A Big Data-Based Integrated Watershed Water Ecological Governance Method and System
By constructing a high-order spatiotemporal perception tensor and a physical information Kuppman autoencoder network, the computational overhead and equipment over-limit problems of nonlinear models in the integrated watershed water ecology governance were solved, realizing linear implicit space control and ecological assessment of watershed water ecology, and improving the system's response speed and hardware security.
Patent Information
- Application Number
- CN202611216133.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-12
- Publication Date
- 2026-11-17
AI Technical Summary
In the existing integrated watershed water ecology management, nonlinear fluid dynamics models are prone to tearing apart the physical topological connectivity structure when processing complex high-dimensional data, causing the prediction model to deviate from reality. Furthermore, the control of massive IoT devices causes an explosion of computational overhead, which cannot meet the requirements of online solution, resulting in sluggish system control response and equipment instability beyond its limits.
A big data-based approach is adopted to construct a high-order spatiotemporal perception tensor by acquiring IoT data. The low-rank core tensor and simplex are extracted by tensor column decomposition. A physical information library Pullman autoencoder network is constructed and mapped to linear latent space state variables. Based on ecological boundary conditions, a latent space control barrier function is constructed, and the control law vector is solved to realize the solution of the linear quadratic programming problem.
It improves upon the problems of computational power explosion and equipment overrunning issues in traditional nonlinear prediction models. Through dual constraints of topology and physics, it achieves linear implicit space control for watershed management, improves system response speed and hardware operation safety, and provides an objective ecological assessment mechanism.
Smart Images

Figure CN123117516A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent watershed water ecology management and control technology, and more specifically, to a watershed water ecology integrated management method based on big data. Background Technology
[0002] Currently, integrated watershed water ecology management mostly relies on nonlinear fluid dynamics models or conventional deep learning networks to process multi-source heterogeneous monitoring data collected by IoT devices in order to deduce hydrodynamic evolution and generate regulation strategies.
[0003] However, this conventional nonlinear control model still has the following drawbacks: When performing dimensionality reduction mapping on complex, high-dimensional watershed environmental data, it is easy to tear apart the original physical topology of the water system, causing the prediction model to deviate from physical reality. Moreover, when facing a massive number of IoT control nodes, it will cause an exponential explosion of computational overhead, which not only fails to meet the requirements of ultra-fast online solution of control commands, but also lacks rigid boundary constraints on the limits of hardware operation, ultimately causing serious delays in system control response and instability of physical equipment execution beyond limits. Summary of the Invention
[0004] To address at least one deficiency or improvement need in existing technologies, this invention provides a big data-based integrated watershed water ecology management method. This method improves upon the fact that traditional watershed management often relies on nonlinear prediction models, which are prone to computational overload due to the collaboration of massive amounts of equipment, leading to issues such as command iteration divergence and equipment exceeding limits and going out of control.
[0005] To achieve the above objectives, according to the first aspect of the present invention, a method for integrated watershed water ecology management based on big data is provided, comprising the following steps: S1 acquires real-time monitoring data of the watershed collected by IoT sensing devices, and constructs the aligned real-time monitoring data into a high-order spatiotemporal sensing tensor. S2 extracts the low-rank core tensor of the high-order spatiotemporal perception tensor through the tensor column decomposition algorithm, constructs a simple complex on the corresponding data point cloud, extracts topological homology features and transforms them into a real physical persistent landscape function. S3 constructs a physical information Koopman autoencoder network, adding the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function to the composite loss function. Through backpropagation, the real-time monitoring data is mapped into linear latent space state variables, and the Koopman operator matrix is output. S4 constructs a latent space control barrier function based on the ecological boundary conditions mapped to the latent space, establishes a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solves for the control law vector. S5 sends the control law vector to the IoT execution device for regulation, and calculates the ecological score based on the rate of change of the Betti number of the simple complex before and after regulation using the projection matrix.
[0006] Further, in step S1, acquiring the real-time monitoring data of the watershed collected by the IoT sensing device and constructing the aligned real-time monitoring data into a high-order spatiotemporal sensing tensor includes: S101 Extracts the water quality monitoring data, hydrodynamic data, and meteorological environment data collected by the IoT sensing device as the real-time monitoring data of the watershed; S102 interpolates and aligns the real-time monitoring data of the watershed according to a preset unified timestamp and spatial latitude and longitude grid; S103 combines the aligned and stitched real-time monitoring data of the watershed along the spatial, temporal, and physical attribute dimensions to construct the high-order spatiotemporal sensing tensor in the form of a multi-dimensional matrix.
[0007] Further, in step S2, the step of extracting the low-rank core tensor of the high-order spatiotemporal perception tensor using a tensor column decomposition algorithm, constructing a simple complex on the corresponding data point cloud, extracting topological homology features, and transforming them into a real physical persistent landscape function includes: S201 performs matrix expansion and singular value decomposition on the high-order spatiotemporal perception tensor to obtain multiple low-rank core tensors that are multiplied consecutively. S202 uses a given distance threshold as a filtering parameter to construct a Vitris-Lipps simplex on the corresponding data point cloud reconstructed by the low-rank core tensor; S203 records the occurrence and disappearance times of the topological homology group generators of each order in the Vitris-Lipps simplex as the filtering parameter monotonically increases, and uses these times as the topological homology features. The topological homology features are mapped into a two-dimensional persistent graph consisting of the occurrence time and the disappearance time; S204 converts the discrete two-dimensional persistent graph into a differentiable continuous function as the real physical persistent landscape function through piecewise linear interpolation.
[0008] Furthermore, in step S3, the construction of the physical information library Pullman autoencoder network, which incorporates the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function into the composite loss function, includes: S301 constructs the physical information Kupman autoencoder network, which includes an encoder and a decoder; S302 establishes an initial network loss function that includes a data reconstruction loss term and a linear evolution prediction loss term; S303 extracts the prediction data of the latent space of the physical information Kupman autoencoder network to generate the predicted persistent landscape function, and calculates the Wasserstein distribution distance between the predicted persistent landscape function and the real physical persistent landscape function as a topology penalty constraint term. S304 extracts the convection-diffusion reaction partial differential equation containing the flow field velocity vector, hydrodynamic dispersion tensor, and biochemical reaction consumption term as the convection-diffusion equation residual; S305 linearly superimposes the initial network loss function, the topology penalty constraint term, and the convection-diffusion equation residual to form the composite loss function.
[0009] Furthermore, in step S3, the process of mapping the real-time monitoring data into linear latent space state variables through backpropagation and outputting the Koopman operator matrix includes: S311 calculates the network gradient with the objective of minimizing the composite loss function; S312 updates the weight parameters of the physical information Koopman autoencoder network through backpropagation using the adaptive moment estimation gradient descent algorithm; S313 The encoder, which has completed the weight parameter update, projects the real-time monitoring data into the linear observable latent space and extracts the latent space feature vector as the state variable of the linear latent space. S314 obtains the Koopman operator matrix by extracting the forward difference transfer operator between the linear latent space state variables of adjacent time steps.
[0010] Further, in step S4, the implicit space control barrier function is constructed based on the ecological boundary conditions mapped to the implicit space. A linear quadratic programming problem is established, driven by the Koopman operator matrix and constrained by the implicit space control barrier function. The control law vector is solved, including: S401 extracts the safe water level threshold and environmental water quality concentration threshold of the watershed physical space as the ecological boundary conditions. S402 maps the ecological boundary conditions to the latent space where the linear latent space state variables are located through the encoder, and generates the latent space control barrier function for the linear latent space state variables; S403 transforms the latent space control barrier function into a discrete-time forward invariance inequality constrained by the extended class function; S404 establishes the linear quadratic programming problem that includes the target reference control law instruction and the weight diagonal matrix; S405 substitutes the physical operation limit boundary of the underlying hydraulic execution equipment and the discrete-time forward invariance inequality as hard constraints into the linear quadratic programming problem, and solves the control law vector through an interior point optimization algorithm.
[0011] Furthermore, in step S5, the step of sending the control law vector to the IoT execution device for regulation, and calculating the ecological score based on the rate of change of the Betti number of the simplex before and after regulation using the projection matrix, includes: S501 parses the control law vector into a physical control command electrical signal corresponding to the IoT execution device and sends it to the control node for execution; S502 Obtains the latest real-time monitoring data after the control operation is completed and reconstructs the corresponding latest simple complex shape; S503 extracts the simplex before and after the execution of the control operation, as well as the zeroth-order Betti number and the first-order Betti number in the latest simplex that characterize the physical connectivity branches and closed circulation of the watershed, and calculates the rate of change of the Betti number. S504 calculates the control input resource consumption norm of the control law vector issued within a single control cycle; S505 inputs the control input resource consumption norm and the Betty number change rate into the preset projection matrix, and outputs a single scalar value as the ecological score.
[0012] According to a second aspect of the present invention, a watershed water ecology integrated management system based on big data is also provided, comprising the following modules: The tensor construction module is configured to acquire real-time monitoring data of the watershed collected by IoT sensing devices and construct a high-order spatiotemporal sensing tensor from the aligned real-time monitoring data. The topology extraction module is configured to extract the low-rank core tensor of the high-order spatiotemporal perception tensor through the tensor column decomposition algorithm, construct a simple complex on the corresponding data point cloud, extract topological homology features and transform them into a real physical persistent landscape function. The network mapping module is configured to construct a physical information Koopman autoencoder network, add the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the real physical persistent landscape function to the composite loss function, map the real-time monitoring data into linear latent space state variables through backpropagation, and output the Koopman operator matrix. The dimension reduction control solution module is configured to construct a latent space control barrier function based on the ecological boundary conditions mapped to the latent space, establish a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solve for the control law vector. The control execution and evaluation module is configured to send the control law vector to the Internet of Things execution device for regulation operation, and calculate the ecological score based on the rate of change of the Betti number of the simple complex before and after regulation through the projection matrix.
[0013] According to a third aspect of the present invention, a computer device is also provided, comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of the big data-based integrated watershed water ecology management method.
[0014] According to a fourth aspect of the invention, a storage medium is also provided, which stores a computer program executable by an access authentication device, which, when run on the access authentication device, causes the access authentication device to perform the steps of the big data-based integrated watershed water ecology management method.
[0015] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: (1) The present invention provides a watershed water ecology integrated management method and system based on big data. Through the topological and physical dual constraint Koppman network, the nonlinear ecological optimization is reduced to the latent space linear control solution, thereby improving the traditional watershed management which mostly adopts nonlinear prediction model. Due to the massive equipment collaboration, it is easy to cause computing power explosion, resulting in command iteration divergence and equipment over-limit loss of control.
[0016] (2) The method and system for integrated watershed water ecology management based on big data provided by the present invention decomposes the high-order perception tensor to extract the low-rank core tensor and constructs a simple complex, thereby transforming the topological homology features into physical persistent landscape functions. This improves the problem that traditional data fusion mostly adopts ordinary linear dimensionality reduction mechanism, which ignores physical spatial features, resulting in feature extraction distortion and water system network structure breakage.
[0017] (3) The method and system for integrated watershed water ecology management based on big data provided by the present invention extracts the change rate of the Betty number of the simple complex before and after regulation, and then uses the projection matrix to map it into a single scalar to calculate the objective ecological score. This improves the problem that traditional watershed management evaluation mostly adopts manual inspection and sampling qualitative analysis, which is extremely dependent on subjective assignment by personnel, resulting in the lagging evaluation of management effectiveness and difficulty in standard quantification. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1A flowchart illustrating a big data-based integrated watershed water ecology management method provided in this application embodiment; Figure 2 A schematic diagram of the network mapping and operator solution process of a big data-based integrated watershed water ecology management method provided in this application embodiment; Figure 3 A schematic diagram of the hidden space security management solution process for a big data-based integrated watershed water ecology governance method provided in this application embodiment; Figure 4 A schematic diagram of the modular architecture of a watershed water ecology integrated management system based on big data, provided for an embodiment of this application; Figure 5 This is a schematic diagram of the computer equipment hardware structure for a big data-based integrated watershed water ecology management method provided in an embodiment of this application. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.
[0021] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.
[0022] Example 1: like Figure 1 As shown, in the first embodiment of the present invention, a method for integrated watershed water ecology management based on big data is provided, including the following steps: S1. Obtain real-time monitoring data of the watershed collected by IoT sensing devices, and construct a high-order spatiotemporal sensing tensor from the aligned real-time monitoring data.
[0023] Furthermore, in step S1, real-time monitoring data of the watershed collected by IoT sensing devices is acquired, and the aligned real-time monitoring data is constructed into a high-order spatiotemporal sensing tensor, including: S101 extracts water quality monitoring data, hydrodynamic data, and meteorological environmental data collected by IoT sensing devices as real-time monitoring data for the watershed. S102 interpolates and aligns the real-time monitoring data of the watershed according to a preset unified timestamp and spatial latitude and longitude grid; S103 combines the aligned and stitched real-time monitoring data of the watershed along the spatial, temporal, and physical attribute dimensions to construct a high-order spatiotemporal sensing tensor in the form of a multi-dimensional matrix.
[0024] Specifically, the system input receives raw, heterogeneous monitoring streams transmitted in real time from discretely distributed IoT sensing nodes. This data stream undergoes resampling and spatial interpolation operations via a spatiotemporal grid alignment engine. After computation, the output generates a multidimensional high-order tensor matrix with strict dimensional mapping relationships, and this matrix stream is then transmitted completely to the downstream topology feature extraction module.
[0025] In traditional watershed monitoring systems, various sensor data are often stored in isolated relational databases. This flattened storage approach completely isolates the coupling relationships between different physical quantities across specific hydrological spaces and continuous time spans. This embodiment specifically extracts water quality monitoring data, hydrodynamic data, and meteorological environmental data to establish a multi-source evaluation benchmark for the watershed environment across media. Because the sampling frequencies of the field-deployed sensors vary and their spatial distribution is random, the system uses a preset unified timestamp and spatial latitude and longitude grid as a benchmark for spatial interpolation and temporal alignment. This processing method eliminates the spatiotemporal misalignment defects caused by heterogeneous sensing devices from the data source.
[0026] Preferably, after alignment, the system combines the aforementioned data along the spatial, temporal, and physical attribute dimensions to generate a high-order spatiotemporal sensing tensor. Upscaling discrete data into a multidimensional tensor structure serves to maintain the original physical entanglement and nonlinear dependencies between multimodal monitoring data. This tensor quantization process provides a faithful data foundation for subsequent system processing, preventing the loss of spatiotemporal features that is easily caused by conventional linear data fusion methods when dealing with complex water network environments.
[0027] Specifically, the high-order spatiotemporal awareness tensor in multidimensional matrix form constructed in this step follows the following mathematical expression: ; in This represents the constructed high-order spatiotemporal perception tensor entity. It represents the domain of the set of real numbers. This represents the total dimension of the longitude division in the preset spatial latitude and longitude grid. This represents the total dimension of the latitude direction division in the preset spatial latitude and longitude grid. It represents the total step size of the time series after discretization by a unified timestamp. This represents the total number of categories in the physical attribute dimension, which is the sum of the feature types of water quality monitoring data, hydrological and dynamic data, and meteorological and environmental data.
[0028] Arbitrary local elements within a higher-order spacetime perception tensor The corresponding representation indicates the watershed in the th The first longitude grid, the first The geographic coordinates of the latitude grid are located at the th latitude grid. A unified timestamp, for the first The objective and true value measured by a specific physical property.
[0029] S2. Extract the low-rank core tensor of the high-order spatiotemporal perception tensor using the tensor column decomposition algorithm, construct a simple complex on the corresponding data point cloud, extract topological homology features, and transform them into a real physical persistent landscape function.
[0030] Specifically, in step S2, the low-rank core tensor of the high-order spatiotemporal perception tensor is extracted using the tensor column decomposition algorithm. A simple complex is then constructed on the corresponding data point cloud. Topological homology features are extracted and transformed into a real physical persistent landscape function, including: S201 performs matrix expansion and singular value decomposition on the high-order spatiotemporal perception tensor to obtain multiple low-rank core tensors that are multiplied consecutively. S202 uses a given distance threshold as a filtering parameter to construct a Vitris-Lipps simplex on the corresponding data point cloud reconstructed by the low-rank core tensor. S203 records the occurrence and disappearance times of topological homology group generators of each order in the Vitris-Lipps simplex as the filtering parameter monotonically increases, serving as topological homology features. S204 maps topological homology features into a two-dimensional persistent graph consisting of appearance time and disappearance time; S205 transforms a discrete two-dimensional persistent graph into a differentiable continuous function as a real physical persistent landscape function through piecewise linear interpolation.
[0031] In this embodiment, the system input receives the high-order spatiotemporal perception tensor processed in the previous stage. This tensor data undergoes spatial algebraic transformation and feature extraction through an internal tensor decomposition module and topology analysis module. The final output generates a differentiable real physical persistent landscape function, and this continuous function stream is transmitted to the subsequent physical information Kupperman autoencoder network, serving as the topological constraint benchmark for loss function calculation.
[0032] Directly using the original high-order spatiotemporal perception tensor from watershed environmental big data in computation can lead to severe dimensionality curse. This step forces the high-order spatiotemporal perception tensor to be matrix-expanded, relying on singular value decomposition to remove redundant environmental noise from the underlying data. This operation decomposes the complex multidimensional matrix into multiple low-rank core tensors that are multiplied consecutively. The retained core tensors maintain the most fundamental algebraic framework of the water ecological data, reducing the computational overhead of subsequent models.
[0033] The underlying mathematical model corresponding to the tensor decomposition process is: ; in It represents the discrete set of elements of a higher-order spatiotemporal perception tensor under a specific coordinate mapping. to These refer to the subscripts of the spatial dimension, temporal dimension, and physical attribute dimension, respectively. to Refers to the first to the second generated after singular value decomposition. The elements of a low-rank core tensor matrix.
[0034] Specifically, the system reconstructs the dimensionality-reduced low-rank core tensor into a point cloud in the metric space, assigning a monotonically increasing distance threshold starting from zero as a filtering parameter. Once the distance constraint between any two or more discrete data points is less than or equal to this filtering parameter, the system immediately connects them to form a simplex, and then aggregates them to generate a Vitris-Lipps simplex complex. This complex construction mechanism transcends the boundary between discrete data and continuous geometric space, mathematically replicating the physical connectivity of real watershed systems.
[0035] The geometric constraint rules for the Vitris-Lipps simplex are defined as follows: ; in The current distance threshold filtering parameter is The Vitry-Lips simple complex generated at that time. This represents the complete set of point clouds generated by reconstructing the data from the low-rank core tensor. It refers to a simplex subset consisting of data points whose distances satisfy certain conditions. as well as It refers to any two distinct discrete data points within a simplex subset. Refers to data points With data points The actual spatial Euclidean distance in the corresponding metric space.
[0036] In this embodiment, as the filtering parameters increase, various topological holes emerge and subsequently close within the connected structure of the point cloud. The system captures the appearance and disappearance times of these topological homology group generators of various orders and establishes them as topological homology features. Based on the appearance and disappearance time mapping, a set of discrete scattered points in a two-dimensional plane is generated, i.e., a two-dimensional persistent graph. Backpropagation of gradients in deep learning networks requires the input features to be differentiable. Discrete two-dimensional persistent graphs cannot directly satisfy the differentiation condition. The system intervenes with a piecewise linear interpolation algorithm to forcibly transform the discrete scattered points into a differentiable continuous function form. A real physical persistent landscape function is established, breaking down the computational barrier that prevents non-Euclidean topological features from being incorporated as penalty terms into the neural network loss function.
[0037] The functional relationship for the interpolation transformation from a two-dimensional persistent graph to a real physically persistent landscape is as follows: ; in Representing the The generators of a topological homology group are independent variables. The landscape function mapping values at that location. Represents a continuous, independent parameter. Representing the The occurrence time of generators of a topological homology group. Representing the The disappearance time of a topological homology group generator. and These represent algebraic operations that determine the maximum and minimum values of the two values, respectively.
[0038] S3. Construct a physical information Koopman autoencoder network, incorporating the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function into a composite loss function. Through backpropagation, map real-time monitoring data into linear latent space state variables and output the Koopman operator matrix, such as... Figure 2 As shown.
[0039] Furthermore, in step S3, a physical information library Pullman autoencoder network is constructed, incorporating the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function into the composite loss function, including: S301 constructs a physical information Koopman autoencoder network containing encoders and decoders; S302 establishes an initial network loss function that includes a data reconstruction loss term and a linear evolution prediction loss term; S303 extracts the prediction data of the latent space of the physical information Kupman autoencoder network to generate the prediction persistent landscape function, and calculates the Wasserstein distribution distance between the prediction persistent landscape function and the real physical persistent landscape function as the topological penalty constraint term. S304 extracts the convection-diffusion partial differential equation containing the flow field velocity vector, hydrodynamic dispersion tensor, and biochemical reaction consumption term as the convection-diffusion equation residual. S305 linearly superimposes the initial network loss function, topological penalty constraint term, and convection-diffusion equation residuals to form a composite loss function.
[0040] Preferably, in step S3, the real-time monitoring data is mapped to linear latent space state variables through backpropagation, and the Koopman operator matrix is output, including: S311 calculates the network gradient with the goal of minimizing the composite loss function; S312 updates the weight parameters of the physical information Koopman autoencoder network through backpropagation using the adaptive moment estimation gradient descent algorithm. S313 projects real-time monitoring data into the linear observable latent space by updating the weight parameters through the encoder, and extracts the latent space feature vector as the linear latent space state variable. S314 obtains the Koopman operator matrix by extracting the forward difference transfer operator between the linear latent space state variables of adjacent time steps.
[0041] In this embodiment, the system acquires the real physical persistent landscape function and real-time monitoring data transmitted in the preceding steps as the underlying input parameters. The physical information Koopman autoencoder network receives the above input parameters and performs forward computation and backward propagation. After the network training converges, the system's autoencoder outputs linear latent space state variables, and the forward difference calculation module outputs the Koopman operator matrix. These two core output data are directed to the control node for constructing the underlying quadratic programming boundary conditions.
[0042] Conventional deep learning models severely lack the constraints of physical conservation laws when processing large amounts of water ecological data, and are prone to disrupting the connectivity of the real water network during dimensionality reduction projection. In this embodiment, a data mapping path with both topological and hydrodynamic constraints is established by constructing a physical information Kuppman autoencoder network containing an encoder and a decoder.
[0043] Specifically, the system establishes an initial network loss function that includes a data reconstruction loss term and a linear evolution prediction loss term to prevent the loss of information entropy during data dimensionality reduction. Predictive data from the network latent space is extracted to generate a predicted persistent landscape function, and the Wasserstein distribution distance between this function and the real physical persistent landscape function is calculated. This distribution distance is set as a topological penalty constraint term, forcing the characteristic manifold of the latent space to maintain consistency with the algebraic topology of the real watershed.
[0044] Optionally, a convection-diffusion partial differential equation (PDE) is introduced, which includes the flow field velocity vector, hydrodynamic dispersion tensor, and biochemical reaction consumption term. The algebraic residual of this PDE is extracted as the convection-diffusion equation residual. The initial network loss function, topology penalty constraint term, and convection-diffusion equation residual are linearly superimposed to form a composite loss function.
[0045] In this embodiment, the network gradient is calculated with the goal of minimizing the aforementioned composite loss function. The system calls the adaptive moment estimation gradient descent algorithm to perform backpropagation, iteratively updating the weight parameters of the physical information Koopman autoencoder network. After completing the weight replacement, the encoder directly projects the complex real-time monitoring data onto the linearly observable latent space, extracting the latent space feature vectors as linear latent space state variables. The Koopman operator matrix is obtained by extracting the forward difference transfer operator between the linear latent space state variables of adjacent time steps. This matrix directly overcomes the computational barrier of nonlinear dynamics, reconstructing the complex watershed evolution into pure matrix multiplication in an infinite-dimensional linear space.
[0046] Specifically, the underlying mathematical expression of the network composite loss function is defined as follows: ; in This represents the overall composite loss function generated by superposition. The data reconstruction loss term represents the measurement of encoding and decoding errors. The linear evolution prediction loss term represents the measure of the time series extrapolation error of latent variables. This represents the Wasserstein distribution distance between two persistent landscape functions. This represents the residual of the convection-diffusion equation.
[0047] The physical partial differential expression of the convection-diffusion equation residuals is as follows: ; in This represents the operation of taking the square of the matrix's L2 norm on the internal elements. This represents the concentration matrix of water environment features that are inversely mapped from the decoder to the three-dimensional physical space. The time-dependent variable represents continuous flow. This represents the Hamiltonian space differential operator. This represents the velocity field vector of the flow field within the physical space of the watershed. The hydrodynamic dispersion tensor represents the intensity of water quality diffusion. This represents the consumption items resulting from naturally occurring biochemical reactions within the aquatic ecosystem.
[0048] In this embodiment, the linear derivation relation for extracting the Koopman operator matrix through forward difference is: ; in This represents the linear latent space state variable acquired at the current computation time step. This represents the linear latent space state variable extracted in the next adjacent time step. This represents the Koopman operator matrix obtained from the final solution.
[0049] S4. Based on the ecological boundary conditions mapped to the latent space, construct the latent space control barrier function, establish a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solve for the control law vector, such as... Figure 3 As shown.
[0050] In step S4, a latent space control barrier function is constructed based on the ecological boundary conditions mapped to the latent space. A linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function is established, and the control law vector is solved. Specifically, this may include: S401 extracts the safe water level threshold and environmental water quality concentration threshold of the watershed physical space as ecological boundary conditions. S402 maps ecological boundary conditions to the latent space where linear latent space state variables reside through an encoder, generating a latent space control barrier function for the linear latent space state variables; S403 transforms the latent space control barrier function into a discrete-time forward invariance inequality constrained by extended class functions; S404 establishes a linear quadratic programming problem that includes the target reference control law command and the weight diagonal matrix; S405 substitutes the physical operation limit boundary of the underlying hydraulic execution equipment and the discrete-time forward invariance inequality as hard constraints into the linear quadratic programming problem, and solves the output control law vector through the interior point optimization algorithm.
[0051] In this embodiment, optionally, the system's underlying data input terminal receives the safe water level threshold and the environmental water quality concentration threshold extracted from the watershed's physical space. The system synchronously imports the linear latent space state variables and the Koopman operator matrix output from the preceding algorithm steps. All of the above data streams are uniformly integrated into the quadratic programming optimization solution engine. After convex optimization matrix operations under internal hard constraints, the system output terminal directly generates the corresponding control law vector. This control law vector data is then directed to the IoT execution device to perform hardware actions.
[0052] Traditional nonlinear model predictive control suffers from the fundamental flaw of excessive computational overhead when dealing with the regulation of massive numbers of equipment nodes at the watershed level. Furthermore, conventional control algorithms are highly susceptible to hardware physical limit violations during instruction iteration and optimization. The core objective of this embodiment is to establish a rigid safety baseline for system instruction execution. The system extracts the safe water level threshold and environmental water quality concentration threshold from the watershed's physical space to establish actual ecological boundary conditions. The encoder then forcibly projects the ecological boundary conditions obtained from the physical dimensions onto the latent space dimension where the linear latent space state variables reside. Based on the reconstructed boundary data, the system generates a latent space control barrier function for the linear latent space state variables. This barrier function constructs a virtual mathematical firewall, defining an absolute red line that the system's state evolution cannot cross.
[0053] In this embodiment, the system immediately transforms the latent space control barrier function into a discrete-time forward invariant inequality, which is strictly constrained by a preset extended class function. This algebraic transformation mechanism mathematically forces the system to ensure that its evolution trajectory in any subsequent time series cannot deviate from the safe set domain. The system then establishes a linear quadratic programming problem, which includes a target reference control law instruction and a weight diagonal matrix. The physical operation limit boundary of the underlying hydraulic execution equipment, along with the aforementioned discrete-time forward invariant inequality, is integrated and defined as a hard constraint condition, which is forcibly substituted into the linear quadratic programming equation. Under this architecture, the highly complex nonlinear ecological regulation optimization operation is completely reduced in dimensionality and transformed into a linear convex optimization problem. The system calls an interior-point optimization algorithm to perform millisecond-level ultra-fast optimization operations, solving and outputting the final control law vector. This mechanism eliminates the iterative divergence risk faced by nonlinear control and establishes the safety boundary of the physical equipment hardware regulation link.
[0054] Specifically, the system establishes the generalized functional expression for the objective function of the linear quadratic programming problem as follows: ; in The vector of the control law to be solved With target reference control law instructions The difference vector.
[0055] The mathematical expression for the discrete-time forward invariance inequality is: ; Optionally, substituting the linear dynamic evolution operator into the above discrete-time forward invariance inequality results in the following hard-constrained boundary inequality: ; in This represents the numerical value of the objective functional optimization during the solution process of a quadratic programming problem. This represents the spatial difference vector obtained by subtracting the target reference control law command from the vector of the control law to be solved. This represents the matrix transpose operation corresponding to the spatial difference vector. The semi-positive definite weighted diagonal matrix represents the smoothness of the adjustment and control operation. This represents the control law vector to be solved within the current time step. The system represents the target reference control law instruction that is pre-set based on historical experience. The value of the hidden space control barrier function represents the value at the current time step node. The predicted value of the hidden space control barrier function represents the value at the next adjacent time step node. The adjustment coefficient is an extended class function representing the rate at which the system state converges to the interior of the safe set. This represents the transpose of the spatial gradient vector of the latent space control barrier function at the current system state node. This represents the globally optimal Koopman operator matrix generated by the preceding steps. This represents the linear latent space state variable extracted at the current time step. This represents the control gain matrix that maps the physical electrical signals input to the actuator to the state space.
[0056] S5. The control law vector is sent to the IoT execution device for regulation and control operations, and the ecological score is calculated by the projection matrix based on the rate of change of the Betti number of the simple complex before and after regulation.
[0057] Furthermore, in step S5, the control law vector is sent to the IoT execution device for regulation, and the ecological score is calculated using the projection matrix based on the rate of change of the Betti number of the simplex before and after regulation. S501 parses the control law vector into the corresponding physical control command electrical signal of the IoT execution device and sends it to the control node for execution. S502 acquires the latest real-time monitoring data after the control operation is completed and reconstructs the corresponding latest simple complex shape; S503 extracts the simplex before and after the execution of the control operation, as well as the zeroth and first Betti numbers in the latest simplex that characterize the physical connectivity branches and closed circulation of the watershed, and calculates the rate of change of the Betti number. S504 calculates the control input resource consumption norm of the control law vector issued within a single control cycle. The S505 takes the control input resource consumption norm and the rate of change of the Betty number as inputs to a preset projection matrix and outputs a single scalar value as an ecological score.
[0058] Specifically, in this embodiment, the system's input end acquires the control law vector generated by the preceding operator solution steps and simultaneously receives the latest real-time monitoring data collected by the underlying sensor network. These two types of data streams enter the control execution and evaluation module. After parsing, complex reconstruction, and matrix mapping calculation, the system's output end generates an ecological score in a single scalar numerical format.
[0059] Preferably, the control law vector is parsed into physical control command electrical signals corresponding to the IoT execution devices and sent to the control nodes for execution. This step converts the purely mathematical solution of the upper-layer algorithm into electrical signals that drive the lower-level pumps or gates. After the control operation is completed, the system's front-end sensing network collects and transmits the latest real-time monitoring data in real time. Based on this, the data processing center reconstructs the latest simplex complex reflecting the current physical connectivity of the watershed system.
[0060] Optionally, the simplex complexes before and after the regulation operation, as well as the latest simplex complex, are extracted to isolate the zeroth-order Betti number, representing the physical connectivity of the watershed, and the first-order Betti number, representing the closed circulation. By comparing the specific numerical differences in the corresponding Betti numbers before and after the regulation, the rate of change of the Betti number is calculated. A decrease in the zeroth-order Betti number indicates that isolated polluted water masses are merging or being washed away by flowing water, while an increase in the first-order Betti number indicates an increase in the number of unobstructed loops in the water circulation. By extracting these two algebraic topological invariants, the degree of physical improvement in the water network structure is objectively quantified directly from the perspective of geometric connectivity.
[0061] It should be noted that the system synchronously calculates the control input resource consumption norm of the control law vector issued within a single control cycle. Calculating the control input resource consumption norm quantifies the total mechanical and electrical energy consumed in this hardware control. Subsequently, the control input resource consumption norm and the rate of change of the Betti number are input into a preset projection matrix, outputting a single scalar value as the final ecological score. By introducing a projection matrix analysis method based on the mapping of algebraic topological invariants to energy consumption input, the traditional technical shortcomings of conventional watershed governance evaluation systems—which fragment the underlying physical control process and rely on manual sampling for qualitative assessment—are improved. This avoids the problem that the effectiveness of watershed ecological management cannot be objectively measured in a closed-loop, automated manner.
[0062] Specifically, the calculation of the rate of change of the Betty number follows the algebraic expression: ; in Representing the The relative rate of change of the order Betti number, subscript The value can be either zero or one. Represents the first simple complex corresponding to the The numerical value of the order Betty number. The first simple complex shape corresponding to the control operation before execution. The original value of the order Betti number.
[0063] The quantitative formula for controlling the input resource consumption norm is as follows: ; in This represents the scalar value of the control input resource consumption norm obtained from the solution. This represents an algebraic operator that calculates the L2 norm of the elements of an internal vector. This represents the control law vector issued within the current regulatory cycle.
[0064] Optionally, the matrix projection formula for the final ecological score is as follows: ; in The ecological score represents the final output in scalar form. The projection matrix represents a pre-labeled weight based on historical governance data, with its algebraic dimension set to one row and three columns. This represents a three-dimensional feature column vector composed of the control input resource consumption norm, the rate of change of the zeroth-order Betti number, and the rate of change of the first-order Betti number, vertically concatenated.
[0065] Example 2: In a second embodiment of the present invention, the present invention provides a watershed water ecology integrated management system based on big data, such as... Figure 4 As shown, it includes the following modules: The tensor construction module is configured to acquire real-time monitoring data of the watershed collected by IoT sensing devices and construct a high-order spatiotemporal sensing tensor from the aligned real-time monitoring data. The topology extraction module is configured to extract the low-rank core tensor of the high-order spatiotemporal perception tensor through the tensor column decomposition algorithm, construct a simple complex on the corresponding data point cloud, extract topological homology features and transform them into a real physical persistent landscape function. The network mapping module is configured to construct a physical information Koopman autoencoder network, add the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the real physical persistent landscape function to the composite loss function, map the real-time monitoring data into linear latent space state variables through backpropagation, and output the Koopman operator matrix. The dimension reduction control solution module is configured to construct a latent space control barrier function based on the ecological boundary conditions mapped to the latent space, establish a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solve for the control law vector. The control execution and evaluation module is configured to send the control law vector to the Internet of Things execution device for regulation operation, and calculate the ecological score based on the rate of change of the Betti number of the simple complex before and after regulation through the projection matrix.
[0066] Specifically, in this embodiment, the front end reconstructs the IoT monitoring flow through tensor construction and topology extraction modules, directly extracting the nonlinear entanglement relationships of multi-source heterogeneous environmental elements and the physical connectivity characteristics of the real water network. In the core mapping stage, the extremely complex nonlinear dynamic evolution is rigorously reduced to a linear latent space by injecting the hydrodynamic convection-diffusion residuals and the topological homology distribution distance into the autoencoder loss function. This method improves upon the traditional shortcomings of conventional data early warning models, which are prone to tearing up the topological connectivity of the water system and lack physical mechanism constraints, thus avoiding the problem of system early warning divergence failure under extreme conditions. In the control solution stage, the ecological safety boundary is established based on the linearized latent space state, directly converting the complex nonlinear regulation optimization into a linear quadratic programming calculation with hard constraints from the control barrier function, cutting off the risks of nonlinear iterative divergence and equipment physical operation exceeding limits. The end control module converts the rapidly calculated instructions into actions of the hydraulic equipment and captures the algebraic topological Betti number drift ratio before and after the treatment operation to perform matrix mapping calculation, completely replacing the evaluation mechanism relying on manual inspection and subjective qualitative assessment, achieving a unified approach of rapid safety response of massive equipment nodes and objective quantitative evaluation of treatment effectiveness.
[0067] Examples 3 and 4: In the third and fourth embodiments of the present invention, as Figure 5 As shown, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.
[0068] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.
[0069] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0070] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.
[0071] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0072] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.
[0074] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0075] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.
[0076] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0077] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for integrated watershed water ecology management based on big data, characterized in that, Includes the following steps: S1 acquires real-time monitoring data of the watershed collected by IoT sensing devices, and constructs the aligned real-time monitoring data into a high-order spatiotemporal sensing tensor. S2 extracts the low-rank core tensor of the high-order spatiotemporal perception tensor through the tensor column decomposition algorithm, constructs a simple complex on the corresponding data point cloud, extracts topological homology features and transforms them into a real physical persistent landscape function. S3 constructs a physical information Koopman autoencoder network, adding the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function to the composite loss function. Through backpropagation, the real-time monitoring data is mapped into linear latent space state variables, and the Koopman operator matrix is output. S4 constructs a latent space control barrier function based on the ecological boundary conditions mapped to the latent space, establishes a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solves for the control law vector. S5 sends the control law vector to the IoT execution device for regulation, and calculates the ecological score based on the rate of change of the Betti number of the simple complex before and after regulation using the projection matrix.
2. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S1, acquiring real-time monitoring data of the watershed collected by IoT sensing devices and constructing a high-order spatiotemporal sensing tensor from the aligned real-time monitoring data includes: S101 Extracts the water quality monitoring data, hydrodynamic data, and meteorological environment data collected by the IoT sensing device as the real-time monitoring data of the watershed; S102 interpolates and aligns the real-time monitoring data of the watershed according to a preset unified timestamp and spatial latitude and longitude grid; S103 combines the aligned and stitched real-time monitoring data of the watershed along the spatial, temporal, and physical attribute dimensions to construct the high-order spatiotemporal sensing tensor in the form of a multi-dimensional matrix.
3. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S2, the step of extracting the low-rank core tensor of the high-order spatiotemporal perception tensor using a tensor column decomposition algorithm, constructing a simple complex on the corresponding data point cloud, extracting topological homology features, and transforming them into a real physical persistent landscape function includes: S201 performs matrix expansion and singular value decomposition on the high-order spatiotemporal perception tensor to obtain multiple low-rank core tensors that are multiplied consecutively. S202 uses a given distance threshold as a filtering parameter to construct a Vitris-Lipps simplex on the corresponding data point cloud reconstructed by the low-rank core tensor; S203 records the occurrence and disappearance times of the topological homology group generators of each order in the Vitris-Lipps simplex as the filtering parameter monotonically increases, and uses these times as the topological homology features. S204 maps the topological homology features into a two-dimensional persistent graph composed of the occurrence time and the disappearance time; S205 converts the discrete two-dimensional persistent graph into a differentiable continuous function as the real physical persistent landscape function through piecewise linear interpolation.
4. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S3, the construction of the physical information Kuppman autoencoder network, which incorporates the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the actual physical persistent landscape function into the composite loss function, includes: S301 constructs the physical information Kupman autoencoder network, which includes an encoder and a decoder; S302 establishes an initial network loss function that includes a data reconstruction loss term and a linear evolution prediction loss term; S303 extracts the prediction data of the latent space of the physical information Kupman autoencoder network to generate the predicted persistent landscape function, and calculates the Wasserstein distribution distance between the predicted persistent landscape function and the real physical persistent landscape function as a topology penalty constraint term. S304 extracts the convection-diffusion reaction partial differential equation containing the flow field velocity vector, hydrodynamic dispersion tensor, and biochemical reaction consumption term as the convection-diffusion equation residual; S305 linearly superimposes the initial network loss function, the topology penalty constraint term, and the convection-diffusion equation residual to form the composite loss function.
5. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S3, mapping the real-time monitoring data into linear latent space state variables through backpropagation and outputting the Koopman operator matrix includes: S311 calculates the network gradient with the objective of minimizing the composite loss function; S312 updates the weight parameters of the physical information Koopman autoencoder network through backpropagation using the adaptive moment estimation gradient descent algorithm; S313 The encoder, which has completed the weight parameter update, projects the real-time monitoring data into the linear observable latent space and extracts the latent space feature vector as the state variable of the linear latent space. S314 obtains the Koopman operator matrix by extracting the forward difference transfer operator between the linear latent space state variables of adjacent time steps.
6. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S4, the implicit space control barrier function is constructed based on the ecological boundary conditions mapped to the implicit space. A linear quadratic programming problem is established, driven by the Koopman operator matrix and constrained by the implicit space control barrier function. The control law vector is solved by finding: S401 extracts the safe water level threshold and environmental water quality concentration threshold of the watershed physical space as the ecological boundary conditions. S402 maps the ecological boundary conditions to the latent space where the linear latent space state variables are located through the encoder, and generates the latent space control barrier function for the linear latent space state variables; S403 transforms the latent space control barrier function into a discrete-time forward invariance inequality constrained by the extended class function; S404 establishes the linear quadratic programming problem that includes the target reference control law instruction and the weight diagonal matrix; S405 substitutes the physical operation limit boundary of the underlying hydraulic execution equipment and the discrete-time forward invariance inequality as hard constraints into the linear quadratic programming problem, and solves the control law vector through an interior point optimization algorithm.
7. The integrated watershed water ecology management method based on big data as described in claim 1, characterized in that, In step S5, the step of sending the control law vector to the IoT execution device for regulation and control operations, and calculating the ecological score based on the rate of change of the Betti number of the simplex before and after regulation using the projection matrix, includes: S501 parses the control law vector into a physical control command electrical signal corresponding to the IoT execution device and sends it to the control node for execution; S502 Obtains the latest real-time monitoring data after the control operation is completed and reconstructs the corresponding latest simple complex shape; S503 extracts the simplex before and after the execution of the control operation, as well as the zeroth-order Betti number and the first-order Betti number in the latest simplex that characterize the physical connectivity branches and closed circulation of the watershed, and calculates the rate of change of the Betti number. S504 calculates the control input resource consumption norm of the control law vector issued within a single control cycle; S505 inputs the control input resource consumption norm and the Betty number change rate into the preset projection matrix, and outputs a single scalar value as the ecological score.
8. A watershed water ecology integrated management system based on big data, characterized in that, Includes the following modules: The tensor construction module is configured to acquire real-time monitoring data of the watershed collected by IoT sensing devices and construct a high-order spatiotemporal sensing tensor from the aligned real-time monitoring data. The topology extraction module is configured to extract the low-rank core tensor of the high-order spatiotemporal perception tensor through the tensor column decomposition algorithm, construct a simple complex on the corresponding data point cloud, extract topological homology features and transform them into a real physical persistent landscape function. The network mapping module is configured to construct a physical information Koopman autoencoder network, add the convection-diffusion equation residuals and the distribution distance between the predicted persistent landscape function and the real physical persistent landscape function to the composite loss function, map the real-time monitoring data into linear latent space state variables through backpropagation, and output the Koopman operator matrix. The dimension reduction control solution module is configured to construct a latent space control barrier function based on the ecological boundary conditions mapped to the latent space, establish a linear quadratic programming problem driven by the Koopman operator matrix and constrained by the latent space control barrier function, and solve for the control law vector. as well as The control execution and evaluation module is configured to send the control law vector to the Internet of Things execution device for regulation operation, and calculate the ecological score based on the rate of change of the Betti number of the simple complex before and after regulation through the projection matrix.
9. A computer device, characterized in that, It includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of the big data-based integrated watershed water ecology management method according to any one of claims 1 to 7.
10. A storage medium, characterized in that, It stores a computer program that can be executed by an access authentication device. When the computer program is run on the access authentication device, it causes the access authentication device to perform the steps of the big data-based integrated watershed water ecology management method as described in any one of claims 1 to 7.
