Data-driven road flood warning system

Abstract

This data-driven road flood warning system uses distributed optical fiber sensing (DFOS) / distributed acoustic sensing (DAS) and machine learning (ML) technologies and techniques to predict flood conditions on roads along telecommunication optical fiber cable routes using DFOS / DAS data and ML models. Operationally, DFOS / DAS interrogators collect and transmit vibration data resulting from rain events, while an online web server provides a user interface to end users. Two machine learning models are built for rain intensity prediction and flood level prediction, respectively. These machine learning models function as rain intensity and flood level prediction models based on the provided data, including historical data on rain intensity, rain duration, and flood level.

Description

[Technical Field]

[0001] This application relates to distributed fiber optic sensing (DFOS) / distributed acoustic sensing (DAS) systems, methods, structures, and artificial intelligence, machine learning (ML) techniques. More specifically, this application relates to a data-driven road flood warning system using DFOS and ML techniques. [Background technology]

[0002] Flooding has become increasingly common in many U.S. cities, causing devastating damage to modern society, including infrastructure, economic damage, social disruption, housing inequity, and loss of life. Between 1978 and 2015, urban flooding contributed to overall flood damage, resulting in 3,345 deaths and approximately $285 billion in direct damage. Numerous communities across the United States face similar challenges, and this increasing trend will continue with the increase in extreme weather events due to climate change.

[0003] As is readily apparent, predicting urban floods is difficult. The main reasons are a lack of data and rapidly changing meteorological phenomena. Urban flood events, especially less severe ones, are rarely recorded. This lack of data is partly due to the high cost of installing and maintaining sensing networks over large urban areas and partly due to technical challenges in remote sensing. For example, satellite images are affected by clouds and complex road geometries, and the re-scanning intervals are too long (typically once every 14 days). As a result, satellite images typically cannot capture meteorologically induced flood events due to satellite orbital constraints. These knowledge and data gaps prevent meteorological researchers from systematically investigating events, conclusively identifying their occurrence mechanisms, and efficiently developing numerical models. As a result, decision makers lack timely information on flood mitigation measures, flood risk, and prevention strategies, preventing them from taking proactive measures. Summary of the Invention

[0004] Aspects of the present disclosure directed to a data-driven road flood warning system using distributed fiber optic sensing (DFOS) / distributed acoustic sensing (DAS) and machine learning (ML) technologies and techniques solve the above problems and advance the technology.

[0005] In contrast to the prior art, systems and methods according to aspects of the present disclosure use DFOS / DAS and ML models to predict flood conditions for roads along telecommunications fiber optic cable routes.

[0006] Operationally, the DFOS / DAS interrogator collects and transmits vibration data resulting from rain events, while an online web server provides a user interface to end users. According to aspects of the present disclosure, two machine learning models are constructed for rain intensity prediction and flood level prediction, respectively. These machine learning models function as rain intensity and flood level prediction models based on provided data, including historical data of rain intensity, rain duration, and flood level. [Brief explanation of the drawings]

[0007] [Figure 1(A)] FIG. 1 is a schematic diagram illustrating an exemplary prior art uncoded DFOS system.

[0008] [Figure 1(B)] FIG. 1 is a schematic diagram illustrating an exemplary prior art coded DFOS system.

[0009] [Figure 2] FIG. 1 is a schematic flow diagram illustrating an overall set of key operational features of systems and methods according to aspects of the present disclosure.

[0010] [Figure 3] FIG. 1 is a schematic flow diagram illustrating an exemplary method for predicting rain intensity according to aspects of the present disclosure.

[0011] [Figure 4] FIG. 1 is a schematic flow diagram illustrating an exemplary random forest model for flood level prediction according to aspects of the present disclosure.

[0012] [Figure 5] FIG. 1 is a schematic diagram illustrating an example online web server for flood level monitoring according to an aspect of the present disclosure.

[0013] [Figure 6] FIG. 1 is a schematic flow diagram illustrating an example operation of rain intensity and flood monitoring using DFOS / DAS and ML, according to an aspect of the present disclosure.

[0014] [Figure 7] FIG. 1 is a schematic diagram illustrating exemplary operational characteristics of rain intensity and flood monitoring using DFOS / DAS and ML, according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0015] The following is merely illustrative of the principles of the present disclosure, and it will thus be appreciated that those skilled in the art will be able to devise various arrangements which, although not explicitly described or shown herein, embody the principles of the present disclosure and are included within its spirit and scope.

[0016] Furthermore, all examples and conditional language set forth herein are intended to be for educational purposes only to aid the reader in understanding the concepts contributed by the inventors to further the principles and techniques of the present disclosure, and should not be construed as being limited to such specifically recited examples and conditions.

[0017] Moreover, all statements herein reciting principles, aspects, and embodiments of the present disclosure, as well as specific examples thereof, are intended to encompass both structural and functional equivalents thereof. Additionally, such equivalents are intended to include both currently known equivalents as well as equivalents developed in the future, i.e., elements developed that perform the same function, regardless of structure.

[0018] Thus, for example, it will be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative circuitry embodying the principles of the disclosure.

[0019] Unless otherwise specified herein, the figures comprising the drawings are not drawn to scale.

[0020] As some additional background, note that a distributed fiber optic sensing system interconnects an optoelectronic integrator to an optical fiber (or cable), transforming the fiber into an array of sensors distributed along the fiber. In effect, the fiber becomes the sensor, and the interrogator generates / injects laser light energy into the fiber to sense / detect events along the fiber.

[0021] As those skilled in the art will understand and appreciate, DFOS technology can be deployed to continuously monitor vehicle movement, human traffic, drilling activity, seismic activity, temperature, structural integrity, liquid and gas leaks, and many other conditions and activities. It is used worldwide to monitor power plants, communication networks, railroads, roads, bridges, borders, critical infrastructure, onshore and offshore power lines and pipelines, and downhole applications in oil, gas, and enhanced geothermal power generation. Advantageously, distributed fiber optic sensing is not constrained by line of sight or remote power access and, depending on the system configuration, can be deployed over continuous lengths of more than 30 miles, with sensing / detection possible at every point along that length. Therefore, the cost per sensing point over long distances is typically incomparable to competing technologies.

[0022] Distributed fiber optic sensing measures changes in the "backscatter" of light that occurs within an optical sensing fiber when the fiber encounters environmental changes, including vibration, strain, or temperature change events. As previously mentioned, the optical sensing fiber acts as a sensor along its entire length, providing real-time information about the physical and environmental surroundings and the integrity and security of the fiber. Additionally, distributed fiber optic sensing data pinpoints the precise location of events and conditions occurring on or near the sensing fiber.

[0023] A schematic diagram illustrating the generalized arrangement and operation of a distributed optical fiber sensing system that may advantageously include artificial intelligence / machine learning (AI / ML) analysis is illustratively shown in Figure 1(A). Referring to Figure 1(A), it can be seen that the optical sensing fiber is connected to an interrogator. Although not shown in detail, the interrogator can include a coded DFOS system that can employ a coherent receiver arrangement known in the art, such as that shown in Figure 1(B).

[0024] As is well known, a modern interrogator is a system that generates an input signal into an optical sensing fiber and detects and analyzes the reflected / backscattered signal that is then received. The received signal is analyzed and an output is generated that is indicative of the environmental conditions encountered along the fiber. The received backscattered signal may be due to reflections within the fiber, such as Raman backscattering, Rayleigh backscattering, or Brillouin backscattering.

[0025] As will be appreciated, modern DFOS systems include an interrogator that periodically generates optical pulses (or any coded signal) and launches them into an optical sensing fiber, which then transmits the optical pulse signal along the optical fiber.

[0026] At certain locations along the fiber, a small portion of the signal is backscattered / reflected back to the interrogator where it is received. The backscattered / reflected signal carries information that the interrogator uses to detect, such as changes in power level that indicate mechanical vibrations.

[0027] The received backscattered signal is converted to the electrical domain and processed within the interrogator. Based on the time of pulse incidence and the time the received signal is detected, the interrogator can determine from which location along the optical sensing fiber the received signal returned, thereby sensing activity at each location along the optical sensing fiber. According to aspects of the present disclosure, classification methods may further be used to detect and locate events or other environmental conditions, including acoustic and / or vibration and / or heat, along the optical sensing fiber.

[0028] 2 is a schematic flow diagram illustrating an overall set of key operational features of systems and methods according to embodiments of the present disclosure. As shown, the systems and methods of the present invention include structures and circuitry configured to receive vibration data from an airborne telecommunications fiber optic cable using DFOS / DAS systems and structures such as those illustratively shown above. As will be understood and appreciated, such DFOS / DAS data may include ambient vibration data (without rain) and rain-related vibration data resulting from rainfall, as well as rainfall duration and rainfall intensity.

[0029] The existing rainfall data database provides rainfall and flood level data, including rainfall intensity and duration data, as well as historical flood level data.

[0030] Finally, the above data will be used to train machine learning (ML) models that will provide flood forecasting and warning data / notifications.

[0031] Although not specifically shown in FIG. 2, the system and method of the present invention employs at least three circuit elements that provide much of its functionality: a data collection circuit designed / used to collect rain data and historical flood level data; a data processing circuit that provides hosting operations and resources for machine learning models that predict flood levels; and finally, an online web server circuit that provides data visualization and user interfaces for end users and other systems and methods, as needed.

[0032] Data collection operation

[0033] A distributed fiber optic sensing / distributed acoustic sensing (DFOS / DAS) interrogator is optically connected to a fiber optic cable, thereby providing an optical sensor fiber along a target path. As is well known, a DFOS / DAS system can detect and measure dynamic strain changes that occur along the length of the optical sensor fiber by detecting the optical phase shift of the backscattered light relative to a local optical oscillator. When raindrops strike the optical sensor fiber of a fiber optic cable, the vibrations caused by the raindrop impact produce a time-varying phase shift in the backscattered light, which can be directly detected and located by the DFOS / DAS system. As will be further understood by those skilled in the art, the fiber optic cable and any individual optical sensor fibers used can simultaneously carry live telecommunications traffic in addition to transmitting DFOS / DAS interrogation signals and backscattered light.

[0034] Identifying Cable Sections: The location of fiber optic cable sections is determined by the location of the utility poles that suspend at least a portion of the optical cable and optical sensor fiber. The GPS location of the utility poles can be obtained from the pole owners, such as electric utilities, if such data / information is available. Advantageously, the GPS location of the utility poles can be determined at the time of installation. Therefore, the GPS information can be preloaded into the graphical user interface (GUI) that is part of the DFOS / DAS system. Alternatively, if GPS location is not used, the distance converted to the sensor fiber length from the interrogator can be used by performing a hammer test on the utility pole and mapping the longitudinal position along the sensor fiber.

[0035] For example, when a utility pole is struck by a hammer or other striking instrument, the resulting vibrations propagate (in two directions) from the pole to the optical sensing fiber, forming a "V" shape. In the waterfall image received / generated by the DFOS / DAS system, the position corresponding to the tip of the "V" is the location of the struck utility pole in terms of optical sensor fiber length from the interrogator. Of course, such distance measurement and location determination differs from GPS coordinate location, which is independent of fiber length.

[0036] Historical flood level data collection: Historical flood level data and corresponding rainfall intensity and rainfall duration can be obtained from the National Weather Service. These data serve as labels for machine learning development.

[0037] Data Processing Operations

[0038] Rain intensity prediction based on linear regression.

[0039] 3 is a schematic flow diagram illustrating an exemplary method for predicting rain intensity according to an embodiment of the present disclosure. The prediction is based on a linear regression model. According to the present invention, both training data and holdout test data are used. Features of the training data are extracted and a regression model is fitted. During testing, features of test acoustic waves are extracted as one of four classes: no rain, light rain, medium rain, and heavy rain, and the trained regression model is expected to predict the corresponding rain intensity.

[0040] Flood level monitoring based on random forests

[0041] A supervised learning technique (random forest model) according to an aspect of the present disclosure is implemented. Random forest uses data in a tabular format, which is an ensemble of decision trees. Each decision tree in the ensemble processes sampled data and predicts an output label ("flood level" in this example). The decision trees in the ensemble are independent and can predict the final response.

[0042] The random forest model used in this disclosure is implemented in Scikit-learn. The mathematical details of the random forest model are as follows: for each decision tree, Scikit-learn calculates the importance of the nodes based on Gini importance:

number

[0043] where

[0044] ni j is the importance of node j

[0045] W j is the weighted number of samples arriving at node j

[0046] C j is the impurity of node j

[0047] left(j) is the child node from the left split of node j

[0048] right(j) is the child node from the right split of node j.

[0049] We can calculate the sum of feature importances in each tree and divide by the total number of trees:

number

[0050] where

[0051]

number

[0052]

number

[0053] T is the total number of trees.

[0054] When identifying flood levels, we need to know which flood level (group) an observation belongs to. This is a typical case of a multi-class classification problem, since there are two or more flood levels to predict. We can use the Random Forest classification function built into the Scikit-learn library to predict flood levels.

[0055] 4 is a schematic flow diagram illustrating an exemplary random forest model for flood level prediction according to an embodiment of the present disclosure. From this diagram, the following operations can be understood.

[0056] Acquisition of raw data sets: The raw DFOS / DAS vibration signals resulting from vibration events (no rain, light rain, moderate rain, heavy rain) occurring along the optical sensor fiber can be stored in a networked, cloud, or local storage system to capture and store the raw signals.

[0057] Creating dependent variable classes: Since Random Forest can only predict numeric values, we convert the flood levels from "Level 1 (Alert, Wait)", "Level 2 (Prepare)", "Level 3 (Evacuate)..." to numeric levels, i.e., [0,1,2,...].

[0058] Feature Extraction and Output: Split the dataset into independent and dependent variables. In the dataset organized in tabular form, the first three columns are the independent variables (rain intensity, rain duration, and past flood level) and the last column "flood level" is the dependent variable, and these values ​​are converted from a data frame to an array for future use.

[0059] Splitting training and test data: The amount of data is large enough that 80% of the data is used for training and the remaining 20% ​​is used as test data.

[0060] Feature scaling: A standard scale operation is used, which subtracts the mean of the observations and divides it by the unit variance of the observations.

[0061] Train the model: Define the parameters for random forest training, for example, define five trees for the random forest, define a loss function to measure the quality of the splits, and define a seed for randomizing the dataset. Finally, train the random forest using both the dependent and independent datasets.

[0062] Calculating the model score: First, we use the test feature set to predict the "flood level" class of the test data. To predict the class, we use the prediction function of a random forest classifier. Then, we convert the predicted values ​​and the numerical classes of the test observations into text data. We evaluated the performance of the classifier using a confusion matrix.

[0063] Online Web Server

[0064] The online web server provides the user interface for end users. As previously mentioned, the system and method of the present invention can transmit three different flood level alerts. These alerts can advantageously be transmitted from a cloud service in communication with the system.

[0065] The first such flood level alert is for Flood Level 1 (a condition indicating that water levels are increasing more rapidly than normal). The second flood level alert is for Flood Level 2 (a notice to government agencies and the public to prepare to evacuate). Finally, the third flood level alert is for Flood Level 3 (a notice to the public and others to evacuate immediately).

[0066] FIG. 5 is a schematic diagram illustrating an exemplary online web server for flood level monitoring according to an embodiment of the present disclosure. As shown in FIG. 5, the web server of the present invention is connected to the Internet from which it can obtain historical data and also obtain real-time data from DFOS / DAS operations. As previously described, predicted flood levels are determined from meteorological conditions and historical data evaluated by the operational model of the present invention. If it is determined that the predicted flood level exceeds a threshold level, an appropriate flood level alert is generated and an alert notification is sent to the appropriate parties.

[0067] FIG. 6 is a schematic flow diagram illustrating an exemplary operation of rain intensity and flood monitoring using DFOS / DAS and ML according to an embodiment of the present disclosure. With reference to FIG. 6, the overall operation of the system and method of the present invention can be understood. In a first step, a DFOS / DAS interrogator is connected to an optical sensor fiber that is at least partially an aerial cable path. In a second step, utility pole location is performed in conjunction with DFOS / DAS operation, using GPS or DFOS / DAS to determine location by mechanical impact on the utility pole. In a second step, meteorological data, such as whether rain will fall and the amount of rain, is collected. In a second step, rain intensity is predicted. In a further step, rain duration is determined. Finally, a flood forecast is presented and alerts / warnings are sent to relevant parties.

[0068] 7 is a schematic diagram illustrating exemplary operational features of rain intensity and flood monitoring using DFOS / DAS and ML, according to an embodiment of the present disclosure. As previously mentioned, the systems and methods of the present invention solve problems associated with predicting flood events resulting from unpredictable weather-related events. The data-driven road flood warning system and method of the present invention advantageously operates without requiring external sensors (other than deployed fiber optic cables) for data collection. Flood control and warnings are provided without the need for extra communication channels for data collection / control / transfer.

[0069] As described above, the system and method of the present invention uses DFOS / DAS to detect / measure rain noise and vibration data in real time. Advantageously, the DAS utilizes the response of the fiber optic cable to the physical disturbance caused by falling raindrops. The raw signals are filtered, normalized, and thresholded to remove noise, and the resulting clean rain noise / vibration waveforms are used for flood forecasting.

[0070] Acoustic waveforms are collected / recorded under different rain intensities. A regression model is then used to predict rain intensity. A random forest model is used to predict flood levels, and the prediction results are reported in real time on a flood level map along the fiber optic cable route. The system and method of the present invention operates autonomously and can prepare personnel, assistance, and emergency services for predicted floods.

[0071] While the present disclosure has been presented above using some specific examples, those skilled in the art will recognize that the present teachings are not so limited. Accordingly, the present disclosure should be limited only by the scope of the claims appended hereto.

Claims

1. operating a distributed fiber optic sensing (DFOS) system along the route of interest to receive rain-related vibration data from the aerial cable; Predicting rain intensity using the trained linear regression model; and Predicting flood levels using random forest models; and outputting an alert if the predicted flood level exceeds a threshold.

2. The method of claim 1 , further comprising training the linear regression model using rain intensity and duration training data.

3. The method of claim 2 , further comprising extracting features from the training data according to four classes including no rain, light rain, medium rain, and heavy rain.

4. The method of claim 3 , wherein the flood levels predicted by the random forest model include levels 1 (wait), 2 (prepare), and 3 (evacuate).

5. 5. The method of claim 4, further comprising dividing the dataset into dependent and independent variables, wherein rain intensity, rain duration, and historical flood level are the independent variables and flood level is the dependent variable.

6. 6. The method of claim 5, further comprising training the random forest model using a dependent variable data set and an independent variable data set.

7. The method of claim 6 , further comprising outputting the alert using a real-time flood map along the target route.

Images