Monthly runoff prediction method and system coupled with physical mechanism

Abstract

The invention discloses a monthly runoff prediction method and system coupled with a physical mechanism, and the method comprises the steps: inputting the hydro meteorological data of the last month into a monthly runoff prediction model, and obtaining the drainage basin outlet section runoff of the current month; the monthly runoff prediction model comprises an overland runoff module, a soil module, an underground module and a confluence module which are connected in sequence, elevation, early-stage rainfall, early-stage temperature, soil humidity, groundwater reserves, inflow runoff and the like are used as forecast factors to be input into the monthly runoff prediction model, each module takes a convolution encoder as a basic framework, and a coder-decoder structure is combined, so that the prediction model is established. And spatial correlation characteristics of rainfall-runoff and other elements are described. And thirdly, taking the water balance constraint as a penalty term of a loss function, and adding physical rule penalty term constraints except data fitting errors for the model so as to improve the logic inference reliability of the model.

Description

Technical Field

[0001] The invention relates to the field of runoff prediction, in particular to a monthly runoff prediction method and system with coupled physical mechanisms. Background Art

[0002] Affected by the special natural geographical and climatic conditions, my country's water resources are unevenly distributed in time and space, the contradiction between water supply and demand is prominent, and water and drought disasters are frequent and serious. Runoff is the most direct manifestation of water resources abundance and scarcity. Accurate and reliable runoff forecasting is of great significance to alleviating the contradiction between water supply and demand and reducing losses caused by water and drought disasters. However, the monthly runoff process involves multiple spheres such as the hydrosphere, atmosphere, and lithosphere, with many influencing factors and complex action mechanisms. The existing forecasting methods are affected by the nonlinear characteristics of the hydrological process and the uncertainty of precipitation, and the forecasting performance is not strong, which seriously lags behind the actual production requirements. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a deep learning monthly runoff forecasting method and system that comprehensively applies the physical structure framework and physical rules to couple the physical mechanism, seeks a balance between deep learning constraints and forecasting performance improvement, and a monthly runoff forecasting method based on a convolutional encoder and physical mechanism guidance to realize the model's consideration of the physical mechanism of the hydrological process and improve the prediction performance of monthly runoff.

[0004] Technical solution: The monthly runoff prediction method of the coupled physical mechanism of the present invention comprises the following steps: inputting the hydrological and meteorological data of the previous month into the monthly runoff prediction model to obtain the runoff of the watershed outlet section of the current month;

[0005] The monthly runoff prediction model includes a surface runoff module, a soil module, an underground module and a confluence module connected in sequence, and the surface runoff module, the soil module, the underground module and the confluence module are all convolutional encoder models;

[0006] The surface runoff module predicts the precipitation, evapotranspiration and surface runoff data of the current month based on the evapotranspiration, potential evapotranspiration, temperature, precipitation and elevation data of the previous month;

[0007] The soil module predicts the soil moisture change and soil midflow data for the current month based on the soil moisture change data of the previous month;

[0008] The underground module predicts the groundwater reserve change and underground runoff data for the current month based on the soil moisture change data for the current month and the groundwater reserve change data for the previous month.

[0009] The confluence module predicts the runoff at the basin outlet section of the current month based on the inflow and runoff data of the previous month, the groundwater storage changes of the current month, the underground runoff, the surface runoff and the subsoil flow data.

[0010] Furthermore, the loss function of the monthly runoff prediction model is constrained by the water balance relationship, which is:

[0011] ET+ΔSM+ΔGW+Q=P+IQ

[0012] Q=P+IQ-ET-ΔSM-ΔGW

[0013] Among them, Q refers to the theoretical value of runoff at the basin outlet section that meets the water balance constraint in the current month, and P, ET, ΔSM, ΔGW and IQ represent precipitation, evapotranspiration, soil moisture change, groundwater storage change and basin inflow in the current month, respectively.

[0014] Furthermore, the loss function of the monthly runoff prediction model is:

[0015] Loss = λMSE physics +(1-λ)MSE data

[0016]

[0017]

[0018] Among them, Loss is the loss function, λ represents the weight coefficient of the penalty for violating the water balance, Q pre Refers to the predicted runoff value of the basin outlet section, Q obs Refers to the observed runoff value at the outlet section of the basin, MSE data Indicates Q pre With Q obs The root mean square error, MSE physics Indicates Q pre The root mean square error between the theoretical value Q of the runoff at the basin outlet section that meets the water balance constraint, n is the number of observations.

[0019] Furthermore, the architectures of the surface runoff module, soil module, underground module and confluence module are all convolutional encoder-convolutional decoder architectures.

[0020] Furthermore, the convolutional encoder-convolutional decoder architecture includes a convolutional encoder, a convolutional decoder and a fully connected layer; the convolutional encoder and the convolutional decoder both include two sequences of convolution operations, which are respectively used to increase and reduce the number of feature maps.

[0021] Furthermore, the hydrological and meteorological data of the previous month were preprocessed, and the hydrological and meteorological data of the previous month included the evapotranspiration, potential evapotranspiration, temperature, precipitation, elevation, soil moisture change and groundwater storage change data of the previous month; the preprocessing included integrating the hydrological and meteorological data of the previous month into the same spatial scale through spatial interpolation method and performing standardization processing.

[0022] Furthermore, the Z-Score standardization method was used for standardization.

[0023] The monthly runoff prediction system of the coupled physical mechanism of the present invention comprises a monthly runoff prediction model; the hydrological and meteorological data of the previous month are input into the monthly runoff prediction model to obtain the runoff volume of the basin outlet in the current month;

[0024] The monthly runoff prediction model includes a surface runoff module, a soil module, an underground module and a confluence module connected in sequence, and the surface runoff module, the soil module, the underground module and the confluence module are all convolutional encoder models;

[0025] The surface runoff module predicts the precipitation, evapotranspiration and surface runoff data of the current month based on the evapotranspiration, potential evapotranspiration, temperature, precipitation and elevation data of the previous month;

[0026] The soil module predicts the soil moisture change and soil midflow data for the current month based on the soil moisture change data of the previous month;

[0027] The underground module predicts the groundwater reserve change and underground runoff data for the current month based on the soil moisture change data for the current month and the groundwater reserve change data for the previous month.

[0028] The confluence module predicts the runoff at the basin outlet section of the current month based on the inflow and runoff data of the previous month, the groundwater storage changes of the current month, the underground runoff, the surface runoff and the subsoil flow data.

[0029] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the monthly runoff prediction method based on the coupled physical mechanism is implemented.

[0030] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the monthly runoff prediction method of the coupled physical mechanism is implemented.

[0031] Beneficial effects: Compared with the prior art, the advantages of the present invention are:

[0032] (1) The present invention couples physical mechanisms to achieve deep learning monthly runoff prediction guided by physical mechanisms. The physical reliability of deep learning runoff forecasting models is not strong, and it is easy to produce forecast results that do not conform to physical laws. In order to improve the quality of deep learning monthly runoff forecasting, a nonlinear mapping relationship between monthly runoff and land-atmosphere influencing factors is constructed, and a water balance relationship is established between runoff and hydrological cycle elements such as evapotranspiration and precipitation. Based on this, the loss function is improved and a physical architecture of the hydrological process is designed to characterize the hydrological response relationship as a physical rule constraint, and the model's causal logic inference process is strengthened, thereby establishing a deep learning forecasting model coupled with physical mechanisms, improving the physical reliability of the model and taking into account the forecasting performance.

[0033] (2) The present invention uses convolutional networks to achieve spatial modeling of the hydrological process in the watershed. Traditional deep learning is mostly based on single-point single-station prediction. The present invention uses convolutional networks to consider the spatial correlation of hydrological and meteorological elements such as precipitation and temperature, and to characterize the spatial structural relationship between rainfall and runoff, so that the spatial information characteristics of hydrological and meteorological elements can be more fully utilized in the runoff prediction process, reflecting the spatial relationship of hydrological elements.

[0034] In summary, the coupling physical mechanism and convolutional deep network of the present invention can not only improve the physical reliability of the deep learning model while taking into account the forecasting performance, but also help to improve the application value of deep learning in hydrological forecasting, and realize the modeling of hydrological processes at the basin scale, which is reasonable and effective. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a flow chart of the monthly runoff prediction method of the present invention;

[0036] Figure 2 A structural diagram of a monthly runoff prediction model according to an embodiment of the present invention;

[0037] Figure 3 1 is a comparison chart of the runoff measured value and the predicted value according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] To facilitate understanding of the present invention, the following description is given:

[0039] Deep Learning:

[0040] Definition 1: Deep learning usually refers to deep neural networks, which are a type of model with a multi-layer network structure developed on the basis of neural networks, including common structures such as long short-term memory networks (LSTM) and convolutional neural networks (CNN).

[0041] Convolutional Encoder:

[0042] Definition 2: Convolutional Autoencoder (CAE) is a deep learning model that can automatically extract the features of input data and compress it into a low-dimensional representation through the combination of an encoder and a decoder.

[0043] Water balance relationship:

[0044] Definition 3: Water balance means that for any selected area, the difference between the amount of water it receives and the amount of water it spends during any period of time must be equal to the change in water storage in the area during that period of time.

[0045] like Figure 1 As shown, the monthly runoff prediction method of the coupled physical mechanism comprises the following steps:

[0046] (1) Selection of prediction factors: It is proposed to select meteorological factors such as previous precipitation, temperature, evapotranspiration, and underlying surface factors such as topography, soil moisture, and groundwater storage as potential influencing factors of regional precipitation. Comprehensive comparison is performed to select factors with physical significance as model prediction factors.

[0047] (2) Data preprocessing: Based on the selected prediction factors, various data are collected and preprocessed. First, the spatial resolution of the data is unified and the data is integrated into the same spatial scale based on the spatial interpolation method; second, all data are standardized.

[0048] (3) Design the basic framework of the model: Apply the principle of module combination and superposition, and use four sub-modules to represent the hydrological cycle of surface processes, subsoil flow, groundwater, and confluence. Finally, the outputs of each module are connected to obtain the runoff forecast result.

[0049] (4) Design of submodule architecture: Each submodule uses a convolutional encoder as the basic architecture and an encoder-decoder structure. The encoder includes multiple sequences of convolutions, and after each convolution, the number of filters in the convolution layer is doubled. The decoder is similar, including the same sequence of deconvolutions, and the feature map is reduced by half after each deconvolution. The last layer uses convolution operations to output the results.

[0050] (5) Modify the loss function: Use the water balance constraint as the penalty term of the loss function and add physical rule penalty term constraints to the model in addition to the data fitting error.

[0051] (6) Model training: Hydrological and meteorological data are used to form a sample set. The samples are divided into three subsets: training, validation, and testing. The training set is used for model parameter fitting, the validation set is used to evaluate the forecasting performance of the trained model to optimize hyperparameters, and the test set is used to evaluate the model fitting performance after optimization. Evaluation indicators such as root mean square error, mean relative error, and Nash efficiency coefficient are used, and strategies such as the Adam algorithm, regularization, and early stopping are used to optimize model parameters and select hyperparameters.

[0052] According to the constructed model, the measured precipitation, temperature and other data at the current moment are used to predict the runoff at the basin outlet section at the next moment.

[0053] Taking the outlet runoff of the upper reaches of the Yangtze River Basin as an example, the deep learning monthly runoff forecasting method with coupled physical mechanism is used to predict the monthly runoff of the outlet section (Cuntan Station) of the upper reaches of the Yangtze River Basin.

[0054] 1. Measured data

[0055] Monthly precipitation and monthly temperature observation data are from the China Surface Precipitation and Temperature Monthly Value 0.5°×0.5° Grid Dataset (V2.0) released by the National Meteorological Information Center. Hydrological station runoff data are from the Information Center of the Ministry of Water Resources. Potential evaporation, actual evaporation, soil moisture, groundwater storage and other data are from the GLDAS Noah Land Surface Model product. All hydrological and meteorological data are from January 1961 to December 2016. 1km digital elevation model data are from the Resources and Environmental Science and Data Center of the Chinese Academy of Sciences.

[0056] 2. Prediction factor selection

[0057] The potential influencing factors of the model are selected as elevation data, previous precipitation, temperature, potential evaporation, actual evaporation, soil moisture, and groundwater storage data, that is, the hydrological and meteorological elements at time t-1, and the model output is the basin outlet runoff at time t, where t represents the current month and t-1 represents the previous month.

[0058] 3. Data preprocessing

[0059] The data in step 1 above are converted back to a 0.5°×0.5° grid data set through bilinear interpolation. The mean and variance of each type of data are calculated to complete data standardization.

[0060] In this embodiment, the Z-Score standardization method is used, as shown in the following formula:

[0061]

[0062] 4. Model basic architecture design

[0063] The basic structure of the constructed model is as follows Figure 2 The model includes four modules that represent the hydrological cycle of surface processes, subsoil flow, groundwater, and confluence.

[0064] The surface runoff module takes the previous evapotranspiration ET(t-1), potential evapotranspiration PET(t-1), temperature T(t-1), precipitation P(t-1) and elevation data as input, and the submodule outputs are the current precipitation P(t), evapotranspiration ET(t), and surface runoff RD(t) (all as intermediate variables of the model);

[0065] The soil module takes the previous soil moisture change ΔSM(t-1) as input, and the submodule outputs the current soil moisture change ΔSM(t) and soil flow RS(t) (both as model intermediate variables);

[0066] The underground module takes the soil moisture change ΔSM(t) at the current moment and the groundwater storage change ΔGW(t-1) at the previous moment output by the soil module as input, and the submodule outputs the groundwater storage change ΔGW(t) at the current moment and the underground runoff RG(t) (both as model intermediate variables);

[0067] The confluence module takes the previous inflow runoff IQ(t-1), the groundwater storage change ΔGW(t) and underground runoff RG(t) at the current moment output by the underground module, the surface runoff RD(t) output by the surface module, and the subsoil flow RS(t) output by the soil module as input, and finally outputs the runoff Q(t) at the outlet section of the basin at the current moment.

[0068] Among them, the current evapotranspiration, precipitation, soil moisture changes, groundwater storage changes, etc. are added to the model training as penalty items of the loss function.

[0069] 5. Sub-module architecture design.

[0070] Each submodule uses the convolutional encoder as the basic architecture, such as Figure 2 (b). The encoder includes two sequences of convolution operations, with two 3×3 convolution kernels in a single sequence and a stride of 1. After each operation, the number of filters in the convolution layer is doubled (8->16). The decoder is similar, including two sequences of convolution operations, with 3×3 convolution operations performed in a single sequence, reducing the feature map by half (16->8). Then, a 1×1 convolution operation and a fully connected layer are used to output the result. In the convolution operation, ReLU is used for activation; in the last convolution layer and the fully connected layer, a linear function is used for activation.

[0071] 6. Modify the loss function Loss.

[0072] The water balance relationship used in the model is expressed as:

[0073] ET+ΔSM+ΔGW+Q=P+IQ

[0074] Q=P+IQ-ET-ΔSM-ΔGW

[0075] Among them, Q refers to the theoretical value of runoff at the outlet section of the basin that meets the water balance constraint; P, ET, ΔSM, ΔGW and IQ represent precipitation, evapotranspiration, soil moisture change, groundwater storage change and basin inflow, respectively.

[0076] In order to meet the water balance constraint, the model predicts the outlet cross-sectional runoff (Q pre ) should be close to the theoretical value (Q), then the model loss function is as follows:

[0077] Loss = λMSE physics +(1-λ)MSE data

[0078]

[0079]

[0080] Among them, Q pre Refers to the runoff forecast value, Q obs refers to the runoff observation value, MSE stands for root mean square error, MSE data represents the runoff prediction value (Q pre ) and runoff observation value (Q obs ) error between physics represents the runoff prediction value (Q pre ) is the error between the theoretical value of runoff satisfying the water balance constraint (Q); n is the number of observations; λ is the weight coefficient of the penalty for violating the water balance.

[0081] 7. Model training

[0082] The sample set is composed of monthly precipitation, temperature, etc., and the model parameters are optimized. The learning rate is set to 1e-3, and the sample set is divided into training set, validation set, and test set. The corresponding period is 1961-1992 as the model training period, 1993-2000 as the validation period, and 2001-2016 as the model testing period.

[0083] 8. Model application and evaluation

[0084] The root mean square error, relative error, Nash coefficient, and correlation coefficient are used to compare and evaluate the model effect. The specific results are shown in Table 1 and Figure 3 shown.

[0085] Table 1 The monthly runoff prediction effect of the constructed model in the upper reaches of the Yangtze River

[0086]

[0087] It can be seen from the above results that the present invention uses coupled physical mechanism and deep learning to predict monthly runoff, which can take into account the constraints of physical rules and achieve accurate prediction of monthly runoff.

[0088] The monthly runoff prediction system of the coupled physical mechanism of the present invention comprises a monthly runoff prediction model; the hydrological and meteorological data of the previous month are input into the monthly runoff prediction model to obtain the runoff volume of the basin outlet in the current month;

[0089] The monthly runoff prediction model includes a surface runoff module, a soil module, an underground module and a confluence module connected in sequence, and the surface runoff module, the soil module, the underground module and the confluence module are all convolutional encoder models;

[0090] The surface runoff module predicts the precipitation, evapotranspiration and surface runoff data of the current month based on the evapotranspiration, potential evapotranspiration, temperature, precipitation and elevation data of the previous month;

[0091] The soil module predicts the soil moisture change and soil midflow data for the current month based on the soil moisture change data of the previous month;

[0092] The underground module predicts the groundwater reserve change and underground runoff data for the current month based on the soil moisture change data for the current month and the groundwater reserve change data for the previous month.

[0093] The confluence module predicts the runoff at the basin outlet section of the current month based on the inflow and runoff data of the previous month, the groundwater storage changes of the current month, the underground runoff, the surface runoff and the subsoil flow data.

[0094] The electronic device described in the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded into the processor, the monthly runoff prediction method based on the coupled physical mechanism is implemented.

[0095] The computer-readable storage medium of the present invention stores a computer program, and when the computer program is executed by a processor, the monthly runoff prediction method of the coupled physical mechanism is implemented.

[0096] The computer-readable storage media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, flash memory, or any other medium that can be used to store the desired program code in the form of instructions or data structures and that can be accessed by a computer.

[0097] The processor is used to execute the computer program stored in the memory to implement each step of the method involved in the above embodiment.

Claims

1. A monthly runoff prediction method based on coupled physical mechanisms, characterized in that: The method comprises the following steps: inputting the hydrological and meteorological data of the previous month into the monthly runoff prediction model to obtain the runoff of the watershed outlet section of the current month; The monthly runoff prediction model includes a surface runoff module, a soil module, an underground module and a confluence module connected in sequence, and the surface runoff module, the soil module, the underground module and the confluence module are all convolutional encoder models; The surface runoff module predicts the precipitation, evapotranspiration and surface runoff data of the current month based on the evapotranspiration, potential evapotranspiration, temperature, precipitation and elevation data of the previous month; The soil module predicts the soil moisture change and soil flow data of the current month based on the soil moisture change data of the previous month; The underground module predicts the groundwater reserve change and underground runoff data for the current month based on the soil moisture change data for the current month and the groundwater reserve change data for the previous month. The confluence module predicts the runoff at the basin outlet section of the current month based on the inflow and runoff data of the previous month, the groundwater storage changes of the current month, the underground runoff, the surface runoff and the subsoil flow data.

2. The monthly runoff prediction method based on coupled physical mechanism according to claim 1 is characterized in that: The loss function of the monthly runoff prediction model is constrained by the water balance relationship, which is: ET+ΔSM+ΔGW+Q=P+IQ Q=P+IQ-ET-ΔSM-ΔGW Among them, Q refers to the theoretical value of runoff at the basin outlet section that meets the water balance constraint in the current month, and P, ET, ΔSM, ΔGW and IQ represent precipitation, evapotranspiration, soil moisture change, groundwater storage change and basin inflow in the current month, respectively.

3. The monthly runoff prediction method based on coupled physical mechanism according to claim 2 is characterized in that: The loss function of the monthly runoff prediction model is: Loss=λMSE physics +(1-λ)MSE data Among them, Loss is the loss function, λ represents the weight coefficient of the penalty for violating the water balance, Q pre Refers to the predicted runoff value of the basin outlet section, Q obs Refers to the observed runoff value at the outlet section of the basin, MSE data Indicates Q pre With Q obs The root mean square error, MSE physics Indicates Q pre The root mean square error between the theoretical value Q of the runoff at the basin outlet section that meets the water balance constraint, n is the number of observations.

4. The monthly runoff prediction method based on coupled physical mechanism according to claim 1 is characterized in that: The architectures of the surface runoff module, soil module, underground module and confluence module are all convolutional encoder-convolutional decoder architectures.

5. The monthly runoff prediction method based on coupled physical mechanism according to claim 4 is characterized in that: The convolutional encoder-convolutional decoder architecture includes a convolutional encoder, a convolutional decoder and a fully connected layer; the convolutional encoder and the convolutional decoder both include two sequences of convolution operations, which are used to increase and reduce the number of feature maps, respectively.

6. The monthly runoff prediction method based on coupled physical mechanism according to claim 1 is characterized in that: The hydrological and meteorological data of the previous month were preprocessed, and the hydrological and meteorological data of the previous month included the evapotranspiration, potential evapotranspiration, temperature, precipitation, elevation, soil moisture change and groundwater storage change data of the previous month; the preprocessing included integrating the hydrological and meteorological data of the previous month into the same spatial scale through spatial interpolation method and performing standardization processing.

7. The monthly runoff prediction method based on coupled physical mechanism according to claim 6 is characterized in that: The Z-Score standardization method was used for standardization.

8. A monthly runoff prediction system with coupled physical mechanism, characterized in that: Including monthly runoff prediction model; input the hydrological and meteorological data of the previous month into the monthly runoff prediction model to obtain the basin outlet runoff of the current month; The monthly runoff prediction model includes a surface runoff module, a soil module, an underground module and a confluence module connected in sequence, and the surface runoff module, the soil module, the underground module and the confluence module are all convolutional encoder models; The surface runoff module predicts the precipitation, evapotranspiration and surface runoff data of the current month based on the evapotranspiration, potential evapotranspiration, temperature, precipitation and elevation data of the previous month; The soil module predicts the soil moisture change and soil flow data of the current month based on the soil moisture change data of the previous month; The underground module predicts the groundwater reserve change and underground runoff data for the current month based on the soil moisture change data for the current month and the groundwater reserve change data for the previous month. The confluence module predicts the runoff at the basin outlet section of the current month based on the inflow and runoff data of the previous month, the groundwater storage changes of the current month, the underground runoff, the surface runoff and the subsoil flow data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the monthly runoff prediction method based on the coupled physical mechanism according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the monthly runoff prediction method based on the coupled physical mechanism according to any one of claims 1 to 7 is implemented.

Images