Inland lake hydrological change driving factor identification method based on trend term convolutional neural network

Abstract

The invention discloses an inland lake hydrological change driving factor identification method based on a trend term convolutional neural network, and belongs to the technical field of hydrological simulation and change attribution. Comprising the following steps: searching and collecting runoff observation data of a hydrometric station of a keiske lake and a data set of natural and human activity factors influencing water resources of an inland lake, and preprocessing the data set to form a time sequence data set; an STL trend decomposition method is adopted to extract trend components of the time sequence reflecting the amount of water entering the lake and the influence factors of the water; establishing and training a convolutional neural network (CNN) model reflecting the relationship between the lake water volume trend component and the influence factor trend component; and identifying hydrological change key driving elements by utilizing factorial analysis of contribution of FA influence factors to lake inflow trend change. According to the method, the simulation precision is remarkably improved, and the method has an important decision support effect on regional lake water resource management and environmental protection.

Description

Technical Field

[0001] The present invention relates to the technical field of hydrological simulation and change attribution, and in particular to a method for identifying driving factors of hydrological changes in inland lakes based on a trend term convolutional neural network. Background Art

[0002] Lakes contain 87% of liquid surface freshwater and play a vital role in providing freshwater supply and ecosystem services. However, about 53% of large lakes have experienced a significant decline in water storage. Inland lakes are usually formed at the end of rivers in arid or semi-arid climate regions, and river water inflow is their main water source. Therefore, proposing reliable methods to predict the amount of water inflow into the lake and identify the main driving factors of its changes is of great significance for mitigating climate change and human activities to reduce the risk of shrinking inland lakes and promote regional ecological sustainable development. When using deep learning / machine learning methods to conduct lake hydrological research, meteorological variables (such as temperature and precipitation) or historical data are usually used as modeling factors to train the model, while important trend / seasonal characteristics in the time series of runoff and its predictors are ignored. The lack of this important information limits the performance improvement of statistical hydrological models. In addition, data-driven models such as deep learning / machine learning and time series analysis methods such as STL trend decomposition lack the ability to quantify factor influences and identify the main driving factors. A method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural networks is needed to solve the above problems. Summary of the invention

[0003] The purpose of the present invention is to propose a method for identifying driving factors of inland lake hydrological changes based on a trend term convolutional neural network, comprising the following steps:

[0004] Find and collect the runoff observation data of the Weilu Lake hydrological station and the data set of natural and human activity factors that affect inland lake water resources, and preprocess them to form a time series data set;

[0005] The STL trend decomposition method is used to extract the trend component of the time series reflecting the amount of water entering the lake and its influencing factors;

[0006] Establish and train a convolutional neural network (CNN) model that reflects the relationship between the trend component of water inflow and the trend component of influencing factors;

[0007] The key driving factors of hydrological changes were identified by factorial analysis of the contribution of FA influencing factors to the trend changes of water inflow into the lake.

[0008] Furthermore, the dataset is a global climate model scenario simulation dataset.

[0009] Furthermore, the steps of establishing a convolutional neural network (CNN) model that reflects the relationship between the trend component of the amount of water entering the lake and the trend component of the influencing factors include:

[0010] A 1D convolutional layer is used to extract local features, with 64 filters of kernel size 2 and activation function as ReLU;

[0011] A 1D pooling layer with a pool size of 2 is used to process the feature set to retain key information and improve computational efficiency;

[0012] Use the Flatten layer to transform the collected feature maps;

[0013] Use the fully connected layer to complete feature integration and simulation.

[0014] Furthermore, the steps of using factorial analysis to analyze the contribution of FA influencing factors to the trend change of water inflow to identify the key driving factors of hydrological changes include:

[0015] The trend components of the time series reflecting the amount of water entering the lake and its influencing factors are set to high and low levels respectively;

[0016] Use the convolutional neural network (CNN) model to predict and simulate future water demand in various countries;

[0017] Calculate the impact of a single factor and two factors on water demand;

[0018] Explore in detail the impact of factors on future changes in the trend of water inflow into the lake under different scenarios, and identify the key driving factors of the changes.

[0019] Furthermore, the calculation formula for the influence of a single factor on water demand under the three-factor situation is:

[0020]

[0021] Among them, SS A SS B SS C Represents the sum of squares of factors A, B, and C respectively; Y ijt It represents the system response value when factor A is at the i-th level, factor B is at the j-th level, and factor C is at the t-th level; I, J, and T represent the level numbers of factors A, B, and C, respectively.

[0022] Furthermore, the calculation formula for the influence of two factors on water demand under the three-factor situation is:

[0023]

[0024]

[0025] Among them, SS A×B SS A×C SS B×C Represents the sum of squares of the two-factor interaction.

[0026] The beneficial effects of the present invention are:

[0027] 1. The present invention analyzes the relationship between the amount of water entering the lake and the long-term trend of its factors, uses STL trend decomposition to extract variable trend features to train the convolutional neural network CNN to construct a convolutional neural network based on trend terms, and significantly improves the simulation accuracy.

[0028] 2. The present invention combines the trend-based convolutional neural network with factorial analysis FA, conducts multi-scenario simulation through full factorial design, systematically explores the impact of main factors on the amount of water entering the lake, identifies the main driving factors of hydrological changes, and plays an important role in decision-making support for regional lake water resources management and environmental protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of the framework of the method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network of the present invention;

[0030] Figure 2 Schematic diagram of individual effect of factors;

[0031] Figure 3 Schematic diagram of the contribution of factors to the change of water inflow into the lake. DETAILED DESCRIPTION

[0032] The present invention proposes a method for identifying driving factors of inland lake hydrological changes based on a trend term convolutional neural network. The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0033] Figure 1 The schematic diagram of the framework of the method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network of the present invention is shown in the figure. The implementation process includes the following steps:

[0034] Step A: Search and collect runoff observation data of the tail lake hydrological station and data sets of natural and human activity factors that affect inland lake water resources, give priority to scenario simulation data such as the Global Climate Model (GCM), select the datasets of the Sixth International Coupled Model Intercomparison Project (CMIP6) and the International Interoperable Simulation Impact Projection Project (ISIMIP), perform necessary preprocessing on them and organize them into time series datasets in formats such as Excel;

[0035] Step B: Use the STL trend decomposition method to extract the trend component of the time series of lake water inflow and its influencing factors.

[0036] Step C: Establish and train a convolutional neural network (CNN) model that reflects the relationship between the trend component of water inflow and the trend component of influencing factors; the details are as follows:

[0037] A 1D convolutional layer is used to extract local features, with 64 filters of kernel size 2 and activation function as ReLU;

[0038] A 1D pooling layer with a pool size of 2 is used to process the feature set to retain key information and improve computational efficiency;

[0039] Use the Flatten layer to transform the collected feature maps;

[0040] Use fully connected layers to complete feature integration and simulation;

[0041] Train CNN to complete the construction of trend item convolutional neural network in the target area.

[0042] In this embodiment, three typical study areas were selected, and the monthly water inflow and influencing factor data from 1960 to 2014 were used to calibrate and verify the parameters of the trend item convolutional neural network, which was decomposed into two typical periods of 1960-1990 and 1991-2014. 70% of the data in the two typical periods were used for calibration, and 30% were used for the verification period. The Nash coefficient (NSE) was selected to reflect the relationship between the simulated value and the measured value to evaluate the simulation effect of the Bayesian support vector machine model. The calculation formula of the above evaluation index is as follows:

[0043]

[0044] Where n is the total number of sample points; y obs,i and sim,i denote the observed and simulated values, respectively; is the mean of the observed values.

[0045] Step D: Use factorial analysis to analyze the contribution of FA influencing factors to the trend change of water inflow into the lake and identify the key driving factors of hydrological changes.

[0046] Set the obtained factor trend items to high and low levels respectively;

[0047] The constructed trend convolutional neural network is used to predict and simulate the future water demand of various countries;

[0048] Calculate the effect of a single factor and the effect of each factor on water demand. Taking three factors as an example, the formula for calculating the individual effect of the factors is:

[0049]

[0050] Among them, SS A SS B SS C Represents the sum of squares of factors A, B, and C respectively; Y ijtIt represents the system response value of factor A at the i-th level, factor B at the j-th level, and factor C at the t-th level; I, J, and T represent the level numbers of factors A, B, and C respectively. The formula for calculating the sum of squares of the interaction between two factors is:

[0051]

[0052]

[0053] Among them, SS A×B SS A×C SS B×C Represents the sum of squares of the two-factor interaction.

[0054] Explore in detail the impact of factors on future changes in the trend of water inflow into the lake under different scenarios, and identify the key driving factors of the changes.

[0055] Figure 2 The results show that reservoir water storage is the main driving factor in the three study areas, and large-scale water conservancy project construction has significantly reduced the amount of water entering the lake in the three study areas. Evapotranspiration also has a significant negative impact on the amount of water entering the lake in the three study areas. In 1960-1990, snow cover was the main driving factor in study areas 1 and 3, with a value of -52m 3 / s and -72m 3 / s. This indicates that snowmelt runoff plays a major role in replenishing the amount of water entering the lake, replacing the dominant role of human activity factors.

[0056] Figure 3 Schematic diagram of the contribution of factors to changes in water inflow to the lake. The results show that snowmelt-related factors (surface snow cover and snowmelt) and evapotranspiration are important contributing factors in the three study areas, and their impact varies by study area and period (17.4% ± 24.6%). The impact of reservoir water storage (16.9% ± 7.0%) and irrigation water withdrawal (5.1% ± 23.2%) highlights the significant human intervention in water resource management. Under the future scenario, the contribution of evapotranspiration and snow cover dynamics in all rivers is increasing, indicating the increasing impact of climate change on water resources. Evapotranspiration and irrigation water withdrawal in study area 2 increased significantly, indicating an increased sensitivity to climate change and agricultural practices. Runoff changes in study area 1 are strongly affected by rainfall and snowmelt, reflecting its vulnerability to climate variability.

[0057] The present invention analyzes the relationship between the long-term trend of the amount of water entering the lake and its factors, uses STL trend decomposition to extract variable trend features to train the convolutional neural network CNN to construct a convolutional neural network based on trend terms, which significantly improves the simulation accuracy. The convolutional neural network based on trend terms is combined with factorial analysis FA, and multi-scenario simulation is performed through full factorial design. The influence of the main factors on the amount of water entering the lake is systematically explored, and the main driving factors of hydrological changes are identified. It plays an important role in decision-making support for regional lake water resources management and environmental protection.

Claims

1. A method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network, characterized in that: The following steps are involved: Find and collect the runoff observation data of the Weilu Lake hydrological station and the data set of natural and human activity factors that affect inland lake water resources, and preprocess them to form a time series data set; The STL trend decomposition method is used to extract the trend component of the time series reflecting the amount of water entering the lake and its influencing factors; Establish and train a convolutional neural network (CNN) model that reflects the relationship between the trend component of water inflow and the trend component of influencing factors; The key driving factors of hydrological changes were identified by factorial analysis of the contribution of FA influencing factors to the trend changes of water inflow into the lake.

2. The method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network according to claim 1 is characterized in that: The dataset is a global climate model scenario simulation dataset.

3. The method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network according to claim 2 is characterized in that: The steps of establishing a convolutional neural network (CNN) model that reflects the relationship between the trend component of water inflow and the trend component of influencing factors include: A 1D convolutional layer is used to extract local features, with 64 filters of kernel size 2 and activation function as ReLU; A 1D pooling layer with a pool size of 2 is used to process the feature set to retain key information and improve computational efficiency; Use the Flatten layer to transform the collected feature maps; Use the fully connected layer to complete feature integration and simulation.

4. The method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network according to claim 3 is characterized in that: The steps of using factorial analysis to identify the key driving factors of hydrological changes by analyzing the contribution of FA influencing factors to the trend change of water inflow to the lake include: The trend components of the time series reflecting the amount of water entering the lake and its influencing factors are set to high and low levels respectively; Use the convolutional neural network (CNN) model to predict and simulate future water demand in various countries; Calculate the impact of single factor and two factors on water demand; Explore in detail the impact of factors on future changes in the trend of water inflow into the lake under different scenarios, and identify the key driving factors of the changes.

5. The method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network according to claim 4 is characterized in that: The calculation formula for the effect of a single factor on water demand under the three-factor situation is: Among them, SS A SS B SS C Represents the sum of squares of factors A, B, and C respectively; Y ijt It represents the system response value of factor A at the i-th level, factor B at the j-th level, and factor C at the t-th level; I, J, and T represent the level numbers of factors A, B, and C, respectively.

6. The method for identifying driving factors of inland lake hydrological changes based on trend term convolutional neural network according to claim 5 is characterized in that: The calculation formula for the influence of two factors on water demand under the three-factor situation is: Among them, SS A×B SS A×C SS B×C Represents the sum of squares of the two-factor interaction.

Images