CN114925923B - Method for improving accuracy of depth forecast of ponding in non-monitoring unit - Google Patents
Method for improving accuracy of depth forecast of ponding in non-monitoring unit Download PDFInfo
- Publication number
- CN114925923B CN114925923B CN202210592618.6A CN202210592618A CN114925923B CN 114925923 B CN114925923 B CN 114925923B CN 202210592618 A CN202210592618 A CN 202210592618A CN 114925923 B CN114925923 B CN 114925923B
- Authority
- CN
- China
- Prior art keywords
- monitoring
- monitoring unit
- unit
- value
- waterlogging
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 293
- 238000000034 method Methods 0.000 title claims abstract description 20
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims abstract description 61
- 238000013528 artificial neural network Methods 0.000 claims abstract description 46
- 239000002352 surface water Substances 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000009825 accumulation Methods 0.000 claims description 14
- 238000006243 chemical reaction Methods 0.000 claims description 12
- 230000009466 transformation Effects 0.000 claims description 11
- 238000002474 experimental method Methods 0.000 claims description 6
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 230000001788 irregular Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000004904 shortening Methods 0.000 claims description 2
- 230000002265 prevention Effects 0.000 abstract description 4
- 238000003062 neural network model Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000012806 monitoring device Methods 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/26—Government or public services
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
- Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Economics (AREA)
- Tourism & Hospitality (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Mathematical Optimization (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Pure & Applied Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Operations Research (AREA)
- Mathematical Analysis (AREA)
- Development Economics (AREA)
- Computational Mathematics (AREA)
- Game Theory and Decision Science (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Entrepreneurship & Innovation (AREA)
- Databases & Information Systems (AREA)
- Quality & Reliability (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
Abstract
The invention discloses a method for improving the accuracy of depth prediction of ponding of a non-monitoring unit, which obtains a ponding depth prediction value of a monitoring point by using a STARMA model; obtaining a surface water depth forecast value by utilizing an InfoWorks ICM model; dividing the earth surface unit into a monitoring unit and a non-monitoring unit according to whether the earth surface unit contains a monitoring point; determining a monitoring point corresponding to the monitoring unit; for the monitoring unit, a predicted value obtained by fusing an InfoWorks ICM model and a STARMA model is used as a final predicted value; and constructing a neural network for improving the accumulated water depth forecasting precision of the non-monitoring units, training the neural network by using the final forecast value of the monitoring units, the forecast value of the InfoWorks ICM model and the forecast value of the STARMA model, and improving the accumulated water depth forecasting precision of the non-monitoring units based on the neural network. The invention has wide application prospect in the fields of urban drainage waterlogging prevention, sponge cities, emergency management and the like.
Description
Technical Field
The invention relates to a method for improving the accumulated water depth forecasting precision of a non-monitoring unit, and belongs to the technical application fields of urban drainage waterlogging prevention, emergency management, sponge cities and the like.
Background
Urban inland inundation seriously threatens the life and property safety of people. The accurate surface water depth forecast can provide reliable decision basis for urban drainage waterlogging prevention, emergency dispatching and the like, so that waterlogging harm is effectively relieved.
The depth of surface water is generally simulated and forecasted by utilizing an urban rainfall FLOOD model, and the commonly used urban rainfall FLOOD model mainly comprises models such as MIKE FLOOD, infoWorks ICM, LISFLOOD-FP, XP-SWMM 2D and the like. Due to the fact that urban rainfall and underlying surfaces have strong spatial heterogeneity, high-precision data of the urban rainfall and underlying surfaces are difficult to obtain, and underground pipe network data are missing, wrong and the like, surface water depth prediction precision of an existing urban rainfall flood model is still to be improved. In order to meet the requirement of urban drainage and waterlogging prevention, a series of waterlogging monitoring devices are distributed in many cities and used for monitoring the waterlogging depth of urban waterlogging points. Actually, a monitoring point water accumulation forecasting model is established by using data such as precipitation, water accumulation and the like of waterlogging monitoring points, and high water accumulation depth forecasting precision can be obtained. The urban rainfall flood model and the monitoring point ponding forecast model have complementarity, and the two models are effectively combined to obtain a more accurate surface ponding depth forecast value.
The urban land surface is discretized into a series of land surface units, and the land surface units are divided into monitoring units and non-monitoring units according to whether waterlogging monitoring points are included. For the monitoring unit, the forecasting values of the urban rainfall flood model and the monitoring point ponding forecasting model can be fused by using a model fusion method (such as weighted average, bagging, boosting, stacking and the like), and the fused forecasting value is used as a final forecasting value, so that the ponding depth forecasting precision is improved. However, an effective technical means is lacked if the accumulated water depth prediction accuracy of the non-monitoring units is improved by using the prediction results of the urban rainfall flood model and the accumulated water prediction model of the monitoring points.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a method for improving the accuracy of the depth forecast of the ponding of a non-monitoring unit, wherein the method divides a surface unit into a monitoring unit and a non-monitoring unit; for the monitoring unit, a Bayesian model is used for weighted average to obtain a final accumulated water depth forecast value; and constructing a neural network for improving the prediction precision of the water depth of the non-monitoring unit on the basis.
The method for improving the accuracy of the depth forecast of the ponding of the non-monitoring unit specifically comprises the following steps:
step (1) waterlogging depth forecast at waterlogging monitoring point
Calculating by using a STARMA model of rainfall water to obtain a water accumulation depth forecast value of each waterlogging monitoring point, wherein the water accumulation depth forecast value of the waterlogging monitoring point based on the model is hereinafter referred to as a STARMA forecast value;
step (2) depth prediction and discretization of prediction value of surface water
Carrying out analog prediction on the surface water depth by utilizing an InfoWorks ICM model, carrying out spatial discretization on a surface water depth prediction value of the model, and shortening the surface water depth prediction value based on the InfoWorks ICM model to be an InfoWorks ICM prediction value;
step (3) surface discretization and classification
Discretizing the earth surface into a series of earth surface units in a discretization mode consistent with the step (2); dividing the surface unit into a surface unit containing the waterlogging monitoring point and a surface unit not containing the waterlogging monitoring point according to whether the surface unit has the waterlogging monitoring point or not; the surface unit containing the waterlogging monitoring point is hereinafter referred to as a monitoring unit, and the surface unit not containing the waterlogging monitoring point is hereinafter referred to as a non-monitoring unit;
step (4) determining the waterlogging monitoring points corresponding to the monitoring units
The invention assumes that one monitoring unit corresponds to one waterlogging monitoring point: for a monitoring unit, if the unit only comprises one waterlogging monitoring point, the waterlogging monitoring point is used as the waterlogging monitoring point corresponding to the unit; if the unit comprises a plurality of waterlogging monitoring points, one waterlogging monitoring point is 'virtually' taken out, the monitoring value of the virtual waterlogging monitoring point is the mean value of the monitoring values of the plurality of waterlogging monitoring points, the forecast value of the virtual waterlogging monitoring point is the mean value of the forecast values of the plurality of waterlogging monitoring points, and the virtual waterlogging monitoring point is taken as the waterlogging monitoring point corresponding to the unit.
Step (5) improving the accumulated water depth forecasting precision of the monitoring unit by using Bayesian model weighted average
For each monitoring unit, carrying out weighted average on an InfoWorks ICM predicted value of the monitoring unit and a STARMA predicted value of a corresponding monitoring point of the monitoring unit by using Bayesian model weighted average to obtain a fused predicted value, wherein the fused predicted value is used as a final accumulated water depth predicted value of the monitoring unit; calculating the weight of the InfoWorks ICM model and the STARMA model of the rainfall ponding by using an expectation-maximization algorithm;
step (6) constructing a neural network for improving the accuracy of non-monitoring unit water depth prediction
Constructing a neural network for improving the water depth forecasting precision of the non-monitoring units, and training the neural network by using the final forecast value, the InfoWorks ICM forecast value and the STARMA forecast value of the monitoring points corresponding to the monitoring units of the monitoring units; for any non-monitoring unit, the input of the neural network is the InfWorks ICM predicted value of the non-monitoring unit, the InfWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and the STARMA predicted values of the monitoring points corresponding to the n monitoring units, and the output of the neural network is the final accumulated water depth predicted value of the non-monitoring unit.
Preferably, the discretization mode in the step (2) is a regular grid or irregular triangular grid mode.
Preferably, the accuracy of the ponding depth forecast of the monitoring unit is improved by using a bayesian model weighted average, which specifically comprises the following steps:
and (5) for each monitoring unit, improving the water accumulation depth forecasting precision by executing (1) to (4).
(1) Normal conversion
Acquiring an InfWorks ICM predicted value sequence of a monitoring unit, a water accumulation depth monitoring data sequence of a monitoring point corresponding to the monitoring unit and a STARMA predicted value sequence of the monitoring point corresponding to the monitoring unit; carrying out normal conversion on the sequences by using Box-Cox transformation to obtain related sequences conforming to normal distribution;
(2) determining a Bayes model weighted average forecasting formula of the ponding depth as follows:
wherein D is the forecast amount, namely the depth of accumulated water; o is accumulated water depth monitoring data of a monitoring point corresponding to the monitoring unit; p (d) 1 I O) is the STARMA predicted value d of the monitoring point corresponding to the monitoring unit under the condition of given monitoring data O 1 Posterior probability of (2), i.e. weight value w of STARMA model of rainfall ponding 1 ;p(d 2 I O) is the predicted value d of the monitoring unit InfoWorks ICM under the condition of given monitoring data O 2 The posterior probability of (1), i.e. the weight value w of the InfoWorks ICM model 2 ;
Is a mean value d 1 Variance->Normal distribution of (2); />Is a mean value d 2 Variance->Normal distribution of (2); suppose O, d in the formula (1) 1 、d 2 All are subjected to normal conversion and accord with normal distribution;
(3) weight calculation
Calculating the data in the formula (1) by an expectation-maximization algorithm by using the related sequence data in the formula (1) which conforms to the normal distributionw 1 And w 2 ;
(4) Application of the method
For the monitoring unit, first, the Box-Cox transformation is usedCarrying out normal conversion on the STARMA predicted value of the monitoring point corresponding to the monitoring unit and the InfWorks ICM predicted value of the monitoring unit to obtain d 1 'and d' 2 Calculating to obtain a weighted average predicted value d' of the Bayes model by using a formula (2); and performing inverse Box-Cox transformation on the d', wherein the value after inverse transformation is used as a final accumulated water depth forecast value of the monitoring unit.
d'=w 1 d 1 '+w 2 d' 2 (2)
Preferably, the constructing of the neural network model for improving the accuracy of non-monitoring unit water depth prediction specifically comprises the following steps:
(1) determining neural network structure
The neural network adopts a three-layer structure and comprises an input layer, a hidden layer and an output layer, wherein the number of neurons in the hidden layer is a hyper-parameter and is determined through experiments;
(2) determining inputs and outputs of a neural network
For any non-monitoring unit, the input of the neural network is the InfoWorks ICM predicted value of the non-monitoring unit, infoWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and STARMA predicted values of monitoring points corresponding to the n monitoring units; the output of the neural network is the final accumulated water depth forecast value of the non-monitoring unit; n is a hyper-parameter and is determined through experiments;
(3) training neural networks
Firstly, a data set used by the neural network is constructed: for each monitoring unit, taking the InfoWorks ICM predicted value of the monitoring unit, the InfoWorks ICM predicted values of other n monitoring units closest to the monitoring unit and the STARMA predicted values of corresponding monitoring points of the n monitoring units as input parts of a data set, and taking the final accumulated water depth predicted value of the monitoring unit as an output part of the data set;
then, randomly dividing the data set into a training set and a test set, training a neural network by using the training set, and optimizing the neural network by adopting a gradient descent method;
(4) application of the method
For each non-monitoring unit, inputting the InfWorks ICM predicted value of the non-monitoring unit, infWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and STARMA predicted values of corresponding monitoring points of the n monitoring units into a trained neural network, and calculating to obtain a final ponding depth predicted value of the non-monitoring unit.
The invention has the beneficial effects that:
1. the method can effectively improve the accumulated water depth forecasting precision of the non-monitoring unit;
2. the method has strong portability and can be suitable for other types of rainfall flood models and monitoring point ponding forecasting models.
Drawings
FIG. 1 is a flow chart of an implementation of the present invention;
FIG. 2 is a schematic diagram of a neural network model for improving accuracy of non-monitoring unit water depth prediction.
Detailed Description
The following describes a specific implementation method of the present invention with reference to the flowchart shown in fig. 1:
step (1) waterlogging depth forecast at waterlogging monitoring point
Calculating a water accumulation depth forecast value of each waterlogging monitoring point by using a STARMA model of the rainfall water, wherein the water accumulation depth forecast value of each waterlogging monitoring point based on the model is referred to as the STARMA forecast value hereinafter, and the STARMA model of the rainfall water is referred to in the literature 'Zhengsai, wanqing, & Jaminyuan' (2014) ', urban rainstorm water accumulation point water accumulation short-time forecast based on the STARMA model, geographical science progress, 33 (7), 949-957.';
step (2) depth prediction and prediction value discretization of surface water
The method comprises the steps that an InfoWorks ICM model is utilized to carry out simulation forecasting on surface water depth, the surface water depth forecast value of the model is subjected to space discretization, the discretization mode supports modes such as regular grids and irregular triangular nets, and the surface water depth forecast value based on the InfoWorks ICM model is hereinafter referred to as the InfoWorks ICM forecast value;
step (3) surface unit discretization and classification
Dividing the urban ground surface into a series of ground surface units by adopting a space discretization mode consistent with the step (2); according to whether the surface unit contains the waterlogging monitoring point or not, dividing the surface unit into a surface unit (hereinafter referred to as a monitoring unit) containing the waterlogging monitoring point and a surface unit (hereinafter referred to as a non-monitoring unit) not containing the waterlogging monitoring point;
step (4) determining waterlogging monitoring points corresponding to the monitoring units
The invention assumes that one monitoring unit corresponds to one waterlogging monitoring point: for one monitoring unit, if the monitoring unit only comprises one waterlogging monitoring point, the waterlogging monitoring point is used as the waterlogging monitoring point corresponding to the monitoring unit; if the monitoring unit comprises a plurality of waterlogging monitoring points, one waterlogging monitoring point is 'virtually' taken out, the monitoring value of the virtual waterlogging monitoring point is the mean value of the monitoring values of the plurality of waterlogging monitoring points, the forecast value of the virtual waterlogging monitoring point is the mean value of the forecast values of the plurality of waterlogging monitoring points, and the virtual waterlogging monitoring point is taken as the waterlogging monitoring point corresponding to the monitoring unit.
Step (5) improving the accumulated water depth forecasting precision of the monitoring unit by using Bayesian model weighted average
For each monitoring unit, model forecast value fusion is realized by executing (1) to (4), wherein (1) to (3) are modeling steps, and (4) are application steps:
(1) obtaining an InfoWorks ICM forecast value sequence of a monitoring unit, a ponding depth monitoring data sequence of a monitoring point corresponding to the monitoring unit and a STARMA forecast value sequence of the monitoring point corresponding to the monitoring unit; carrying out normal conversion on the time sequence data by using Box-Cox transformation to obtain a related sequence conforming to normal distribution;
(2) determining a ponding depth BMA forecasting formula as follows:
wherein D is the forecast amount, namely the depth of accumulated water; o is accumulated water depth monitoring data of a monitoring point corresponding to the monitoring unit; p (d) 1 I O) is the STARMA predicted value d of the monitoring point corresponding to the monitoring unit under the condition of given monitoring data O 1 A posteriori probability of (i.e. weight value w of STARMA model of rainfall water) 1 ;p(d 2 I O) is the InfoWorks ICM predicted value d of the monitoring unit under the condition of given monitoring data O 2 The posterior probability of (1), i.e. the weight value w of the InfoWorks ICM model 2 ;Is a mean value d 1 Variance->Normal distribution of (2); />Is a mean value d 2 Variance->Normal distribution of (2); suppose O, d in formula (1) 1 、d 2 All are subjected to normal conversion and accord with normal distribution;
(3) weight calculation using the normally distributed sequence data in (1) to calculate the data in equation (1) by the expectation-maximization algorithmw 1 And w 2 ;
(4) For the monitoring unit, firstly, the Box-Cox conversion is utilized to carry out normal conversion on the STARMA predicted value of the monitoring point corresponding to the monitoring unit and the InfWorks ICM predicted value of the monitoring unit to obtain d 1 'and d' 2 Calculating by using a formula (2) to obtain a Bayesian model weighted average predicted value d'; performing inverse Box-Cox transformation on the d', wherein the numerical value after the inverse transformation is the final accumulated water depth forecast value of the monitoring unit;
d'=w 1 d 1 '+w 2 d' 2 (2)
step (6) constructing a neural network for improving the accuracy of non-monitoring unit water depth prediction
For a non-monitoring unit, besides the InfoWorks ICM predicted value of the non-monitoring unit, the InfoWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and the STARMA predicted values of monitoring points corresponding to the n monitoring units can also provide extra 'information quantity' for improving the accuracy of the predicted values of the non-monitoring unit, and the three 'information quantities' are actually the predicted values of an InfoWorks ICM model and a STARMA model of rainfall ponding; in order to fully utilize the three information quantities (forecast values of two types of models) to improve the accumulated water depth forecasting precision of the non-monitoring units and construct a neural network for improving the accumulated water depth forecasting precision of the non-monitoring units, the concrete steps are as follows (wherein (1) - (3) are modeling steps, and step (4) is an application step):
(1) determining a neural network structure, wherein the neural network adopts a three-layer structure and consists of an input layer, a hidden layer and an output layer (figure 2), the number of neurons in the hidden layer is a hyper-parameter and is determined through experiments;
(2) for any non-monitoring unit, the input of the neural network is the InfoWorks ICM predicted value of the non-monitoring unit, the InfoWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and the STARMA predicted values of corresponding monitoring points of the n monitoring units, and the output of the neural network is the final accumulated water depth predicted value of the non-monitoring unit; n is used as a hyper-parameter and is determined through experiments;
(3) training a neural network:
firstly, a data set used by the neural network is constructed: for each monitoring unit, taking the InfoWorks ICM predicted value of the monitoring unit, the InfoWorks ICM predicted values of n other monitoring units closest to the monitoring unit and the STARMA predicted values of monitoring points corresponding to the n monitoring units as input parts of a data set, and taking the final accumulated water depth predicted value of the monitoring unit (obtained by calculating the step (5) of improving the accumulated water depth prediction precision of the monitoring unit by using Bayesian model weighted average) as an output part of the data set;
then, randomly dividing the data set into a training set and a testing set, training a neural network by using the training set, and optimizing the neural network by adopting a gradient descent method;
(4) the method is applied that for each non-monitoring unit, the InfWorks ICM predicted value of the non-monitoring unit, the InfWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and the STARMA predicted values of monitoring points corresponding to the n monitoring units are input into a trained neural network, and the final ponding depth predicted value of the non-monitoring unit is obtained through calculation.
Claims (3)
1. A method for improving the accuracy of water depth forecast of a non-monitoring unit is characterized by comprising the following steps:
step (1) waterlogging depth forecast at waterlogging monitoring point
Calculating by using a STARMA model of rainfall water to obtain a water accumulation depth forecast value of each waterlogging monitoring point, wherein the water accumulation depth forecast value of the waterlogging monitoring point based on the model is referred to as the STARMA forecast value for short;
step (2) depth prediction and prediction value discretization of surface water
Carrying out analog prediction on the surface water depth by utilizing an InfoWorks ICM model, carrying out spatial discretization on a surface water depth prediction value of the model, and shortening the surface water depth prediction value based on the InfoWorks ICM model to be an InfoWorks ICM prediction value;
step (3) surface discretization and classification
Discretizing the earth surface into a series of earth surface units in a discretization mode consistent with the step (2); dividing the surface unit into a surface unit containing the monitoring points and a surface unit not containing the monitoring points according to whether the surface unit contains the waterlogging monitoring points or not; the earth surface unit containing the monitoring point is hereinafter referred to as a monitoring unit, and the earth surface unit not containing the monitoring point is hereinafter referred to as a non-monitoring unit;
step (4) determining the waterlogging monitoring points corresponding to the monitoring units
Assuming that one monitoring unit corresponds to one waterlogging monitoring point: for one monitoring unit, if the unit only comprises one waterlogging monitoring point, the waterlogging monitoring point is used as the waterlogging monitoring point corresponding to the unit; if the unit comprises a plurality of waterlogging monitoring points, virtually generating a waterlogging monitoring point, wherein the monitoring value of the virtual waterlogging monitoring point is the average value of the monitoring values of the plurality of waterlogging monitoring points, the forecast value of the virtual waterlogging monitoring point is the average value of the forecast values of the plurality of waterlogging monitoring points, and the virtual waterlogging monitoring point is used as the waterlogging monitoring point corresponding to the unit;
step (5) improving the accumulated water depth forecasting precision of the monitoring unit by using Bayesian model weighted average
For each monitoring unit, improving the accumulated water depth forecasting precision by executing (1) - (4);
(1) normal conversion
Acquiring an InfWorks ICM predicted value sequence of a monitoring unit, a water accumulation depth monitoring data sequence of a monitoring point corresponding to the monitoring unit and a STARMA predicted value sequence of the monitoring point corresponding to the monitoring unit; respectively carrying out normal conversion on the sequences by using Box-Cox transformation to obtain related sequences conforming to normal distribution;
(2) determining a weighted average forecasting formula of the ponding depth Bayes model as follows:
wherein D is the forecast amount, namely the depth of accumulated water; o is accumulated water depth monitoring data of a monitoring point corresponding to the monitoring unit; p (d) 1 I O) is the STARMA predicted value d of the corresponding monitoring point of the monitoring unit under the condition of given monitoring data O 1 A posteriori probability of (i.e. weight value w of STARMA model of rainfall water) 1 ;p(d 2 I O) is the predicted value d of the monitoring unit InfoWorks ICM under the condition of given monitoring data O 2 The posterior probability of (1), i.e. the weight value w of the InfoWorks ICM model 2 ;To give a predicted value d 1 : |, variance |)>Normal distribution of (2); />To give a predicted value d 2 : |, variance |)>Normal distribution of (2); suppose O, d in the formula (1) 1 、d 2 All are subjected to normal conversion and accord with normal distribution;
(3) weight calculation
Calculating the data in the formula (1) by an expectation-maximization algorithm by using the related sequence data in the formula (1) which conforms to the normal distributionw 1 And w 2 ;
(4) Application of the method
For the monitoring unit, firstly, carrying out normal conversion on a STARMA predicted value of a monitoring point corresponding to the monitoring unit and an InfoWorks ICM predicted value of the monitoring unit by using Box-Cox transformation to obtain d' 1 And d' 2 Calculating to obtain a weighted average predicted value d' of the Bayes model by using a formula (2); and performing inverse Box-Cox transformation on the d', wherein the value after inverse transformation is used as a final accumulated water depth forecast value of the monitoring unit:
d'=w 1 d′ 1 +w 2 d′ 2 (2)
step (6) constructing a neural network for improving the accuracy of non-monitoring unit water depth prediction
Constructing a neural network for improving the water depth forecasting precision of the non-monitoring units, and training the neural network by using the final forecast value, the InfoWorks ICM forecast value and the STARMA forecast value of the monitoring points corresponding to the monitoring units of the monitoring units; for any non-monitoring unit, the input of the neural network is the InfWorks ICM predicted value of the non-monitoring unit, the InfWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and the STARMA predicted values of the monitoring points corresponding to the n monitoring units, and the output of the neural network is the final accumulated water depth predicted value of the non-monitoring unit.
2. The method for improving the accuracy of the non-monitoring unit ponding depth forecast of claim 1, characterized by: and (3) the discretization mode in the step (2) is a regular grid mode or an irregular triangular grid mode.
3. The method for improving the accuracy of the non-monitoring unit ponding depth forecast of claim 1, characterized by: the method for constructing the neural network for improving the prediction accuracy of the non-monitoring unit water depth specifically comprises the following steps:
(1) determining neural network structure
The neural network adopts a three-layer structure and comprises an input layer, a hidden layer and an output layer, wherein the number of neurons in the hidden layer is a hyper-parameter and is determined through experiments;
(2) determining inputs and outputs of a neural network
For any non-monitoring unit, the input of the neural network is an InfoWorks ICM predicted value of the non-monitoring unit, infoWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and STARMA predicted values of monitoring points corresponding to the n monitoring units; the output of the neural network is the final accumulated water depth forecast value of the non-monitoring unit; n is a hyper-parameter and is determined through experiments;
(3) training neural networks
Firstly, a data set used by the neural network is constructed by utilizing a monitoring unit: for each monitoring unit, taking the InfoWorks ICM predicted value of the monitoring unit, the InfoWorks ICM predicted values of n other monitoring units nearest to the monitoring unit and STARMA predicted values of monitoring points corresponding to the n monitoring units as input parts of a data set, and taking the final ponding depth predicted value of the monitoring unit as an output part of the data set;
then, randomly dividing the data set into a training set and a test set, training a neural network by using the training set, and optimizing the neural network by adopting a gradient descent method;
(4) application of the method
For each non-monitoring unit, inputting the InfWorks ICM predicted value of the non-monitoring unit, infWorks ICM predicted values of n monitoring units nearest to the non-monitoring unit and STARMA predicted values of corresponding monitoring points of the n monitoring units into a trained neural network, and calculating to obtain a final ponding depth predicted value of the non-monitoring unit.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210592618.6A CN114925923B (en) | 2022-05-27 | 2022-05-27 | Method for improving accuracy of depth forecast of ponding in non-monitoring unit |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210592618.6A CN114925923B (en) | 2022-05-27 | 2022-05-27 | Method for improving accuracy of depth forecast of ponding in non-monitoring unit |
Publications (2)
Publication Number | Publication Date |
---|---|
CN114925923A CN114925923A (en) | 2022-08-19 |
CN114925923B true CN114925923B (en) | 2023-04-07 |
Family
ID=82810576
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210592618.6A Active CN114925923B (en) | 2022-05-27 | 2022-05-27 | Method for improving accuracy of depth forecast of ponding in non-monitoring unit |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114925923B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111882830A (en) * | 2020-07-31 | 2020-11-03 | 珠江水利委员会珠江水利科学研究院 | Urban waterlogging monitoring, forecasting and early warning method, device and system and storage medium |
CN114254561A (en) * | 2021-12-17 | 2022-03-29 | 奥格科技股份有限公司 | Waterlogging prediction method, waterlogging prediction system and storage medium |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN205091884U (en) * | 2015-10-30 | 2016-03-16 | 南京汉森思物联网科技有限公司 | Flood prevention system of intelligent city based on torrential rain forecasting model |
CN108107488B (en) * | 2017-12-21 | 2020-08-21 | 广州积雨云科技有限公司 | Urban waterlogging early warning method |
KR102225241B1 (en) * | 2019-04-30 | 2021-03-11 | 경북대학교 산학협력단 | Apparatus and method for real-time flooding prediction |
CN110334740A (en) * | 2019-06-05 | 2019-10-15 | 武汉大学 | The electrical equipment fault of artificial intelligence reasoning fusion detects localization method |
CN110533885B (en) * | 2019-08-15 | 2024-06-28 | 江苏省气象服务中心 | Urban waterlogging monitoring and early warning system and method |
CN113344305B (en) * | 2021-08-05 | 2021-10-29 | 中国水利水电科学研究院 | Rapid prediction method for rainstorm waterlogging event |
-
2022
- 2022-05-27 CN CN202210592618.6A patent/CN114925923B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111882830A (en) * | 2020-07-31 | 2020-11-03 | 珠江水利委员会珠江水利科学研究院 | Urban waterlogging monitoring, forecasting and early warning method, device and system and storage medium |
CN114254561A (en) * | 2021-12-17 | 2022-03-29 | 奥格科技股份有限公司 | Waterlogging prediction method, waterlogging prediction system and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN114925923A (en) | 2022-08-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Hung et al. | An artificial neural network model for rainfall forecasting in Bangkok, Thailand | |
Bussi et al. | Distributed sediment yield modelling: Importance of initial sediment conditions | |
CN114997541B (en) | Urban waterlogging prediction method and early warning platform based on digital twin technology | |
CN110852577A (en) | Urban flood assessment method based on urban toughness and urban drainage basin hydrological model | |
Prasad et al. | Estimating net aquifer recharge and zonal hydraulic conductivity values for Mahi Right Bank Canal project area, India by genetic algorithm | |
Minglei et al. | Classified real-time flood forecasting by coupling fuzzy clustering and neural network | |
CN111507375A (en) | Urban waterlogging risk rapid assessment method and system | |
Ahmadi et al. | Assessment of climate change impact on surface runoff, statistical downscaling and hydrological modeling | |
CN113191582B (en) | Road torrential flood vulnerability evaluation method based on GIS and machine learning | |
CN116502775B (en) | Hydrologic sequence enhancement and prediction method | |
Poretti et al. | An approach for flood hazard modelling and mapping in the medium Valtellina | |
Wu et al. | Local and global Bayesian network based model for flood prediction | |
Bahrami et al. | Evaluation of SCS model for flood characteristic prediction in an ungauged catchment considering effects of excess rainfall and base flow separation | |
CN114841417B (en) | High-precision salt tide forecasting method and system and readable storage medium | |
Zhao et al. | Risk assessment of flood disaster and forewarning model at different spatial-temporal scales | |
Lai et al. | Waterlogging risk assessment based on self-organizing map (SOM) artificial neural networks: a case study of an urban storm in Beijing | |
Jin et al. | Integration of an improved transformer with physical models for the spatiotemporal simulation of urban flooding depths | |
Ai-di et al. | A sampled method of classification of susceptibility evaluation unit for geological hazards based on GIS | |
CN117648878A (en) | Flood rapid evolution and flooding simulation method based on 1D-CNN algorithm | |
CN114925923B (en) | Method for improving accuracy of depth forecast of ponding in non-monitoring unit | |
Tayfur et al. | Predicting hourly-based flow discharge hydrographs from level data using genetic algorithms | |
Gidado et al. | Application of Ca-Markov model for the analysis of urban growth in Kenyir Basin | |
Yan et al. | A novel integrated urban flood risk assessment approach based on one-two dimensional coupled hydrodynamic model and improved projection pursuit method | |
Dewals et al. | Integrated assessment of flood protection measures in the context of climate change: hydraulic modelling and economic approach | |
Perera et al. | Identification of the spatial variability of runoff coefficients of three wet zone watersheds of Sri Lanka |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |