System and method for ai-based water treatment prediction and evaluation
An AI-based system analyzes water treatment data as multivariate time series with HRT to predict future water quality and operation status, addressing prediction accuracy and stability challenges, and supports smart water treatment operations.
Patent Information
- Application Number
- KR1020260125611
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-07-08
- Publication Date
- 2026-11-10
- Estimated Expiration
- 2046-07-08
AI Technical Summary
Existing water treatment systems struggle to accurately predict future water quality changes due to the influence of influent water quality, flow rate, and process operating conditions, and fail to reflect time-delay relationships effectively.
An artificial intelligence-based system that chronologically analyzes water treatment data as multivariate time series data, incorporating hydraulic retention time (HRT), to predict future water quality, flow rate, and process operation status, and evaluates prediction performance by comparing predicted values with actual values, with model retraining when necessary.
Enables accurate prediction of water treatment data at multiple future points, reflecting time-delay characteristics, maintaining prediction accuracy and stability in changing environments, and supports smart water treatment operations.
Smart Images

Figure 112026083247792-PAT00011_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a water treatment prediction and evaluation technology, and more specifically, to an artificial intelligence-based water treatment prediction and evaluation system and method that predicts water treatment data including at least one of water quality status, water level, flow rate, power consumption, or process operation status data corresponding to a plurality of future time points using an artificial intelligence prediction model, evaluates prediction performance by comparing the predicted value with the actual value, and retrains the artificial intelligence prediction model as necessary. Background Technology
[0003] In general, in water treatment processes such as sewage treatment facilities, wastewater treatment facilities, and water purification facilities, the treatment status and effluent water quality may vary depending on the influent water quality, influent flow rate, process operating conditions, and changes in the external environment. To manage this, water quality data such as COD, TOC, SS, TN, and TP, as well as process operation data such as flow rate, airflow rate, and equipment operating status, may be measured.
[0004] Conventionally, methods primarily used to determine process status based on current measurements, managerial experience, or established reference values. However, since water treatment conditions change continuously over time and multiple water quality, flow rate, and process operation variables influence each other, it is difficult to accurately predict future water quality changes based solely on measurements at a specific point in time.
[0005] Meanwhile, in water treatment processes, a Hydraulic Retention Time (HRT) is required for the influent to reach the downstream end of the process or the discharge point; therefore, the influent water quality, influent flow rate, and process operating conditions at a specific point in time can affect the effluent water quality or outflow flow rate after the HRT has elapsed. There is a limitation in that it is difficult to adequately reflect these time-delay relationships when analyzing based solely on data from the same point in time. The problem to be solved
[0007] Derived from this technical background, the present invention aims to provide an artificial intelligence-based water treatment prediction and evaluation system and method capable of predicting water quality, flow rate, power consumption, or process operation status at multiple future points in time by chronologically analyzing water treatment data, such as water quality data, flow rate data, power consumption, and process operation data, that change over time, and reflecting the time delay characteristics according to the retention time (HRT) of the water treatment process.
[0008] Furthermore, the present invention aims to provide a water treatment prediction and evaluation system and method capable of quantitatively evaluating prediction performance by residence time by comparing predicted water treatment data values with actual water treatment data values at a corresponding future point in time, and retraining an artificial intelligence prediction model when a decrease in prediction performance or a change in data patterns is detected. means of solving the problem
[0010] The present invention for achieving the above objectives includes the following configuration.
[0011] That is, an artificial intelligence-based water treatment prediction and evaluation method according to one embodiment of the present invention is performed on a computing device having one or more processors and a memory storing one or more programs executed by said one or more processors, the method comprising: collecting a plurality of water treatment data measured over time; sorting the collected plurality of water treatment data by time to generate multivariate time series data; setting past time series data included in a predetermined time interval prior to a past reference point among the generated multivariate time series data as input data; setting water treatment measurement data at a point in time after each of a plurality of residence times has elapsed from said past reference point as target data corresponding to each of said residence times; generating a plurality of training data sets by matching the set input data with the target data corresponding to each of said residence times; generating an artificial intelligence prediction model that learns the temporal change pattern of the input data and the correlation between the water treatment data using the plurality of training data sets; and inputting multivariate time series data included in a predetermined time interval prior to a current reference point into the artificial intelligence prediction model to generate water treatment prediction values corresponding to a plurality of future points in time after each of said residence times has elapsed from said current reference point. and after each of the plurality of future points in time has arrived, the method includes a step of evaluating the prediction performance of the artificial intelligence prediction model by comparing the predicted water treatment value with the actual water treatment value measured at the corresponding future point in time.
[0012] The above water treatment data may include at least one of water quality data, flow rate data, water level data, supply volume data, discharge volume data, power consumption data, process operation data, equipment status data, and external environment data. Effects of the invention
[0014] According to the present invention, water treatment data including at least one of water quality, water level, flow rate, power consumption, and process operation data is configured as multivariate time series data, and by learning by linking past time series data with water treatment data corresponding to multiple residence times, an effect can be derived in which future water treatment data can be predicted by reflecting both the time delay characteristics within the process and the water treatment change pattern according to changes over time.
[0015] In addition, by generating water treatment prediction values for multiple future time points corresponding to multiple residence times, it is possible to verify the trends of water treatment data changes over time more specifically compared to a single time point prediction, and the effect of enabling preemptive response to changes in water treatment data is derived.
[0016] In addition, by quantitatively evaluating prediction performance by retention time by comparing predicted values with actual water treatment values at corresponding future points in time, and retraining the AI prediction model when a decline in prediction performance or a change in data patterns is detected, prediction accuracy and model stability can be continuously maintained even in changing process environments.
[0017] In addition, the prediction and evaluation results of the present invention can be utilized as decision-making information for smart water treatment operations, and can be used as basic information for determining water quality stability, responding to changes in flow rate, analyzing equipment operating status, reviewing energy consumption reduction, managing carbon neutrality operation indicators, and evaluating water quality stability for water reuse. Brief explanation of the drawing
[0019] FIG. 1 is a block diagram illustrating the configuration of a water treatment prediction and evaluation system according to one embodiment of the present invention. Figure 2 is an example of a graph of the results of a multi-point prediction for multiple dwell times. Figure 3 is an example diagram showing a graph comparing actual values and predicted values according to an embodiment. Figure 4 is an example graph showing a comparative analysis based on time of stay. FIG. 5 is a flowchart illustrating an artificial intelligence-based water treatment prediction and evaluation method according to one embodiment of the present invention. Specific details for implementing the invention
[0020] Hereinafter, embodiments disclosed in this specification will be described in detail with reference to the attached drawings. Identical or similar components are assigned identical or similar reference numerals, and redundant descriptions thereof will be omitted. The suffixes "module" and "part" used for components in the following description are assigned or used interchangeably solely for the ease of drafting the specification and do not inherently possess distinct meanings or roles. Furthermore, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of related prior art could obscure the essence of the embodiments disclosed in this specification, such detailed description will be omitted. Additionally, the attached drawings are intended only to facilitate understanding of the embodiments disclosed in this specification; the technical concept disclosed in this specification is not limited by the attached drawings, and it should be understood that they include all modifications, equivalents, and substitutions that fall within the concept and technical scope of the embodiments.
[0021] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. These terms are used solely for the purpose of distinguishing one component from another.
[0022] When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between.
[0023] Singular expressions include plural expressions unless the context clearly indicates otherwise.
[0024] In this application, terms such as “comprising” or “having” are intended to specify the existence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.
[0025] A water treatment prediction and evaluation system according to one embodiment can predict the state of water treatment data at a future point in time by collecting and analyzing multivariate time series data generated based on water quality status, water level information, process operation information, inflow rate, power consumption, and changes in the external environment. It can be implemented to manage data generated throughout the water treatment process within an integrated structure by utilizing water treatment data change patterns and interrelationships between variables as training data.
[0026] In addition, a hydraulic retention time (HRT)-based prediction structure is applied to reflect the time-series variation characteristics of water treatment data and the time lag characteristics within the process, and changes in water treatment at future points in time can be analyzed through a model training process linked with time-series learning data.
[0027] Furthermore, as performance evaluation results and error analysis information are reflected in the retraining process, it can be implemented to continuously manage prediction performance even in changing process environments. In this case, the prediction results can be utilized as operational support information to analyze the process operating status and identify future changes in water treatment data in advance.
[0029] Hereinafter, preferred embodiments according to the present invention will be described in detail with reference to the attached drawings.
[0030] FIG. 1 is a block diagram illustrating the configuration of a water treatment prediction and evaluation system according to one embodiment of the present invention.
[0031] As illustrated in FIG. 1, a water treatment prediction and evaluation system (10) according to one embodiment specifically includes a communication interface (110), a memory (120), an input / output interface (130), and a processor (140).
[0032] The communication interface (110) can provide a function for the water treatment prediction and evaluation system (10) to communicate with other devices through a network. For example, the processor (140) of the water treatment prediction and evaluation system (10) can transmit requests, commands, data, files, etc., generated according to program code stored in a recording device such as memory (120) to at least one user terminal (30) or an integrated monitoring system (20) through a network under the control of the communication interface (110).
[0033] Conversely, signals, commands, data, files, etc. from other devices may be received via a network through the communication interface (110) of the water treatment prediction and evaluation system (10). Signals, commands, data, etc. received through the communication interface (110) may be transmitted to a processor (140) or memory (120), and files, etc. may be stored in a storage medium (the permanent storage device described above) that the water treatment prediction and evaluation system (10) may further include.
[0034] The network may include any one or more networks such as PAN (personal area network), LAN (local area network), CAN (campus area network), MAN (metropolitan area network), WAN (wide area network), BBN (broadband network), and the Internet. Additionally, the network may include any one or more network topologies such as bus networks, star networks, ring networks, mesh networks, star-bus networks, tree or hierarchical networks, but is not limited thereto.
[0035] The integrated monitoring system (20) may be a system that comprehensively collects and manages water quality data, water level data, flow rate data, power consumption, process operation data, and equipment status generated in a water treatment process. The integrated monitoring system (20) may be connected to communicate with at least one of a water treatment sensor, a flow meter, a blower, a pump, an agitator, a chemical injection device, a treatment tank, and other process equipment.
[0036] The integrated monitoring system (20) can receive field measurement data or equipment status data from at least one of a field control panel (RCS: Remote Control Station), a field control terminal unit (RTU: Remote Terminal Unit, 21), or a remote monitoring / control station (TM / TC: Tele-Metering / Tele-Control, 22). The integrated monitoring system (20) can operate in conjunction with the field control panel, the field control terminal unit (21), and the remote monitoring / control station (22).
[0037] The field control panel (RCS) and the field control terminal (21) are installed at the site of the water treatment process and can be interfaced with at least one of a water treatment analyzer, a water level gauge, a flow meter, a turbidity meter, a concentration meter, a pressure gauge, and process equipment. The remote monitoring and control station (TM / TC) (22) can be linked with the field control terminal (21), measuring instrument, or process equipment placed remotely to transmit remote field data to the integrated monitoring system (20).
[0038] The integrated monitoring system (20) receives a data request from the water treatment prediction and evaluation system (10) and can provide the requested water treatment data, process operation data, or facility status data to the water treatment prediction and evaluation system (10). In another embodiment, the integrated monitoring system (20) may automatically transmit data to the water treatment prediction and evaluation system (10) according to a preset cycle or data generation time.
[0039] The user terminal (30) can access the water treatment prediction and evaluation system (10) or the integrated monitoring system (20) via a wired network or a wireless network. The user terminal (30) can receive water treatment prediction and evaluation information using at least one of a web browser, a dedicated application, or a remote monitoring program.
[0040] The user terminal (30) may select or change at least one of the predicted water treatment item, inquiry period, residence time range, process or facility, performance evaluation indicator, and output format according to the user's input. If there is a possibility that the predicted water treatment value may exceed the set water treatment standard, the user terminal (30) may provide warning information in at least one of a pop-up, notification message, sound, vibration, email, or push notification. The user terminal (30) may provide at least one of a simple inquiry function, a setting change function, a data download function, a retraining request function, or an operating model selection function according to the user's authority.
[0041] Memory (120) is a computer-readable recording medium and may include RAM (random access memory), ROM (read only memory), and permanent mass storage devices such as disk drives and servers. Here, permanent mass storage devices such as ROM and disk drives may be included in the water treatment prediction and evaluation system (10) as separate permanent storage devices distinct from memory (120).
[0042] Additionally, an operating system and at least one program code may be stored in the memory (120). These software components may be loaded into the memory (120) from a computer-readable recording medium separate from the memory (120). This separate computer-readable recording medium may include a floppy drive, disk, tape, DVD / CD-ROM drive, memory card, SSD, USB, etc.
[0043] In another embodiment, software components may be loaded into memory (120) via a communication interface (110) rather than a computer-readable recording medium. For example, software components may be loaded into memory (120) of the water treatment prediction and evaluation system (10) based on a computer program installed by files received over a network.
[0044] The input / output interface (130) may be a means for interfacing with an input / output device. For example, the input device of the input / output device may include a device such as an operation button, keyboard, or mouse provided on the front of the device, and the output device may include a device such as a display or speaker. As another example, the input / output interface (130) may be a means for interfacing with a device in which the functions for input and output are integrated into one, such as a touchscreen. The input / output device may be physically configured as a single device with the water treatment prediction and evaluation system (10).
[0045] The processor (140) may be configured to process instructions of a computer program by performing basic arithmetic, logic, and input / output operations. Instructions may be provided to the processor (140) via memory (120) or a communication interface (110). For example, the processor (140) may be configured to execute instructions received according to program code stored in a recording device such as memory (120).
[0046] As illustrated in FIG. 1, the processor (140) of the water treatment prediction and evaluation system (10) according to one embodiment may more specifically include a data collection unit (1410), a time series data generation unit (1415), an input data setting unit (1420), a target data setting unit (1430), a training data generation unit (1440), a prediction model generation unit (1450), a water treatment prediction unit (1460), a performance evaluation unit (1470), a data verification unit (1480), and a model retraining unit (1490).
[0047] The data collection unit (1410) can collect water treatment data and external environment data, including at least one of water quality data, water level data, process operation data, flow rate data, and power consumption data that are generated or measured over time in a water treatment process. The data collection unit (1410) can collect data from at least one of a water treatment sensor, a flow meter, process equipment, an integrated monitoring system (20), an external data providing system, or an external linkage interface.
[0048] In one embodiment, the data collection unit (1410) may collect at least one of field measurement data, process operation data, and equipment status data through at least one of a field control terminal device (21) or a remote monitoring control station (22) that is data-linked with the integrated monitoring system (20). Field measurement data may include measurement values generated from a water quality analyzer, a water level gauge, a flow meter, a turbidity meter, a concentration meter, a pressure gauge, or other measuring instruments, and equipment status data may include data indicating the operating status of a pump, a blower, a stirrer, a valve, a chemical injection device, a power supply, power consumption, or other process equipment. The data collection unit (1410) may further receive control history data including one of a history of operating condition changes, control commands, manual control, automatic control, or remote control generated or performed by the integrated monitoring system (20), a central operation station, a field control panel (RCS), or a field control terminal device (21).
[0049] Water quality data may include at least one of total organic carbon (TOC), chemical oxygen demand (COD), suspended solids (SS), total nitrogen (TN), total phosphorus (TP), dissolved oxygen (DO), mixed liquid suspended solids concentration (MLSS), oxidation-reduction potential (ORP), nitrate nitrogen concentration (NO3), ammonia nitrogen concentration (NH4), and hydrogen ion concentration (pH). Process operation data may include at least one of inflow rate, effluent rate, internal circulation rate, airflow rate, stirring conditions, equipment operating status, water level in the treatment tank, treatment temperature, and process control value. Flow rate data may include at least one of inflow rate, effluent rate, discharge rate, supply rate, river flow rate, and flow rate change information by basin. Water level data may include at least one of the following: treatment tank water level, inflow channel water level, outflow channel water level, sedimentation tank water level, retention tank water level, reservoir water level, purified water reservoir water level, river water level, water reservoir water level, pipeline water level, tide level, and water level change information by water system. Equipment status data may include at least one of the following: pump status, blower status, agitator status, valve status, chemical injector, and power consumption information. External environment data may include at least one of the following: temperature, precipitation, humidity, seasonal information, weather information, and external water system information. External environment data may be collected from an administrative network, a public data linkage network, or an external data providing server.
[0050] The data collection unit (1410) can directly collect data generated in real time from water treatment sensors, flow meters, power usage, and process equipment, or collect data by querying past data stored in an integrated monitoring system (20), an integrated database, or an external system.
[0051] The data collection unit (1410) can collect metadata including at least one of a measurement time, data type, measurement location, sensor identification information, process identification information, and data source along with each data. The data collection unit (1410) can collect data generated at different intervals from different sensors or systems. For example, some water quality, water level, and flow rate data may be measured in minutes, process operation data may be measured in seconds or minutes, and external environment data may be collected in hours or days. The data collection unit (1410) can receive setting information regarding at least one of a data collection cycle, target water treatment item, target process, data collection period, and storage conditions from the integrated monitoring system (20) or user terminal (30).
[0052] The data collection unit (1410) can store the collected raw data in an integrated database. The raw data can be managed based on the measurement time so that it can be reused in subsequent preprocessing, learning, prediction, performance evaluation, and retraining processes.
[0053] The time series data generation unit (1415) can perform preprocessing on multiple water quality and water level data, power consumption data, multiple process operation data, flow rate data, and external environment data collected from the data collection unit (1410), and generate multivariate time series data by aligning the preprocessed data based on time.
[0054] The time series data generation unit (1415) can identify at least one of anomalies, missing values, abnormal zero values, null values, duplicate data, sensor errors, communication errors, abnormal sudden changes, and distortion values that deviate from process characteristics included in the collected data. Anomalies can be detected based on at least one of a preset normal range, data distribution, difference from adjacent time point data, rate of change, moving average, standard deviation, and process operating conditions. The time series data generation unit (1415) can remove the detected anomalies or correct them based on the change characteristics of the normal data.
[0055] The time series data generation unit (1415) can distinguish between missing data occurring at a single point in time and missing data intervals occurring at multiple consecutive points in time. Missing data can be corrected by at least one of interpolation based on adjacent point in time data, moving average, trend-based estimation, multiple regression, or estimation based on functional correlation with related process variables.
[0056] In one embodiment, the time series data generation unit (1415) can correct missing values or outliers by combining multiple regression, interpolation, and moving window methods to maintain a functional correlation between the water treatment data to be predicted and the input measurement data. Accordingly, the distortion of the actual water treatment data change pattern caused by simple average value substitution can be reduced.
[0057] The time series data generation unit (1415) can convert data having different measurement cycles into a common time unit. For example, the time series data generation unit (1415) can set a common time interval in minutes or hours and resample each data to correspond to the common time interval.
[0058] The time series data generation unit (1415) can align and synchronize water quality data, water level data, process operation data, flow rate data, power consumption, and external environment data based on the same time axis. Accordingly, multiple variable values at the same point in time can be composed of a single multivariate data record. The time series data generation unit (1415) can normalize or scale the data to convert different units and ranges of values for each variable according to a certain standard. In one embodiment, min-max normalization (MinMax Scaling) may be used. The normalization process can prevent excessive influence on model learning due to a specific input variable having a relatively large range of values, and can ensure that the relationship between multiple variables is reflected in a balanced manner.
[0059] The time series data generation unit (1415) can generate a derived variable from a raw variable. The derived variable may include at least one of an average value, a sum value, a change amount, a rate of change, a cumulative value, a moving average, and a moving variance over a certain period of time.
[0060] In one embodiment, the time series data generation unit (1415) can generate at least one of the following as a derived variable: water treatment data such as water quality data, water level data, flow rate data, power consumption, or process operation data, as a time difference, rate of change, moving statistical value, cumulative value, or sum of multiple time points.
[0061] The time series data generation unit (1415) can generate multivariate time series data by continuously arranging preprocessed data based on time information. The generated multivariate time series data may include at least one of data value, measurement time, data item, quality status, correction status, normalization information, and derived variable information. The time series data generation unit (1415) can store the multivariate time series data in an integrated database so that it can be used based on the same time standard during the learning, prediction, evaluation, and relearning processes.
[0062] The input data setting unit (1420) can set past time series data included in a predetermined time interval prior to a past reference point among the multivariate time series data generated by the time series data generation unit (1415) as input data for learning the artificial intelligence prediction model.
[0063] A predetermined time interval may represent the length of an input sequence that an artificial intelligence prediction model refers to in order to analyze the temporal pattern of water treatment changes. In one embodiment, the predetermined time interval may be set according to the process characteristics of the water treatment facility, the data collection cycle, the residence time, or the purpose of prediction, and may be set, for example, to the past 6 hours.
[0064] The input data setting unit (1420) can set multiple input data by shifting past reference points at regular time intervals. For example, by shifting the reference point by one hour, data from six hours prior to each reference point can be set as a single input sequence. The input data can be configured in the form of a multidimensional array having a time axis and a variable axis. Multiple points included in a predetermined time interval are arranged on the time axis, and water quality data, water level data, process operation data, flow rate data, power consumption, and external environment data for each point can be arranged on the variable axis.
[0065] The input data setting unit (1420) may set different types of input variables depending on the water treatment item to be predicted. For example, if the water treatment item to be predicted is any one of an organic matter indicator, a nutrient indicator, a suspended solids indicator, a disinfection-related indicator, or other water treatment management indicator, the input variable may be set to at least one of water quality data, water level data, flow rate data, power consumption data, process operation data, external environment data, and derived variables related to the water treatment item. In one embodiment, for prediction related to organic matter, suspended solids, or nutrients, at least one of ORP, NO3, NH4, DO, MLSS, pH, or a derived variable thereof may be included as an input variable.
[0067] The input data setting unit (1420) can distinguish between input data for training and input data for actual prediction. Input data for training consists of time series data from a past period, and input data for actual prediction can consist of time series data included in a newer predetermined time interval prior to the current reference point.
[0068] The input data setting unit (1420) may change the input variable or input time interval according to at least one of the water treatment item, application process, residence time, data quality and prediction purpose.
[0069] The target data setting unit (1430) can set water treatment measurement data at a point in time when each of the plurality of retention times has elapsed from a past reference point as target data corresponding to each of the plurality of retention times. The retention time (HRT: Hydraulic Retention Time) may represent the time required for influent water or raw water to reach the downstream end of the process or the discharge point after it has entered the water treatment process, or a corresponding time delay.
[0070] Figure 2 is an example graph of a multi-point prediction result for multiple residence times. Future water treatment values for each residence time (HRT) interval can be predicted based on measured input data. Each prediction result is generated for future time intervals corresponding to the residence time within the process, and is configured to allow verification of the trend of water treatment data changes over time.
[0071] The target data setting unit (1430) can calculate a target time point by adding a residence time or a set time delay value at a past reference time point, and set the treated water, discharge water, or water treatment data of the predicted target point actually measured at the target time point as the target data. A plurality of residence times can be set within a range from 0 hours corresponding to the current time point to a set maximum residence time. In one embodiment, a plurality of residence times can be set according to the process characteristics of the water treatment facility, flow rate conditions, data collection cycle, or prediction purpose, and, for example, can be set at time intervals set from 0 hours to 12 hours.
[0072] Accordingly, for the same input data, water treatment data measured at a reference point, 1 hour later, and 2 to 12 hours later, respectively, can be set as different target data. The target data may include at least one of the water treatment items measured at the treated water, discharged water, raw water, process water, or the prediction target point.
[0073] Each target data point can be configured based on the prediction target point, the water treatment item to be predicted, and the target time point. For example, specific target data can be configured as the value of a water treatment item measured at the prediction target point at the target time point, such as the organic matter concentration of effluent, the turbidity of treated water, or the nutrient concentration of process water.
[0074] The target data setting unit (1430) may exclude the target data from the learning target or set the data corrected by the time series data generation unit (1415) as the target data if the water treatment measurement data at the target time point does not exist or does not meet the quality standards. The target data setting unit (1430) may store the correspondence between the input reference time point, the target time point, the residence time, the target water treatment item, and the target value.
[0075] The learning data generation unit (1440) can generate multiple learning data sets by matching the input data set by the input data setting unit (1420) with the multiple target data set by the target data setting unit (1430).
[0076] The training data generation unit (1440) can correspond multiple target data corresponding to multiple residence times to a single input data. Accordingly, multiple training data sets with different target times can be generated while maintaining the same structure of the input data. For example, for input data included in a set time interval prior to a past reference time, water treatment data measured at a reference time or multiple target times after the reference time can be corresponded to each. The training data generation unit (1440) can generate different training data sets for each water treatment item. Each training data set can be distinguished according to the water treatment item to be predicted, the configuration of input variables, the application process, or the target time. The training data generation unit (1440) can include at least one of an input sequence, a target value, a past reference time, a target time, a residence time, a water treatment item, input variable information, and quality information in each training data set.
[0077] The training data generation unit (1440) can divide the training data into a training period, a verification period, and an evaluation period while maintaining the time series order of the training data. A continuous period in the past may be set as the training period, and a subsequent period may be set as the verification period or the evaluation period. For example, past time series data accumulated for a certain period (from June 2020 to December 2025) may be used as training data. The data period may be changed depending on the water treatment facility, the amount of data secured, and the purpose of application.
[0078] The training data generation unit (1440) checks the number of training data, variable configuration, time series length, number of data per dwell time, and number of data per water processing item, and can exclude data sets that do not meet the set minimum training data criteria from the training target.
[0079] The prediction model generation unit (1450) can generate an artificial intelligence prediction model that learns the correlation between the temporal change pattern of input data and water treatment data, such as multiple water quality, water level, flow rate data, power consumption, and multiple process operation data, using a plurality of training data sets generated by the training data generation unit (1440). The artificial intelligence prediction model may be configured as an artificial intelligence model structure for time series prediction and may include at least one of a convolutional neural network layer, a batch normalization layer, a recurrent neural network layer, an attention layer, and an output layer. In one embodiment, the artificial intelligence prediction model may be configured as a time series prediction model that includes at least one of a Conv1D layer, a Batch Normalization layer, an LSTM layer, and an Attention layer, depending on the water treatment item to be predicted, data characteristics, or the applied process. The artificial intelligence prediction model may be configured to learn the change pattern of time series data, long-term dependency, interaction between multiple variables, or important point-in-time information.
[0080] The prediction model generation unit (1450) can learn the interaction between variables and temporal change patterns by using water quality data, water level data, power consumption, process operation data, flow rate data, and external environment data as integrated inputs. The prediction model generation unit (1450) can generate different artificial intelligence prediction models for each water treatment item. In addition, different prediction models can be generated for each water treatment item based on retention time. For example, different prediction models can be generated for specific water treatment items to be predicted, such as discharge water quality COD, TN, TP, and SS, based on retention time.
[0081] In another embodiment, a single artificial intelligence prediction model may be configured to simultaneously output multiple water treatment values corresponding to multiple residence times.
[0082] The prediction model generation unit (1450) can apply a loss function to reduce the error between the predicted value and the target data and iteratively adjust the model parameters. In one embodiment, the loss function may be set to at least one of a loss function based on mean squared error (MSE), mean absolute error (MAE), or prediction error.
[0083] The prediction model generation unit (1450) can set at least one of the learning rate, batch size, and number of iterations (Epoch) as a learning condition.
[0084] The prediction model generation unit (1450) monitors the training loss and validation loss to determine whether overfitting has occurred, and can terminate the training if the validation loss does not improve more than a set number of times.
[0085] The prediction model generation unit (1450) can verify the generalization performance of the model using a verification interval or an evaluation interval that was not used for training. The generated artificial intelligence prediction model can be stored in memory (120) along with at least one of model identification information, applied water treatment items, applied residence time, input variables, input sequence length, normalization information, model structure, and model parameters.
[0086] The water treatment prediction unit (1460) inputs multivariate time series data included in a predetermined time interval prior to the current reference point into an artificial intelligence prediction model to generate water treatment prediction values corresponding to multiple future points in time after each of the multiple residence times or time delay values has elapsed from the current reference point.
[0087] In one embodiment, the water treatment prediction unit (1460) inputs measurement data from 6 hours prior to the current reference point for a sewage or wastewater treatment process and can generate at least one predicted value among COD, SS, TN, and TP corresponding to a plurality of future points in time from the reference point up to 12 hours later. In another embodiment, the input time interval, the prediction time range, the time interval, and the water treatment item to be predicted may be set differently depending on the water purification process, the river / water system management environment, the data collection cycle, the retention time, or the purpose of prediction.
[0088] The water treatment prediction unit (1460) can configure actual prediction input data under conditions identical to the order of input variables, normalization criteria, and input sequence length applied when creating the prediction model.
[0089] The water treatment prediction unit (1460) can update the input sequence with the latest measurement data whenever the current reference time is updated and repeatedly generate water treatment prediction values for each residence time.
[0090] The prediction result may be stored along with at least one of the following: a prediction reference time, a prediction target time, a residence time, a water treatment item, a predicted value, model identification information, and data quality information. The water treatment prediction unit (1460) may generate time series data representing the trend of water treatment change by aligning the prediction result based on a time axis. The water treatment prediction unit (1460) may provide at least one of the following to an integrated monitoring system (20) or a user terminal (30): a predicted value, a trend of change by residence time, an expected peak time, an expected time exceeding the reference, and whether the expected water treatment reference is exceeded. The prediction result may be output in at least one form among a monitoring screen, a time series graph, a table by water treatment item, a table by residence time, warning information, a mobile screen, a file, or data linked to an external system.
[0091] The water treatment prediction unit (1460) can be applied to at least one of a sewage treatment process, a wastewater treatment process, a water purification process, and a river and water system management environment. In a sewage or wastewater treatment process, changes in effluent treatment and process load can be predicted based on influent water quality, flow rate, and process operating conditions. The prediction results are provided in a form that reflects changes in water treatment and process load according to the retention time (HRT), and can be used as a standard for monitoring process status and water treatment management in a sewage or wastewater treatment operating environment. In a water purification process, future water treatment of treated water can be predicted based on raw water quality, treatment process status, and environmental conditions. The prediction results are provided in a form that reflects the operating status of treatment processes such as coagulation, sedimentation, and filtration, and characteristics of water treatment changes, and can be used as a standard for water purification quality management and process operating status judgment.
[0092] In the context of river and water system management, trends in water treatment changes over time across the entire system can be predicted using meteorological data and water treatment data, such as water quality, water level, flow rate, and power consumption. Predicted changes in flow rate or water level can be utilized as operational support information to identify flood-risk areas, areas requiring drought response, potential supply shortages, or the need for discharge volume adjustments in advance. The prediction results are provided in a format that reflects continuous water treatment changes and impact ranges across upstream and downstream sections, and can be used as a standard for water system operation management and responding to changes in water treatment.
[0093] In addition, the prediction results can be utilized as smart water treatment operational information linked with IoT-based field measurement data, and can serve as reference information for reviewing potential energy consumption reductions, managing carbon neutrality-related operational indicators, and assessing water quality stability for water reuse by analyzing the operating status of key process equipment such as blowers, pumps, and chemical injection facilities.
[0094] The performance evaluation unit (1470) can evaluate the prediction performance of the artificial intelligence prediction model by comparing the water treatment prediction value generated by the water treatment prediction unit (1460) with the actual water treatment value measured at the corresponding future time point after each of the plurality of future time points has arrived.
[0095] The performance evaluation unit (1470) can identify a target time for prediction using a prediction reference time and a residence time or time delay value, and can extract the actual water treatment value of the target time for prediction from a database. The performance evaluation unit (1470) can calculate a basic error by subtracting the predicted value from the actual value, and can calculate at least one of an absolute error, a squared error, and a relative error based on the basic error. The performance evaluation unit (1470) can calculate at least one of a coefficient of determination (R²), mean absolute error (MAE), mean squared error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and median absolute error (MedAE). At this time, the performance indicator may be selectively applied according to the value range, unit, data distribution, or evaluation purpose of the target water treatment item.
[0096] The coefficient of determination (R²) can indicate the extent to which the predicted value explains the characteristics of the change in the actual value. The coefficient of determination can be calculated as shown in Equation 1 to show how similarly the predicted value changes to the actual data.
[0097]
[0098] The closer the coefficient of determination (R²) is to 1, the higher the degree of agreement between the model's prediction results and the actual data; the closer it is to 0, the lower the agreement.
[0099] can be the average of the actual water treatment values.
[0100] Mean Absolute Error (MAE) can represent the average magnitude of the error between the predicted value and the actual value.
[0101] The mean absolute error can be calculated as shown in Equation 2.
[0102]
[0103] Mean Absolute Error (MAE) can be suitable for understanding overall prediction accuracy because it reflects the magnitude of the error.
[0104] The mean squared error can be calculated as shown in Equation 3.
[0105]
[0106] y t is the actual value, is the predicted value, and N can be the number of samples.
[0107] Mean squared error can be used to identify unstable prediction intervals by assigning high weights to relatively large errors.
[0108] The error can be calculated as a ratio to the actual value using the Mean Absolute Percentage Error (MAPE). The Mean Absolute Percentage Error can be calculated as shown in Equation 4.
[0109]
[0110] MAPE can represent the relative ratio of the prediction error to the measured value. However, if the measured water treatment value is zero or below a set threshold value, the data at that point in time may be excluded from the MAPE calculation, or other performance indicators may be applied depending on the characteristics of the water treatment item being predicted. RMSE can be used to interpret the error in the same unit as the actual water treatment value. MedAE can be used as a median-based error indicator that is relatively less affected by outliers.
[0111] The performance evaluation unit (1470) can analyze at least one of the following: the numerical difference between the predicted value and the actual value, as well as the upward / downward trend, the time of occurrence of the highest or lowest value, the time of occurrence of the lowest point, the rate of change, the leading or lagging of the reaction time, the error distribution, and the prediction stability. The performance evaluation unit (1470) can analyze whether the prediction error increases with the residence time and can distinguish between a stable prediction section and a section where performance deteriorates for each residence time.
[0112] The performance evaluation unit (1470) can analyze at least one of a plurality of performance indicators, error distribution, error change by residence time, error concentration by time interval, and prediction stability. Additionally, if the prediction error repeatedly increases in a specific time interval, residence time interval, or water treatment item, the performance evaluation unit (1470) can identify the corresponding interval as a performance degradation interval or an interval subject to review. The derived results can be used for data correction, setting intervals subject to retraining, or operator review information.
[0113] The performance evaluation results may be provided in at least one of the following forms: a time series graph, a graph comparing actual and predicted values, an error graph, a performance table by residence time, a performance table by water treatment item, a performance grade, and a downloadable file.
[0114] FIG. 3 is an example diagram showing a graph comparing actual values and predicted values according to an embodiment, and FIG. 4 is an example diagram showing a graph of comparative analysis based on dwell time.
[0115] The time-based comparison graph in Fig. 3 allows for the comparison of trends over time by continuously arranging predicted and measured values on the same time axis. Through this, it is possible to comprehensively verify the degree of agreement between the two values over time, including not only simple numerical differences at specific points in time but also the flow and rate of change in rising and falling sections. By comparing the points of change between the measured and predicted values, the appropriateness of reflecting the set residence time or time delay value can be verified. In some sections, a time deviation may occur where the predicted value leads or lags behind the measured value; this can be interpreted as occurring when the actual time delay required for the influent or raw water to pass through the treatment process and reach the predicted target point differs from the set residence time or time delay value.
[0116] In addition, the concentration of errors in specific time intervals can be checked through time-based error graphs (actual value vs. predicted value). The stability of the prediction can be evaluated by analyzing the level of error variation across the entire time interval.
[0117] The residence time-based comparative analysis graph in Fig. 4 can represent the results of a residence time-based comparative analysis by correlating the time at which the predicted value is generated with the time at which the corresponding water treatment state is measured in the actual process based on the residence time. By reflecting the time delay from the inflow point to the prediction target point and correlating the predicted value and the actual value based on the same water flow standard, a consistent comparison can be performed that considers the movement process within the process. The prediction accuracy for each segment can be verified by comparing the actual value (orange solid line) and the predicted value (blue dotted line), reflecting the set residence time (6 hours) or the time delay value, on the same time axis. The level of error can be identified by analyzing the change in deviation between the predicted value and the actual value as the prediction reference time elapses. Stable and unstable prediction segments can be distinguished by comparing the numerical proximity of the predicted value and the actual value for each time segment. Based on the error level and stability for each segment, a residence time range applicable to the actual process can be established, and the practical application range of the prediction results can be determined by focusing on segments where the deviation between the two values is small and the direction of change coincides.
[0118] The data verification unit (1480) can determine the necessity of retraining the artificial intelligence prediction model by verifying the performance indicators, error analysis results, and data patterns of the time series data calculated by the performance evaluation unit (1470). The data verification unit (1480) can determine that retraining of the artificial intelligence prediction model is necessary if the prediction performance of the artificial intelligence prediction model does not meet the set performance criteria or if a change in the data pattern of the multivariate time series data is detected.
[0119] The data verification unit (1480) can compare at least one of the mean, variance, distribution, trend, periodicity, rate of change, cumulative change, and correlation between variables of at least one of the water treatment data and external environment data, including at least one of the current water quality data, water level data, flow rate data, power consumption, and process operation data, with the corresponding characteristics of the existing training data.
[0120] The data verification unit (1480) can detect at least one of a data pattern change, a data distribution shift, a change in mean, an increase in volatility, a change in trend, a change in seasonal pattern, a change in correlation between variables, the occurrence of a new data pattern, and a deterioration in sensor data quality.
[0121] The data verification unit (1480) can perform verification based on at least one analysis result among the coefficient of determination (R²), mean absolute error (MAE), mean squared error (MSE), root mean square error (RMSE), mean absolute percentage error (MAPE), and median absolute error (MedAE). At least one of an increase in the error fluctuation range, an increase in the variability of performance indicators, and a decrease in the consistency of prediction results can be determined as a decrease in prediction performance.
[0122] The data verification unit (1480) may determine whether performance degradation or data change persists for longer than a set time interval. That is, data and performance change detection can identify changes different from existing patterns by analyzing the statistical characteristics of the input data and changes in the performance indicators of the prediction results together over time. Specifically, the data verification unit (1480) can distinguish between temporary fluctuations and continuous performance degradation or data change by analyzing whether a degradation of performance indicators, an increase in prediction error, or a change in data patterns appears repeatedly in multiple consecutive time intervals. The degree of performance degradation or the degree of data change calculated in each time interval is accumulated, and if the accumulated value exceeds a set threshold value, it can be determined that continuous performance degradation or data change has occurred. In one embodiment, the data verification unit (1480) can calculate the degree of performance degradation of the current interval relative to the threshold interval. The ratio of performance degradation relative to the threshold can be defined as Equation 5.
[0123]
[0124] Here is the mean absolute error (MAE) in the current interval k, is the Mean Absolute Error (MAE) of the reference interval (initial training interval or stable interval), This may be the ratio of performance degradation relative to the standard. If performance degradation exceeding a certain level persists in a continuous interval, it may be determined that retraining is required.
[0125] The data verification unit (1480) can compare at least one of the mean, variance, trend, rate of change, cumulative change, correlation between variables, and change in performance indicators among at least one of the water treatment data including at least one of the current water quality data, water level data, flow rate data, power consumption, and process operation data, and at least one of the external environment data with the corresponding characteristics of the existing training data to determine the change in data patterns. In addition, the data verification unit (1480) can determine the necessity of retraining by distinguishing it from temporary fluctuations when an increase in prediction error, a decrease in performance indicators, or a change in data patterns repeatedly appears in multiple consecutive time intervals. The result of this determination can be used as a criterion for selecting the interval to be retrained, the water treatment item to be retrained, or the retention time to be retrained.
[0126] The data verification unit (1480) may determine the method of performing retraining as immediate retraining, periodic retraining, or retraining when conditions are met. The data verification unit (1480) may select a residence time, water treatment item, or time interval where prediction performance has deteriorated as the retraining target interval. The data verification unit (1480) may selectively configure retraining data centered on the performance deterioration interval and the data pattern change interval. Accordingly, unnecessary training can be reduced compared to using the entire data for retraining every time. The data verification unit (1480) may set the range of the retraining data to include both the data before and after the retraining target interval. Accordingly, the time flow and process state changes before and after the point of error occurrence can be continuously reflected.
[0127] The data verification unit (1480) may perform at least one of outlier correction, missing value processing, signal stabilization, time alignment, and normalization on the data to be retrained, or request the time series data generation unit (1415) to perform the above processing. The data verification unit (1480) may construct a residence time-based retraining data set by re-establishing the time relationship between the input data of the retraining target section and the target water processing data after the residence time has elapsed.
[0128] The data verification unit (1480) can adjust the reflection range or reflection ratio of existing training data and retraining data according to at least one of the level of performance degradation, error magnitude, degree of data pattern change, and possibility of overfitting. At this time, the data verification unit (1480) can increase the training weight of the performance degradation section or new data, while maintaining a balance between existing data and new data to prevent overfitting in a specific section.
[0129] The data verification unit (1480) can reset at least one of the learning range, learning objective, input structure, and learning condition information according to the level of change in the performance indicator. Sections where errors are concentrated are preferentially included in the retraining data, and the learning weight for each section may be set differently depending on the level of error occurrence. If a change in the data pattern occurs, at least one of the ratio between the existing training data and the new data, the input variable, and the input sequence structure may be changed.
[0130] And the model retraining unit (1490) can retrain the artificial intelligence prediction model using retraining data to include actual water treatment values corresponding to water treatment prediction values. That is, the model retraining unit (1490) can retrain the artificial intelligence prediction model using retraining data when it is determined by the data verification unit (1480) that retraining is necessary.
[0131] The model retraining unit (1490) can reflect the level of change in indicators of the performance degradation section, adjust the learning goal and criteria according to the error magnitude, and configure the learning conditions by considering the performance deviation by section.
[0132] The model retraining unit (1490) can perform additional training using the model parameters of an existing artificial intelligence prediction model as initial values, or create a new artificial intelligence prediction model using retraining data and existing training data. The model retraining unit (1490) can selectively retrain only the model corresponding to a specific water treatment item, a specific retention time, or a specific process where performance degradation has occurred. In another embodiment, the model retraining unit (1490) can retrain the entire model corresponding to a plurality of water treatment items and a plurality of retention times in a batch.
[0133] The model retraining unit (1490) can reset at least one of the learning rate, batch size, number of iterations, loss function, optimizer, input variable, input sequence length, ratio of existing data to new data and early termination condition during the retraining process.
[0134] The model retraining unit (1490) can divide the retraining data into a training section, a verification section, and an evaluation section according to the time series order. The training section for retraining is used for adjusting model parameters, and the verification section can be used to check whether the retraining process is overfitted and the appropriateness of the training conditions. The evaluation section can be used to evaluate the final performance of the retrained model.
[0135] The model retraining unit (1490) can compare at least one of performance indicators such as R², MAE, MSE, RMSE, MAPE, and MedAE by applying the same evaluation data to the retrained model and the existing model. The model retraining unit (1490) can compare performance before and after retraining by retention time, water treatment item, and time interval. The model retraining unit (1490) can verify at least one of whether the performance of the retrained model has improved compared to the performance of the existing model, whether it meets the set minimum performance criteria, and whether new performance degradation has not occurred in a specific interval.
[0136] If the retrained model meets the set criteria, the model retraining unit (1490) may set the retrained model as the operational model. If the retrained model does not meet the set criteria, the model retraining unit (1490) may maintain the existing model or perform retraining again by changing the training conditions, data range, or data reflection ratio. The model retraining unit (1490) may analyze the retraining effect to calculate at least one of the degree of improvement in performance indicators, the degree of reduction in error, the degree of improvement in prediction stability, and the degree of response to changes in data patterns. The model retraining unit (1490) may provide the retrained model adopted as the operational model to the water treatment prediction unit (1460). Subsequently, the water treatment prediction unit (1460) may generate new water treatment prediction values using the retrained model.
[0137] The model retraining unit (1490) can store at least one of the retraining history, the time of retraining, the data to be retrained, the changed learning conditions, the results of the performance comparison with the existing model, and whether the operating model is adopted. Retraining may be performed at regular intervals or when a decrease in prediction performance or a change in data patterns is detected. The retraining cycle may be set according to at least one of the data generation speed, the frequency of process changes, the rate of performance degradation, and the system operating conditions. Accordingly, a cyclic structure may be formed in which evaluation, determination of the need for retraining, resetting of learning conditions, retraining, performance verification, and reflection in the operating model are repeated.
[0139] FIG. 5 is a flowchart illustrating an artificial intelligence-based water treatment prediction and evaluation method according to one embodiment of the present invention.
[0140] First, the water treatment prediction and evaluation system can collect water treatment data such as multiple water quality, water level, flow rate, power consumption, and multiple process operation data measured over time in the water treatment process (S200).
[0141] Multiple water quality data may be data indicating the water treatment status of raw water, influent water, process water, treated water, or effluent water. Multiple process operation data may be data indicating the operation status of a water treatment process. A water treatment prediction and evaluation system may collect water treatment data, such as the multiple water quality data, water level data, flow rate data, power consumption, and the multiple process operation data, from at least one of a water treatment sensor, flow meter, process equipment, integrated monitoring system, database, or external data provision server. In this case, the integrated monitoring system may provide field measurement data or equipment status data collected through a field control panel / field control terminal (RCS / RTU) or a remote monitoring and control station (TM / TC).
[0142] The water treatment prediction and evaluation system can collect time information indicating the time at which the data was measured or generated, along with each data point. Additionally, the water treatment prediction and evaluation system can store at least one of the data type, measurement location, sensor identification information, process identification information, and data source corresponding to each data point.
[0143] In one embodiment, the water treatment prediction and evaluation system may further collect external environment data including at least one of temperature, precipitation, humidity, seasonal information, and weather information.
[0144] Next, the water treatment prediction and evaluation system can generate multivariate time series data by aligning collected water treatment data, such as multiple water quality, water level, flow rate data, power consumption, and multiple process operation data, based on time (S210). The water treatment prediction and evaluation system can convert data collected according to different devices or different measurement cycles into a common time unit. For example, data collected in seconds, minutes, or hours can be resampled to a preset common time interval. The water treatment prediction and evaluation system can synchronize water treatment data, such as multiple water quality, water level, flow rate data, power consumption, and multiple process operation data, based on the same time axis. Accordingly, multiple variable values at the same or corresponding time points can be composed of a single multivariate data record.
[0145] Next, the water treatment prediction and evaluation system may set past time series data included in a predetermined time interval prior to a past reference point among the generated multivariate time series data as input data (S220). The past reference point may be a reference time set on the past time series data to generate a training data set. The predetermined time interval may represent the length of an input sequence referenced by an artificial intelligence prediction model to learn the temporal pattern of water treatment changes. The water treatment prediction and evaluation system may set multiple input data by shifting the past reference point at regular time intervals.
[0146] Next, the water treatment prediction and evaluation system can set water treatment measurement data at a point in time after each of the multiple retention times has elapsed from a past reference point as target data corresponding to each of the multiple retention times (S230). Retention time may represent the time required for influent water or raw water to reach the downstream end of the water treatment process, the treated water point, or the discharge point, or a corresponding time delay. The water treatment prediction and evaluation system can determine multiple target points in time by adding each retention time at a past reference point, and set the water treatment data actually measured at each target point in time as target data corresponding to the corresponding retention time.
[0147] Next, the water treatment prediction and evaluation system can generate multiple training data sets by matching the set input data with target data corresponding to each of the multiple residence times (S240). The water treatment prediction and evaluation system can match multiple target data for each residence time to a single input data. Accordingly, multiple training data sets can be generated in which the content of the input data is identical but the target time points or target data are different. The step of generating multiple training data sets may include the step of matching different target data at the time points after each of the multiple residence times has elapsed to the same input data. Additionally, the training data can be configured by matching the input data at a specific time point with future water treatment data after the residence time (HRT).
[0148] Next, the water treatment prediction and evaluation system can generate an artificial intelligence prediction model using multiple training data sets (S250). The artificial intelligence prediction model can learn the correlation between the temporal change patterns of the input data and water treatment data, such as multiple water quality, water level, flow rate data, power consumption, and multiple process operation data. In one embodiment, the artificial intelligence prediction model may include a convolutional neural network layer, a recurrent neural network layer, and an attention layer. The convolutional neural network layer can extract time interval features from the input data. The convolutional neural network layer may include a one-dimensional convolutional layer. The recurrent neural network layer can learn the temporal dependency of the time interval features extracted by the convolutional neural network layer. The recurrent neural network layer may include a Long Short-Term Memory (LSTM) layer. The attention layer can assign a high weight to features of past time points that have a relatively high influence on water treatment changes among the multiple past time point features output from the recurrent neural network layer. The artificial intelligence prediction model may further include a batch normalization layer or an output layer. An artificial intelligence prediction model can generate water treatment prediction values corresponding to multiple future points in time after each of the multiple residence times has elapsed from the current reference point.
[0149] Next, the water treatment prediction and evaluation system can generate water treatment prediction values corresponding to multiple future points in time by inputting multivariate time series data included in a predetermined time interval prior to the current reference point into an artificial intelligence prediction model (S260). The current reference point may be the point in time when actual water treatment prediction is performed. The water treatment prediction and evaluation system can set multivariate time series data included in the latest predetermined time interval prior to the current reference point as input data for actual prediction.
[0150] Next, the water treatment prediction and evaluation system can evaluate the prediction performance of the artificial intelligence prediction model by comparing the predicted water treatment value with the actual water treatment value measured at each future point in time after each of the multiple future points in time has arrived (S270). The water treatment prediction and evaluation system can identify the target time for prediction corresponding to each predicted water treatment value using the prediction reference time and the residence time. The water treatment prediction and evaluation system obtains the actual water treatment value measured at the identified target time for prediction from the database and can align the predicted water treatment value with the actual water treatment value based on the same time reference.
[0151] Next, the water treatment prediction and evaluation system can determine the necessity of retraining by analyzing whether the prediction performance of the artificial intelligence prediction model has deteriorated or whether there has been a change in data patterns (S280). The water treatment prediction and evaluation system can determine whether the prediction performance of the artificial intelligence prediction model meets the set performance criteria.
[0152] If it is determined that retraining is necessary, the water treatment prediction and evaluation system may configure retraining data (S290). The retraining data may include actual water treatment values corresponding to water treatment prediction values. The water treatment prediction and evaluation system may include input data and actual water treatment values corresponding to retention time, water treatment items, or time intervals where prediction performance has deteriorated in the retraining data. The water treatment prediction and evaluation system may prioritize including data from time intervals where the prediction error is greater than a set error criterion in the retraining data.
[0153] Next, the water treatment prediction and evaluation system can retrain an artificial intelligence prediction model using retraining data (S295). The water treatment prediction and evaluation system can perform additional training using the model parameters of an existing artificial intelligence prediction model as initial values, or generate a new artificial intelligence prediction model using existing training data and retraining data. The water treatment prediction and evaluation system can selectively retrain only the artificial intelligence prediction model corresponding to a specific water treatment item or a specific retention time where performance degradation has occurred.
[0154] The water treatment prediction and evaluation system can compare prediction performance before and after retraining by applying the same evaluation data to the retrained AI prediction model and the existing AI prediction model. If the prediction performance of the retrained AI prediction model improves compared to the existing AI prediction model and meets the set performance criteria, the water treatment prediction and evaluation system may set the retrained AI prediction model as the operational model. If the prediction performance of the retrained AI prediction model does not meet the set performance criteria, the water treatment prediction and evaluation system may maintain the existing AI prediction model or perform retraining again by changing the training conditions or the range of the retraining data. Accordingly, data collection, generation of multivariate time series data, generation of training data sets, creation of AI prediction models, future water treatment prediction, prediction performance evaluation, and model retraining can be performed cyclically.
[0155] The above-described method may be implemented as an application or in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., either individually or in combination.
[0156] The program instructions recorded on the above-mentioned computer-readable recording medium are those specifically designed and configured for the present invention, but may also be those known and available to those skilled in the art of computer software.
[0157] Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, and flash memory.
[0158] Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware device may be configured to operate as one or more software modules to perform processing according to the present invention, and vice versa.
[0159] Although the embodiments have been described in detail above, the scope of the rights to the embodiments is not limited thereto, and various modifications and improvements by those skilled in the art using the basic concepts of the embodiments defined in the following claims also fall within the scope of the rights to the embodiments.
[0160] Accordingly, the detailed description above should not be interpreted restrictively in all respects but should be considered as an example. The scope of the embodiments should be determined by a reasonable interpretation of the appended claims, and all modifications within the equivalent scope of the embodiments are included within the scope of the embodiments. Explanation of the symbols
[0162] 10: Water Treatment Prediction and Evaluation System 20: Integrated Monitoring System 21: Field control terminal device 22: Remote monitoring control station 30: User terminal 110: Communication interface 120: Memory 130: I / O interface 140: Processor 1410: Data acquisition unit 1415: Time series data generation unit 1420: Input data setting unit 1430: Target Data Setting Unit 1440: Training Data Generation Unit 1450: Prediction Model Generation Unit 1460: Water Treatment Prediction Unit 1470: Performance Evaluation Department 1480: Data Verification Department 1490 : Model Retraining Section
Claims
Claim 1 A method performed in a computing device having one or more processors and a memory for storing one or more programs executed by said one or more processors, comprising: collecting water treatment data including at least one of a plurality of water quality data, flow rate data, water level data, supply amount data, discharge amount data, power consumption data, process operation data, facility status data, and external environment data measured over time; aligning the collected water treatment data by time to generate multivariate time series data; and setting past time series data included in a predetermined time interval prior to a past reference point among the generated multivariate time series data as input data. A step of setting water treatment measurement data at a point in time after each of the plurality of residence times has elapsed from the aforementioned past reference point as target data corresponding to each of the plurality of residence times; a step of generating a plurality of training data sets by matching the set input data with the target data corresponding to each of the plurality of residence times; a step of generating an artificial intelligence prediction model that learns the correlation between the temporal change pattern of the input data and the water treatment data using the plurality of training data sets; a step of inputting multivariate time series data included in a predetermined time interval prior to the current reference point into the artificial intelligence prediction model to generate water treatment prediction values corresponding to a plurality of future points in time after each of the plurality of residence times has elapsed from the current reference point.The method includes a step of evaluating the prediction performance of the artificial intelligence prediction model by comparing the water treatment prediction value with the actual water treatment value measured at the corresponding future time point after each of the plurality of future time points has arrived, and the step of generating the artificial intelligence prediction model includes a step of generating a different artificial intelligence prediction model for each of the plurality of residence times, and the step of evaluating the prediction performance includes a step of identifying a prediction target time point corresponding to each of the water treatment prediction value using the current reference time point and each of the plurality of residence times; and a step of obtaining the actual water treatment value measured at the identified prediction target time point and aligning the water treatment prediction value and the actual water treatment value based on the same time standard. The method includes the step of calculating at least one performance indicator among the coefficient of determination (R²), mean absolute error (MAE), mean squared error (MSE), mean absolute percentage error (MAPE), root mean square error (RMSE), and median absolute error (MedAE) for each of the plurality of residence times; the step of selecting a specific residence time among the plurality of residence times for which the performance indicator does not satisfy a set performance standard as a residence time subject to retraining; the step of configuring retraining data to include input data corresponding to the specific residence time and actual water treatment values measured after the specific residence time has elapsed corresponding to the input data, wherein the time relationship between the input data included in the retraining data and the target water treatment data after the specific residence time has elapsed is re-established to configure a residence time-based retraining data set; the step of selectively retraining an artificial intelligence prediction model corresponding to the specific residence time among different artificial intelligence prediction models generated for each of the plurality of residence times using the residence time-based retraining data set; and the step of applying the same evaluation data to the retrained artificial intelligence prediction model and the artificial intelligence prediction model before retraining to the specific A step of comparing the prediction performance of retention time before and after retraining;A method for AI-based water treatment prediction and evaluation comprising the step of: setting the retrained AI prediction model as an operational model when the prediction performance of the retrained AI prediction model is improved compared to the prediction performance of the AI prediction model prior to retraining and satisfies the set performance criteria; and maintaining the AI prediction model prior to retraining when the prediction performance of the retrained AI prediction model does not satisfy the set performance criteria. Claim 2 In claim 1, the collecting step comprises collecting field measurement data or equipment status data measured or generated from at least one of a water treatment analyzer, a water level gauge, a flow meter, a turbidity meter, a concentration meter, a pressure gauge, a thermometer, a power meter, or process equipment through at least one of a field control panel (RCS: Remote Control Station), a field control terminal unit (RTU: Remote Terminal Unit), a remote monitoring / control station (TM / TC: Tele-Metering / Tele-Control) or an IoT sensor that is data-linked with an integrated monitoring system. This is an artificial intelligence-based water treatment prediction and evaluation method. Claim 3 The artificial intelligence-based water treatment prediction and evaluation method according to claim 1, wherein the step of generating multivariate time series data comprises: a step of detecting and removing or correcting outliers included in the plurality of water treatment data; a step of correcting missing data included in the plurality of water treatment data; and a step of normalizing the plurality of water treatment data. Claim 4 In claim 1, the artificial intelligence prediction model comprises: a convolutional neural network layer that extracts time interval features from the input data; a recurrent neural network layer that learns the temporal dependency of the time interval features extracted by the convolutional neural network layer; and an attention layer that assigns weights to past time points among a plurality of past time point features output from the recurrent neural network layer that have a relatively high influence on water treatment changes, thereby providing an artificial intelligence-based method for water treatment prediction and evaluation. Claim 5 delete Claim 6 delete
Citation Information
Patent Citations
Apparatus and method for detecting outliers in multivariate time series data
KR1020250103456A
Apparatus and method for determining optimal sequential prediction model
KR1020250172158A
Realtime water level prediction device, realtime water level prediction method, and computer program
JP2025073811A
Apparatus and Method for Artificial Intelligence Deep Learning-Based Water Supply and Quality Prediction
KR102748052B1
Intelligent Marine Ecosystem Management System
KR102961105B1
