Description:FIELD OF THE INVENTION
[0001] The present invention relates early warning method and system for
prediction of floods. More specifically, the early warning method and system uses
artificial intelligence, machine learning and geospatial data analysis to predict
5 floods.
BACKGROUND OF THE INVENTION
[0002] Flooding is a phenomenon, which causes large scale destruction of human
lives and property. Large scale floods have caused destruction of human lives,
animal lives, property, and environment causing in a huge financial impact. The
10 intensity and frequency of floods are increasing due to climate change, rise in
temperature of earth, haphazard constructions and extreme meteorological
conditions. Flash flooding, which caused by sudden burst of clouds, unprecedented
high rainfall, causes the water bodies to rapidly overflow and cause mass scale
destruction. Climate change scientists and experts are working constantly to
15 accurately predict flood occurrences in order to furnish proper response to mitigate
its impact and save lives and property. To plan and implement effective flood
mitigation strategies, it is necessary to accurately predict floods and their
occurrence. There are various structural or non-structural base methods to prevent
floods and accurate flood forecasting can provide mitigation and flood control
20 strategy at both local level and national level. Real-time data collection mechanisms
coupled to flood control systems can help in accurately predicting the floods.
[0003] Several scientific techniques are used for the forecasting and control of
floods. These techniques are used widely from decades in developed countries,
which provide early information of flood event occurrence. Alerts are triggered
25 from early warning prediction systems that collect and process data captured by the
river gauges, IMD predictions, rainfall threshold at a very coarse level. The machine
learning models disclosed here implement supervised and unsupervised
classification followed up by clustering, regression and classification to predict
floods.
30 SUMMARY OF THE INVENTION
3
[0004] A flood control and prediction system and method is disclosed. The flood
control and prediction system and method uses a set of statistical techniques and
procedures to solve the problem of predicting floods. The flood control and
prediction system comprises a processor and a memory. The memory includes a
5 flood analytics module having software instruction and logical divided into
different modules that when executed by a processor predicts the occurrence of
flood. The flood control and prediction system includes a data collection and
aggregation module that receives data from different sources, which is analyzed in
a statistical analysis module to eliminate statistical dependency in the received data.
10 A feature engineering module extracts relevant features for received data. A data
integration module integrates data from other sources to make prediction of floods.
A relevance vector machine prediction regressor module is uses the received data
for building a model for prediction. In some embodiments, the relevance vector
machine prediction regressor module is provided with a threshold values as an input
15 for at least one variable or features. The output from the relevance vector machine
prediction regressor module is provided to an analytical engine, which implements
artificial intelligence algorithms and machine learning algorithms to predict flood
based on the received data from feature engineering module, the data integration
module, the relevance vector machine prediction regressor module and a geospatial
20 analysis module. A display module integrates the prediction of floods with
geospatial data to provide insight related to floods for a given geographical area.
[0005] In some embodiments, the analytical module includes an analytical
database, an artificial intelligence module, a rule based engine and a
recommendation engine. The artificial intelligence and machine learning model and
25 is set to predict the flood’s susceptibility in river watershed. Machine learning
model(s) may be developed, trained and tested using relevant variables and finally
utilized for predicting the floods ahead of time.
[0006] In embodiments, a digital elevation model (DEM) for study and visual
depiction of geographical area is created by the flood analytics module. A Digital
30 Elevation Model (DEM) is a digital dataset in the form of an image that represents
a continuous topographic elevation surface through a series of pixels. Each pixel in
4
a DEM represents the elevation (z) of a feature at its location (x and y). DEMs are
a “bare earth” representation because they only contain information about the
elevation of ground features.
[0007] In some embodiments, the geographical area is divided into different
5 watersheds using the geospatial techniques. The DEM of each watershed is utilized
for calculation the slope and aspect map. The slope and aspect maps are used to
calculate the direction that the water would take to flow into the river.
[0008] In some embodiments, a separate map of the rivers for each geographical
area is prepared. The historical record of floods for each river is analyzed. In some
10 embodiments, the historical record of floods for each river is utilized for training
artificial intelligence algorithms for prediction of floods.
[0009] In some embodiments, a sparse Bayesian machine learning technique called
the Relevance Vector Machine Predictor (RVMP) is used for the prediction of
floods. The multivariate RVMP is first trained using test data and based on the deep
15 learning is used to predict flood events.
[00010] In some embodiments, the system and method may use highresolution data elevation model for geospatial analysis of the geographical area and
robust machine learning technique for flood prediction. A one or more gauging
stations are installed at one or more locations along the river site. The one or more
20 gauging station collect hydrological data and meteorological data such as rainfall,
water level, water discharge, humidity, temperature, wind speed, and wind direction
etc. The data collected form one or more gauging station are analyzed and along
with geospatial data is feed to the analytical engine for prediction of floods. Based
on the prediction, the computer implemented flood prediction system and method
25 provide timely information to avoid/minimize the losses.
[00011] The flood prediction system improves the forecasting of floods and
provides an early warning system. The invention provides the flood forecasting at
multiple lead times ranging between 1 to 4 days in advance.
[00012] In one embodiment, the flood prediction system includes a device, a
30 GSM module, a server, an electronic device, and a cloud infrastructure. The device
has a microcontroller with a board, a plurality of sensors, and a navigation module.
5
The sensors collect the environmental data and the meteorological data. The
microcontroller receives data collected from the sensors and inserts the location
information from the navigation module. The microcontroller processes the data for
forecasting the flood conditions based on an artificial intelligence module, which
5 implements deep learning algorithms. Based on the prediction of floods, the flood
control and prediction system provides an alert to the different users regarding the
flood conditions.
[00013] In some embodiments, the one or more sensor may be a water level
sensor installed in the gauging station to sense the water levels in a river.
10 [00014] In some embodiments, the one or more sensors may be water-flow
sensor measures the amount of water passing and the flow rate of the water.
[00015] In some embodiments, the one or more sensors may be a wind speed
and wind direction sensor measures wind direction and speed.
[00016] In some embodiments, the one or more sensors may be a rain gauge,
15 which detects precipitation at a set location for a predefined period.
[00017] In some embodiments, the one or more sensors may be a
temperature-humidity sensor records temperature and humidity values.
[00018] In some embodiments, the one or more sensors may be a
combination of at least different sensors the water level sensor, the water-flow
20 sensor, the wind speed and wind direction sensor, and the rain gauge.
[00019] In some embodiments, the GSM module communicates an alert to
at least one electronic device linked to the flood control and prediction system. The
cloud infrastructure stores the collected parameters received from one or more
sensors and data collected by the microcontroller based on time and location of one
25 or more sensors.
[00020] In some embodiments, the flood control and prediction system may
include a server. The server can access the data collected from one or more sensors
along with location information from one or more sensors of different types and
then analyzes it to predict the live fq1lood conditions.
30 [00021] In some embodiments, the flood control and prediction system may
be implemented over a cloud.
6
[00022] In some embodiments, the flood control and prediction system may
be implemented over a distributed network.
[00023] The flood control and prediction system assimilates and collects data
and information, which shows the hydrological status of river basins. The high
5 flood level, danger level, water level and rainfall are provided as inputs to the
RVMP model. The meteorological data like rainfall and temperature are weighted
with the catchment run-off, which is represents computation obtains the forecast
result of water level in gauge.
[00024] In one embodiment, the RVMP model output provides predicted
10 water danger levels in river well in advance of flood event to occur.
[00025] In one embodiment, the alert is an email/text
message/call/audio/radio alert message or a video message in advance based upon
the model output results.
[00026] In one embodiment, the hydro-meteorological parameters include at
15 least one parameter or group of parameters from water level, water discharge, water
flow, temperature, humidity, wind speed, and precipitation in flood-susceptible
region.
[00027] In one embodiment, the navigation module includes a GPS (Global
Positioning System).
20 [00028] The multivariate Relevance vector machine predictor (RVMP) is
able to predict the water levels at least 4 days into the future. If the forecasted result
is more than or equal to the declared danger level, a flood is declared.
[00029] In some embodiments, the machine learning algorithms
implemented in the analytics engine may uses the historical data recorded at a one
25 or more gauging stations. The historical data recorded at each gauging station is
utilized for training the machine learning algorithms for prediction of floods. In
addition, the RVMP model developed facilities the forecasting of water levels on
days on “d”, “d+1”, “d+2” and “d+3” (four days in the future). Additionally, the
RVMP model requires the water level of the river in meters (mts), discharge in
cubic meter per second (m3 30 /sec.), rainfall in millimeters (mm), and temperature in
degree centigrade (°C). The model is produced with a limited set of variables, which
7
can be assimilated and utilized to build a robust generalized model. The invention
provides an opportunity to laws makers, planners and other stakeholders for earlystage adoption of flood control measures to protect life and financial loss.
[00030] In some embodiments, statistical validation has been carried out to
5 optimize the model to evaluate the forecasting performance of the model by using
the Root Mean Square Error (RMSE). In the method, a cross validation technique
is used to determine the model parameters
[00031] This invention provides an improvement to the existing
conventional flood early warning system by integrating new information sources.
10 In addition, the machine learning algorithms implemented by the analytical module
increase warning time before the occurrence of flood.
[00032] In embodiments, the flood control and prediction system
implementing machine learning algorithms may act in two phases. In first phase,
the flood control and prediction system may be trained for prediction referred as a
15 training phase. In second phase, the flood control and prediction system may enter
into a test phase. Training phase require prediction parameters and the test phase of
the model takes trained/learned parameters as an input to generate accurate
predictions.
[00033] In embodiments, the prediction can be provided next 4 days and the
20 predictions may be updated every day. The data is collected from different sources
with are available publicly, one or more sensors and the relevance vector machine
predictor model for a defined period. The results are validated using RMSE and
predicted water levels are used to generate timely warnings.
25 BRIEF DESCRIPTION OF THE DRAWINGS
[00034] Fig. 1 illustrates the environment of a computer implemented flood
prediction system in an embodiment of the present invention;
[00035] Fig. 2 illustrates the different components of a flood prediction
30 system in an embodiment of the present invention;
8
[00036] Fig. 3 illustrates different modules of a flood analytics module in an
embodiment of the present invention;
[00037] Fig. 4 illustrates different modules of a data collection and
aggregation module in an embodiment of the present invention;
5 [00038] Fig. 5 illustrates different modules of a statistical analysis module in
an embodiment of the present invention;
[00039] Fig 6 illustrates the process of prediction of floods using the
computer implemented deep learning algorithms in an embodiment of the
invention;
10 [00040] Fig. 7 illustrates a process of flood prediction using a threshold value
in an embodiment of the present invention;
[00041] Fig. 8 and Fig. 9 illustrate a comparison of predicted water levels by
using a multivariate relevance vector machine prediction model and actual water
levels in an embodiment of the present invention;
15 [00042] Fig. 10 and Fig. 11 illustrates a comparison between the test results
of predicted water levels using a multivariate relevance vector machine prediction
model against real water levels by providing unseen real data input in an
embodiment of the present invention;
[00043] Fig. 12 illustrates process of flood prediction using a threshold
20 parameters of each variable in an embodiment of the present invention;
[00044] Fig. 13 illustrates a multivariate relevance vector machine prediction
model for predicting floods in another embodiment of the present invention; and
[00045] Fig 14 illustrates an exemplary input dataset for prediction of floods
in an embodiment of the present invention.
25
DETAILED DESCRIPTION OF THE INVENTION
[00046] As used herein any reference to “one embodiment” or “an
embodiment” means that a particular element, feature, structure, or characteristic
30 described in connection with the embodiment is included in at least one
embodiment. The appearances of the phrase “in one embodiment” or "in one
9
implementation" or "in variation of the implementation" at various places in the
specification are not necessarily all referring to the same embodiment.
[00047] As used throughout this description, the word “comprises,”
“comprising,” “includes,” “including,” “has,” “having” or any other variation
5 thereof, are intended to cover a non-exclusive inclusion. For example, a process,
method, article, or apparatus that comprises a list of elements is not necessarily
limited to only those elements but may include other elements not expressly listed
or inherent to such process, method, article, or apparatus. The use of “a” or “an” are
employed to describe elements and components of the embodiments herein. This is
10 done merely for convenience and to give a general sense of the invention. This
description should be read to include one or at least one and the singular also
includes the plural unless it is obvious that it is meant otherwise.
[00048] Embodiment of invention discloses an Artificial Intelligence (AI)/
Machine learning (ML) flood forecasting method. In order to solve the problem,
15 the embodiment of invention discloses the artificial intelligence and machine
learning method, the multivariate relevance vector machine predictor (RVMP)
using collected data from flood management information system (FMIS) for the
flood prediction.
[00049] Fig. 1 illustrates the environment of a flood prediction system in an
20 embodiment of the present invention. The flood prediction environment 100
includes a flood prediction system 110 connected with different geographical areas,
a geographical area 102 and a geographical area 104. Each geographical area such
as the geographical area 102 comprises one or more gauging stations 120A and one
or more rain gauge station 120B. Likewise, the geographical area 104 comprises
25 one or more gauging station 120C and a rain gauge station 120D. Furthermore, the
flood prediction system 110 is connected a server 112, a database 114, and a cloud
computing environment 118 and other electronic processing devices through a
network 108. The flood prediction system 110 collects and aggregates data related
to water levels in the river and its adjoining areas in meters (mts) or in centimetres
30 (cms). In addition, the flood prediction system 110 collects and aggregates data
related rainfall in millimeters (mms) for each of the geographical areas such as the
10
geographical area 102 or the geographical area 104. The operating prediction
environment for flood 100 of the flood prediction system 110 is exemplary but in
other implementations the operating prediction environment 100 may include
additional or fewer components than shown.
5 [00050] In some embodiments, the flood prediction system 110 may reside
on the server 112.
[00051] In another embodiment, the flood prediction system 110 may be
implemented over a distributed environment or on a cloud computing environment
118.
10 [00052] In some embodiments, the flood prediction system 110 may be
connected to one or more databases 114. The one or more databases 114 may
include an ancillary database, which may store information related land use and
land cover for a particular geographical region. Land cover data is associated with
the physical characteristics of Earth’s surface such as forest, tidal wetland,
15 urbanized region, or grassland. Likewise, land use data defines the utilization of the
land by people whether it is for agricultural, recreational, residential, commercial,
or industrial purposes. For example, the ancillary database may store information
for each of geographical area related to land use, drainage system, road network,
weather conditions, best management practices for structures etc. The ancillary
20 database in at least one implementation may include additional information related
to geospatial data about weather, funds received by different agencies, flood
threshold indicators and other factors associated with the floods.
[00053] In one embodiment, the flood prediction system 110 may employ
Bayesian statistics for evolutionary computation as a modeling tool and combine it
25 with additional ancillary data related to weather conditions, drainage conditions,
and land use.
[00054] Fig. 2 illustrates the different hardware components of a flood
prediction system in an embodiment of the present invention. The flood prediction
system 110 includes a memory 204, a one or more processor 218, an input/output
30 module 220, a communication module 222, an internal bus 214 and an external
interface 224. The internal bus 214 allows exchange of data and electrical power
11
between the memory 204 and the processor 220, the input/output module 214 and
the communication module 222. Additionally, the external interface 218 allows
communication with external devices such as the server 112 or data received from
database 114 or from the cloud computing environment 118. The memory 204 may
5 include one or more operating systems 208, one or more applications 210, and a
flood analytical module 212. This is one exemplary hardware configuration of the
flood prediction system 110, however, in other implementations, the flood
prediction system 110 may have additional or lesser number of modules.
[00055] The memory 104 may include an operating system 208, one or more
10 applications 210, and a flood analytics module 212 apart from other modules. The
operating system 208 may be a windows OS, Macintosh OS, Linux OS or some
other type of operating system. The one or more applications 210 may aggregate
data related to weather, water level, prediction analytics, flood analysis and water
resource management. The flood analytics module 212 may include machine
15 learning algorithms, artificial intelligence algorithms and other forecasting
algorithms for flood prediction and flood analysis, land use planning, and resource
management. The flood analytics module 212 may other modules for flood
prediction.
[00056] Fig. 3 illustrates different modules of a flood analytics module in an
20 embodiment of the present invention. The flood analytics module 212 may include
a data collection and aggregation module 302, a statistical analysis module 304, a
feature engineering module 306, a data integration module 308, a relevance vector
machine prediction regression module 310, a geospatial analysis module 312, an
analytical engine 320, an external database 328 apart from other modules.
25 [00057] Fig. 4 illustrates different modules of a data collection and
aggregation module in an embodiment of the present invention. The data collection
and aggregation module 302 may collect data from different geographical areas and
regions such as the geographical area 102. In addition, the data collection and
aggregation module 302 may receive data from external sources such as but not
30 limited to external database 328, which includes historical data related to land use,
12
land cover, floods and other flood related data for different geographical areas
and/or geographical regions.
[00058] Referring to Fig. 4, the data collection and aggregation module 302
may include a proprietary data module 402 comprising proprietary data related to
5 floods, land use and land cover and other data related to floods; a hydrologic data
module 404 comprising precipitation liquid and solid (3 or 6 hourly), discharge
(naturalized flow, i.e. stream flow corrected for manmade storage; daily), lakes and
reservoir levels, soil moisture, snow water equivalent, snow cover area, snow depth,
and evapotranspiration data; a gauge location data module 408 comprising data
10 collected from different type of gauges such as rain gauge, stream gauge, radio
gauge and other measuring gauges installed at different locations in a predefined
area such as region, state, district or a block. A demographic data module 410
comprising data collected from different demographic databases related to
demographics of the region. A land use data module 412 comprising data collected
15 from different satellite imageries for recent land cover. The data collection and
aggregation module 302 further includes research data module 414 comprising of
research data and research models from different research institutions related to
improvement of accuracy of flood prediction. A stakeholder data module 418
comprising data about different stakeholders associated with management, control
20 and maintenance of water bodies, floods, and rivers. An ancillary data module 420
comprising data related to geospatial data. The geospatial data describe any data
related to or containing information about a specific location on the Earth’s surface.
The information may be related to descriptive information about a location such as
points, lines and polygons, for example, vectors and attributes; a collection of co25 located charted points that can be recontextured as 3D models, for example, point
clouds; high-resolution images of our world, for example, satellite imagery, census
data tied to specific geographic areas, computer aided design images of buildings
or other structures, delivering geographic information as well as architectural data,
for example, drawn images. In addition, the ancillary data module 420 may include
30 data about weather, the fund received by the resources, the flood threshold
indicators and other parameters related to flood measurement.
13
[00059] Fig. 5 illustrates different modules of a statistical analysis module in
an embodiment of the present invention. The statistical analysis module 304 may
analyze land use data, land cover data, geospatial data for different geographical
regions for deriving statistical inferences related to data, for example, correlated
5 data. The statistical analysis module 304 includes a variable analysis module 502,
a statistical data cleaning module 504, and a dimensionality reduction module 508.
[00060] The variable analysis module 502 removes different data anomalies
in the data, for example, outliners, missing values and other data discrepancies. The
statistical data cleaning module 504 may apply different statistical techniques to
10 identify different relationships in data, for example, correlation in the data. The data
that has been removed of various data discrepancies and relationship is passed to
the data evaluation and extraction of different features to be used for training and
prediction of floods. The dimensionality reduction module 508 reduced the number
of variables so that variables that do not contribute significantly to the predication
15 and data model are eliminated.
[00061] In some embodiments, the statistical analysis module 304 may
receive data in different formats and convert it into ASCII format for analysis and
interpretation. In addition, the data received from remote sensing satellite may
provide additional information related to geospatial data such as data related to
20 weather conditions, road conditions, and land use conditions and other data related
to flood prediction.
[00062] The feature engineering module 306 may identify different features
from the data received from the statistical analysis module 304. In embodiments,
the feature engineering may involves inputting missing values, encoding
25 categorical variables, transforming and discretizing numerical variables, removing
or censoring outliers, and scaling features, among others. In addition, the feature
engineering module 306 may extract features related to geographical points of
interests like, rivers, dams, locations of Water Level (WL) sensors, locations of rain
gauges and other aspects to be used for training the decision tree model to perform
30 bivariate classification. In some embodiments, the flood analytics module 212 may
predict water levels to generate early warning and to identify danger hotspots.
14
[00063] The data integration module 308 may assimilate data for different
sources, for example, ASCII data of maps. In one implementation, the data
integration module 308 may also receive and integrate geospatial data. Likewise,
the data integration module 308 may add ancillary data to create meaningful
5 analysis to produce different analytics related to environment, accessibility,
location, and weather for flood prediction. The data integration module 308 may
also receive additional data from the external database 328, which may include
digital elevation model data is used as an input with predefined levels of granularity.
In some embodiments, the combined use of assimilated data or integrated data and
10 the geospatial may be used to train the machine learning algorithms for accurate
flood forecasting.
[00064] The relevance vector machine prediction regressor module 310 may
receive data from data integration module, the geospatial analysis module 312 and
other modules to allow a user to draw inferences based on the set goals.
15 [00065] In embodiments, the set goals may be prediction of flood, outburst
of flash floods, forecasting of floods based on global warming, water logging and
landscape changes due to forecasted floods or some other prediction related to
floods.
[00066] The relevance vector machine prediction module 310 also is
20 connected with the analytical engine 320. The analytical engine 320 includes an
analytical database 314, an artificial intelligence module 318, a rule-based engine
322, and a recommendation module 324. The rule-based engine 322 implements
different rules for creating one or more flood forecasting models. The
recommendation module 324 provides one or more insights to the user for
25 prediction of floods. The analytical database 314 may include data related to floods
for different geographical area such as the geographical area 102 and stores
different analytical models created using artificial intelligence algorithms. The
artificial intelligence module 318 may apply analytical models prediction of flood
using real time data.
15
[00067] In some embodiments, the relevance vector machine prediction
module 310 and the flood prediction engine 320 may work in tandem to produce
flood prediction.
[00068] The geospatial analysis module 312 receives flood prediction data
5 from the analytical engine 320. The flood data received from different modules
such as the feature engineering module 306, the data integration module 308, and
the relevance vector machine prediction regressor module 310 are analyzed in the
analytical engine 320. The artificial intelligence module 318 and the rule based
engine 322 apply machine learning algorithms to predict the set objectives.
10 Example of set objective may be the occurrence of flash flood in a given
geographical area 102. Based on the models stored in the analytics database 314,
the analytical engine 320 predicts and makes recommendations.
[00069] In some embodiments, the regression and flood forecasting data can
be in raw format such as shape file, tiff, other raster or vector formats after the
15 analysis of the collected data. The resultant data is analyzed to recreate visual
dashboards by converting the tiff format back to digital numbers, which provide
insights related to hotspots linked with other features such as demographics etc.
This recreation of visual data can happen in geospatial analysis module 312 or
alternatively can be processed in the analytical engine 320, which is finally passed
20 to a display module 330. In some embodiments, the analyzed data and processed
data may be aggregated from different modules and provided to the display module
330 for creation one or more dashboards.
[00070] In some embodiments, the geospatial analysis module 312 may also
add additional information based on external sources such as but not limited to
25 intelligence received from remote sensing satellite and may also produce geo
referenced and projected images. In some embodiments, the flood forecasting
engine 320 may be associated with the user interface, which may provide visual and
text information related to flood prediction to the user.
[00071] Fig 6 illustrates the process of prediction of floods using the
30 computer implemented deep learning algorithms in an embodiment of the
invention. The process 600 is initiated at step 602 and moves to step 604. At step
16
604, the process 600 receives one or more meteorological parameters 608A and one
or more hydrological variables 608B. The process 600 may receive additional flood
related data as inputs, for example, geospatial data, global warning data, climate
change data or historical data related to rainfall and floods.
5 [00072] The training phase involves validating the one or more
meteorological parameters 608A and one or more hydrological variables 608B. In
addition, the training phase may incorporate additional flood related data as inputs,
for example, geospatial data, global warning data, climate change data or historical
data related to rainfall and floods to identify feature relevant to prediction of floods.
10 [00073] At step 610, the process 600 pre-process the one or more
meteorological parameters 608A and one or more hydrological variables 608B. In
some embodiments, the preprocessing of data may involve statistical analysis and
feature engineering. At step 612, the process 600 uses the receive data to training
the multivariate relevance vector machine prediction model. Training the
15 multivariate relevance vector machine prediction model is an iterative process and
may involve identifying relevant features that can be used to optimize the prediction
model. The training process involves hyper parameter tuning at step 614, which
involves hyper parameters that cannot be directly learned from the regular training
process and which are usually fixed before the actual training process. At step, 618,
20 the process produces the results for training output.
[00074] At step 620, the process 600 learns the prediction parameters to be
used during testing and/or prediction phase. Once the multivariate relevance vector
machine prediction model has been trained for prediction and the parameters to be
used as inputs are identified. After the training phase concludes the learned
25 parameters are utilized by the test phase. The process continues and the test phase
begins where new and previously unseen inputs are fed to the multivariate relevance
vector machine prediction model for making predictions on the test dataset, which
is the water level in meters in the river. At step 622, the process 600 can be used for
predicting one or more set objectives related to flood prediction.
30 [00075] At step 622, the process 600 is provided with previously unseen data
as input for prediction. At step 624, the process 600 utilizes the trained multivariate
17
relevance vector machine prediction model for prediction. At step 628, the process
600 predicts the river water level based on the unseen data on the basis of previous
training provided to it. The process 600 terminates at step 630.
[00076] Fig. 7 illustrates a process of flood prediction using a threshold value
5 in an embodiment of the present invention. The process 700 starts at step 702 and
moves to step 704. At step 704, the process 700 collects the hydrological parameters
and meteorological parameters 708. In addition, the real time monitoring unit 704
may collect other addition parameters in real time. The hydrological parameters and
meteorological parameters 708 and the data collected in the real time is used as an
10 input to the flood prediction. In some embodiments, the hydrological parameters
and meteorological parameters 708 may be collected using a data sensing unit 708
and other parameters may be collected using a real-time monitoring unit 704. The
real time monitoring unit 704 may also collect data using one or more sensors.
[00077] At step 710, the process 700 may be programmed with a value that
15 represent early warning setting related to flood prediction. In some embodiments,
early warning may be related to river water level, flood prediction, water logging,
flash floods, rain downpour in (mm) and other flood related early warning
indicators. In some embodiments, the early warning flood prediction threshold
values may be determined by the analytical engine 320 or may be manual setup by
20 a user or an operator. At step 712, the process 700 may evaluate if the set threshold
value has been achieved, if the early warning threshold value has not been achieved
then logic unit 712 loops back to step 710 to build up the model accuracy during
until a early warning threshold value(s) for one or more statistics is achieved. The
process 700 continues monitoring until the threshold level has been achieved.
25 Once, the threshold level has been achieved, the process 700 moves to preprocessing data step 718. At step 718, the data pre processing unit may also receive
historical monitored data 714 as input to improve the training of the model and
predict the accurate water levels. At step 720, the analysis and modeling of data is
performed using at least one multivariate model such as relevance vector machine
30 prediction model. At step 722, the trained model provides early warning
18
recommendation based on the previous training with the test data. The process 700
terminates at 724.
[00078] Fig. 8 and Fig. 9 illustrate a comparison of predicted water levels by
using a multivariate relevance vector machine prediction model and actual water
5 levels in an embodiment of the present invention. Fig. 8 and Fig. 9 show the results
of actual measured water level with the predicted water level by the model. In this
process, the relevance vector machine prediction model is trained using the
historical and recorded datasets and learns the pattern in the time series. The data
used for training includes historical data as input and includes parameters such as
10 water level, rainfall, danger level, and high flood level for particular gauging station
available on Flood Management Information System Centre (FMIS). River gauge
stations are fixed stations, set up to measure the river stages and discharge. These
data sets are important for planning, design, and management of water and natural
resources including the design and management of flood.
15 [00079] Fig. 10 and Fig. 11 illustrates a comparison between the test results
of predicted water levels using a multivariate relevance vector machine prediction
model against real water levels by providing unseen real data input in an
embodiment of the present invention. The comparison between the test results,
which is the predicted water levels against real water levels, using previously
20 captured unseen real data as input. The predicted results of the relevance vector
machine prediction model are compared with the known records to establish
accuracy of the results.
[00080] Fig 12 illustrates process of flood prediction using a threshold
parameters of each variable in an embodiment of the present invention. The process
25 1200 is initiated at step 1202 and immediately moves to step 1204. At step 1204,
the process 1200 collects data from the data collection and aggregation module 302
and from other sources such as external database 328. The collected data is passed
to the relevance vector machine prediction model at step 1208. at step 1210, the
process 1200 tunes the hyper parameters and at step 1214 the relevance vector
30 machine prediction model provides results related to prediction of floods, for
example, rise in water levels. At step 1218, the process 1200 validates data using
19
goodness-of fit test, which is conducted to test the degree of association between
the observed values and estimated data values. In addition, at step 1218, the process
1200 tests the mean absolute error (MAE), a linear measure, and root mean square
error (RMSE), a quadratic scoring rule, which are used to measure the average
5 magnitude of error. The index of agreement (IoA) and coefficient of efficiency
(CoE) are used to check the model performance. The IoA is calculated by
comparing an observed group variance with an expected random variance. It varies
from 0 (inferior model) to 1(excellent model). CoE ranges from -8 (inferior model)
to 1 (excellent model). Thus, a value of 0 for the CoE indicates that the observed
10 mean is as good an estimator as the model, whereas negative values indicate that
the observed mean is a better estimator than the model. At step 1220, the process
1200 determines if the threshold values of each of the data validation parameters
(RMSE, IOA, CoE, MAE) are violated (crossed form the minimum permissible
threshold value). If the threshold values are violated, then process 1200 moves back
15 to step 1210 and reevaluates and optimizes the hyper parameters. If at step 1220,
the process 1200 evaluates that the threshold values are within the permissible
threshold values, if so, /then the process 1200 at step 1222 provides result for
prediction of flood levels. The process 1200 terminates at step 1224.
[00081] Fig 13 illustrates a multivariate relevance vector machine prediction
20 model for predicting floods in another embodiment of the present invention. In this
embodiment, extension of 1-day prediction to 4-day prediction is implemented and
the prediction for four days in advance can be provided based on historical and
other data collected from different sources. Different inputs such as but not limited
to water levels, rainfall, and danger levels (thresholds) on the days d-4, d-3, d-2 and
25 d-1 from sensor are collected. The input data from different input sources such as
an input source 1320 (sensor data), an input source 1304 (rainfall data), and an input
source 1308 (water level) is provided to a multivariate relevance vector machine
flood prediction model 1310. Machine learning and artificial algorithm implement
in the analytical engine 320 utilize historical data recorded at one or more gauging
30 stations. The historical data collected at one or more gauging stations are applied to
the multivariate relevance vector machine flood prediction model to forecast the
20
water levels and generate output for four days (“d”, “d+1”, “d+2” and “d+3”) in
advance. An output prediction 1312 is used provided predicted output for (“d”,
“d+1”, “d+2” and “d+3”). The predicted output is utilized for taking corrective
action by relevant stakeholders. This allows policy makers and flood control
5 authorities to provide early-stage adoption of flood control measures to protect life
and financial loss.
[00082] Fig 14 illustrates an exemplary input dataset for prediction of floods
in an embodiment of the present invention. The dataset comprises date 1402, high
flood level 1402, danger level 1408, water level 1410, trend of water level 1412,
10 status of the water level 1414, and rainfall record 1418, in ‘centimeters’. The trend
of water level 1412 status of the water level is recorded daily and classified as rising
(R), falling (F), steady (S) and constant (C) as compared to the day recorded on the
previous day. This collected data is passed to the multivariate relevance vector
machine prediction model, which then provides prediction for four days in advance.
15 [00083] The invention can be modified into many variations in different
implementations and is not limited to different embodiments described herein.
Other variations that can be amended or modified are within the scope of this
invention. , Claims:WE CLAIM:
1. A flood control and prediction system to predict the occurrence of flood in
a geographical area, the flood control and prediction system comprises:
a processor (218);
5 a memory (104) having a flood analytics module (212), which comprises:
a data collection and aggregation module (302) to collected data
from one or more data sources;
a statistical analysis module (304) implementing statistical
transformation of the collected data;
10 a feature engineering module (306) receiving data from the
statistical analysis module (304) and extracting one or more feature for
prediction of floods;
a data integration module (308) configured to integrate data from
one or more sources;
15 a relevance vector machine prediction regressor module (310) for
building flood prediction model, wherein a threshold value for one or more feature
is set;
an analytical engine (320) comprises an analytical database (314),
an artificial intelligence module (318), a rule based engine (322) and a
20 recommendation module
(324), wherein the analytical engine (320) predicts the occurrence of the
flood based the inputs received from the feature engineering module (306), the data
integration module (308), the relevance vector machine prediction regressor
module (310) and a geospatial analysis module (312); and
25 a display module (330), which integrates the geospatial data with the
prediction of occurrence of flood.
2. A flood control and prediction system as claimed in claim 1 includes an
ancillary data module (420) which stores information related land use and land
cover for the geographical area.
30 3. A flood control and prediction system as claimed in claim 1 includes a one
or more real time monitoring data sensors.
22
4. A data collection and aggregation module (302) as claimed in claim 1
receives real time hydrological data and meteorological data.
5. A computer implemented method for prediction to the occurrence of flood
in a geographical area, the computer implemented method for prediction to the
5 occurrence of flood performing the steps of:
collecting data from one or more sources through a data collection and
aggregation module (302);
applying statistical techniques to filter the collected data using a statistical
analysis module (304);
10 extracting relevant feature from the filtered data using a feature engineering
module (306);
combing data from external database (328) in a data integration module
(308), where in external database includes digital elevation data;
analyzing the data received from the data integration module (308) using a
15 relevance vector machine prediction regressor module (310) for building flood
prediction model by pre-defining a threshold value for one or more feature;
applying machine learning algorithms using an analytical engine (320),
wherein the analytical engine (320) predicts the occurrence of the flood based the
inputs received from the feature engineering module (306), the data integration
20 module (308), the relevance vector machine prediction regressor module (310) and
a geospatial analysis module (312); and
providing a visual depiction of the occurrence of flood for the geographical
area using a display module (330), wherein the visual depiction is based on a digital
elevation model.
25 6. The computer implemented method for prediction to the occurrence of flood
in a geographical area as claimed in claim 1 includes an ancillary data module (420)
which stores information related land use and land cover for the geographical area.
7. The computer implemented method for prediction to the occurrence of flood
in a geographical area as claimed in claim 1 includes a one or more real time
30 monitoring data sensors.
23
8. The computer implemented method for prediction to the occurrence of flood
in a geographical area as claimed in claim 1 receives real time hydrological data
and meteorological data.
Dated this 27 August 2023
5 Digitally Signed by:
ARVIND CHOPRA [IN/PA 2675]
AGENT FOR THE APPLICANT
To,
The Controller of Patents,
10 The Patent Office at New Delhi