An ecological flow monitoring method and system based on microwave Doppler effect and optimized bidirectional long short-term memory network

Abstract

The application provides an ecological flow monitoring method and system based on a microwave Doppler effect and an optimized bidirectional long-short term memory network, and relates to the technical field of water conservancy project monitoring. The method comprises the following steps: performing grid discrete processing on a target river and constructing a river partition topological structure; calculating an initial roughness coefficient and interpolating a dynamic roughness spatial distribution field; constructing a Legendre orthogonal function model based on the distribution field constraint; using a bidirectional long-short term memory network to solve and obtain a set of water dynamic field fitting coefficients; mapping the set of fitting coefficients to a space-time relationship graph based on the topological structure, and calculating a global correction coefficient; correcting real-time flow data according to the global correction coefficient to obtain final ecological flow monitoring results. The application realizes fine characterization of river dynamic resistance characteristics and global correction of systematic errors by fusing physical mechanisms and data-driven models, and significantly improves the accuracy of ecological flow monitoring under complex water conditions.

Description

Technical Field

[0001] This invention relates to the field of water conservancy engineering monitoring technology, and in particular to an ecological flow monitoring method and system based on microwave Doppler effect and optimized bidirectional long short-time memory network. Background Technology

[0002] Ecological flow is not only a core element in ensuring the integrity of watershed ecological functions, but also an important indicator for water conservancy projects to fulfill their ecological protection responsibilities. Currently, monitoring mainly relies on contact equipment such as current meters or acoustic Doppler current profilers. While these methods offer a certain level of measurement accuracy, they are extremely difficult to implement during high flood seasons or under complex water conditions, and the equipment maintenance costs are high. With technological advancements, non-contact technologies based on radar wave velocity measurement or image velocity measurement are gradually being applied in real-world scenarios. These technologies are typically used in conjunction with hydraulic models such as Manning's formula to calculate cross-sectional flow using surface velocity, becoming one of the mainstream methods for obtaining flow data in the industry.

[0003] Currently, ecological flow monitoring technology is rapidly evolving towards multi-source sensing fusion and intelligent inversion. The industry is beginning to explore the use of drones, satellite remote sensing, and ground sensors to construct an integrated air-space-ground sensing network, aiming to acquire more refined data on river boundaries and water surface flow fields. Simultaneously, deep learning algorithms are being introduced to process long-sequence hydrological data, attempting to leverage the powerful nonlinear fitting capabilities of neural networks to address the poor adaptability of traditional single hydraulic empirical formulas in complex nonlinear hydrological environments, thus meeting increasingly stringent regulatory requirements.

[0004] Despite advancements in hardware integration and algorithm application, existing technologies still face core challenges. Firstly, many current methods rely on fixed empirical roughness coefficients or simple partitioned roughness coefficients, neglecting the dynamic impact of seasonal vegetation growth and water level fluctuations on riverbed resistance, leading to distorted model input parameters. Secondly, purely hydraulic models struggle to capture random fluctuations caused by environmental noise, while purely neural network black-box models lack physical interpretability and generalization ability. Furthermore, existing systems are mostly unidirectional output models, lacking dynamic feedback mechanisms for real-time alignment and residual analysis of calculated values ​​with central control data. This results in system errors accumulating over time and failing to self-heal, hindering the achievement of precise regulatory requirements. Summary of the Invention

[0005] To overcome the shortcomings of existing technologies, the purpose of this invention is to provide an ecological flow monitoring method and system based on the microwave Doppler effect and an optimized bidirectional long short-term memory network. This invention solves the problems in existing ecological flow monitoring technologies, such as the single roughness parameter characterization, the separation between physical models and data-driven approaches, and the lack of a closed-loop correction mechanism leading to low monitoring accuracy and the inability to adaptively calibrate.

[0006] To achieve the above objectives, the present invention provides the following solution: An ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network includes: The target river channel is discretized into a grid based on the preset monitoring range, a river channel partition topology is constructed, and multiple monitoring section units are determined based on the river channel partition topology. Multispectral remote sensing images and microwave Doppler echo data covering the monitoring range are collected. The microwave Doppler echo data is solved based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit. The cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing images are extracted. The initial roughness coefficient of each monitoring section unit is calculated based on the surface velocity vector, cross-sectional geometric features, and regional texture features. Based on the preset hydrodynamic sensitivity index, the initial roughness coefficient is spatiotemporally interpolated to obtain a dynamic roughness spatial distribution field covering the topology of the river channel partition. Based on the constraints of the dynamic roughness spatial distribution field, a Legendre orthogonal function model is constructed for each monitoring section unit, and the Legendre orthogonal function model is solved using a two-way long short-time memory network to obtain the set of hydrodynamic field fitting coefficients corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. Using the aforementioned river channel partition topology as the basis for the graph structure, all the aforementioned hydrodynamic field fitting coefficient sets are mapped to graph node attributes to construct a spatiotemporal relationship graph. Based on the sequence regression analysis algorithm, the global correction coefficient in the spatiotemporal relationship map is calculated. Based on the global correction coefficient, the real-time collected hydropower station flow data are processed sequentially to obtain the corrected flow sequence. The corrected flow sequence is then smoothed and filtered, and boundary constraint verification is performed to obtain a complete ecological flow monitoring map. Threshold determination detection is performed based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

[0007] An ecological flow monitoring system based on the microwave Doppler effect and an optimized bidirectional long short-time memory network includes: The monitoring section unit construction module is used to perform grid-based discretization of the target river channel based on a preset monitoring range, construct a river channel partition topology, and determine multiple monitoring section units based on the river channel partition topology. The velocity vector calculation module is used to acquire multispectral remote sensing images and microwave Doppler echo data covering the monitoring range, and to solve the microwave Doppler echo data based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit, and to extract the cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing image. The initial roughness coefficient calculation module is used to calculate the initial roughness coefficient of each monitoring section unit based on the surface velocity vector, cross-sectional geometric features, and regional texture features. The dynamic roughness spatial distribution field calculation module is used to perform spatiotemporal interpolation on the initial roughness coefficient according to the preset hydrodynamic sensitivity index to obtain a dynamic roughness spatial distribution field covering the river channel partition topology. The hydrodynamic field fitting model construction module is used to construct a Legendre orthogonal function model for each monitoring section unit based on the constraint of the dynamic roughness spatial distribution field, and solve the Legendre orthogonal function model using a two-way long short-time memory network to obtain the hydrodynamic field fitting coefficient set corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. The spatiotemporal relationship graph construction module is used to map all the hydrodynamic field fitting coefficient sets as graph node attributes based on the river channel partition topology to construct the spatiotemporal relationship graph. The global correction coefficient calculation module is used to calculate the global correction coefficient in the spatiotemporal relationship map based on the sequence regression analysis algorithm. The real-time flow data processing module is used to process the real-time collected hydropower station flow data sequentially according to the global correction coefficient to obtain the corrected flow sequence, and to perform smoothing filtering and boundary constraint verification on the corrected flow sequence to obtain a complete ecological flow monitoring map. The threshold determination and detection module is used to perform threshold determination and detection based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

[0008] The present invention discloses the following technical effects: This invention provides an ecological flow monitoring method and system based on the microwave Doppler effect and an optimized bidirectional long short-term memory network. By fusing multispectral image texture and hydrodynamic sensitivity indicators, this invention constructs a dynamic roughness spatial distribution field, overcoming the limitations of traditional static and singular empirical roughness parameters, and accurately characterizing the dynamic changes in riverbed resistance with seasons and water levels. It innovatively constructs a physics-data dual-drive architecture, using Legendre orthogonal functions to establish the hydrodynamic physical framework, and combining it with a bidirectional long short-term memory network to capture nonlinear residuals. This ensures the interpretability of the model and solves the problems of weak anti-interference ability of pure hydraulic models and poor generalization of pure AI models. Simultaneously, by introducing spatiotemporal relationship maps and global correction coefficients, a closed-loop feedback mechanism between monitored values ​​and the true values ​​of hydropower stations is established, effectively eliminating system cumulative errors and significantly improving the intelligence level and data reliability of ecological flow monitoring under complex hydrological conditions. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart of an ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network is provided for an embodiment of the present invention. Figure 2 This is a flowchart illustrating the construction process of a monitoring section unit provided in an embodiment of the present invention. Figure 3 A flowchart illustrating the construction process of the dynamic roughness spatial distribution field provided in this embodiment of the invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0013] like Figure 1 As shown, this invention provides an ecological flow monitoring method based on the microwave Doppler effect and an optimized bidirectional long short-term memory network, comprising: Step 100: Based on the preset monitoring range, the target river channel is discretized into a grid to construct a river channel partition topology, and multiple monitoring section units are determined based on the river channel partition topology. Step 200: Collect multispectral remote sensing images and microwave Doppler echo data covering the monitoring range, and calculate the microwave Doppler echo data based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit, and extract the cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing images. Step 300: Calculate the initial roughness coefficient of each monitoring section unit based on the surface velocity vector, cross-sectional geometric features, and regional texture features; Step 400: Perform spatiotemporal interpolation on the initial roughness coefficient according to the preset hydrodynamic sensitivity index to obtain a dynamic roughness spatial distribution field covering the topology of the river channel partition. Step 500: Based on the constraints of the dynamic roughness spatial distribution field, construct a Legendre orthogonal function model for each monitoring section unit, and solve the Legendre orthogonal function model using a two-way long short-time memory network to obtain the set of hydrodynamic field fitting coefficients corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. Step 600: Using the river channel partition topology as the basis of the graph structure, map all the hydrodynamic field fitting coefficient sets as graph node attributes to construct a spatiotemporal relationship graph; Step 700: Based on the sequence regression analysis algorithm, calculate the global correction coefficient in the spatiotemporal relationship map; Step 800: Process the real-time collected hydropower station flow data sequentially according to the global correction coefficient to obtain the corrected flow sequence, and perform smoothing filtering and boundary constraint verification on the corrected flow sequence to obtain a complete ecological flow monitoring map; Step 900: Perform threshold determination detection based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

[0014] Furthermore, such as Figure 2As shown, this embodiment first acquires digital elevation model data within the monitoring area using high-precision surveying technology. Based on this topographical foundation, a hydrological analysis algorithm is used to accurately extract the main channel centerline of the target river, thereby identifying the core dynamic axis of the river flow. Subsequently, based on the geometric orientation of the main channel centerline, a spatial partitioning algorithm is used to generate an orthogonal grid set covering the entire region. By traversing the orthogonal grid set, the adjacency relationships between grids and the direction of water flow connectivity are analyzed, thus constructing an adjacency matrix reflecting the hydraulic correlation across the entire region. Furthermore, this embodiment materializes this adjacency matrix into a river channel partition topology structure, using this as the logical framework. Finally, within each independent spatial computational domain, a section perpendicular to the main channel centerline is extracted, and this section is designated as the monitoring section unit, thus completing the accurate discretization modeling from macroscopic topography to microscopic computational units, laying the data foundation for subsequent hydrodynamic parameter inversion.

[0015] The specific implementation process of step 100 is as follows: This embodiment first initiates a spatial discretization process for the target river channel to construct a high-precision computational base. This embodiment acquires digital elevation model (DEM) data within the monitoring range using airborne lidar or UAV oblique photogrammetry technology, ensuring sufficient elevation accuracy in the terrain data to reflect riverbed undulations. Based on this DEM data, this embodiment utilizes depression filling and a single-flow-direction algorithm to calculate flow direction and flow accumulation grids, thereby automatically extracting the continuous main channel centerline of the target river channel and establishing the dominant dynamic axis of the water flow. Furthermore, based on the curve of the main channel centerline, this embodiment employs adaptive grid generation technology to generate an orthogonal grid set covering the entire monitoring range at a preset spatial resolution. In this process, this embodiment explicitly defines each grid in the orthogonal grid set as an independent spatial computational domain, making it the smallest physical control unit for subsequent hydrodynamic parameter inversion and flow calculation, thus effectively solving the problem of discrete and insufficiently representative monitoring points in traditional monitoring methods.

[0016] Subsequently, this embodiment performs a global traversal of the generated orthogonal mesh set to establish the data interaction logic between meshes. This embodiment deeply analyzes the geometric adjacency state between adjacent spatial computational domains and, combined with the upstream and downstream flow directions determined by the elevation gradient, constructs an adjacency matrix reflecting the global connectivity. In this adjacency matrix, non-zero elements represent two spatial computational domains that are physically adjacent and hydraulically connected. Furthermore, this embodiment materializes this adjacency matrix into a river channel partition topology, where each node uniquely represents a specific spatial computational domain, and the directed edges between nodes precisely represent the flow path and momentum transfer direction of the water flow. This method of transforming the physical river channel into a graph topology not only preserves the spatial distribution characteristics of the river channel but also lays a solid data structure foundation for subsequent processing of non-Euclidean hydrodynamic data using graph neural network models.

[0017] Finally, based on the constructed spatial computational domain and the main channel centerline, this embodiment completes the final establishment of the physical monitoring object. Within each independent spatial computational domain, the geometric centroid is calculated, and a cross-section perpendicular to the tangent direction of the main channel centerline is drawn through this centroid. This cross-section is defined as the monitoring section unit. This monitoring section unit, as the core computational carrier of the system, directly carries key geometric attributes such as the cross-sectional area, wetted perimeter, and hydraulic radius within the grid area. It also serves as the spatial reference for subsequent mapping of microwave Doppler velocity vectors and inversion of multispectral image texture features. Through the above steps, this embodiment successfully discretizes continuous and irregular natural channels into a set of monitoring section units with clear geometric definitions and rigorous topological relationships. This achieves precise dimensionality reduction from macroscopic channel morphology to microscopic computational sections, providing standardized boundary conditions for subsequent high-precision step-by-step calculation of ecological flow.

[0018] Furthermore, the specific implementation process of step 200 is as follows: This embodiment first initiates a multi-source sensing data acquisition program, utilizing a microwave Doppler radar array and multispectral imaging equipment mounted at the monitoring station to simultaneously acquire multispectral remote sensing images and microwave Doppler echo data covering the entire monitoring range. For the acquired microwave Doppler echo data, this embodiment employs digital signal processing technology to perform a Fast Fourier Transform, converting the time-domain signal into a frequency-domain spectrum. A peak search algorithm is then used to accurately identify the center frequency corresponding to the spectral energy peak to extract the Doppler frequency shift value. Subsequently, based on the Doppler effect principle, this embodiment constructs a conversion formula combining the speed of light constant and the radar transmission center frequency to calculate the Doppler frequency shift value as the radial flow velocity along the radar beam's line-of-sight direction. To obtain flow velocity data with practical hydraulic physical significance, this embodiment further retrieves the spatial attitude parameters of the monitoring section unit within the river channel topology, obtaining the spatial normal vector of the unit. A vector projection decomposition algorithm is then used to map the one-dimensional radial flow velocity onto a two-dimensional tangent plane on the river surface, thereby obtaining a surface flow velocity vector containing information on both the magnitude and direction of the flow.

[0019] After acquiring the dynamic flow field data, this embodiment simultaneously conducts a refined analysis of the static boundary features of the river channel. Based on the river channel zoning topology constructed in the previous steps, this embodiment retrieves and analyzes the boundary coordinate point set of each monitoring section unit in three-dimensional geographic space. These coordinate point sets accurately depict the undulation morphology and boundary range of the riverbed cross-section. Based on the three-dimensional spatial boundary coordinates and the current water level elevation, this embodiment uses a differential integral algorithm to calculate the effective cross-sectional area of ​​the monitoring section unit and performs line integration along the wetted boundary of the riverbed to calculate its wetted perimeter. Furthermore, this embodiment determines the ratio of the cross-sectional area to the wetted perimeter as the hydraulic radius according to the hydraulic definition. This parameter is a key indicator characterizing the river channel's water conveyance efficiency and resistance characteristics. Finally, this embodiment structurally encapsulates the calculated cross-sectional area, wetted perimeter, and hydraulic radius to construct a cross-sectional geometric feature that comprehensively reflects the river channel's morphological attributes, providing geometric constraints for the subsequent calculation of the roughness coefficient.

[0020] To capture the microscopic morphological features of the land surface that influence water flow resistance, this embodiment further delves into the texture information in multispectral remote sensing images. First, this embodiment establishes a rigorous affine transformation mapping relationship between the pixel coordinate system of the multispectral remote sensing image and the geographic coordinate system of the river channel topology, ensuring precise spatial alignment between the image data and the physical cross-sections. Based on this mapping relationship, this embodiment uses a spatial indexing algorithm to locate and crop regions of interest (ROIs) within the wide-swath remote sensing image, precisely matching the spatial location of each monitoring cross-section unit. Subsequently, this embodiment performs grayscale quantization on the ROI image, calculating the grayscale co-occurrence matrix under different directions and step sizes to statistically analyze the second-order spatial correlation of pixel grayscale. Based on this grayscale co-occurrence matrix, this embodiment quantitatively extracts energy indicators reflecting texture uniformity, entropy indicators reflecting texture complexity, contrast indicators reflecting texture clarity, and correlation statistical indicators reflecting texture directionality. These indicators are then normalized using range standardization to ultimately generate regional texture features characterizing the roughness of the riverbed and bank slopes.

[0021] The expression for the Doppler effect formula is: ; The radial flow velocity obtained from the solution; The speed at which microwave signals propagate in the air; The Doppler frequency shift value is extracted by Fast Fourier Transform; The transmission center frequency of the microwave Doppler sensor is denoted as .

[0022] Specifically, this embodiment establishes a precise inversion logic from frequency domain characteristics to velocity vectors based on the Doppler effect principle. In this calculation process, the system executes specific Doppler effect operation rules, specifically by multiplying the propagation speed of the microwave signal in air by the Doppler frequency shift value extracted via Fast Fourier Transform in the previous step, and then dividing the resulting product by twice the transmission center frequency of the microwave Doppler sensor, thereby calculating the radial flow velocity of each monitoring section unit.

[0023] In this operational logic, the so-called radial velocity is a user-defined vector component concept, specifically referring to the relative motion velocity of surface particles along the line of sight of the microwave Doppler radar beam. It is the fundamental component for the subsequent synthesis of the complete velocity vector. The propagation speed of microwave signals in air is used as a physical constant in the calculation. Its function is to provide a spatiotemporal reference for electromagnetic wave transmission. In this embodiment, its value is usually set to the speed of light constant. The Doppler frequency shift value is a dynamic input variable that reflects the state of water flow and comes from the frequency difference between the echo signal and the transmitted signal. The transmission center frequency of the microwave Doppler sensor is an inherent attribute parameter of the radar hardware, which determines the sensor's detection sensitivity and penetration capability. As a specific application demonstration in this embodiment, this parameter can be set to a value in the industrial common frequency band such as 24GHz or 77GHz, depending on the selected radar model.

[0024] Furthermore, the specific implementation process of step 300 is as follows: This embodiment first quantifies the inherent resistance characteristics of the monitoring section unit based on the static properties of the physical environment. It utilizes regional texture features extracted from multispectral remote sensing images in previous steps, focusing on analyzing energy and contrast indices. These two indices directly reflect the roughness of the riverbed sediment and the vegetation cover density of the bank slopes. This embodiment pre-constructs a vegetation texture mapping table containing various land cover types, establishing an empirical correspondence between texture statistical feature values ​​and the Manning roughness coefficient. By retrieving this mapping table, this embodiment maps the current energy and contrast values ​​of the monitoring section unit to specific physical parameters, thereby determining the static surface roughness baseline value of the monitoring section unit. This baseline value is only related to the physical boundary conditions of the river channel, representing the theoretical structural roughness under conditions without hydrodynamic disturbance, providing a stable physical benchmark for subsequent comprehensive roughness calculations.

[0025] While obtaining the static baseline, this embodiment, based on hydrodynamic principles, inversely calculates the dynamic resistance encountered by the water flow during actual flow. This embodiment obtains the surface velocity vector and cross-sectional geometric characteristics of each monitoring section unit. Considering the premise that the Manning formula is applicable to one-dimensional steady flow calculations, this embodiment first uses a vector projection algorithm to project the three-dimensional surface velocity vector onto the main axis of the river channel, extracting the effective water conveyance velocity component. Subsequently, based on the inverse operation logic of the Manning formula, combined with the hydraulic radius, hydraulic gradient, and projected effective water conveyance velocity of the monitoring section unit, this embodiment inversely solves for the hydraulic friction coefficient at that moment. This coefficient dynamically reflects the interaction intensity between the water flow and the riverbed boundary under specific flow velocity conditions, capturing the instantaneous impact of water level fluctuations on flow resistance, and compensating for the inadequacy of static roughness in reflecting changes in hydrodynamic processes.

[0026] Finally, this embodiment achieves the organic integration of static physical properties and dynamic hydraulic characteristics by constructing an adaptive weighting mechanism. This embodiment performs statistical analysis on the surface velocity vector sequence within a time window, calculating the temporal variance of its modulus to quantify the turbulence intensity and flow stability. Based on this temporal variance, this embodiment constructs a flow stability weighting function. When the velocity fluctuation is small, i.e., the flow is stable, a higher weight is assigned to the dynamic hydraulic friction coefficient; conversely, when the velocity fluctuation is severe, it tends to rely on a stable static surface roughness baseline value to reduce measurement noise interference. This embodiment uses this flow stability weighting to perform a weighted summation calculation on the static surface roughness baseline value and the hydraulic friction coefficient, ultimately obtaining the initial roughness coefficient of the monitoring section unit that balances physical structural stability and hydrodynamic time-varying characteristics, providing more realistic parameter input for subsequent construction of a high-precision hydrodynamic model.

[0027] Specifically, the expression for the Manning formula is: ; Let be the hydraulic friction coefficient of the monitoring section element to be solved; The magnitude of the surface velocity vector after projection onto the main axis of the water flow; To monitor the hydraulic radius of the cross-sectional unit; To monitor the hydraulic gradient of the cross-sectional unit.

[0028] Specifically, this embodiment employs the inverse operation logic of the Manning formula to accurately quantify the dynamic resistance characteristics of the riverbed. In this calculation process, the system executes specific hydraulic inversion rules. Specifically, the calculation involves multiplying the 2 / 3 power of the hydraulic radius of the monitoring section unit by the 1 / 2 power of its hydraulic gradient, and then dividing the product by the magnitude of the surface velocity vector projected onto the main flow axis. This yields the hydraulic friction coefficient of the monitoring section unit. In this operational logic, the hydraulic friction coefficient is a user-defined physical state parameter, representing a comprehensive coefficient that reflects the roughness of the riverbed wall and its resistance to water flow. Its function is to quantify the degree to which the riverbed sediment and vegetation consume water flow. The surface velocity vector magnitude projected onto the main axis of the water flow is derived from the vector decomposition and projection processing of the microwave Doppler data in the previous steps. It represents the effective water transport rate along the main direction of the river channel. This value may be 2.5 m / s during flood season. The hydraulic radius of the monitoring section unit is derived from the ratio of the cross-sectional area to the wetted perimeter calculated in the previous step. Its function is to serve as a geometric measure of the cross-sectional water conveyance efficiency. The hydraulic gradient of the monitoring section unit is derived from the ratio of the water level difference between two adjacent monitoring section units to the channel distance. Its function is to characterize the potential energy gradient driving the water flow. As a typical application example, the value of this parameter in a plain river channel may be approximately 0.0005.

[0029] Furthermore, such as Figure 3 As shown, this embodiment first constructs a topological distance matrix representing the physical connectivity and spatial span between monitoring section units by analyzing the edge connectivity relationships in the river channel partition topology. Based on this, a preset hydrodynamic sensitivity index is defined as a physical attenuation factor, and combined with the aforementioned topological distance matrix, the roughness influence weight of each monitoring section unit on its adjacent nodes in resistance transmission is accurately and quantitatively calculated. At the time dimension, this embodiment extracts the long-period historical time series of the initial roughness coefficients and uses a moving average algorithm to filter out random fluctuation noise, thereby extracting a roughness evolution trend term reflecting the riverbed evolution law. Finally, this embodiment uses the spatial dimension influence weight and the time dimension evolution trend term as co-constraint variables to perform spatiotemporal kriging interpolation on the initial roughness coefficients, ultimately outputting a dynamic roughness spatial distribution field covering the entire region and possessing spatiotemporal continuity, achieving a refined dynamic characterization of river channel resistance features.

[0030] The specific implementation process of step 400 is as follows: This embodiment first constructs the spatial interaction relationships between monitoring section units from the perspective of spatial topological association. This embodiment deeply analyzes the river channel partition topology structure constructed in the previous steps, traversing and reading the stored edge connection relationship data, which accurately describes the hydraulic connectivity paths between upstream and downstream grids. Based on this connectivity, this embodiment uses the shortest path algorithm from graph theory to calculate the path length between any two monitoring section units on this topology structure, thereby constructing a topological distance matrix characterizing the degree of spatial adjacency across the entire domain. Furthermore, this embodiment introduces a preset hydrodynamic sensitivity index, which quantifies the transmission effect of upstream hydrodynamic disturbances on downstream roughness changes; this embodiment defines it as a physical attenuation factor. Combining the topological distance matrix and this attenuation factor, this embodiment calculates the roughness influence weight of each monitoring section unit on adjacent nodes. This weight objectively reflects the spatial continuity and gradual change patterns of the river channel micro-geomorphic features, providing a quantitative basis for spatial interpolation.

[0031] While establishing spatial correlation, this embodiment extracts the evolution characteristics of roughness parameters in the time dimension. This embodiment retrieves initial roughness coefficient records for each monitoring section unit from the historical database over consecutive time periods to construct a high-resolution historical time series. Considering that roughness exhibits slow time-varying characteristics due to factors such as seasonal vegetation growth, riverbed sediment deposition, and water level fluctuations, simple instantaneous values ​​are insufficient to reflect its long-term patterns. Therefore, this embodiment uses a moving average algorithm to smooth the historical time series, setting specific time windows to filter out high-frequency noise caused by instantaneous measurement errors, thereby separating and extracting the roughness evolution trend term that reflects the long-term changes in the river channel's physical properties, establishing an evolution benchmark in the time dimension.

[0032] Finally, this embodiment employs spatiotemporal co-processing technology to generate a continuous physical field covering the entire region. This embodiment uses the calculated spatial dimension roughness influence weight and the temporal dimension roughness evolution trend term as dual constraints, inputting them into a spatiotemporal Kriging interpolation model. This model uses the measured initial roughness of the monitoring cross-section units as sample points, and utilizes a semi-variogram function to fit the spatiotemporal correlation, performing unbiased optimal estimation for the blank areas where no sensors are deployed. Through this process, this embodiment expands the discrete cross-section data into a dynamic roughness spatial distribution field covering the entire river channel topology. In this distribution field, each grid pixel is assigned a spatiotemporally corrected roughness value, thereby achieving a refined and dynamic characterization of the river channel's overall resistance characteristics, providing a seamless parameter field input for subsequent hydrodynamic model calculations.

[0033] Furthermore, the specific implementation process of step 500 is as follows: This embodiment first constructs a mathematical representation model of the hydrodynamic process based on physical mechanisms, aiming to approximate complex nonlinear flow laws using the completeness of orthogonal functions. This embodiment selects water level, flow velocity, and flow rate as three core independent variables describing the hydrodynamic state of the monitoring section unit, and normalizes these three variables to eliminate dimensional differences. Based on this, this embodiment constructs Legendre orthogonal polynomial basis functions of corresponding orders, forming a high-dimensional function space describing the hydrodynamic characteristics. To ensure that the mathematical model conforms to the real river physical environment, this embodiment calls the dynamic roughness space distribution field generated in the previous steps, extracting the roughness value corresponding to the monitoring section unit as a key physical constraint parameter. This embodiment incorporates this roughness value into the boundary condition setting of the basis functions and performs a linear weighted combination of Legendre orthogonal polynomial basis functions of each order, thereby constructing a Legendre orthogonal function model containing several undetermined weight coefficients. This model is essentially a physical analytical equation to be solved; its mathematical structure is determined by Legendre polynomials, while its physical characteristics are controlled by the undetermined weight coefficients.

[0034] After establishing the physical model framework, this embodiment constructs a data-driven intelligent solver to obtain optimal model parameters. This embodiment designs and builds a bidirectional long short-term memory (LSTM) network architecture comprising an input layer, a bidirectional recurrent layer, and a fully connected output layer. Given the significant time lag and sequence correlation inherent in water flow, this embodiment selects historical water level and velocity sequences within a continuous time window for the monitoring section unit as input features. These time-series data enter the bidirectional recurrent layer via the input layer. This layer simultaneously scans the input sequence using two independent forward and reverse time recursive units, capturing not only the impact of past hydrological conditions on the present but also the feedback of future trends, thereby extracting the deep hydrodynamic evolution characteristics implicit in both long and short time series. This embodiment transmits the output feature vector of the bidirectional recurrent layer to the fully connected output layer, mapping its dimensions to an output vector with the same number of undetermined coefficients as the aforementioned Legendre orthogonal function model, achieving a direct mapping from time-series data to physical model parameters.

[0035] Finally, this embodiment performs a joint optimization training process based on physical residual constraints to establish the final hydrodynamic field fitting coefficients. Instead of directly using the neural network output as the prediction result, this embodiment assigns the vector output of the bidirectional long short-term memory network to the undetermined coefficients of the Legendre orthogonal function model, and then uses this function model to calculate the reconstructed hydrodynamic physical quantity values. This embodiment constructs a loss function to calculate the fitting residual between the reconstructed value and the actual monitored measured data, and sets minimizing this residual as the global optimization objective of the network. Through backpropagation and gradient descent strategies, this embodiment continuously adjusts the weights and biases within the neural network until the fitting residual converges to a preset threshold range. At this point, the network is in its optimal state, and this embodiment determines the parameters generated by its output layer as the set of hydrodynamic field fitting coefficients for the monitoring section unit, thus completing the deep integration of physical mechanisms and data-driven approaches, and achieving a precise parameterized expression of complex hydrodynamic fields.

[0036] Specifically, the expression for the Legendre orthogonal function model is: ; in, The output value of the constructed objective function expression; For the highest order of the Legendre orthogonal polynomial; For the first Weight parameters of the basis functions; For the first Legendre orthogonal polynomial basis functions; These are the normalized independent variables, including water level, flow velocity, and flow rate.

[0037] Specifically, this embodiment employs a linear weighted superposition algorithm based on orthogonal bases to mathematically approximate complex nonlinear hydrodynamic processes. During this calculation process, the system executes specific function combination operation rules; The specific steps are as follows: For each specific order, the weight parameters of the k-th order basis function are multiplied by the k-th order Legendre orthogonal polynomial basis function values ​​calculated based on the normalized independent variables. The products of all orders from order 0 up to the maximum order of the Legendre orthogonal polynomial are then summed to finally calculate the output value of the constructed objective function expression. In this operational logic, the output value of the constructed objective function expression is a custom mathematical mapping result, whose function is to quantitatively characterize the hydrodynamic process state of the monitoring section unit at a specific time, such as the fitted value of the velocity distribution. The highest order of the Legendre orthogonal polynomial is a hyperparameter that controls the model fitting accuracy and computational complexity. Its function is to determine the dimensional capacity of the basis function space. As a specific application demonstration in this embodiment, in order to achieve a balance between computational efficiency and fitting accuracy, this parameter is usually set to 6 or 8. The weight parameters of the k-th order basis function are derived from the output layer mapping of the bidirectional long short-term memory network. Their function is to quantify the contribution weight of the corresponding orthogonal basis function in describing the current hydrodynamic field. The k-th order Legendre orthogonal polynomial basis functions are mathematically predefined sets of standard orthogonal functions that provide a mathematical "skeleton" to describe the spatial distribution of physical quantities. The normalized independent variables refer to the water level, velocity, and flow rate data after range standardization. Their function is to eliminate the order-of-magnitude differences between different physical dimensions to accelerate model convergence. As a specific numerical example, when the measured water level is half of the historical highest water level, the value of this variable is 0.5.

[0038] Furthermore, the specific implementation process of step 600 is as follows: This embodiment first initiates the graph structure skeleton construction program, aiming to transform the connectivity of the river channel in physical space into a set of edges in mathematical graph theory. This embodiment deeply analyzes the river channel partition topology constructed in the previous steps, reading the stored node connection relationship data. This data intuitively reflects the upstream and downstream adjacency status of different monitoring section units in the direction of water flow. Based on this node connection relationship, this embodiment constructs a high-dimensional spatial adjacency matrix, where the rows and columns correspond to all monitoring section units within the entire domain. If two monitoring section units have direct hydraulic connectivity in physical topology, a non-zero weight representing the connectivity strength is assigned to the corresponding position in the matrix; otherwise, a value of zero is assigned. Through this process, this embodiment abstracts the complex river network physical structure into a sparse matrix form that can be processed by a computer, establishing the spatial correlation foundation of the spatiotemporal relationship graph.

[0039] While establishing the graph skeleton, this embodiment focuses on constructing graph node attributes that carry rich physical information. This embodiment traverses all monitoring section units across the entire area, extracting the hydrodynamic field fitting coefficient set corresponding to each unit. This coefficient set is obtained by solving a bidirectional long short-term memory network and consists of core parameters that accurately describe the local hydrodynamic characteristics of the unit. This embodiment standardizes and vectorizes this coefficient set, recombining it into a one-dimensional node feature vector. Considering that simple physical parameters lack explicit expression of the temporal evolution dimension, this embodiment introduces a preset time position encoding technique to generate a time-series encoding vector corresponding to the current monitoring time. Subsequently, this embodiment uses a feature concatenation algorithm to fused the time position encoding vector with the aforementioned node feature vector along the feature channel dimension, generating a spatiotemporal node attribute matrix that contains both local hydrodynamic physical characteristics and absolute time position information, thereby endowing static nodes with dynamic temporal perception capabilities.

[0040] Finally, this embodiment performs the physical assembly and encapsulation process of the graph. This embodiment maps the constructed spatial adjacency matrix to an edge set in the graph structure, defining the interaction paths between nodes in the graph; simultaneously, it maps the spatiotemporal node attribute matrix to a node set in the graph structure, defining the internal state of each node at a specific spatiotemporal point. This embodiment organically combines these two through a graph data structure, ultimately constructing a spatiotemporal relationship graph containing complete spatiotemporal evolution characteristics. In this graph, each node is not only a data carrier storing water level, flow velocity, and roughness fitting parameters, but also an intelligent sensing unit in the global hydrodynamic network. They achieve mutual state transmission and coupling through edge sets. This spatiotemporal relationship graph, through this structured approach, fully characterizes the topological correlation of the hydrodynamic parameters of the entire river channel in the spatial dimension and the evolution law in the temporal dimension, providing standardized graph structure input data for subsequent global system error analysis and correction.

[0041] Furthermore, the specific implementation process of step 700 is as follows: This embodiment first initiates a deep aggregation program of graph features, aiming to uncover implicit correlation information of monitoring section units across the entire domain. This embodiment calls a pre-defined graph convolutional neural network operator to perform graph convolution operations on the spatiotemporal relationship graph generated in the previous steps. During this operation, this embodiment utilizes the spatial adjacency matrix in the graph as a message passing channel to control the diffusion and aggregation of spatiotemporal node attributes in the topological structure. Through multi-layer graph convolution operations, each node not only retains its own local hydrodynamic features but also weightedly fuses the attribute information of its upstream and downstream neighboring nodes, thereby mapping the originally discrete node features into a continuous and dense feature space. After this processing, this embodiment generates a high-dimensional spatiotemporal feature matrix containing implicit global correlation information. This matrix mathematically represents the hydrodynamic coupling state of the entire basin at the current moment, providing high-order feature inputs containing spatial context for subsequent global error analysis.

[0042] After extracting the graph features, this embodiment performs a physical-logical transformation from graph-structured data to sequence-structured data. This embodiment re-analyzes the river channel topology and, based on the connectivity between nodes and elevation gradients, accurately identifies the flow sequence from the upstream source to the downstream outlet. Based on this physical flow sequence, this embodiment rearranges and reassembles the row vectors in the high-dimensional spatiotemporal feature matrix. Specifically, this embodiment places the feature vectors corresponding to the upstream monitoring section units at the beginning of the sequence and the feature vectors corresponding to the downstream units at the end of the sequence, thereby constructing a global river state sequence that strictly follows hydraulic causal logic. This sequence transforms the complex spatial graph structure into a time / space sequence with a clear causal orientation, enabling subsequent models to capture the accumulation and propagation patterns of errors along the direction of water flow propagation.

[0043] Finally, this embodiment employs sequential regression analysis to systematically calculate the overall state deviation. A nonlinear sequential regression model is constructed, and the generated overall river state sequence is imported as input. This model utilizes its internal regression algorithm to analyze the implicit nonlinear fluctuation patterns in the state sequence and compares them with the ideal hydrodynamic equilibrium state to identify systematic deviation trends caused by sensor system drift, environmental background noise, or model simplification. This embodiment interprets the model's output as a numerical vector with a specific dimension, where each element quantitatively describes the overall correction magnitude for the corresponding monitoring area. This numerical vector is ultimately determined as the global correction coefficient, used for global dynamic compensation of the measured data in subsequent steps, thereby achieving a closed-loop improvement in the accuracy of ecological flow monitoring.

[0044] Furthermore, the specific implementation process of step 800 is as follows: This embodiment enters the final data correction and compensation stage, aiming to apply the calculated mathematical correction parameters to the physical observation data. First, this embodiment acquires instantaneous flow monitoring data uploaded by the hydropower station in real time through a data interface, and strictly reassembles it into a real-time flow data vector with dimensions completely consistent with the global correction coefficients, following the arrangement order of the global river state sequence determined in the previous steps. Based on this, this embodiment performs bitwise compensation operations, mathematically synthesizing each element in the real-time flow data vector with the corresponding element in the global correction coefficient vector. This operation effectively eliminates additive or multiplicative biases caused by sensor zero-point drift, environmental background noise, or equipment system errors by superimposing a systematic bias compensation amount on the original observation values, thereby generating a corrected flow sequence containing preliminary error compensation values, providing corrected basic data for subsequent processing.

[0045] To further improve the signal-to-noise ratio of the data and ensure its physical compliance, this embodiment performs refined signal processing and logical verification on the corrected flow sequence. First, a Gaussian weighted algorithm is introduced to construct a one-dimensional Gaussian convolution kernel with a preset time window length. This kernel is used to perform temporal convolution processing on the corrected flow sequence. By assigning higher weights to data at the center time and attenuating the weights to data at the edge time, high-frequency random noise caused by water turbulence or instantaneous equipment vibration is effectively smoothed out, resulting in a smoother flow sequence with a more stable trend. Subsequently, this embodiment calls upon the cross-sectional geometric features extracted in the previous steps and, based on the current measured water level and Manning's formula, calculates the physically permissible maximum flow capacity and minimum ecological baseflow for each monitoring cross-sectional unit, establishing a hydraulic limit threshold. This embodiment uses this threshold to perform numerical truncation and legality verification on the smoothed flow sequence, forcibly correcting abnormal abrupt changes exceeding physical limits to a reasonable range, thereby generating an effective flow sequence that conforms to the objective laws of fluid mechanics.

[0046] Finally, this embodiment completes the reverse mapping of data from numerical sequence space to geophysical space, generating visualized monitoring results. Based on the river channel zoning topology constructed in step one, this embodiment retrieves the stored node index information, which establishes a unique correspondence between data sequence locations and physical monitoring section units. Based on this node index, this embodiment accurately projects each flow value in the effective flow sequence back to its corresponding geospatial coordinate point, achieving a re-linking of data and space. Through this process, this embodiment dynamically generates a complete ecological flow monitoring map containing spatiotemporal flow distribution information across the entire region. This map not only intuitively displays the real-time flow status of each river section but also completely preserves the evolution details of hydrodynamic parameters in the spatiotemporal dimension, providing water conservancy regulatory departments with high-precision, comprehensive, and physically interpretable final ecological flow monitoring results.

[0047] Furthermore, the specific implementation process of step 900 is as follows: This embodiment first establishes an ecological flow compliance evaluation benchmark, aiming to transform numerical monitoring maps into a basis for management decision-making. This embodiment calls upon a pre-built ecological flow regulatory database, which stores the statutory ecological base flow thresholds for each monitoring section unit in different seasons and hydrological periods. This embodiment iterates through the complete ecological flow monitoring map generated in the preceding steps, extracts the real-time flow values ​​for each monitoring section unit, and aligns and matches them with the corresponding ecological base flow thresholds in the database. Through this spatiotemporal matching mechanism, this embodiment ensures that each judgment is based on specific regulatory standards at the current time and geographical location, thereby avoiding the risk of misjudgment caused by a "one-size-fits-all" evaluation and establishing an accurate comparison benchmark.

[0048] After establishing the evaluation benchmark, this embodiment executes a multi-level threshold determination and status classification procedure. This embodiment calculates the difference and ratio between the real-time flow value and the ecological base flow threshold, and classifies the monitoring results into different control states according to preset early warning grading rules. Specifically, when the real-time flow is significantly higher than the ecological base flow threshold, this embodiment classifies the monitoring section unit as "compliant status"; when the real-time flow approaches but has not yet fallen below the ecological base flow threshold and is in the buffer zone, it is classified as "early warning status"; when the real-time flow is lower than the ecological base flow threshold, it is directly classified as "violation alarm status". This embodiment encodes these status identifiers into the attribute fields of the map, generating an ecological health status distribution matrix covering the entire region, realizing the quantitative grading of the river's ecological surplus and deficit status.

[0049] Finally, this embodiment summarizes and generates the final regulatory output to trigger the corresponding control response. Based on the aforementioned state distribution matrix, this embodiment comprehensively generates the final ecological flow monitoring result, which includes flow rate curves, state color indicators, and statistics on violation periods. For monitoring section units determined to be in "violation alarm state," this embodiment automatically triggers an abnormal event recording program, generating a violation log containing time, location, water shortage amount, and duration, and sends an immediate alarm signal to the regulatory terminal through a data interface. Simultaneously, this embodiment stores the final ecological flow monitoring result in a historical archive, serving as a legal basis for subsequent water resource scheduling optimization and hydropower station release assessment, thus completing a full technical closed loop from data perception to intelligent supervision.

[0050] This embodiment also provides an ecological flow monitoring system based on the microwave Doppler effect and an optimized bidirectional long short-time memory network, including: The monitoring section unit construction module is used to perform grid-based discretization of the target river channel based on a preset monitoring range, construct a river channel partition topology, and determine multiple monitoring section units based on the river channel partition topology. The velocity vector calculation module is used to acquire multispectral remote sensing images and microwave Doppler echo data covering the monitoring range, and to solve the microwave Doppler echo data based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit, and to extract the cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing image. The initial roughness coefficient calculation module is used to calculate the initial roughness coefficient of each monitoring section unit based on the surface velocity vector, cross-sectional geometric features, and regional texture features. The dynamic roughness spatial distribution field calculation module is used to perform spatiotemporal interpolation on the initial roughness coefficient according to the preset hydrodynamic sensitivity index to obtain a dynamic roughness spatial distribution field covering the river channel partition topology. The hydrodynamic field fitting model construction module is used to construct a Legendre orthogonal function model for each monitoring section unit based on the constraint of the dynamic roughness spatial distribution field, and solve the Legendre orthogonal function model using a two-way long short-time memory network to obtain the hydrodynamic field fitting coefficient set corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. The spatiotemporal relationship graph construction module is used to map all the hydrodynamic field fitting coefficient sets as graph node attributes based on the river channel partition topology to construct the spatiotemporal relationship graph. The global correction coefficient calculation module is used to calculate the global correction coefficient in the spatiotemporal relationship map based on the sequence regression analysis algorithm. The real-time flow data processing module is used to process the real-time collected hydropower station flow data sequentially according to the global correction coefficient to obtain the corrected flow sequence, and to perform smoothing filtering and boundary constraint verification on the corrected flow sequence to obtain a complete ecological flow monitoring map. The threshold determination and detection module is used to perform threshold determination and detection based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0052] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. An ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network, characterized in that, include: The target river channel is discretized into a grid based on the preset monitoring range, a river channel partition topology is constructed, and multiple monitoring section units are determined based on the river channel partition topology. Multispectral remote sensing images and microwave Doppler echo data covering the monitoring range are collected. The microwave Doppler echo data is solved based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit. The cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing images are extracted. The initial roughness coefficient of each monitoring section unit is calculated based on the surface velocity vector, cross-sectional geometric features, and regional texture features. Based on the preset hydrodynamic sensitivity index, the initial roughness coefficient is spatiotemporally interpolated to obtain a dynamic roughness spatial distribution field covering the topology of the river channel partition. Based on the constraints of the dynamic roughness spatial distribution field, a Legendre orthogonal function model is constructed for each monitoring section unit, and the Legendre orthogonal function model is solved using a two-way long short-time memory network to obtain the set of hydrodynamic field fitting coefficients corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. Using the aforementioned river channel partition topology as the basis for the graph structure, all the aforementioned hydrodynamic field fitting coefficient sets are mapped to graph node attributes to construct a spatiotemporal relationship graph. Based on the sequence regression analysis algorithm, the global correction coefficient in the spatiotemporal relationship map is calculated. Based on the global correction coefficient, the real-time collected hydropower station flow data are processed sequentially to obtain the corrected flow sequence. The corrected flow sequence is then smoothed and filtered, and boundary constraint verification is performed to obtain a complete ecological flow monitoring map. Threshold determination detection is performed based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

2. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The target river channel is discretized into a grid based on a preset monitoring range to construct a river channel partition topology, and multiple monitoring section units are determined based on the river channel partition topology, including: Obtain digital elevation model data within the monitoring range; Based on the digital elevation model data, the main channel centerline of the target river is extracted using a hydrological analysis algorithm; Based on the direction of the main centerline, an orthogonal grid set covering the monitoring range is generated according to a preset spatial resolution, and each grid in the orthogonal grid set is defined as an independent spatial computing domain; Traverse the orthogonal grid set, construct an adjacency matrix based on the spatial adjacency relationships between the spatial computing domains and the upstream and downstream water flow directions, and transform the adjacency matrix into the river channel partition topology; wherein, the nodes in the river channel partition topology represent the spatial computing domains, and the edges represent the water flow connectivity; Extract the cross section perpendicular to the main centerline within each of the spatial computing domains, and determine the cross section as the monitoring section unit.

3. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The process involves acquiring multispectral remote sensing images and microwave Doppler echo data covering the monitoring range. Based on the Doppler frequency shift principle, the microwave Doppler echo data is processed to obtain the surface velocity vector for each monitoring section unit. Furthermore, the cross-sectional geometric features of each monitoring section unit and the corresponding regional texture features from the multispectral remote sensing images are extracted, including: Collect multispectral remote sensing images and microwave Doppler echo data covering the monitoring range; The microwave Doppler echo data is subjected to Fast Fourier Transform to extract the Doppler frequency shift value corresponding to the spectral peak, and the Doppler frequency shift value is converted into radial flow velocity using the Doppler effect formula; The radial velocity projection is decomposed based on the spatial normal vector of the monitoring section unit to obtain the surface velocity vector; Analyze the three-dimensional spatial boundary coordinates of the monitoring section unit in the river channel topology; The cross-sectional area and wetted perimeter of the monitoring section unit are calculated based on the three-dimensional spatial boundary coordinates, and the ratio of the cross-sectional area to the wetted perimeter is determined as the hydraulic radius. The cross-sectional area, wetted perimeter, and hydraulic radius are combined to form the cross-sectional geometric features. Establish the spatial coordinate mapping relationship between the multispectral remote sensing image and the river channel zoning topology; Based on the spatial coordinate mapping relationship, the region of interest image that matches the spatial position of each monitoring section unit is cropped from the multispectral remote sensing image; Calculate the gray-level co-occurrence matrix of the region of interest image, and extract energy, entropy, contrast and correlation statistical indicators based on the gray-level co-occurrence matrix. After normalizing the statistical indicators, the region texture features are formed. The expression for the Doppler effect formula is: ； The radial flow velocity obtained from the solution; The speed at which microwave signals propagate in the air; The Doppler frequency shift value is extracted by Fast Fourier Transform; The transmission center frequency of the microwave Doppler sensor is denoted as .

4. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The calculation of the initial roughness coefficient of each monitoring section unit based on the surface velocity vector, cross-sectional geometric features, and regional texture features includes: Based on the energy and contrast indices in the regional texture features, the static surface roughness base value of the monitoring section unit is determined using a preset vegetation texture mapping table. Based on the surface velocity vector and the cross-sectional geometric features, the hydraulic friction coefficient of the monitoring cross-section unit is solved by the inverse operation of the Manning formula, wherein the surface velocity vector is projected onto the direction of the main water flow axis before the calculation. Calculate the temporal variance of the surface velocity vector, and construct flow stability weights based on the temporal variance; The initial roughness coefficient of the monitoring section unit is obtained by weighting and fusing the static surface roughness base value and the hydraulic friction coefficient using the fluid stability weight. The expression for the Manning formula is: ； Let be the hydraulic friction coefficient of the monitoring section element to be solved; The magnitude of the surface velocity vector after projection onto the main axis of the water flow; To monitor the hydraulic radius of the cross-sectional unit; To monitor the hydraulic gradient of the cross-sectional unit.

5. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The step of performing spatiotemporal interpolation on the initial roughness coefficient based on a preset hydrodynamic sensitivity index to obtain a dynamic roughness spatial distribution field covering the river channel topology includes: The edge connection relationships in the river channel partition topology are analyzed to construct a topological distance matrix representing the spatial distance between each monitoring section unit; The preset hydrodynamic sensitivity index is defined as the attenuation factor, and the roughness influence weight of each monitoring section unit on the adjacent node is calculated in combination with the topological distance matrix. Extract the historical time series of the initial roughness coefficient; Based on the historical time series of the initial roughness coefficient, the moving average algorithm is used to extract the roughness evolution trend term. Based on the roughness influence weight and the roughness evolution trend term, spatiotemporal kriging interpolation is performed on the initial roughness coefficient to obtain a continuously distributed dynamic roughness spatial distribution field, wherein each pixel in the dynamic roughness spatial distribution field corresponds to a corrected roughness value.

6. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, Based on the constraints of the dynamic roughness spatial distribution field, a Legendre orthogonal function model is constructed for each monitoring section unit, and the Legendre orthogonal function model is solved using a bidirectional long short-time memory network to obtain the set of hydrodynamic field fitting coefficients corresponding to each monitoring section unit, including: Water level, flow velocity, and flow rate are selected as independent variables, and Legendre orthogonal polynomial basis functions of corresponding orders are constructed respectively. Using the roughness values ​​in the dynamic roughness space distribution field as constraint parameters, the Legendre orthogonal polynomial basis functions are linearly combined to construct a Legendre orthogonal function model containing undetermined coefficients; A bidirectional long short-term memory network consisting of an input layer, a bidirectional recurrent layer, and a fully connected output layer is constructed, and the historical water level sequence and flow velocity sequence of the monitoring section unit are input into the bidirectional long short-term memory network as input features. The output of the bidirectional long short-term memory network is mapped to the undetermined coefficients of the Legendre orthogonal function model. The optimization objective is to minimize the fitting residual between the reconstructed value calculated by the Legendre orthogonal function model and the measured data. The bidirectional long short-term memory network is then trained to obtain the set of fitting coefficients for the hydrodynamic field. The expression for the Legendre orthogonal function model is: ； in, The output value of the constructed objective function expression; For the highest order of the Legendre orthogonal polynomial; For the first Weight parameters of the basis functions; For the first Legendre orthogonal polynomial basis functions; These are the normalized independent variables, including water level, flow velocity, and flow rate.

7. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The method of using the river channel partition topology as the basis for a graph structure to map all the hydrodynamic field fitting coefficient sets as graph node attributes to construct a spatiotemporal relationship graph includes: The node connection relationships in the river channel partition topology are analyzed, and a spatial adjacency matrix representing the connectivity between the monitoring section units is constructed based on the node connection relationships. Traverse all the monitoring section units, extract the hydrodynamic field fitting coefficient set corresponding to each monitoring section unit, and vectorize the hydrodynamic field fitting coefficient set to generate node feature vectors. A preset time position code is introduced, and the time position code is concatenated with the node feature vector to generate a spatiotemporal node attribute matrix containing temporal information; The spatial adjacency matrix is ​​used as the edge set of the graph, and the spatiotemporal node attribute matrix is ​​used as the node set of the graph. By combining them, a spatiotemporal relationship graph containing spatiotemporal evolution characteristics is obtained; wherein, the spatiotemporal relationship graph is used to characterize the coupling relationship of hydrodynamic parameters of the entire river channel in the spatiotemporal dimension.

8. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The algorithm based on sequence regression analysis is used to calculate the global correction coefficient in the spatiotemporal relationship graph, including: The spatiotemporal relationship graph is subjected to graph convolution operation using graph convolution neural network operators to obtain a high-dimensional spatiotemporal feature matrix that contains implicit global correlation information. Based on the high-dimensional spatiotemporal feature matrix, the upstream and downstream flow order in the river channel partition topology is analyzed to obtain the whole-domain river state sequence. Based on the sequence regression analysis algorithm, the entire river channel state sequence is solved to obtain a numerical vector representing the trend of systematic error, and the numerical vector is determined as the global correction coefficient.

9. The ecological flow monitoring method based on microwave Doppler effect and optimized bidirectional long short-term memory network according to claim 1, characterized in that, The process involves sequentially processing the real-time collected hydropower station flow data according to the global correction coefficient to obtain a corrected flow sequence, and then performing smoothing filtering and boundary constraint verification on the corrected flow sequence to obtain a complete ecological flow monitoring map, including: Construct a real-time traffic data vector with the same dimension as the global correction coefficient, and perform bitwise compensation operation on the real-time traffic data vector and the global correction coefficient to obtain a corrected traffic sequence containing preliminary error compensation values; The corrected flow sequence is processed by temporal convolution using a Gaussian weighted algorithm to obtain a smooth flow sequence; Based on the cross-sectional geometric features of the monitoring section unit, the hydraulic limit threshold of each monitoring section unit at the current water level is calculated. The hydraulic limit threshold is then used to numerically truncate and validate the smoothed flow sequence to obtain an effective flow sequence. Based on the node index of the river channel partition topology, the flow value space in the effective flow sequence is mapped to the corresponding monitoring section unit to obtain a complete ecological flow monitoring map containing spatiotemporal flow distribution information of the entire region.

10. An ecological flow monitoring system based on microwave Doppler effect and optimized bidirectional long short-term memory network, characterized in that, include: The monitoring section unit construction module is used to perform grid-based discretization processing on the target river channel based on a preset monitoring range, construct a river channel partition topology, and determine multiple monitoring section units based on the river channel partition topology. The velocity vector calculation module is used to acquire multispectral remote sensing images and microwave Doppler echo data covering the monitoring range, and to solve the microwave Doppler echo data based on the Doppler frequency shift principle to obtain the surface velocity vector of each monitoring section unit, and to extract the cross-sectional geometric features of each monitoring section unit and the regional texture features corresponding to the multispectral remote sensing image. The initial roughness coefficient calculation module is used to calculate the initial roughness coefficient of each monitoring section unit based on the surface velocity vector, cross-sectional geometric features, and regional texture features. The dynamic roughness spatial distribution field calculation module is used to perform spatiotemporal interpolation on the initial roughness coefficient according to the preset hydrodynamic sensitivity index to obtain a dynamic roughness spatial distribution field covering the river channel partition topology. The hydrodynamic field fitting model construction module is used to construct a Legendre orthogonal function model for each monitoring section unit based on the constraint of the dynamic roughness spatial distribution field, and solve the Legendre orthogonal function model using a two-way long short-time memory network to obtain the hydrodynamic field fitting coefficient set corresponding to each monitoring section unit. The Legendre orthogonal function model is a combination of physical basis functions for the mapping relationship between water level, flow velocity and flow rate of the monitoring section unit. The spatiotemporal relationship graph construction module is used to map all the hydrodynamic field fitting coefficient sets as graph node attributes based on the river channel partition topology to construct the spatiotemporal relationship graph. The global correction coefficient calculation module is used to calculate the global correction coefficient in the spatiotemporal relationship graph based on the sequence regression analysis algorithm. The real-time flow data processing module is used to process the real-time collected hydropower station flow data sequentially according to the global correction coefficient to obtain the corrected flow sequence, and to perform smoothing filtering and boundary constraint verification on the corrected flow sequence to obtain a complete ecological flow monitoring map. The threshold determination and detection module is used to perform threshold determination and detection based on the complete ecological flow monitoring map to obtain the final ecological flow monitoring result.

Images