input data of the reverse osmosis membrane prediction model, characterized in that the hidden layer is 1 to 3 layers. According to claim 1,
A method for determining the amount of sample input data in a reverse osmosis membrane prediction model, characterized in that the amount of input data increases as the future prediction time is farther from the present. According to claim 1,
The input data includes missing data, and the amount of input data increases as the number of missing data increases. A method for determining the amount of sample input data in a reverse osmosis membrane contamination prediction model. According to claim 1,
The prediction accuracy is a method for determining the amount of sample input data in a reverse osmosis membrane prediction model, characterized in that it is calculated according to the following equation (1).

(One)
(here,

Is a simulated value at t,

Is the observation at t,

Is the mean of the observations) According to claim 1,
A method for determining the amount of sample input data in a reverse osmosis membrane contamination prediction model, wherein the amount of data is determined in units of days in determining the amount of input data when the highest prediction accuracy is obtained. According to claim 1,
And providing the determined amount of data to the reverse osmosis membrane contamination prediction in real time; further comprising a method for determining the amount of sample input data of the reverse osmosis membrane contamination prediction model. According to claim 1,
Constructing a database of monitoring data collected from sensors of the reverse osmosis process; and constructing a reverse osmosis membrane contamination prediction model using the database; Normalizing the input value between (normalizing); Method for determining the amount of sample input data of the reverse osmosis membrane prediction model further comprising a. A data collection unit for collecting influent and production water factors in a reverse osmosis process;
A driver for receiving the influent factor and the production water factor and driving a reverse osmosis membrane contamination prediction model configured through time delay neural network machine learning using a database of monitoring data collected from sensors of a reverse osmosis process;
In order to prevent a signal from being transmitted from an input layer node to a next layer node due to data loss, and to prevent a decrease in prediction accuracy, data from a reference point to an arbitrary point in time are tied to a reverse osmosis membrane prediction model driven by the driving unit. , Computing unit for calculating the prediction accuracy by changing the input data amount according to the type of the input variable and determining the amount of input data when having the best prediction accuracy; And
And a feedback unit that delivers the amount of data determined by the operation unit in real time to the driving unit. delete The method of claim 13,
When determining the amount of input data when the calculation unit has the highest prediction accuracy, the reverse osmosis membrane contamination prediction apparatus, characterized in that the amount of data is determined in units of days.

Images