KR-102962941-B1 - System and Method for Sensor-less Pipe Water Level Prediction Using Data Assimilation Based on Ensemble Kalman Filter

KR102962941B1KR 102962941 B1KR102962941 B1KR 102962941B1KR-102962941-B1

Abstract

The present invention relates to a sensorless pipe water level prediction system and method using ensemble Kalman filter-based data assimilation, which enables accurate estimation of the water level and flow rate of the entire stormwater pipe network using only a limited number of sensors, thereby allowing for the prediction and response to urban flooding risks in advance. The system of the present invention includes a data collection unit that collects rainfall prediction information, rainfall observation information, tide level information and operation information; a pipe information database that stores structural information of a stormwater pipe network; a data preprocessing unit that preprocesses input data; a SWMM model execution unit that predicts the initial water level and flow rate of each node by running a SWMM-based one-dimensional drainage model; a sensor data receiving unit that receives actual water level data from sensors installed on some nodes; a data assimilation processing unit that estimates the corrected water level and flow rate of the entire network, including nodes without sensors, by optimally combining model prediction values and actual values using a data assimilation technique using an ensemble Kalman filter; and a visualization unit that visualizes the estimation results through a digital twin-based 3D GIS platform. The data assimilation processing unit generates more than 20 ensemble scenarios reflecting uncertainties regarding the initial pipe water level, rainfall, roughness coefficient, inflow, and downstream boundary water level, and runs a SWMM model for each scenario to calculate the mean and covariance of the predicted values. Subsequently, it analyzes the error between the predicted and measured values at the sensor installation nodes to calculate the Kalman gain, and applies this to the entire network to estimate the accurate water level and flow rate up to the point without sensors. The Kalman gain K is calculated using the formula K = P_xy × (P_yy + R)^(-1), and the state vector of each ensemble is updated using the formula x(i) = x_pred(i) + K × (z - y_pred(i)). The present invention may further include a risk analysis unit that generates a flood risk map by identifying overflow risk nodes, confirming poor drainage points, and analyzing freeboard for each pipeline, and a machine learning correction unit that automatically corrects model parameters by analyzing accumulated data using machine learning. The system enables real-time operation by completing a 5-minute prediction within 15 seconds, and generates differential warnings by classifying node water levels into caution, warning, and danger stages relative to the design water level. The present invention dramatically improves the ability to respond to urban flood disasters by providing a learning system that minimizes sensor installation costs while optimally combining physical models and actual measurement data to secure high prediction accuracy and improves performance as data accumulates.

Inventors

박병용
유호동

Assignees

라미랩 주식회사

Dates

Publication Date: 20260512
Application Date: 20260202

Claims (2)

In a sensorless pipeline water level prediction system using ensemble Kalman filter-based data assimilation, A data collection unit that collects input data including rainfall prediction information, rainfall observation information, tide level information, and operational information; Pipeline information database storing structural information of a stormwater pipeline network; A data preprocessing unit that preprocesses the above input data and checks its quality; A SWMM model execution unit that drives a one-dimensional drainage model using preprocessed input data and predicts the initial water level and flow rate for each node of the stormwater pipeline network; A sensor data receiving unit that receives actual water level data from sensors installed at some nodes of the above-mentioned drainage pipe network; A data assimilation processing unit that estimates the corrected water level and flow rate of the entire network, including nodes without sensors, by applying a data assimilation technique using an ensemble Kalman filter and combining the water level and flow rate predicted by the SWMM model execution unit with the actual water level data received by the sensor data reception unit; and A visualization unit that visualizes and outputs the corrected water level and flow rate information estimated by the above data assimilation processing unit through a digital twin-based 3D GIS platform; A sensorless pipeline water level prediction system using ensemble Kalman filter-based data assimilation, characterized by including
In a sensorless pipeline water level prediction method using ensemble Kalman filter-based data assimilation, A step of collecting input data including rainfall forecast information, rainfall observation information, tide level information, and operational information; A step of preprocessing collected input data and inspecting its quality; A step of driving a one-dimensional drainage model using preprocessed input data and predicting the initial water level and flow rate for each node of the stormwater pipeline network; A step of defining the water level and flow rate for each node of the above-mentioned stormwater pipeline network as state vectors, and generating multiple ensemble scenarios within a preset uncertainty range for the initial pipeline water level, rainfall, roughness coefficient, inflow, and downstream boundary water level; A step of calculating predicted water levels and flow rates by driving the one-dimensional drainage model for each of the plurality of ensemble scenarios, and calculating the mean and covariance of the predicted water levels and flow rates of the plurality of ensemble scenarios; A step of receiving actual water level data from sensors installed at some nodes of the above-mentioned drainage network; A step of calculating the error between the predicted water level and the actual water level at the node where the sensor is installed; A step of estimating the corrected water level and flow rate of the entire network including nodes without sensors installed by calculating a Kalman gain based on the above error and updating the state vector of the entire network by applying the Kalman gain to each of the plurality of ensemble scenarios; and A step of visualizing and outputting the above-mentioned corrected water level and flow rate information through a digital twin-based 3D GIS platform; A sensorless pipeline water level prediction method using ensemble Kalman filter-based data assimilation, characterized by including

Description

System and Method for Sensor-less Pipe Water Level Prediction Using Data Assimilation Based on Ensemble Kalman Filter The present invention relates to a drainage pipe water level monitoring system, and more specifically, to a sensorless pipe water level prediction system and method that combines a one-dimensional drainage model (SWMM) and an Ensemble Kalman Filter-based data assimilation technique to predict the water level and flow rate at pipe points where no sensors are installed in real time from a limited number of sensor measurements, and visualizes this through a digital twin-based 3D GIS platform to detect and respond to urban flooding risks in advance. In particular, the present invention relates to a data assimilation technique that receives rainfall prediction and actual measurement data, tide level information, and operational information, predicts the initial water level/flow rate using a SWMM model, calculates the Kalman gain by analyzing the error between the actual values of sensors installed at some nodes and the model prediction values using an ensemble Kalman filter, and applies this to the entire network to estimate accurate water levels and flow rates up to points without sensors. In addition, the present invention belongs to the field of technology that supports real-time operation of urban drainage systems and disaster response decision-making through early warning of nodes (manholes, storm drains) at risk of overflow during heavy rain, identification of points with poor drainage flow, analysis of freeboard for each pipeline, and generation of flood risk maps. Recently, as the frequency and intensity of localized torrential rains caused by climate change have increased, urban flooding damage is occurring frequently. In particular, when rainfall exceeds the drainage capacity of underground storm drains, overflow occurs through manholes and catch basins, leading to serious property and human casualties, including flooded roads and underground spaces. Consequently, there is a growing need for a system that monitors water levels and flow rates within storm drains in real time, predicts flood risks in advance, and enables early response. Conventional stormwater pipeline monitoring technologies are broadly classified into sensor-based direct measurement methods and hydraulic and hydrological model-based simulation methods. The sensor-based method involves collecting actual measurement data by installing sensors, such as water level and flow sensors, inside the pipelines. While it offers the advantage of high measurement accuracy, it has limitations, including the enormous cost required to install and maintain sensors across a vast pipeline network throughout the city, as well as potential issues such as sensor failures, communication disruptions, and malfunctions caused by sediment. On the other hand, simulation methods using one-dimensional drainage models such as the Storm Water Management Model (SWMM) are economical as they can calculate water levels and flow rates across the entire network based on rainfall input and pipeline characteristic information; however, they suffer from reduced prediction accuracy due to model parameter uncertainty, errors in rainfall forecast data, and discrepancies between actual pipeline conditions and design information. In particular, since parameters such as pipeline roughness coefficient, infiltration rate, and evapotranspiration vary significantly depending on actual field conditions, it is difficult to provide reliable prediction results necessary for real-time operation using only the model. To address these issues, some prior art has proposed methods that combine sensor measurements and model predictions. For example, Korean Patent Publication No. 10-2023-0108430 discloses a system that detects flooding situations and guides evacuation routes using flood detection sensors and CCTVs; however, this focuses on post-flooding response after flooding has already occurred and does not include a function to predict water levels in advance at locations without sensors. In addition, Korean Registered Patent Publication No. 10-2542353 proposes a flood prediction system using deep learning, but this is a data-based approach that relies on past flood history data, and thus has limitations such as reduced prediction performance in areas with insufficient training data or new rainfall patterns, and low interpretability of the model due to the inability to reflect physical causal relationships. Meanwhile, data assimilation techniques are widely used in the meteorological and oceanographic fields. Data assimilation is a technique that statistically optimizes the combination of predicted values from numerical models and actual observations to more accurately estimate the state of a system. In particular, the Ensemble Kalman Filter (EnKF) can be effectively applied even to nonlinear systems, demonstrating excellent performance in weather forecasting and ocean circulati