A distributed flood simulation method for the Loess Plateau with temporal and spatial differentiation of runoff patterns

Abstract

The present invention discloses a distributed flood simulation method for the Loess Plateau with spatiotemporal differentiation of runoff patterns, which belongs to the technical field of model building methods, including constructing a technical framework for identifying spatiotemporal differentiation of runoff patterns, improving plant interception modules, infiltration modules, evapotranspiration modules, silt dam / terrace modules, coupling grid water droplet confluence methods and deep learning confluence calculations, and improving the model in combination with variable parameter theory and experience. Based on the mechanism of the influence of changing environment on the process of rainstorm runoff, the present invention further reveals the spatiotemporal analysis characteristics of runoff patterns on the Loess Plateau under a changing environment, combines deep learning theory, constructs a distributed flood simulation model that couples spatiotemporal differentiation of runoff patterns and deep learning theory, and performs confluence calculations of the model through grid water droplet confluence methods and deep learning theory; finally, in combination with variable parameter theory and experience, improves the parameter module of the model to realize the function of transforming model parameters from static to dynamic.

Description

Technical Field

[0001] The present invention relates to the technical field of model building methods, and in particular to a method for building a distributed flood simulation model. Background Art

[0002] The flood simulation model is based on the terrain data, river network data, and hydrological data of the study area. It constructs a rainstorm-flood coupling model and uses measured data to set parameters and verify the model, providing model support for the simulation of rainstorm-flood risk factors in the study area.

[0003] The Loess Plateau has a fragmented terrain and complex landforms. At the same time, changes in the underlying surface caused by human activities (such as water and soil conservation measures such as silt dams and terraces) have caused drastic changes in the underlying surface of the Loess Plateau, making the runoff process pattern in the basin have significant spatiotemporal heterogeneity. The spatiotemporal heterogeneity of runoff patterns in the Loess Plateau has been ignored in traditional flood models, and the accuracy of flood simulation needs to be improved. Based on this, a distributed flood simulation method for the Loess Plateau with spatiotemporal heterogeneity of runoff patterns is proposed. On the basis of the "influence mechanism of changing environment on rainstorm runoff generation and convergence process", the spatiotemporal analysis characteristics of runoff patterns in the Loess Plateau under a changing environment are further revealed. Combined with deep learning theory, a distributed flood simulation model that couples spatiotemporal heterogeneity of runoff patterns and deep learning theory is constructed. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the prior art and to propose a distributed flood simulation method suitable for the temporal and spatial differentiation of runoff generation patterns in the Loess Plateau.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A distributed flood simulation method for the Loess Plateau with spatiotemporal differentiation of runoff patterns, including the construction of a technical framework for identifying spatiotemporal differentiation of runoff patterns, plant interception module, infiltration module, evapotranspiration module, silt dam / terrace module improvement, coupled grid droplet confluence method and deep learning confluence calculation and combining variable parameter theory and experience to improve the model;

[0007] The specific steps include:

[0008] S1. Constructing a technical framework for identifying the spatiotemporal differentiation of runoff patterns: Combining the soil property detection and runoff observation test results of different underlying surface types, and based on the historical rainstorms and floods in the basin and the underlying surface characteristics, further reveal the spatiotemporal differentiation of runoff patterns in the Loess Plateau under a changing environment, establish a technical framework for spatiotemporal analysis and identification of runoff patterns, and then study the spatial distribution characteristics of the basin's runoff mechanism and divide the basin into blocks with different runoff mechanisms;

[0009] S2. Improvement of plant interception module, infiltration module, evapotranspiration module, and silt dam / terrace module: Combine the key parameters of different underlying surface types with the response process mechanism model of key parameters of rainstorm runoff, and improve the plant interception module, rainfall infiltration module, evapotranspiration module, and silt dam / terrace module in the flood simulation model;

[0010] S3, coupled grid water droplet confluence method and deep learning confluence calculation: coupled grid water droplet confluence method and deep learning theory to perform model confluence calculation;

[0011] S4. Improve the model by combining variable parameter theory and experience: Improve the parameter module of the model by combining variable parameter theory and experience, realize the transformation of model parameters from static to dynamic, and complete the construction of a distributed flood simulation model that is coupled with the spatiotemporal differentiation of runoff patterns and deep learning theory accompanied by underlying surface changes.

[0012] Preferably, in S1, remote sensing images and Google Maps are combined to extract the specific distribution locations of silt dams and terraces;

[0013] Specifically, based on a series of conditions such as land use, silt dams, terraces, topography, soil and vegetation coverage in the basin, GIS spatial analysis technology is used to unify the scale of soil type, land use, slope, etc. in the basin into a unit grid consistent with the model resolution, and then the runoff pattern of the grid unit is determined based on the constructed spatiotemporal differentiation identification technology framework, and the basin is divided into blocks with different runoff mechanisms.

[0014] For the full storage runoff mode, the model selects the water storage capacity allocation curve for runoff calculation, while for the over-seepage runoff mode, the model selects the infiltration capacity allocation curve for runoff calculation.

[0015] Preferably, the plant interception module in S2 is calculated by establishing a canopy-litter interception dynamic model based on the results of many years of forest management interception experiments in the Nanxiao River Basin;

[0016] The infiltration module uses the Philip infiltration model for calculation, and the evapotranspiration module uses the three-layer evapotranspiration model for calculation. Based on the response process mechanism model of the key parameters of rainstorm runoff to the key parameters of different underlying surface types, the relevant parameters of the infiltration module and the evapotranspiration module are adjusted and optimized.

[0017] Preferably, the response process mechanism model adds a silt dam / terrace module to reflect the impact of silt dam / terrace and terrace;

[0018] In the silt dam / terrace module, the model quantitatively describes the role of the silt dam / terrace based on the exponential decay weighted function and the linear storage-discharge relationship equation; the amount of water entering the silt dam / terrace is calculated using the exponential decay weighted function.

[0019] Q in,k =(1-exp(-1 / δ g ))×β×Q soil,k +exp(-1 / δ g )×Q in,k-1

[0020] In the formula, Q in,k and Q in,k-1 is the amount of water entering the silt dam / terrace in the kth and k-1th periods, in mm; δ g is the time lag parameter, in days; Q soil,k is the water supply distribution coefficient of the silt dam / terrace (0-1); Q soil,k is the outflow of the soil profile in the kth period, in mm;

[0021] The amount of water discharged from the check dam or terrace is calculated using the linear storage-discharge relationship equation

[0022]

[0023] In the formula, Q out,k and Q out,k-1 Indicates the amount of water flowing out from the silt dam / terrace in the kth and k-1th periods, in m 3 ; α m It is a parameter that characterizes the drainage performance of the silt dam / terrace; a j is the area of ​​the jth grid cell, in m 2 .

[0024] Preferably, in said S3, a grid water droplet confluence method is used to calculate the confluence process;

[0025] Then, a long short-term memory neural network (LSTM) deep learning model is used to learn the error characteristics between the confluence calculation results and the measured results. The confluence results are corrected based on the error law characteristics to obtain the final confluence results.

[0026] Grid droplet confluence method: The slope confluence formula in the grid droplet confluence calculation is:

[0027]

[0028] Where V slope is the slope flow velocity; K v is the velocity coefficient; n slope is the surface roughness of the slope; m is a parameter related to the roughness;

[0029] Since the hydraulic conditions of river confluence and slope confluence are different, the Manning formula is used in river confluence:

[0030]

[0031] Where: V h is the velocity of the grid water drop in the river channel, C v is the velocity coefficient, velocity coefficient, unit is m / s; n c R is the river channel roughness; c is the hydraulic radius of the river channel; after obtaining the flow velocity of the grid water droplets on the slope and the river channel, the confluence time of the grid water droplets is obtained according to the confluence path length;

[0032] Long short-term memory neural network LSTM deep learning model: Using LSTM based on a large amount of data, find the relationship between the flow rate of each grid unit in each time period and the error ΔQ between the flow rate at the watershed outlet section calculated based on the grid droplet confluence method and the measured flow rate, so as to correct the confluence results based on the error law characteristics;

[0033] The LSTM model calculation steps are as follows:

[0034] f t =σ[W f (ΔQ t-1 ,R t )+b f ]

[0035] i t =σ[W f (ΔQ t-1 ,R t )+b i ]

[0036]

[0037] o t =σ[W0(ΔQ t-1 ,R t )+b0]

[0038] ΔQ t =o t ⊙tanh(C t )

[0039] In the formula, f t ,i t , o t They are the forget gate, input gate and output gate of the LSTM model respectively; W f represents the weight of the input gate; b f Characterizes the offset of the input gate; C t is the state variable of the model, and b c yes The offset vector.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. Based on the influence mechanism of changing environment on rainstorm runoff process, the present invention further reveals the spatiotemporal analysis characteristics of runoff pattern in the Loess Plateau under changing environment, and constructs a distributed flood simulation model that couples spatiotemporal differentiation of runoff pattern and deep learning theory based on deep learning theory. The runoff calculation of the model is performed through the grid droplet runoff method and deep learning theory. Finally, the parameter module of the model is improved by combining variable parameter theory and experience to realize the function of transforming model parameters from static to dynamic. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic diagram of the flow chart of a distributed flood simulation method for the Loess Plateau with temporal and spatial differentiation of runoff patterns proposed by the present invention;

[0043] Figure 2 This is a technical framework diagram of identifying the spatiotemporal differentiation of runoff patterns in a distributed flood simulation method for the Loess Plateau with spatiotemporal differentiation of runoff patterns proposed by the present invention;

[0044] Figure 3 This is a flow chart of the LSTM model in the distributed flood simulation method for the Loess Plateau with spatiotemporal differentiation of runoff patterns proposed by the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0046] Example, see Figure 1-Figure 3 , a distributed flood simulation method for the Loess Plateau with spatiotemporal differentiation of runoff patterns, including the construction of a technical framework for identifying spatiotemporal differentiation of runoff patterns, plant interception module, infiltration module, evapotranspiration module, silt dam / terrace module improvement, coupled grid droplet confluence method and deep learning confluence calculation and combining variable parameter theory and experience to improve the model;

[0047] It is worth noting that the overall research idea of ​​this plan is: taking the Nanxiaohegou Basin in the Loess Plateau as the research object, systematically analyzing the variation patterns of regional rainstorm floods and underlying surface influencing factors, carrying out soil property detection and runoff observation experiments on different underlying surface types, and establishing a response process mechanism model of key parameters of rainstorm runoff to key parameters of different underlying surface types, thereby revealing the spatiotemporal differentiation of runoff patterns under a changing environment. On this basis, combined with deep learning theory, a distributed flood simulation model for small and medium-sized basins in the Loess Plateau that is adapted to the background of a changing environment is developed to evaluate the multi-scale response of flood processes to changes in composite underlying surfaces.

[0048] Construct a technical framework for identifying the spatiotemporal differentiation of runoff patterns: Combine the soil property detection of different underlying surface types with the runoff observation test results, and according to the historical rainstorms and floods in the basin and the underlying surface characteristics, establish a technical framework for identifying the spatiotemporal differentiation of runoff patterns.

[0049] Combined with remote sensing images and Google Maps, the specific distribution locations of silt dams and terraces were extracted. According to the land use, silt dams, terraces, topography, soil and vegetation coverage of the basin, GIS spatial analysis technology was used to unify the scale of soil type, land use, slope, etc. in the basin to a unit grid consistent with the model resolution, and then the runoff pattern of the grid unit was determined based on the constructed spatiotemporal differentiation identification technology framework, and the basin was divided into blocks with different runoff mechanisms.

[0050] For the full storage runoff mode, the model selects the water storage capacity allocation curve for runoff calculation, while for the over-seepage runoff mode, the model selects the infiltration capacity allocation curve for runoff calculation.

[0051] The technical framework for identifying the spatiotemporal differentiation of runoff patterns is as follows Figure 2 As shown;

[0052] Improvement of plant interception module, infiltration module, evapotranspiration module, and silt dam / terrace module: Combined with the response process mechanism model of key parameters of rainstorm runoff to key parameters of different underlying surface types, the plant interception module, infiltration module, evapotranspiration module, and silt dam / terrace module in the flood simulation model are improved.

[0053] The plant interception module uses the canopy-litter interception dynamic model calculated according to the results of many years of forest management interception experiments in the Nanxiaohegou Basin. The infiltration module uses the Philip infiltration model, and the evapotranspiration module uses the three-layer evapotranspiration model. Based on the response process mechanism model of the key parameters of rainstorm runoff to the key parameters of different underlying surface types, the relevant parameters of the infiltration module and the evapotranspiration module are adjusted and optimized.

[0054] In particular, the model adds a check dam / terracing module to reflect the impact of check dams / terraces and terraces. In the check dam / terracing module, the model quantitatively describes the role of check dams / terraces based on an exponential decay weighting function and a linear storage-discharge relationship equation.

[0055] The amount of water entering the silt dam / terraces is calculated using an exponential decay weighted function:

[0056] Q in,k =(1-exp(-1 / δ g ))×β×Q soil,k +exp(-1 / δ g )×Q in,k-1

[0057] In the formula, Q in,k and Q in,k-1 is the amount of water entering the silt dam / terrace in the kth and k-1th periods, in mm; δ g is the time lag parameter, the unit is day; β is the water supply distribution coefficient of the silt dam / terrace (0~1); Q soil,k is the outflow rate of the soil profile in the kth period, in mm.

[0058] The amount of water discharged from the check dam or terrace is calculated using the linear storage-discharge relationship equation:

[0059]

[0060] In the formula, Q out,k and Q out,k-1 Indicates the amount of water flowing out from the silt dam / terrace in the kth and k-1th periods, in m 3 ; α m It is a parameter that characterizes the water discharge performance of the silt dam / terrace; Q karst,k =(1-β)×Q soil,k is the area of ​​the jth grid cell, in m 2 .

[0061] Confluence calculation by coupling grid water droplet confluence method and deep learning: The confluence calculation of the model is performed by coupling the grid water droplet confluence method and deep learning theory. In the model, the grid water droplet confluence method is first used to calculate the confluence process. Then the long short-term memory neural network LSTM deep learning model is used to learn the error characteristics of the confluence calculation results and the measured results, and the confluence results are corrected based on the error law characteristics to obtain the final confluence results.

[0062] Grid water droplet confluence method: This theory regards the grid cells that produce runoff after rainfall in the basin as water droplets. The water droplets move along the ground slope under the combined effect of gravity and resistance. The time it takes for each water droplet to reach the outlet section depends on the path length and the speed of movement on the path trajectory. The slope confluence formula in the grid water droplet confluence calculation is:

[0063]

[0064] Where V slope is the slope flow velocity; K v is the velocity coefficient; n slope is the surface roughness of the slope; m is a parameter related to the roughness.

[0065] Since the hydraulic conditions of river confluence and slope confluence are different, the Manning formula is used in river confluence:

[0066]

[0067] Where: V h is the velocity of the grid water drop in the river channel, C v is the velocity coefficient, velocity coefficient, unit is m / s; n c R is the river channel roughness; c is the hydraulic radius of the river channel. After obtaining the flow velocity of the grid water droplets on the slope and the river channel, the confluence time of the grid water droplets is obtained according to the confluence path length.

[0068] Long Short-Term Memory Neural Network LSTM Deep Learning Model: Using LSTM, we can find the relationship between the flow rate of each grid unit in each time period and the error ΔQ between the flow rate at the watershed outlet section calculated based on the grid water droplet confluence method and the measured flow rate, based on a large amount of data, so as to correct the confluence results based on the error law characteristics. The LSTM model process is as follows Figure 3 shown.

[0069] The LSTM model calculation steps are as follows:

[0070] f t =σ[W f (ΔQ t-1 ,R t )+b f ]

[0071] i t =σ[W f (ΔQ t-1 ,R t )+b i ]

[0072]

[0073] o t =σ[W0(ΔQ t-1 ,R t )+b0]

[0074] ΔQ t =o t ⊙tanh(C t )

[0075] In the formula, f t ,i t , o t They are the forget gate, input gate and output gate of the LSTM model respectively; W f represents the weight of the input gate; b f Characterizes the offset of the input gate; C t is the state variable of the model, and b c yes The offset vector.

[0076] Finally, by combining the variable parameter theory and experience, the parameter module of the model is improved to realize the transformation of the model parameters from static to dynamic, thereby completing the construction of a distributed flood simulation model that is coupled with the spatiotemporal differentiation of runoff patterns and deep learning theory accompanied by underlying surface changes.

[0077] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A distributed flood simulation method for the Loess Plateau with temporal and spatial differentiation of runoff patterns, characterized by: It includes the construction of a technical framework for identifying the spatiotemporal differentiation of runoff patterns, the improvement of plant interception modules, infiltration modules, evapotranspiration modules, silt dam / terracing modules, coupled grid droplet confluence methods and deep learning confluence calculations, and the improvement of models combining variable parameter theory and experience; The specific steps include: S1. Constructing a technical framework for identifying the spatiotemporal differentiation of runoff patterns: Combining the soil property detection and runoff observation test results of different underlying surface types, and based on the historical rainstorms and floods in the basin and the underlying surface characteristics, further reveal the spatiotemporal differentiation of runoff patterns in the Loess Plateau under a changing environment, establish a technical framework for spatiotemporal analysis and identification of runoff patterns, and then determine the spatial distribution characteristics of the basin's runoff mechanism, and divide the basin into blocks with different runoff mechanisms; S2. Improvement of plant interception module, infiltration module, evapotranspiration module, and silt dam / terrace module: Combine the key parameters of different underlying surface types with the response process mechanism model of key parameters of rainstorm runoff, and improve the plant interception module, rainfall infiltration module, evapotranspiration module, and silt dam / terrace module in the flood simulation model; S3, coupled grid water droplet confluence method and deep learning confluence calculation: coupled grid water droplet confluence method and deep learning theory to perform model confluence calculation; S4. Improve the model by combining variable parameter theory and experience: Improve the parameter module of the model by combining variable parameter theory and experience, realize the transformation of model parameters from static to dynamic, and complete the construction of a distributed flood simulation model that is coupled with the spatiotemporal differentiation of runoff patterns and deep learning theory with underlying surface changes; The response process mechanism model adds a check dam / terracing module to reflect the impact of check dams and terraces; In the silt dam / terrace module, the model quantitatively describes the role of the silt dam / terrace based on the exponential decay weighted function and the linear storage-discharge relationship equation; the amount of water entering the silt dam / terrace is calculated using the exponential decay weighted function: Q in,k =(1-exp(-1 / δ g ))×β×Q soil,k +exp(-1 / δ g )×Q in,k-1 In the formula, Q in,k and Q in,k-1 is the amount of water entering the silt dam / terrace in the kth and k-1th periods, in mm; δ g is the time lag parameter, in days; β is the water supply distribution coefficient of the silt dam / terrace, which is (0,1); Q soil,k is the outflow of the soil profile in the kth period, in mm; The amount of water discharged from the check dam or terrace is calculated using the linear storage-discharge relationship equation: In the formula, Q out,k and Q out,k-1 Indicates the amount of water flowing out from the silt dam / terrace in the kth and k-1th periods, in m 3 ; α m It is a parameter that characterizes the drainage performance of the silt dam / terrace; a j is the area of ​​the jth grid cell, in m 2 , Q in,kj is the water volume of the check dam / terrace in the basin at the jth grid unit area in the kth time period, in mm; Δt is the time interval; In said S3, a grid water droplet confluence method is used to calculate the confluence process; Then, a long short-term memory neural network (LSTM) deep learning model is used to learn the error characteristics between the confluence calculation results and the measured results. The confluence results are corrected based on the error law characteristics to obtain the final confluence results. Grid droplet confluence method: The slope confluence formula in the grid droplet confluence calculation is: Where V slope is the slope flow velocity; K v is the velocity coefficient; n slope is the surface roughness of the slope; m is a parameter related to the roughness; S is the surface slope along the direction of water flow; μ is an empirical parameter that reflects the change of the surface slope along the direction of water flow; Since the hydraulic conditions of river confluence and slope confluence are different, the Manning formula is used in river confluence: Where: V h is the velocity of the grid water drop in the river channel, C v is the velocity coefficient, unit is m / s; n c R is the river channel roughness; c is the hydraulic radius of the river channel; after obtaining the flow velocity of the grid water droplets on the slope and the river channel, the confluence time of the grid water droplets is obtained according to the confluence path length; Long short-term memory neural network LSTM deep learning model: Using LSTM based on a large amount of data, find the relationship between the flow rate of each grid unit in each time period and the error ΔQ between the flow rate at the watershed outlet section calculated based on the grid droplet confluence method and the measured flow rate, so as to correct the confluence results based on the error law characteristics; The LSTM model calculation steps are as follows: f t =σ[W f (ΔQ t-1 ,R t )+b f ] i t =σ[W f (ΔQ t-1 ,R t )+b i ] o t =σ[W0(ΔQ t-1 ,R t )+b0] ΔQ t =o t ⊙tanh(C t ) In the formula, f t ,i t , o t They are the forget gate, input gate and output gate of the LSTM model respectively; W f represents the weight of the input gate; b f Characterizes the bias term of the forget gate; C t is the state variable of the model, is the memory unit state; and b c yes The offset vector of R t is the recursive input at time t; b i is the bias term of the input gate; W c for ; W0 is the weight matrix of the output gate; b0 is the bias term of the output gate; σ is the sigmoid activation function; tanh is the hyperbolic tangent activation function.

2. The method for simulating distributed floods in the Loess Plateau with temporal and spatial differentiation of runoff patterns according to claim 1 is characterized in that: In said S1, remote sensing images and Google Maps are combined to extract the specific distribution locations of silt dams and terraces; Specifically, according to the land use, silt dam, terraces, topography, soil and vegetation cover conditions of the basin, GIS spatial analysis technology is used to uniformly downscale the soil type, land use and slope of the basin to a unit grid consistent with the model resolution, and then the runoff pattern of the grid unit is determined based on the superimposed attributes of the constructed spatiotemporal differentiation identification technology framework of runoff pattern, and the basin is divided into blocks with different runoff mechanisms; For the full storage runoff mode, the model selects the water storage capacity allocation curve for runoff calculation, while for the over-seepage runoff mode, the model selects the infiltration capacity allocation curve for runoff calculation.

3. The method for simulating distributed floods in the Loess Plateau with temporal and spatial differentiation of runoff patterns according to claim 1 is characterized in that: The plant interception module in S2 is calculated by establishing a canopy-litter interception dynamic model based on the results of many years of forest management interception experiments in the Nanxiaohegou Basin; The infiltration module adopts Philip infiltration model for calculation, and the evapotranspiration module adopts three-layer evapotranspiration model for calculation. Based on the response process mechanism model of the key parameters of rainstorm runoff to the key parameters of different underlying surface types, the relevant parameters of the modules are adjusted and optimized.

Images