Hydrometric site arrangement method and system based on satellite remote sensing and artificial intelligence

Abstract

The invention relates to a hydrological station arrangement method and system based on satellite remote sensing and artificial intelligence, and belongs to the technical field of hydrological monitoring. The method comprises the following steps: firstly, acquiring terrain, water system and vegetation information by using Sentinel-1 / Landsat and the like, and after radiation correction, extracting features by using a CNN-LSTM coupled hydrological model; evaluating a potential region by information entropy-spatial autocorrelation, and introducing variable structure mimicry calculation to dynamically optimize resources; and finally, a station is optimized under error and cost constraints by using a multi-target algorithm, and future representativeness is verified by using a DI exponential emergence constraint model. According to the invention, large-range, high-precision and low-cost intelligent arrangement of hydrometric stations can be realized.

Description

Technical Field

[0001] This invention relates to a method and system for the layout of hydrological stations based on satellite remote sensing and artificial intelligence, belonging to the field of hydrological monitoring technology. Background Technology

[0002] Hydrological stations are of great significance for water resource monitoring, management, and related research. Their rational layout can more accurately obtain hydrological information, providing a reliable basis for water conservancy project construction, water resource allocation, flood control, and drought relief. Traditional methods of hydrological station layout have many limitations. On the one hand, they mostly rely on human experience and limited ground observation data. In areas with complex terrain and sparse population, it is difficult to comprehensively and accurately grasp the hydrological characteristics, resulting in a lack of scientific and rational station layout. For example, in mountainous areas, due to the large topographic relief, it is difficult to capture the hydrological differences at different altitudes and slope orientations using only ground observations. On the other hand, traditional methods cannot fully utilize the advantages of modern information technology. When facing the needs of large-scale, long-term hydrological monitoring, they are inefficient, and the monitoring accuracy cannot meet the ever-increasing demands.

[0003] The European Space Agency (ESA) is actively promoting the application of satellite remote sensing technology in hydrological monitoring. Its Sentinel series satellites provide a wealth of data sources for acquiring high-precision, wide-area surface information. For example, Sentinel-1 radar remote sensing data can effectively penetrate cloud cover and clearly detect water system distribution and topographic features in areas with complex terrain and variable climates, providing basic data for the preliminary selection of hydrological station locations. NASA, using the Landsat series of optical remote sensing images, has been monitoring global surface changes over the long term, accurately classifying vegetation cover and land use types in different regions. This provides important evidence for analyzing regional hydrological characteristics, and when determining potential hydrological station locations, these characteristics can be used to select areas with minimal human interference and relatively stable hydrological processes.

[0004] A research team at Stanford University in the United States has developed advanced deep learning algorithms that, through learning from a large amount of satellite remote sensing imagery and historical hydrological data, can quickly and accurately extract hydrological features. For example, they can use convolutional neural networks (CNNs) to identify river boundaries and calculate river network density, achieving significantly improved accuracy compared to traditional methods. A collaboration between the University of Oxford in the UK and the Image Processing Laboratory of the University of Valencia in Spain has constructed a flood monitoring model based on neural networks. Through satellite remote sensing and artificial intelligence technologies, this model enables near real-time monitoring and analysis of floods, providing crucial information for optimizing the layout of hydrological stations in flood-prone areas to better capture dynamic changes in floods. Domestic universities such as Tsinghua University and Wuhan University have made significant progress in applying artificial intelligence algorithms to hydrological research. They have used recurrent neural networks (RNNs) and their variant, long short-term memory networks (LSTMs), to model hydrological time-series data and analyze precipitation-runoff relationships, water level and discharge variation patterns, etc.

[0005] Inspired by mimicry phenomena in nature, variable structure biomimetic computing technology can dynamically adjust the structure according to different computing needs. Through software and hardware co-design (such as software-defined interconnect technology and software-defined on-chip system), it can achieve dynamic resource coupling and intelligent system reconfiguration, which can significantly improve computing efficiency and reduce energy consumption, providing a new approach to solving the problems of efficient data processing and model optimization in hydrological station layout.

[0006] With the development of satellite remote sensing technology, it is possible to acquire large-scale, continuous surface information, including topography, water systems, and vegetation, providing a rich data foundation for the layout of hydrological stations. Meanwhile, artificial intelligence technology has demonstrated powerful capabilities in data analysis and processing, enabling efficient analysis of massive amounts of remote sensing data to uncover hidden hydrological information and patterns. Therefore, combining satellite remote sensing with artificial intelligence technology for hydrological station layout has become an effective way to address the shortcomings of traditional methods. Summary of the Invention

[0007] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for the layout of hydrological stations based on satellite remote sensing and artificial intelligence, so as to solve the problems of insufficient scientificity and rationality, low efficiency and difficulty in meeting the monitoring accuracy requirements of existing hydrological station layout methods.

[0008] The technical solution adopted in this invention is as follows:

[0009] A method for locating hydrological stations based on satellite remote sensing and artificial intelligence includes the following steps:

[0010] S1. Collect and preprocess multi-source satellite remote sensing data of the study area, and collect existing hydrological stations and historical monitoring data of the study area;

[0011] S2. Use CNN-LSTM coupled hydrological models to extract topographic, water system and vegetation features from preprocessed satellite remote sensing data; use historical monitoring data from hydrological stations to establish various hydrological models, including precipitation-runoff relationship models, water level change law models and flow change characteristic models, to analyze the hydrological processes in the study area;

[0012] S3. Using multiple artificial intelligence models, areas with significant changes in hydrological characteristics are identified, and these areas are designated as potential hydrological station deployment areas. Information entropy theory and spatial autocorrelation analysis methods are used to evaluate the information richness and spatial representativeness of each potential area. Variable structure mimicry computing technology is used to dynamically adjust the structure according to different computing needs to select potential stations with rich information and strong spatial representativeness.

[0013] S4. Based on artificial intelligence genetic algorithms or particle swarm optimization algorithms, with the goal of minimizing monitoring errors and maximizing spatial representativeness and information content, the potential stations selected are optimized and laid out. Under the constraints of economic cost and construction difficulty, the optimal hydrological station layout scheme is determined while meeting the monitoring accuracy requirements.

[0014] S5. Construct a multi-scale emergent constraint model for the hydrological station layout area, and use historical hydrological data and simulation data to evaluate and verify the future representativeness of the determined hydrological station layout scheme.

[0015] In the above method, the terrain features mentioned in step S2 include slope, aspect, and elevation; the water system features include river length, river network density, and watershed area; and the vegetation features include vegetation index and vegetation coverage.

[0016] The artificial intelligence model mentioned in step S3 includes convolutional neural networks, long short-term memory models, support vector machines, random forests, and artificial neural networks; the artificial intelligence model is trained using the gradient descent method to obtain the calibrated artificial intelligence model.

[0017] In step S3, the information richness of the potential region is evaluated using information entropy theory. The calculation formula is as follows:

[0018]

[0019] Where p(x) i H(X) represents the probability of the i-th hydrological feature occurring. The larger the value of H(X), the richer the hydrological information in the area.

[0020] Spatial autocorrelation analysis uses Moran's I index, calculated as follows:

[0021]

[0022] Where wij Let x be the spatial weight matrix. i Let i be the hydrological characteristic value of region i. This is the average value.

[0023] The variable structure mimicry computing technique described in step S3 is as follows:

[0024] (1) Calculate the richness of information of the indicator I i The total amount of effective information that site i can provide is represented as:

[0025]

[0026] Where n is the information dimension, w k f represents the weights for each dimension. k (x i Let be the representation function of site i in dimension k;

[0027] (2) Spatial representativeness Si: measures the representativeness of station i to the surrounding area, expressed as:

[0028] Si=1-1 / A∑d(χi,χj)j∈Ωi,

[0029] Where A is the total area of ​​the region, Ω i Let d(xi,xj) be the coverage area of ​​station i, and let d(xi,xj) be the spatial distance between station i and point j within the area.

[0030] (3) Calculate the comprehensive score Fi: Combining the above two factors, the comprehensive value of site i is:

[0031] Fi=α·I i +(1-α)·S i ,

[0032] Where α is the balance coefficient, 0 < α < 1.

[0033] By dynamically adjusting the calculated weight w k Coverage area Ω i Improve computational efficiency by accelerating I / O through parallel computing. i and S i Batch calculation; or by optimizing the distance d(x) i ,x j Improve calculation accuracy.

[0034] The objective function established in step S4 above is the fitness function, which minimizes the monitoring error E, maximizes the spatial representativeness S and the information content I, and considers the constraints of cost C and construction difficulty D. The fitness function can be expressed as:

[0035]

[0036] Where x is the site layout scheme, which is binary encoded, with 1 indicating the establishment of a site and 0 indicating the absence of a site, and ω1-ω5 are weight coefficients used to optimize the selected potential sites.

[0037] In step S5, for the preferred site layout, a linear relationship is constructed between the historical temperature change trend and the future DI index change trend simulated by the temperature change model, specifically as follows:

[0038] DItrend k =a·Ttrend k +b,

[0039] Where: DItrend k and Ttrend k Let a and b represent the future trend of the DI index and the historical trend of the monthly average temperature at the k-th grid point, respectively; a and b represent the model parameters; the parameters a and b of the emergence constraint model are solved using the least squares method.

[0040] Furthermore, a linear relationship is constructed between the historical precipitation change trend and the future DI index change trend simulated by the precipitation change model, specifically:

[0041] DItrend2 k =a2·exp(Ptrend k )+b2,

[0042] In the formula: DItrend2 k and Ptrend k denoted as the future trend of the DI index and the historical trend of monthly precipitation at the kth grid point, respectively; a2 and b2 represent the model parameters, which are solved using the least squares method.

[0043] The trend of the DI index was obtained by optimizing the two models mentioned above, as follows:

[0044] DItrendCOR k =w1·DItrend k +w2·DItrend2 k

[0045] Where: DItrendCOR k The final result of the DI trend; w1 and w2 are the weights.

[0046] Finally, the optimal solution is selected based on the changes in the DI index.

[0047] Another objective of this invention is to provide a hydrological station layout system based on satellite remote sensing and artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the hydrological station layout method based on satellite remote sensing and artificial intelligence as described above.

[0048] The beneficial effects of this invention are:

[0049] (1) This invention utilizes multi-source remote sensing data such as Sentinel-1 / Landsat to overcome terrain and climate limitations and achieve complete identification and spatial coverage of potential site areas;

[0050] (2) This invention extracts multiple features by coupling a hydrological model with CNN-LSTM, which significantly improves the accuracy and reduces the RMSE of daily flow simulation from 12.3m in traditional ARIMA. 3 / s decreased to 8.6m 3 / s;

[0051] (3) Adopting variable structure mimicry computing to reconstruct hardware as needed during the site selection stage, thereby achieving dual optimization of computing efficiency and energy consumption;

[0052] (4) Through multi-objective optimal layout: GA / PSO joint optimization, 85% spatial representative coverage can be achieved under the constraints of ≤5% monitoring error and ≤1 million yuan cost.

[0053] (5) Validation through future representativeness: By using the emergence constraint model of DI index under historical climate and future scenarios, the uncertainty of the future representativeness prediction of the site is significantly reduced.

[0054] This invention enables the intelligent deployment of hydrological stations on a large scale, with high precision and low cost. Attached Figure Description

[0055] Figure 1 This is a flowchart of the method of the present invention; Detailed Implementation

[0056] The present invention will be further described below with reference to specific embodiments.

[0057] Example 1: A method for locating hydrological stations based on satellite remote sensing and artificial intelligence, comprising the following steps:

[0058] S1. Collect and preprocess multi-source satellite remote sensing data of the study area, and collect existing hydrological station and historical monitoring data of the study area:

[0059] Multi-source satellite remote sensing data, including optical remote sensing imagery and radar remote sensing data, was collected for the study area to obtain information on topography, water systems, and vegetation cover. Existing hydrological station data, including historical monitoring data on water level, flow rate, and water quality, were also collected. Preprocessing operations such as radiometric and geometric corrections were performed on the satellite remote sensing data to improve data quality. Quality control and outlier handling were implemented for the hydrological station data to ensure accuracy and reliability.

[0060] Taking the Yangtze River Basin in Hubei Province as an example, Landsat series optical remote sensing images and Sentinel-1 radar remote sensing data were collected for the region, spanning from 2000 to 2025. Remote sensing image processing software such as ENVI and Erdas were used to perform radiometric and geometric corrections on the satellite remote sensing data to achieve high-precision geolocation and radiometric calibration. Simultaneously, monitoring data on water level, flow rate, and water quality from three existing hydrological stations in the region were collected from 2000 to 2025. Data quality control was implemented to remove outliers and erroneous data.

[0061] The variables obtained or derived from ERA5-Land include air temperature, relative humidity, snowfall, precipitation, runoff depth, shortwave radiation, and longwave radiation.

[0062] S2. A CNN-LSTM coupled hydrological model was used to extract topographic, river system, and vegetation features from the preprocessed satellite remote sensing data. Using historical monitoring data from hydrological stations, various hydrological models were established, including a precipitation-runoff relationship model, a water level change pattern model, and a flow change characteristic model, to analyze the hydrological processes in the study area.

[0063] Artificial intelligence algorithms are used to analyze preprocessed satellite remote sensing data and extract topographic features (such as slope, aspect, elevation, etc.), water system features (such as river length, river network density, watershed area, etc.), and vegetation features (such as vegetation index, vegetation coverage, etc.).

[0064] The CNN-LSTM coupled hydrological model first uses a modified U-Net architecture CNN model to extract hydrological features, extracts topographic, water system and vegetation features from preprocessed satellite remote sensing data, and then uses an LSTM (Long Short-Term Memory) network model to analyze hydrological time series data.

[0065] Hydrological feature extraction using a CNN model with the U-Net architecture involves first performing convolutional layer calculations, followed by pooling layer calculations.

[0066] The calculation process of the convolutional layer is as follows:

[0067]

[0068] Where W∈RM×N×C×K The kernel is M×N; (M×N is the kernel size, K is the number of output channels), b k The bias term is σ, and the activation function is ReLU (σ(x) = max(0,x)).

[0069] Pooling layer calculation:

[0070] Downsampling is performed using max pooling.

[0071] Y i,j,k =max{X i×s+m,j×s+n,k |0≤m,n <s}

[0072] Where s is the pooling step size.

[0073] In this embodiment, a 2×2 max pooling operation with s=2 is used.

[0074] Input data: Landsat-8 remote sensing imagery (30m resolution, containing 7 spectral bands),

[0075] Network structure: 4-layer encoder (each containing 2 convolutional layers + 1 pooling layer) and 4-layer decoder (each containing 2 convolutional layers + 1 upsampling layer).

[0076] Training parameters: learning rate 0.001, batch size 16, 50 iterations.

[0077] Feature extraction: Hydrological features such as slope α, aspect β, and water system density ρ of the study area are obtained through the feature map output by the model.

[0078] Hydrological process modeling based on recurrent neural networks (RNNs) employs LSTM (Long Short-Term Memory) models to analyze hydrological time-series data, addressing the gradient vanishing problem inherent in traditional RNNs.

[0079] f t =σ(W f ·[h t-1 , χ t ]+b f Forgotten Gate

[0080] i t =σ(W i ·[h t-1 ,x t ]+b i Input gate

[0081]

[0082] σ t =σ(W o ·[h t-1 , χ t ]+bo Output gate

[0083] h t =o t ☉tanh(C t Hidden state

[0084] Where (☉) denotes element-wise multiplication, (σ) is the sigmoid function, and W and b are the weight matrix and bias term, respectively.

[0085] In this embodiment, 10 years of daily flow data (3650 records) from 20 existing hydrological stations in the study area were used. A 2-layer LSTM (64 hidden units per layer) + 1 fully connected layer was employed, including precipitation Pt, air temperature Tt, and anterior flow Q. t-1 The vegetation index NDVIt extracted by remote sensing is used to predict daily flow Q. t .

[0086] By combining historical monitoring data from hydrological stations, various hydrological models are established to analyze the hydrological processes in the study area, including precipitation-runoff relationship models, water level change patterns models, and flow change characteristic models.

[0087] For example, the precipitation-runoff relationship model uses the SCS (Soil Conservation Service) model, which is a commonly used precipitation-runoff estimation model. The core formula is as follows:

[0088] When the rainfall (p≥0.2S) is equal to the runoff depth, the runoff depth is equal to the runoff depth.

[0089] When the rainfall (P < 0.2S), the runoff depth (Q = 0) is...

[0090] in:

[0091] Q is the runoff depth (mm).

[0092] P represents rainfall (mm).

[0093] S is the maximum retention capacity of the watershed (mm), and

[0094] CN represents the curve number (reflecting the characteristics of the underlying surface of the watershed, ranging from 0 to 100; the larger the value, the stronger the runoff generation capacity).

[0095] For example, in a water level change model, the relationship between water level (H) and flow rate (Q) can be represented by a linear regression model, and then the change pattern of water level over time can be derived:

[0096] Water level-flow rate relationship:

[0097] Q = aH + b (a and b are fitting parameters, determined by measured data).

[0098] Water level change pattern (assuming flow rate changes linearly with time):

[0099]

[0100] in:

[0101] H(t) is the water level (m) at time t.

[0102] Q(t) = ct + d is the flow rate (m³) at time t. 3 / s),

[0103] k = c / a and m = (db) / a are parameters for how the water level changes over time.

[0104] Flow change characteristic model:

[0105] Taking the Muskingum confluence model as an example,

[0106] The flow rate calculation equation is used to describe the variation of flow rate along the course of a river flood.

[0107] Q2 = C0I2 + C1I1 + C2Q1,

[0108] in:

[0109] Q2 is the outflow rate (m³) at the end of the calculation period. 3 / s),

[0110] I2 represents the inflow rate (m³) at the end of the calculation period. 3 / s),

[0111] I1 represents the inflow rate (m³) at the beginning of the calculation period. 3 / s),

[0112] Q1 is the outflow rate (m³) at the beginning of the calculation period. 3 / s),

[0113] coefficient

[0114] K is the propagation time parameter (h), x is the flow weight factor (0≤x≤0.5), and Δt is the calculation period (h).

[0115] S3. Employing multiple artificial intelligence models, areas with significant changes in hydrological characteristics are identified, and these areas are designated as potential hydrological station deployment areas. Using information entropy theory and spatial autocorrelation analysis, the information richness and spatial representativeness of each potential area are evaluated. Variable structure mimicry computing technology is employed to dynamically adjust the structure according to different computational needs, selecting potential stations with rich information and strong spatial representativeness.

[0116] For the hydrological station deployment area, five artificial intelligence models are used to identify areas with significant changes in hydrological characteristics, and these areas are selected as potential hydrological station deployment areas. Traditional machine learning simulation methods generally calculate the regional average hydrological parameter variables first, and then use them to calibrate the artificial intelligence model. This invention uses meteorological and hydrological variables of all grid points in the study area as input, which is more consistent with the physical driving mechanism of terrestrial water storage. The five artificial intelligence models mentioned in this invention are convolutional neural networks, long short-term memory models, support vector machines, random forests, and artificial neural networks. The gradient descent method is used to train the five artificial intelligence models to obtain the calibrated models. The calibration method of the artificial intelligence models is a conventional technique in this field.

[0117] For five artificial intelligence models, derive the weight parameters for the multi-model weighted average method. Specifically, this includes:

[0118] A year has 12 months, from January to December. In the weighting calculation scheme, each month has the same weight in different years. Therefore, it is necessary to calculate the weights for each of the 12 months separately, as follows:

[0119] For each month, the weight parameters of the 5 AI models satisfy:

[0120]

[0121] Where i represents the sequence number of the artificial intelligence model, w o This represents the weight parameters of the artificial intelligence model, and M represents the number of artificial intelligence models.

[0122] Calculate the independence weight parameters and skill-based weight parameters for each AI model. The independence weight parameter w for each AI model... u It can be represented as:

[0123]

[0124] Wherein, S(δ) i,j The similarity () represents the similarity between the current AI model and other AI models, with a value range of (0,1). The calculation formula is as follows:

[0125]

[0126] Where, δ i,j The Euclidean difference between different artificial intelligence models is obtained by standardizing the average difference of simulated rainfall under artificial intelligence models i and j; D u The similarity radius is obtained by iteratively calculating the difference between simulated and measured rainfall under the optimal combination scenario.

[0127] Skill weight parameter w of artificial intelligence modelq It can be represented as:

[0128]

[0129] Where, δ i,obs The error of the artificial intelligence model i relative to the actual observation is obtained by standardizing the root mean square error of the simulated rainfall relative to the measured rainfall; D q The mass radius of the model is represented by the difference between the simulated rainfall and the measured rainfall under the optimal combination scenario; when D q When the value is close to 0, only the optimal AI model has a high skill weight.

[0130] Using information entropy theory to assess the information richness of potential regions:

[0131]

[0132] Where p(x) i H(X) represents the probability of the i-th hydrological feature occurring. The larger the value of H(X), the richer the hydrological information in the area.

[0133] Spatial autocorrelation analysis uses Moran's I index:

[0134]

[0135] Where w ij Let x be the spatial weight matrix. i Let i be the hydrological characteristic value of region i. This is the average value.

[0136] Employing variable-structure mimicry computing technology, the structure is dynamically adjusted according to different computing needs. Through hardware and software co-design (such as software-defined interconnect technology and software-defined on-chip systems), dynamic resource coupling and intelligent system reconfiguration are achieved, which can improve computing efficiency and reduce energy consumption, ultimately selecting potential sites with rich information and strong spatial representativeness; the process is as follows:

[0137] (1) Calculate the richness of information of the indicator I i The total amount of effective information that site i can provide is represented as:

[0138]

[0139] Where n represents the information dimension (such as data accuracy, timeliness, and coverage), and w k f represents the weights for each dimension. k (x i ) represents the performance function of site i in dimension k (such as data collection frequency, error rate);

[0140] (2) Spatial representativeness Si: measures the representativeness of station i to the surrounding area, expressed as:

[0141] Si = 1 - 1 / A∑d(xii,χj) j∈Ωi ,

[0142] Where A is the total area of ​​the region, Ω i Let d(xi,xj) be the coverage area of ​​station i, and let d(xi,xj) be the spatial distance between station i and point j within the area (the smaller the distance, the stronger the representativeness).

[0143] (3) Calculate the comprehensive score Fi: Combining the above two factors, the comprehensive value of site i is:

[0144] Fi=α·I i +(1-α)·S i ,

[0145] Where α is the balance coefficient, 0 < α < 1.

[0146] By dynamically adjusting the calculated weight w k Coverage area Ω i Improve computational efficiency by accelerating I / O through parallel computing. i and S i Batch calculation; or by optimizing the distance d(x) i ,x j Improve calculation accuracy.

[0147] S4. Based on artificial intelligence genetic algorithms or particle swarm optimization algorithms, with the objectives of minimizing monitoring errors and maximizing spatial representativeness and information content, the layout of the selected potential stations is optimized. Under the constraints of economic cost and construction difficulty, the optimal hydrological station layout scheme is determined while meeting the monitoring accuracy requirements.

[0148] Based on artificial intelligence genetic algorithms, which simulate biological evolution, populations are optimized through selection, crossover, and mutation operations. The core formula is as follows:

[0149] Fitness function (objective function): For hydrological station optimization, the objectives are to minimize monitoring error (E), maximize spatial representativeness (S) and information content (I), while also considering cost (C) and construction difficulty (D) constraints. The fitness function can be expressed as:

[0150]

[0151] Where x represents the site layout scheme (binary encoding, 1 indicates a site is set up, 0 indicates no site is set up).

[0152] ω1-ω5 are weighting coefficients (adjusted according to the importance of the target).

[0153] A certain watershed has 20 potential hydrological stations, and 8 need to be selected, meeting the requirements of monitoring error ≤5% and total cost ≤1 million yuan. The population size is 50, crossover probability is 0.8, mutation probability is 0.05, and the number of iterations is 100.

[0154] Initialize the population: randomly generate 50 binary schemes (length 20, containing 8 ones).

[0155] Calculate fitness: Substitute indicators such as monitoring error, spatial coverage, and construction cost of each scheme into the fitness function.

[0156] Selection: Select 30 individuals with high fitness through roulette.

[0157] Crossover and mutation: Selected individuals are crossovered (generating 20 offspring) and mutated to form a new population.

[0158] Iteration: Repeat steps 2-4 to finally obtain the optimal solution: the 8 selected stations have a monitoring error of 4.2%, a cost of 950,000 yuan, and spatial representativeness covering 85% of the entire basin.

[0159] The Particle Swarm Optimization (PSO) algorithm uses PSO to simulate the foraging behavior of bird flocks and searches for the optimal solution by updating particle position and velocity.

[0160] In the layout of hydrological stations, the location can be encoded as a continuous variable representing the station coordinates, or a binary variable representing whether a station is established. Here, we take continuous coordinates as an example.

[0161] Similar to GA, the objective function is constructed based on factors such as monitoring error, spatial representativeness, and cost.

[0162] f(χ i = minimize(E(x) i ))+maximize(S(χ i )+I(χ i ))-λ·C(χ i )

[0163] Where λ is the cost penalty coefficient.

[0164] Based on artificial intelligence genetic algorithms and particle swarm optimization algorithms, the potential stations selected are optimized to minimize monitoring errors and maximize spatial representativeness and information content. Considering constraints such as economic cost and construction difficulty, the optimal hydrological station layout scheme is determined under the premise of meeting the monitoring accuracy requirements.

[0165] S5. Construct a multi-scale emergent constraint model for the hydrological station layout area, and use historical hydrological data and simulation data to evaluate and verify the future representativeness of the determined hydrological station layout scheme:

[0166] For the area where hydrological stations are located, a multi-scale emergent constraint model is constructed, specifically as follows:

[0167] First, for the preferred site layout, a linear relationship is constructed between the historical temperature change trend (1940-2014) and the future (2030-2100) DI index change trend simulated by the temperature change model, specifically:

[0168] DItrend k =a·Ttrend k +b,

[0169] Where: DItrend k and Ttrend k Let a and b represent the future trend of the DI index and the historical trend of the monthly average temperature at the k-th grid point, respectively; a and b represent the model parameters; the parameters a and b of the emergence constraint model are solved using the least squares method.

[0170] Furthermore, a linear relationship is constructed between the historical precipitation change trend (1940-2014) and the future (2030-2100) DI index change trend simulated by precipitation change models, specifically:

[0171] DItrend2 k =a2·exp(Ptrend k )+b2

[0172] In the formula: DItrend2 k and Ptrend k denoted as the future trend of the DI index and the historical trend of monthly precipitation at the kth grid point, respectively; a2 and b2 represent the model parameters, which are solved using the least squares method.

[0173] Furthermore, the changing trend of the DI index was obtained by optimizing the two models mentioned above, as follows:

[0174] DItrendCOR k =w1·DItrend k +w2·DItrend2 k

[0175] Where: DItrendCOR k The final result of the DI trend; w1 and w2 are the weights.

[0176] Finally, based on the changes in the DI index, the optimal layout of hydrological stations is predicted.

[0177] Example 2: A hydrological station layout system based on satellite remote sensing and artificial intelligence, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the hydrological station layout method based on satellite remote sensing and artificial intelligence as described in Example 1 above.

[0178] The above is a further description of the present invention in conjunction with embodiments, and the scope of protection of the present invention is not limited thereto.

Claims

1. A method for arranging hydrological stations based on satellite remote sensing and artificial intelligence, characterized in that, The steps include the following: S1. Collect and preprocess multi-source satellite remote sensing data of the study area, and collect existing hydrological stations and historical monitoring data of the study area; S2. Use CNN-LSTM coupled hydrological models to extract topographic, water system and vegetation features from preprocessed satellite remote sensing data; use historical monitoring data from hydrological stations to establish various hydrological models, including precipitation-runoff relationship models, water level change law models and flow change characteristic models, to analyze the hydrological processes in the study area; S3. Using multiple artificial intelligence models, areas with significant changes in hydrological characteristics are identified, and these areas are designated as potential hydrological station deployment areas. Information entropy theory and spatial autocorrelation analysis methods are used to evaluate the information richness and spatial representativeness of each potential area. Variable structure mimicry computing technology is used to dynamically adjust the structure according to different computing needs to select potential stations with rich information and strong spatial representativeness. S4. Based on artificial intelligence genetic algorithms or particle swarm optimization algorithms, with the goal of minimizing monitoring errors and maximizing spatial representativeness and information content, the potential stations selected are optimized and laid out. Under the constraints of economic cost and construction difficulty, the optimal hydrological station layout scheme is determined while meeting the monitoring accuracy requirements. S5. Construct a multi-scale emergent constraint model for the hydrological station layout area, and use historical hydrological data and simulation data to evaluate and verify the future representativeness of the determined hydrological station layout scheme.

2. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, The terrain features mentioned in step S2 include slope, aspect, and elevation; the water system features include river length, river network density, and drainage area; and the vegetation features include vegetation index and vegetation coverage.

3. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, The artificial intelligence model mentioned in step S3 includes convolutional neural networks, long short-term memory models, support vector machines, random forests, and artificial neural networks; the artificial intelligence model is trained using the gradient descent method to obtain the calibrated artificial intelligence model.

4. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, In step S3, the information richness of the potential region is evaluated using information entropy theory. The calculation formula is as follows: Where p(x) i H(X) represents the probability of the i-th hydrological feature occurring. The larger the value of H(X), the richer the hydrological information in the area. Spatial autocorrelation analysis uses Moran's I index, calculated as follows: Where w ij Let x be the spatial weight matrix. i Let i be the hydrological characteristic value of region i. This is the average value.

5. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, The variable structure mimicry computing technique described in step S3 is as follows: (1) Calculate the richness of information of the indicator I i The total amount of effective information that site i can provide is represented as: Where n is the information dimension, w k f represents the weights for each dimension. k (x i Let be the representation function of site i in dimension k; (2) Spatial representativeness Si: measures the representativeness of station i to the surrounding area, expressed as: Si=1-1 / A∑d(xi,xj)j∈Ωi, Where A is the total area of ​​the region, Ω i Let d(xi,xj) be the coverage area of ​​station i, and let d(xi,xj) be the spatial distance between station i and point j within the area. (3) Calculate the comprehensive score Fi: Combining the above two factors, the comprehensive value of site i is: Fi=α·I i +(1-α)·S i , Where α is the balance coefficient, 0 < α < 1.

6. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, The objective function established in step S4 is the fitness function, which minimizes the monitoring error E, maximizes the spatial representativeness S and the information content I, and considers the constraints of cost C and construction difficulty D. The fitness function can be expressed as: Where x is the site layout scheme, which is binary encoded, with 1 indicating the establishment of a site and 0 indicating the absence of a site, and ω1-ω5 are weight coefficients used to optimize the selected potential sites.

7. The method for arranging hydrological stations based on satellite remote sensing and artificial intelligence according to claim 1, characterized in that, In step S5, for the preferred site layout, a linear relationship is constructed between the historical temperature change trend and the future DI index change trend simulated by the temperature change model, specifically as follows: DItrend k =a·Ttrend k +b, Where: DItrend k and Ttrend k Let a and b represent the future trend of the DI index and the historical trend of the monthly average temperature at the k-th grid point, respectively; a and b represent the model parameters; the parameters a and b of the emergence constraint model are solved using the least squares method. A linear relationship is constructed between historical precipitation trends and future DI index trends simulated by precipitation change models, specifically as follows: DItrend2 k =a2·exp(Ptrend k +b2 In the formula: DItrend2 k and Ptrend k denoted as the future trend of the DI index and the historical trend of monthly precipitation at the kth grid point, respectively; a2 and b2 represent the model parameters, which are solved using the least squares method. The trend of the DI index was obtained by optimizing the two models mentioned above, as follows: DItrendCOR k =w1·DItrend k +w2·DItrend2 k Where: DItrendCOR k The final result of the DI trend; w1 and w2 are the weights. Finally, the optimal solution is selected based on the changes in the DI index.

8. A hydrological station layout system based on satellite remote sensing and artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the hydrological station layout method based on satellite remote sensing and artificial intelligence as described in any one of claims 1-7.

Images