Water habitat multi-granularity data alignment method and device based on machine learning, computer equipment and readable storage medium

Abstract

The invention discloses a water habitat multi-granularity data alignment method and device based on machine learning, computer equipment and a readable storage medium. The method comprises the steps that firstly, hydrodynamic force, water quality and water ecological parameters of a target area are acquired; performing feature space segmentation on the target area through a support vector machine to obtain divided water quality space data and water ecology space data; based on the segmented data, a hydrodynamic water quality model and a hydrodynamic water ecological model are constructed by using a multi-layer sensor, and data upscaling is realized; and finally, performing multi-granularity data fusion according to the upscaled data to obtain a multi-granularity data alignment result of the aquatic environment. According to the design, the relationship among the multi-source data of the aquatic environment can be comprehensively analyzed, and a scientific basis is provided for aquatic ecology protection.

Description

Technical Field

[0001] The present invention relates to the field of ecological monitoring, and in particular to a method, device, computer equipment and readable storage medium for aligning multi-granularity data of aquatic habitats based on machine learning. Background Art

[0002] In the field of water habitat monitoring and analysis, accurate understanding of water quality and water ecology is crucial to protecting aquatic ecosystems. However, when dealing with complex water habitat data, traditional methods have problems such as difficulty in fully considering the relationship between water dynamics, water quality and water ecology, as well as inconsistent data granularity, which leads to insufficient in-depth and accurate understanding of water habitats and inability to effectively support ecological protection and management decisions. Summary of the invention

[0003] The object of the present invention is to provide a method, device, computer equipment and readable storage medium for aligning multi-granularity data of aquatic habitats based on machine learning.

[0004] In a first aspect, an embodiment of the present invention provides a method for aligning multi-granularity data of aquatic habitats based on machine learning, comprising: Obtain the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area; Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, where m and n are both positive integers; Based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed through a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; Based on the water quality data and water ecological data that have completed data upscaling, multi-granularity data fusion is performed to obtain the multi-granularity data alignment results of the water habitat.

[0005] In a possible implementation, the obtaining of the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area includes: Obtaining water flow velocity, water flow direction and water depth in the target area as original hydrodynamic parameters; Obtaining the pH value, dissolved oxygen, conductivity and turbidity of the target area as original water quality parameters; Obtaining the abundance of benthic animals, zooplankton, phytoplankton and fish in the target area as original water ecological parameters; Based on the PCA algorithm, data features are extracted from the hydrodynamic original parameters, water quality original parameters, and water ecological original parameters, redundant variables are eliminated, key features are retained, and decorrelated high-dimensional feature data are obtained; The hydrodynamic parameters, water quality parameters and water ecological parameters of the target area are obtained based on the decorrelated high-dimensional feature data.

[0006] In a possible implementation, based on the hydrodynamic parameters, the water quality parameters, and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, including: Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented in feature space by a support vector machine, and a hyperplane that maximizes the classification boundary is determined in the feature space: ;in, : hyperplane weight vector; : bias; : Slack variable, used to deal with classification errors; : Penalty factor, used to control model complexity and balance classification accuracy; according to the determined hyperplane that maximizes the classification boundary, the water quality data divided into m water quality spaces and the water ecological data divided into n water ecological spaces are obtained.

[0007] In a possible implementation, based on the water quality data of the m water quality spaces, a hydrodynamic water quality model is constructed by a multi-layer perceptron to achieve data upscaling for the water quality data, including: The input layer to the hidden layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the activation function; the hydrodynamic parameters include velocity V, flow Q, tidal amplitude T, and water depth H; is the hydrodynamic eigenvector; is the weight matrix from the input layer to the hidden layer; For the hidden layer The bias of each neuron; The hidden layer to output layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, is the predicted value of water quality parameter; is the activation function of the output layer; is the weight from the hidden layer to the output layer; is the output layer bias; The trained hydrodynamic water quality model is used to map low-resolution water quality data onto a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity, so as to achieve data upscaling for the water quality data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, The predicted water quality parameters.

[0008] In a possible implementation, based on the water ecological data of the n water ecological spaces, a hydrodynamic water ecological model is constructed by a multi-layer perceptron to achieve data upscaling for the water ecological data, including: The input layer to the hidden layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the hydrodynamic characteristic vector, V is the velocity, Q is the flow, T is the tidal amplitude, and H is the water depth; The hidden layer to output layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, is the predicted value of water ecological parameters; is the output layer activation function; is the weight from the hidden layer to the output layer; is the output layer bias; The hydrodynamic and water ecological model that has been trained is used to map the hydrodynamic parameters of the unobserved area to water ecological parameters to achieve data upscaling for the water ecological data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, For the predicted water ecological parameters.

[0009] In a possible implementation, the multi-granularity data fusion is performed based on the water quality data and the water ecological data after the data upscaling is completed to obtain the multi-granularity data alignment result of the water habitat, including: A two-way chain feedback verification is carried out based on the water quality data and water ecological data that have completed data upscaling, and the multi-granularity data alignment results of the water habitat are determined in combination with the chain feedback error.

[0010] In one possible implementation, the chain feedback error is expressed by the formula: Calculated, among which, is the chain feedback error, is the original input water quality parameter; is the water quality prediction value after reverse calculation; is the sample size.

[0011] In a second aspect, an embodiment of the present invention provides a device for aligning multi-granularity data of aquatic habitats based on machine learning, comprising: An acquisition module is used to acquire hydrodynamic parameters, water quality parameters and water ecological parameters of a target area; based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, wherein m and n are both positive integers; based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed by a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; The analysis module is used to perform multi-granularity data fusion based on the water quality data and water ecological data that have completed data upscaling, and obtain the multi-granularity data alignment results of the water habitat.

[0012] In a third aspect, an embodiment of the present invention provides a computer device, comprising a processor and a non-volatile memory storing computer instructions, wherein when the computer instructions are executed by the processor, the computer device executes the method described in the first aspect.

[0013] In a fourth aspect, an embodiment of the present invention provides a readable storage medium, wherein the readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method described in the first aspect.

[0014] Compared with the prior art, the beneficial effects provided by the present invention include: adopting a method, device, computer equipment and readable storage medium for aligning multi-granularity data of water habitats based on machine learning disclosed by the present invention, by obtaining the hydrodynamics, water quality and water ecological parameters of the target area; performing feature space segmentation on the target area by a support vector machine to obtain the divided water quality space data and water ecological space data; based on the segmented data, constructing a hydrodynamic water quality model and a hydrodynamic water ecological model by a multi-layer perceptron to achieve data upscaling; using the trained hydrodynamic water quality model to map low-resolution water quality data to a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity; finally, performing multi-granularity data fusion according to the upscaled data to obtain the alignment result of multi-granularity data of water habitats. Such a design can comprehensively analyze the relationship between multi-source data of water habitats and provide a scientific basis for aquatic ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative work.

[0016] Figure 1 A schematic diagram of the steps of a method for aligning multi-granularity data of aquatic habitats based on machine learning provided in an embodiment of the present invention; Figure 2 A schematic block diagram of the structure of a device for aligning multi-granularity data of aquatic habitats based on machine learning provided in an embodiment of the present invention; Figure 3 A schematic block diagram of the structure of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0018] The specific implementation modes of the present invention are described in detail below in conjunction with the accompanying drawings.

[0019] In order to solve the technical problems in the aforementioned background technology, Figure 1A flow chart of a method for aligning multi-granularity data of aquatic habitats based on machine learning is provided in an embodiment of the present disclosure. The method for aligning multi-granularity data of aquatic habitats based on machine learning is introduced in detail below.

[0020] Step S201, obtaining the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area; Step S202, based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, where m and n are both positive integers; Step S203, based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed through a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; Step S204, multi-granularity data fusion is performed based on the water quality data and water ecological data that have completed data upscaling to obtain a multi-granularity data alignment result of the water habitat.

[0021] In the embodiment of the present invention, for example, in an actual water habitat monitoring project, the server, as a core processing unit, undertakes the important task of data collection and collation. The target area is set as a large lake basin, which has a vast area and a complex and diverse water ecosystem.

[0022] Acquisition of hydrodynamic parameters: The server first collects hydrodynamic parameters from sensor nodes distributed at key locations in the lake basin. These sensors include current meters, water level meters and other equipment, which monitor and record data such as water flow velocity, water flow direction, and water depth in real time. For example, the current meter located at the water inlet of the lake transmits water flow velocity data to the server once an hour, and the server organizes and stores this data according to time and space. Assume that at a certain moment, the water flow velocity at the water inlet is 2 meters per second, the water flow direction is southeast, and the water depth is 5 meters. The server accurately records this data to form an ordered data set.

[0023] Acquisition of water quality parameters: At the same time, the server obtains water quality parameters from water quality monitoring stations. These monitoring stations are equipped with various professional water quality testing instruments, such as pH meters, dissolved oxygen meters, conductivity meters, turbidity meters, etc., which can detect and upload water quality data such as pH value, dissolved oxygen, conductivity, turbidity, etc. in real time. For example, the water quality monitoring station located in the center of the lake sends water quality data to the server every half an hour, including the current pH value of 7.5, dissolved oxygen content of 8mg / L, conductivity of 500μS / cm, turbidity of 5NTU, etc. The server also organizes and stores these data according to time and space location for subsequent analysis and processing.

[0024] Acquisition of water ecological parameters: For water ecological parameters, the server collects data from multiple ecological monitoring points. These monitoring points obtain data such as benthic animal abundance, zooplankton abundance, phytoplankton abundance, and fish abundance through manual sampling and laboratory analysis. For example, ecological monitoring personnel regularly collect water samples and biological samples in different areas of the lake. After laboratory analysis, the data such as the abundance of benthic animals is 100 per square meter, the abundance of zooplankton is 50 per liter, the abundance of phytoplankton is 1000 per liter, and the abundance of fish is 5 per cubic meter are uploaded to the server. The server classifies and organizes these data, and together with the hydrodynamic parameters and water quality parameter data, they constitute the basic data set for the target area.

[0025] After acquiring rich data on hydrodynamics, water quality and water ecological parameters, the server began to use the support vector machine (SVM) algorithm to perform feature space segmentation on the target area.

[0026] Data preprocessing: The server first decorrelates the collected high-dimensional feature data to eliminate redundancy and correlation between the data. For example, through the principal component analysis (PCA) algorithm, it identifies and eliminates redundant variables that have little impact on the results, retains key parameters, and improves the efficiency of data expression. Suppose that after PCA analysis, the server determines that the water flow velocity in the hydrodynamic parameters, the pH value in the water quality parameters, and the phytoplankton abundance in the water ecological parameters are key parameters.

[0027] SVM model training: The server uses decorrelated high-dimensional feature data as input and combines the spatial information of feature elements to start training the SVM model. During the training process, the server determines that the input parameters include screened hydrodynamic parameters, water quality parameters, and water ecological parameters. For example, the server sets the water flow velocity, water flow direction, and water depth in the hydrodynamic parameters, the pH value, dissolved oxygen, conductivity, and turbidity in the water quality parameters, and the abundance of benthic animals, zooplankton, phytoplankton, and fish in the water ecological parameters as the input parameters of the SVM model. At the same time, the server optimizes the linear kernel function, regularization parameter (C), and kernel function parameters to improve model performance.

[0028] Spatial segmentation process: The SVM model projects high-dimensional data into a higher-dimensional space by constructing a decision hyperplane that maximizes the classification boundary, thereby achieving accurate classification of complex data. In the example of this large lake basin, the server uses the SVM model to segment the water quality data into m spatial regions and the water ecology data into n spatial regions. Assume that m=5 and n=4, that is, the water quality data is segmented into 5 spatial regions and the water ecology data is segmented into 4 spatial regions. The data within each region has a high degree of homogeneity, while the differences between regions are obvious. For example, in one of the water quality spatial regions, the water quality characteristics are characterized by a relatively stable pH value between 7-7.5 and a high dissolved oxygen content; while in another water quality spatial region, the pH value may be between 6-6.5 and the dissolved oxygen content is low.

[0029] Model output and evaluation: After SVM spatial segmentation is completed, the server outputs the support vector (i.e., the key data points selected during the training process), the decision boundary (the hyperplane that separates different categories of data), the classification label (the classification results of each region after spatial segmentation), the weight vector, and the bias term (the parameter that defines the decision boundary). At the same time, the server also outputs model accuracy indicators (such as precision, recall rate, etc.) to evaluate the performance and reliability of the model. For example, the server calculated that the model has a precision of 90% and a recall rate of 85%, indicating that the model has high accuracy and reliability in spatial segmentation monitoring.

[0030] After completing the feature space segmentation, the server uses the water quality space and water ecological space data obtained through the segmentation to construct a hydrodynamic water quality model and a hydrodynamic water ecological model respectively through a multi-layer perceptron (MLP) to achieve data upscaling.

[0031] Hydrodynamic-water quality model construction: Data preparation: The server uses the aligned hydrodynamic data and water quality data as samples, and divides them into 80% training sets and 20% test sets in the m water quality spaces. For example, in the first water quality space, the server selects 800 sets of 1,000 sets of hydrodynamic and water quality data as training sets and 200 sets as test sets.

[0032] Model training: In each SVM segmentation space, the server applies the MLP algorithm to construct the hydrodynamic → water quality spatial response relationship. The hydrodynamic parameters (flow velocity, flow, tidal amplitude and water depth) are used as input layer variables, and the key features are extracted through the nonlinear mapping of the hidden layer, and the predicted values ​​of water quality indicators are finally output. For example, in a certain water quality spatial area, the server sets the hydrodynamic parameters of the input layer to flow velocity 2 m / s, flow 100 m3 / s, tidal amplitude 1 m, and water depth 5 m. After the nonlinear mapping and calculation of the hidden layer, the output water quality indicators such as Chl-a are predicted to be 5 μg / L. For the design of the hidden layer, the server selects the appropriate number of neurons and activation functions according to the complexity of the water quality characteristics. For water quality parameters that show strong nonlinear relationships, such as Chl-a, the server uses the ReLU activation function to better capture the boundary effect; for smoother water quality indicators, such as dissolved oxygen, a linear activation function is applied. During model training, the server iteratively adjusts the network weights through the back propagation algorithm to gradually minimize the prediction error and ensure the fitting effect of the model.

[0033] Upscaling: Using the trained MLP model, the server maps low-resolution water quality data to a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity. For example, the server uses upscaling to extrapolate the original water quality data monitored every 1 km to obtain a high-resolution water quality data set with a grid point every 100 meters. During the extrapolation process, the server adopts a subspace-by-subspace extrapolation strategy, inputs the hydrodynamic parameters of the unobserved area, and uses the MLP model to calculate the corresponding water quality parameter output values ​​one by one. For example, for an unobserved area, the server inputs the hydrodynamic parameters of the area (flow velocity 1.5 m / s, flow 80 m3 / s, tidal amplitude 0.8 m, water depth 4 m), and the MLP model outputs the water quality parameter values ​​of the area (DO is 7 mg / L, TN is 2 mg / L, TP is 0.1 mg / L, etc.). In order to verify the reliability of the extrapolation results, the server compares some of the observed data with the extrapolation results and calculates error indicators (such as MSE and R^2). When the deduction results are highly consistent with the actual observations, it indicates that the model's predictions in unobserved areas have high credibility.

[0034] Construction of hydrodynamic-water ecology model: Data preparation: The server uses the aligned hydrodynamic data and water quality data as samples, and divides them into 80% training sets and 20% test sets in the n water ecological spaces. For example, in the first water ecological space, the server selects 640 sets of 800 sets of hydrodynamic and water ecological data as training sets and 160 sets as test sets.

[0035] Model training: In each SVM segmentation space, the server applies the MLP algorithm to construct the hydrodynamic → water ecological spatial response relationship. The input layer of MLP includes hydrodynamic parameters such as flow velocity, flow, tidal amplitude and water depth, and the output layer predicts water ecological parameters (such as BM, ZP, PP, FM). For example, the server inputs the hydrodynamic parameters of a certain area (flow velocity 1.8 m / s, flow 90 m3 / s, tidal amplitude 0.9 m, water depth 4.5 m), and after calculation by the MLP model, outputs the predicted values ​​of the water ecological parameters of the area (benthic abundance is 80 per square meter, zooplankton abundance is 40 per liter, phytoplankton abundance is 800 per liter, fish abundance is 4 per cubic meter, etc.). The hidden layer captures the nonlinear relationship through the activation function, and the number of neurons is determined according to the complexity of the ecological characteristics. During the model training process, the server gradually approaches the response law of the real ecosystem by optimizing the learning weights and bias parameters layer by layer.

[0036] Upscaling: Using the trained MLP model, the server maps the hydrodynamic parameters of the unobserved area to the water ecological parameters, achieving the goal of water ecological indicators from local observation to large-scale deduction. For example, for an unobserved area, the server inputs the hydrodynamic parameters of the area (flow velocity 1.6 m / s, flow 75 m3 / s, tidal amplitude 0.7 m, water depth 4 m), and the MLP model outputs the water ecological parameters of the area (benthic abundance 70 per square meter, zooplankton abundance 35 per liter, phytoplankton abundance 700 per liter, fish abundance 3 per cubic meter, etc.). The prediction results are verified by comparing with some observed data, and the mean square error (MSE) and determination coefficient (R²) of the model are calculated to evaluate the applicability and accuracy of the model in unobserved areas.

[0037] After completing the upscaling of water quality data and water ecology data, the server begins to perform multi-granularity data fusion and construct a water quality-water ecology chain feedback reaction model.

[0038] Bidirectional chain feedback verification: The server uses water quality to drive water ecology prediction, and then uses water ecology to reversely infer water quality through a bidirectional chain process, verifying the consistency of water quality and water ecology models in bidirectional prediction and the rationality of the mutual feedback mechanism. For example, in the water quality-driven water ecology succession model, water quality parameters drive the growth of phytoplankton through nutrient concentrations (such as TN and TP), and dissolved oxygen supports the survival of benthic organisms and fish through oxygen supply. The server uses the nonlinear structure of MLP to capture these complex relationships, such as using the quantitative indicators of phytoplankton as output to directly reflect the driving effects of TN and TP. In the water ecology-driven water quality change model, the water ecosystem reacts to water quality parameters through metabolism, such as phytoplankton photosynthesis increases dissolved oxygen, and the activities of benthic organisms may increase the organic matter content in the water body, thereby affecting COD and BOD. The server extracts the nonlinear pattern of water ecological characteristics through hidden layers and deduces the details of water quality changes.

[0039] Chain feedback error calculation: The server uses the predicted water ecological parameters as input to infer the water quality parameters, compares the initial water quality input and the inferred water quality output, and calculates the feedback error. For example, the server uses the predicted water ecological parameters such as phytoplankton density (PP), zooplankton density (ZP), benthic abundance (BM) and fish abundance (FM) as input to infer water quality parameters (such as TN, TP, DO, COD, etc.). Assuming that the TN in the original input water quality parameter is 2 mg / L, the TN in the water quality prediction value obtained after inversion is 1.8 mg / L, and the number of samples is 100, the server calculates the feedback error according to the formula. If the feedback error is small, it means that the water quality and water ecological models maintain logical consistency during the prediction and inversion process; if the error is large, the server analyzes the source of the deviation, which may indicate that some input features are not important enough or the weights of the model need to be re-optimized. For example, if the prediction error of phytoplankton density (PP) is large, the server analysis may be that the model is not sensitive enough to total phosphorus (TP) or flow velocity (V), thereby further optimizing the feature selection and training weights of the model. By identifying key parameters, the server can formulate more precise intervention measures for regional water quality management, such as controlling total phosphorus emissions or optimizing hydrodynamic conditions, and ultimately obtain multi-granularity data alignment results for water habitats, providing a scientific basis for the protection and management of regional water habitats.

[0040] In the embodiment of the present invention, the acquisition of the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area can be implemented through the following examples.

[0041] Obtaining water flow velocity, water flow direction and water depth in the target area as original hydrodynamic parameters; Obtaining the pH value, dissolved oxygen, conductivity and turbidity of the target area as original water quality parameters; Obtaining the abundance of benthic animals, zooplankton, phytoplankton and fish in the target area as original water ecological parameters; Based on the PCA algorithm, data features are extracted from the hydrodynamic original parameters, water quality original parameters, and water ecological original parameters, redundant variables are eliminated, key features are retained, and decorrelated high-dimensional feature data are obtained; The hydrodynamic parameters, water quality parameters and water ecological parameters of the target area are obtained based on the decorrelated high-dimensional feature data.

[0042] In the embodiment of the present invention, for example, in a water habitat monitoring project for a large river basin, the server undertakes the key task of data collection and processing. In order to obtain the original hydrodynamic parameters of the target area, a large number of sensor devices are installed at key locations in the river basin, such as the river inlet, bends, and tributary intersections.

[0043] Among them, the flow meter is used to measure the water flow velocity. These flow meters are distributed at different monitoring points, and send the real-time monitored water flow velocity data to the server at fixed time intervals (for example, every 15 minutes). For example, the flow meter at the river inlet measures a water flow velocity of 3 meters per second at a certain moment. It will package this data together with the location information and measurement time of the monitoring point and send it to the server. After receiving this data, the server will organize and store it according to the location and time sequence of the monitoring points to form an orderly water flow velocity data set.

[0044] The monitoring of water flow direction relies on compass-type water flow direction sensors. These sensors are also distributed at various monitoring points. They can accurately measure the direction of water flow and send the data to the server. For example, at a bend in the river, the sensor measures that the water flow direction is southeast. The server will associate this information with other hydrodynamic data of the monitoring point for subsequent comprehensive analysis.

[0045] The water depth data is obtained through the pressure water level sensor installed at the bottom of the river. The sensor calculates the current water depth based on the water pressure and transmits the data to the server. For example, at a certain location in the river, the water depth measured by the water level sensor is 8 meters. The server will store this data together with the water flow speed and direction data at that location to form a complete data set of hydrodynamic raw parameters.

[0046] In order to obtain the original parameters of water quality in the target area, multiple water quality monitoring stations are set up in the river basin. These monitoring stations are equipped with various professional water quality testing instruments.

[0047] The pH value is measured by a pH meter. The pH meter is installed at the water sample collection point of the monitoring station. It can measure the pH value of the water sample in real time and send the data to the server. For example, at a monitoring station in the middle reaches of a river, the pH meter measures the pH value of the current water sample to be 7.2. The server will record this data as well as the corresponding monitoring station location and measurement time.

[0048] The dissolved oxygen content is monitored by a dissolved oxygen meter. The dissolved oxygen meter uses electrochemical or optical methods to measure the dissolved oxygen concentration in the water sample and transmits the data to the server. For example, at a monitoring station downstream of the river, the dissolved oxygen meter measures the dissolved oxygen content as 6 mg / L, and the server will include this data in the water quality original parameter data set.

[0049] The measurement of conductivity relies on a conductivity meter. The conductivity meter can measure the conductivity of water samples, reflecting the content of dissolved electrolytes in water. The conductivity meter at the monitoring station sends the measured data to the server. For example, if the conductivity measured at a certain monitoring point is 400μS / cm, the server will organize and store this data.

[0050] Turbidity is monitored using a turbidity meter. The turbidity meter determines turbidity by measuring the degree of light scattering by suspended particles in the water sample. For example, in a polluted area of ​​a river, the turbidity measured by the turbidity meter is 10 NTU. The server will save this data together with other water quality parameters to form a complete data set of original water quality parameters.

[0051] In order to obtain the original parameters of water ecology, ecological monitoring personnel will regularly collect biological samples in different areas of the river basin.

[0052] For the benthic animal abundance survey, monitoring personnel will set up multiple sampling points at the bottom of the river and use professional sampling tools to collect benthic animal samples. The samples are then classified and counted in the laboratory to obtain the benthic animal abundance data for each sampling point. For example, in a certain section of the river, after sampling and analysis, the benthic animal abundance is 50 per square meter, and these data will be uploaded to the server.

[0053] Zooplankton abundance is measured by collecting water samples and counting the zooplankton in the water samples using a microscope or a plankton counter. For example, in a lake area of ​​a river, after analyzing the collected water samples, the zooplankton abundance is 30 per liter, and the server will receive and store this data.

[0054] Phytoplankton abundance is also monitored by water sampling and laboratory analysis. Phytoplankton abundance data is obtained by counting the phytoplankton in the water sample. For example, in a eutrophic area of ​​a river, the phytoplankton abundance is 800 per liter, and the server will record this data.

[0055] The survey of fish abundance uses a variety of methods, such as netting and mark-recapture. Monitoring personnel sample fish in different areas, count the number of fish, and obtain fish abundance data. For example, in a certain fish gathering area in a river, after investigation, the fish abundance is 8 per cubic meter. These data will be uploaded to the server and together with other water ecological parameters, form a water ecological original parameter data set.

[0056] After receiving the original hydrodynamic parameters, water quality parameters and water ecology parameters, the server will start the data feature extraction process based on the PCA algorithm.

[0057] First, the server integrates and normalizes all the raw parameter data to make different types of data comparable. For example, the water velocity, pH value, benthic animal abundance and other data are normalized so that their values ​​range from 0 to 1.

[0058] Then, the server uses the PCA algorithm to analyze the standardized data. The PCA algorithm finds the main direction of change in the data, namely the principal component, by calculating the covariance matrix, eigenvalues, and eigenvectors of the data. For example, in the hydrodynamic parameters, it may be found that there is a certain correlation between water velocity and water depth. The PCA algorithm can extract a principal component that can represent the main information of these two parameters, thereby eliminating redundant variables.

[0059] Among the water quality parameters, some parameters may have relatively little impact on water quality. The PCA algorithm can identify these less important variables and remove them. For example, in a specific river area, the change in conductivity has little impact on the overall assessment of water quality. The server will remove it based on the analysis results of the PCA algorithm and retain key features such as pH value and dissolved oxygen.

[0060] For water ecological parameters, the PCA algorithm is also used for analysis. For example, in some cases, there may be a strong correlation between zooplankton abundance and phytoplankton abundance. The server can extract the principal components that represent their main information, eliminate redundant information, and retain the features that have a key impact on the water ecosystem.

[0061] After the PCA algorithm extracts data features, the server obtains decorrelated high-dimensional feature data. These data eliminate redundant variables and retain key features, providing a more efficient and accurate data foundation for subsequent analysis and model building.

[0062] After obtaining the decorrelated high-dimensional feature data, the server will further organize and classify these data to obtain the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area.

[0063] For hydrodynamic parameters, the server extracts key feature data related to water velocity, water direction and water depth from the decorrelated high-dimensional feature data. For example, after PCA algorithm processing, a principal component that can comprehensively reflect water velocity and water depth information may be obtained. The server will store this principal component as a new hydrodynamic parameter for subsequent analysis.

[0064] For water quality parameters, the server will select key feature data related to pH, dissolved oxygen, conductivity and turbidity. For example, key features such as pH and dissolved oxygen that have been screened by the PCA algorithm are retained as the final water quality parameters for subsequent water quality assessment and model building.

[0065] For water ecological parameters, the server will extract key feature data related to benthic abundance, zooplankton abundance, phytoplankton abundance, and fish abundance from the decorrelated high-dimensional feature data. For example, after PCA algorithm processing, a principal component that can represent the main information of zooplankton abundance and phytoplankton abundance may be obtained, and the server will save and analyze this principal component as a new water ecological parameter.

[0066] Through the above steps, the server successfully obtained the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area, providing accurate and effective data support for the subsequent multi-granularity data feedback analysis of the water habitat.

[0067] In an embodiment of the present invention, based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the feature space of the target area is segmented by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, which can be implemented through the following examples.

[0068] Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented in feature space by a support vector machine, and a hyperplane that maximizes the classification boundary is determined in the feature space: ;in, : hyperplane weight vector; : bias; : Slack variable, used to deal with classification errors; : Penalty factor, used to control model complexity and balance classification accuracy; according to the determined hyperplane that maximizes the classification boundary, the water quality data divided into m water quality spaces and the water ecological data divided into n water ecological spaces are obtained.

[0069] In the embodiment of the present invention, for example, in a water habitat monitoring and analysis project for a large lake, the server has acquired a wealth of hydrodynamic parameters (such as water velocity, water flow direction, water depth), water quality parameters (such as pH value, dissolved oxygen, conductivity, turbidity) and water ecological parameters (such as benthic abundance, zooplankton abundance, phytoplankton abundance, fish abundance). Now, the server will use a support vector machine (SVM) to segment the target area into feature spaces to determine the hyperplane that maximizes the classification boundary.

[0070] The server first integrates and preprocesses these parameter data to ensure that the data format is unified and comparable. For example, the data obtained from different monitoring points and at different times are arranged and standardized according to certain rules so that the data of each parameter is within a reasonable value range.

[0071] Next, the server starts the SVM algorithm. The goal of SVM is to find a hyperplane in the feature space that can best separate different categories of data. In this lake example, different categories of data may represent areas with different water quality and water ecological characteristics.

[0072] In order to find this hyperplane, the server needs to determine the hyperplane weight vector, bias, slack variable, and penalty factor. The hyperplane weight vector determines the direction of the hyperplane, which is determined by learning and analyzing the data features. For example, when analyzing water quality data, the server may find that the two features of pH value and dissolved oxygen content are important for distinguishing different water quality areas. Then, when determining the hyperplane weight vector, the weights corresponding to these two features will be relatively large.

[0073] The bias is the position offset of the hyperplane in the feature space. The server will adjust the bias value according to the distribution of the data so that the hyperplane can better adapt to the characteristics of the data. For example, if most of the data with good water quality is concentrated in a certain area of ​​the feature space, the server will adjust the bias to make the hyperplane closer to this area, thereby more accurately separating the areas with good water quality from those with poor water quality.

[0074] Slack variables are used to deal with classification errors. In actual data, there may be some data points that are difficult to classify completely correctly. These data points are called outliers or noise points. Slack variables allow these data points to deviate from the hyperplane to a certain extent, thereby avoiding the model from overfitting these abnormal data. For example, in some local areas of the lake, due to special geographical environment or human factors, water quality parameters may fluctuate abnormally, and the data points in these areas may not fully conform to the division rules of the hyperplane. At this time, slack variables can play a certain regulatory role.

[0075] The penalty factor is used to control model complexity and balance classification accuracy. If the penalty factor is set to a large value, the model will pay more attention to classification accuracy, which may cause the model to be too complex and prone to overfitting; if the penalty factor is set to a small value, the model will be relatively simple, but it may sacrifice a certain degree of classification accuracy. The server will find a suitable penalty factor based on actual conditions through continuous trial and adjustment. For example, in the initial model training, the server set a large penalty factor and found that the model had high classification accuracy on the training data, but performed poorly on new data and had an overfitting problem. Therefore, the server gradually reduced the penalty factor and observed the performance of the model on the validation data set until a penalty factor value was found that could both ensure classification accuracy and avoid overfitting.

[0076] In this process, the server will continuously iterate and optimize the parameters of the hyperplane so that the hyperplane can separate data of different categories as much as possible while minimizing the classification error. For example, the server will start from the initial random parameters, continuously adjust the hyperplane weight vector, bias, slack variables and penalty factors, calculate the classification results after each adjustment, and update the parameters according to the difference between the classification results and the actual labels. After multiple iterations, the server finally determines the hyperplane that maximizes the classification boundary in the feature space.

[0077] After determining the hyperplane that maximizes the classification boundary, the server will spatially segment the water quality data and water ecology data of the target area according to this hyperplane.

[0078] For water quality data, the server will divide the entire lake area into m water quality spaces according to the position and direction of the hyperplane. For example, assuming m=3, the server divides the lake into three water quality spaces according to the hyperplane. In the first water quality space, the water quality data shows characteristics such as high pH value, rich dissolved oxygen content, and low turbidity, which may represent areas with better water quality in the lake, such as near the upstream source of the lake, where there is less pollution and a relatively superior water ecological environment. In the second water quality space, the water quality data may be characterized by moderate pH value, relatively low dissolved oxygen content, and increased turbidity. This may be the middle reaches of the lake, which has been affected by a certain degree of human activities and has deteriorated water quality. In the third water quality space, the water quality data may show the characteristics of low pH value, low dissolved oxygen content, and high turbidity. This may be the downstream area of ​​the lake, where the water quality is poor due to the acceptance of a large number of pollutants.

[0079] The server will organize and store the water quality data in each water quality space to form independent data sets. These data sets will be used for subsequent analysis and model building, such as building water quality prediction models and evaluating water quality change trends.

[0080] Similarly, for water ecological data, the server divides the lake area into n water ecological spaces according to the hyperplane. Assuming n=4, the server divides the lake into four water ecological spaces. In the first water ecological space, the water ecological data show that the abundance of benthic animals is high, the abundance of zooplankton is moderate, the abundance of phytoplankton is rich, and the abundance of fish is high, which indicates that the water ecosystem in this area is relatively healthy and stable, and the biodiversity is rich. This may be an area with a good ecological environment in the lake, such as some shallow water areas with lush aquatic plants and relatively gentle water flow. In the second water ecological space, there may be a low abundance of benthic animals, a decrease in zooplankton abundance, a relatively low abundance of phytoplankton, and a decrease in fish abundance. This may be due to certain pollution of water quality, which affects the survival and reproduction of organisms. In the third water ecological space, the water ecological data may show the characteristics of excessive reproduction of phytoplankton, unstable abundance of zooplankton, threatened survival of benthic animals, and low abundance of fish. This may be an area where eutrophication occurs in the lake. In the fourth water ecological space, the water ecological data may show an overall low abundance of organisms due to relatively harsh environmental conditions.

[0081] The server will classify and store the data in each water ecological space to form corresponding data sets. These data sets will provide important basis for subsequent water ecological analysis and protection measures, such as assessing the health status of water ecosystems and formulating ecological restoration plans.

[0082] Through the above steps, the server successfully performed feature space segmentation of the target area based on the support vector machine, and obtained water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, laying the foundation for further feedback analysis of multi-granularity data of water habitats.

[0083] In an embodiment of the present invention, based on the water quality data of the m water quality spaces, a hydrodynamic water quality model is constructed through a multi-layer perceptron to achieve data upscaling for the water quality data, which can be implemented through the following examples.

[0084] The input layer to the hidden layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the activation function; the hydrodynamic parameters include velocity V, flow Q, tidal amplitude T, and water depth H; is the hydrodynamic eigenvector; is the weight matrix from the input layer to the hidden layer; For the hidden layer The bias of each neuron; The hidden layer to output layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, is the predicted value of water quality parameter; is the activation function of the output layer; is the weight from the hidden layer to the output layer; is the output layer bias; The trained hydrodynamic water quality model is used to map low-resolution water quality data onto a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity, so as to achieve data upscaling for the water quality data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, The predicted water quality parameters.

[0085] In an embodiment of the present invention, illustratively, in a water habitat monitoring and analysis project for a coastal estuary area, the server has obtained water quality data of m divided water quality spaces, and now needs to construct a hydrodynamic water quality model through a multi-layer perceptron (MLP).

[0086] The server first processes the input layer to the hidden layer of the model. In this estuary area, hydrodynamic parameters (flow velocity V, flow Q, tidal amplitude T, water depth H) are important input information of the model. For example, at a specific monitoring point in the estuary, the real-time hydrodynamic data received by the server are: flow velocity V = 2.5 m / s, flow Q = 500 m3 / s, tidal amplitude T = 1.2 m, water depth H = 6 m. These data constitute the hydrodynamic feature vector.

[0087] The server inputs these hydrodynamic feature vectors into the input layer of the MLP model. The number of neurons in the input layer corresponds to the number of hydrodynamic parameters, and each neuron is responsible for receiving the value of a hydrodynamic parameter. Next, these data are connected to the neurons in the hidden layer through the weight matrix. The weight matrix is ​​continuously adjusted and optimized during the model training process, and it determines the degree of influence of each input parameter on the hidden layer neurons.

[0088] The activation function plays a key role in this process. For water quality characteristics that exhibit strong nonlinear relationships, the server may choose the ReLU activation function. For example, in this estuary area, there is a complex nonlinear relationship between certain water quality parameters (such as chlorophyll-a concentration) and hydrodynamic conditions. After the input hydrodynamic feature vector is calculated through the weight matrix and the neurons in the hidden layer, the ReLU activation function processes the calculation result. If the calculation result is greater than 0, the result is directly output; if it is less than 0, 0 is output. This can effectively capture the nonlinear characteristics in the data, allowing the model to better fit the actual situation.

[0089] The bias of the i-th neuron in the hidden layer is also an important parameter. It can adjust the activation threshold of the neuron so that the neuron can have different responses when receiving inputs of different strengths. For example, in some cases, even if the input hydrodynamic parameter value is small, the neuron may still be activated due to the bias, thus affecting the output of the model.

[0090] Assume that in a water quality space in this estuary area, after calculation from the input layer to the hidden layer, the output of the first neuron in the hidden layer is 0.8. This output value will serve as the basis for subsequent calculations and continue to be passed in the model.

[0091] The output of the hidden layer is passed as input to the output layer. The goal of the output layer is to predict the values ​​of water quality parameters, such as dissolved oxygen (DO), chemical oxygen demand (COD), total nitrogen (TN), etc. The server determines the number of neurons in the output layer based on the water quality characteristics and research needs, and each neuron corresponds to a water quality parameter that needs to be predicted.

[0092] The activation function of the output layer will be selected according to the characteristics of the water quality parameters. For some water quality parameters with a certain range of values, such as dissolved oxygen content, which is usually between 0-10 mg / L, the server may choose a linear activation function. This ensures that the output value is within a reasonable range and can directly reflect the actual value of the water quality parameter.

[0093] The weights from the hidden layer to the output layer and the output layer bias are also continuously adjusted and optimized during the model training process. These parameters determine how the outputs of the hidden layer are combined and converted into the final water quality parameter prediction values.

[0094] For example, in a water quality space in this estuary area, after calculation from the hidden layer to the output layer, the predicted value for the water quality parameter dissolved oxygen (DO) is 7.2 mg / L. This predicted value is based on the input hydrodynamic parameters and the weights and biases learned by the model, and it reflects a possible state of water quality in the area under the current hydrodynamic conditions.

[0095] After completing the training of the hydrodynamic water quality model, the server will use the model to upscale the water quality data.

[0096] In this estuary area, the original water quality monitoring data may be obtained at some discrete monitoring points, and the data resolution is relatively low. For example, a monitoring point is set every 5 kilometers to obtain the water quality parameter data at that point. However, in order to understand the water quality status of the entire estuary area more comprehensively and in detail, these low-resolution data need to be mapped to a high-resolution hydrodynamic data grid.

[0097] The server first obtains a high-resolution hydrodynamic data grid, which contains detailed spatial information of the entire estuary area, and the grid point spacing may be as small as 100 meters. For each grid point, the server has corresponding hydrodynamic parameters (such as flow rate, flow, tidal amplitude, and water depth).

[0098] The server then inputs the hydrodynamic parameters of each grid point into the trained hydrodynamic and water quality model. The model will predict the water quality parameter value corresponding to the grid point based on these hydrodynamic parameters, following the calculation process from the input layer to the hidden layer and from the hidden layer to the output layer.

[0099] For example, at a grid point in the estuary area, the spatial coordinates are (x=1000m, y=2000m), the hydrodynamic parameters are velocity V=2.2m / s, flow Q=450m3 / s, tidal amplitude T=1.1m, and water depth H=5.5m. The server inputs these parameters into the hydrodynamic water quality model, and after calculation, predicts the water quality parameter values ​​of the grid point, such as dissolved oxygen (DO) of 6.8mg / L, chemical oxygen demand (COD) of 5mg / L, and total nitrogen (TN) of 1.2mg / L.

[0100] The server will perform the same operation on all grid points in the entire estuary area, upscaling the low-resolution water quality data into a high-resolution water quality dataset. This high-resolution dataset has spatiotemporal continuity and can reflect the spatial distribution and change trend of water quality in the estuary area in more detail.

[0101] In order to verify the reliability of the upscaling results, the server will compare some of the original monitoring data with the predicted data after upscaling. For example, select some water quality data of the original monitoring points, compare them with the predicted data of the corresponding grid points after upscaling, and calculate the error index (such as mean square error, determination coefficient, etc.). If the error index is within an acceptable range, it means that the upscaling results are relatively reliable and can be used for further water quality analysis and research.

[0102] Through the above steps, the server successfully used the multi-layer perceptron to build a hydrodynamic water quality model and achieved the upscaling of water quality data, providing strong support for a deeper understanding and research of the water quality conditions in the estuary area.

[0103] In an embodiment of the present invention, based on the water ecological data of the n water ecological spaces, a hydrodynamic water ecological model is constructed through a multi-layer perceptron to achieve data upscaling for the water ecological data, which can be implemented through the following examples.

[0104] The input layer to the hidden layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the hydrodynamic characteristic vector, V is the velocity, Q is the flow, T is the tidal amplitude, and H is the water depth; The hidden layer to output layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, is the predicted value of water ecological parameters; is the output layer activation function; is the weight from the hidden layer to the output layer; is the output layer bias; The hydrodynamic and water ecological model that has been trained is used to map the hydrodynamic parameters of the unobserved area to water ecological parameters to achieve data upscaling for the water ecological data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, In an embodiment of the present invention, for example, in a water habitat monitoring project for a large wetland, the server constructs a hydrodynamic water ecology model through a multi-layer perceptron (MLP) based on the existing water ecology data of n water ecological spaces to achieve data upscaling.

[0105] For the input layer to the hidden layer of the hydrodynamic and water ecology model, the server receives the hydrodynamic feature vector, such as the flow velocity of 1.5 m / s, the flow of 300 m3 / s, the tidal amplitude of 0.8 m, and the water depth of 4 m monitored in a certain area of ​​the wetland. These data are used as input, and after being calculated with the weight matrix from the input layer to the hidden layer, they are processed by the activation function. The activation function captures the potential relationship between hydrodynamics and water ecology based on the complex nonlinear characteristics of wetland water ecology, and obtains the output of each neuron in the hidden layer. For example, the output of the first neuron in the hidden layer may be 0.6, which represents the comprehensive impact of hydrodynamic factors at the neuron level.

[0106] In the hidden layer to output layer stage, the output of the hidden layer is used as a new input, combined with the weight from the hidden layer to the output layer, and the output layer bias is added, and then processed by the output layer activation function to obtain the predicted value of the water ecological parameter. For example, the predicted abundance of phytoplankton is 800 per liter.

[0107] When data upscaling is to be performed on an unobserved area, the server obtains the hydrodynamic parameters of the grid points in the area. For example, if the coordinates of a grid point are (x=500 meters, y=600 meters), the hydrodynamic parameters are a velocity of 1.3 meters per second, a flow of 280 cubic meters per second, etc. The server inputs these parameters into the trained hydrodynamic water ecological model, and the model outputs the predicted values ​​of the water ecological parameters of the grid point, such as the abundance of zooplankton is 40 per liter, etc. In this way, the upscaling of water ecological data is achieved, providing support for a comprehensive understanding of the water ecological status of the wetland.

[0108] In the embodiment of the present invention, the multi-granularity data fusion is performed on the water quality data and the water ecological data after the data upscaling to obtain the multi-granularity data alignment result of the water habitat, which can be implemented through the following examples.

[0109] A two-way chain feedback verification is carried out based on the water quality data and water ecological data that have completed data upscaling, and the multi-granularity data alignment results of the water habitat are determined in combination with the chain feedback error.

[0110] In an embodiment of the present invention, exemplarily, in a comprehensive monitoring project of aquatic habitats for a city lake, the server undertakes the important task of multi-granularity data fusion based on the water quality data and water ecological data that have completed data upscaling to obtain multi-granularity data alignment results for the aquatic habitats.

[0111] First, the server conducts two-way chain feedback verification. In terms of water quality driving water ecology prediction, the server uses upscaled water quality data, such as nutrient concentration data such as total nitrogen (TN) and total phosphorus (TP) in specific areas, to drive phytoplankton growth predictions. For example, when the TN concentration in a certain area increases, the model predicts that the abundance of phytoplankton will increase accordingly. At the same time, in terms of water ecology driving water quality changes, the server reversely infers water quality changes based on upscaled water ecology data, such as phytoplankton increasing dissolved oxygen content through photosynthesis, benthic biological activities affecting the organic matter content of water bodies, etc. For example, in areas where phytoplankton reproduce in large numbers, the dissolved oxygen content is predicted to rise.

[0112] Next, the server calculates the chain feedback error. It uses the predicted water ecological parameters as input to infer the water quality parameters and compares them with the initial water quality input. For example, the dissolved oxygen content in a certain area in the initial water quality data is 6 mg / L, and the predicted dissolved oxygen value obtained after water ecological inversion is 5.8 mg / L. The server determines the feedback error by calculating the difference between the two.

[0113] Finally, the server determines the alignment result of multi-granularity data of water habitats in combination with the chain feedback error. If the feedback error is small, it indicates that the water quality and water ecological models are consistent in the two-way prediction and the mutual feedback mechanism is reasonable. The prediction result at this time can be used as a reliable alignment result of multi-granularity data of water habitats, providing a scientific basis for the ecological protection and management of lakes; if the error is large, the model needs to be further optimized.

[0114] In the embodiment of the present invention, the chain feedback error is expressed by the formula: Calculated, among which, is the chain feedback error, is the original input water quality parameter; is the water quality prediction value after reverse calculation; is the sample size.

[0115] In an embodiment of the present invention, exemplarily, in a project for monitoring and analyzing the water habitat of an inland lake, when processing multi-granularity data fusion, the server calculates the chain feedback error according to specific steps to determine the reliability of the multi-granularity data alignment results of the water habitat.

[0116] The server first identifies the data that needs to be calculated. The original input water quality parameters are the water quality data of different areas of the lake that were accurately monitored and recorded in the early stage. For example, at a certain sampling point in the lake, the recorded dissolved oxygen content is 8mg / L, the chemical oxygen demand (COD) is 20mg / L, etc. These data are obtained from actual observations and are the basis for subsequent comparisons.

[0117] The water quality prediction value after inference is the water quality parameter prediction value obtained by inference from the water ecological data. For example, based on the water ecological data of phytoplankton, zooplankton and other water ecology in the area, combined with the relationship model between them and water quality, the server infers that the dissolved oxygen content of the sampling point may be 7.8mg / L, COD is 19mg / L, etc.

[0118] The number of samples is the total number of samples involved in this calculation. Assuming that in this lake monitoring, a total of 100 sampling points in different locations were selected for data collection and analysis, then the number of samples is 100.

[0119] After obtaining these data, the server starts to calculate the chain feedback error. For each sampling point, it calculates the difference between the original input water quality parameters and the water quality prediction value after inversion, such as the difference in dissolved oxygen is 8mg / L-7.8mg / L=0.2mg / L. Then the differences of all sampling points are comprehensively processed, such as calculating the average difference. The final chain feedback error value can reflect the consistency of the model in two-way prediction. If the error is small, it means that the mutual feedback mechanism of the model is reasonable, and the obtained multi-granularity data alignment result of the water habitat is more reliable; if the error is large, the model needs to be adjusted and optimized.

[0120] In order to more clearly describe the solution provided by the embodiment of the present invention, a relatively complete implementation method is provided below.

[0121] Spatial heterogeneous water quality-ecological chain feedback algorithm framework construction: Feature space segmentation: With decorrelated high-dimensional feature data as input, combined with the spatial information of feature elements, the SVM algorithm is applied for spatial segmentation. The basic idea of ​​SVM is to separate data with different spatial characteristics by finding the best decision boundary. Water quality and water ecological characteristics are usually distributed in different geographical spaces, and their internal data show high homogeneity, while the differences between spaces are obvious. In order to accurately capture these differences, support vector machine (SVM) is used to separate data with different spatial characteristics.

[0122] SVM projects high-dimensional data into a higher-dimensional space by constructing a decision hyperplane that maximizes the classification boundary, thereby achieving accurate classification of complex data. Combining the high-dimensional characteristics of water quality and water ecological data, SVM can identify the spatial characteristic distribution patterns of each region, thereby achieving a reasonable division of data. The goal of this step is to divide the water quality data into m spatial regions and the water ecological data into n spatial regions. Through such segmentation, the spatial differences of water quality and water ecological data can be reflected, providing a regionally homogeneous input data basis for subsequent model construction.

[0123] In the process of spatial segmentation monitoring using support vector machine (SVM), the input parameters include hydrodynamic parameters (water flow velocity, water flow direction, water depth, etc.), water quality parameters (pH value, dissolved oxygen, conductivity, turbidity, etc.) and water ecological parameters (benthic animal abundance, zooplankton abundance, phytoplankton abundance, fish abundance, etc.). The linear kernel function, regularization parameter (C) and kernel function parameters are optimized to optimize model performance. SVM spatial segmentation outputs support vectors (i.e. key data points selected during training), decision boundaries (hyperplanes that separate data of different categories), classification labels (classification results of each region after spatial segmentation), weight vectors and bias terms (parameters that define decision boundaries); and outputs model accuracy indicators (such as precision, recall rate, etc.) to evaluate the performance and reliability of the model. In spatial segmentation monitoring, SVM eliminates redundant variables, identifies key parameters, and improves the efficiency of data expression by combining the principal component analysis (PCA) algorithm. Spatial segmentation is performed, and the monitoring area is divided into several spatial parts by the SVM algorithm to identify the spatial differences of each part. The hyperplane to which the water quality and water ecological monitoring data belong is determined through the classification decision function.

[0124] Construction of SVM classification model: The support vector machine optimizes the objective function and searches for the hyperplane that can maximize the classification boundary in the feature space: ;in, : hyperplane weight vector; : bias; : Slack variables, used to deal with classification errors; : Penalty factor, used to control model complexity and balance classification accuracy.

[0125] The model training process includes data input and labeling (labeling category labels according to the preliminary division of regions); using RBF kernel function to improve nonlinear classification capabilities; optimizing penalty factor C and kernel function parameter γ to balance model performance.

[0126] Results Water quality data were divided into m spatial regions, and water ecology data were divided into n spatial regions.

[0127] Through the space segmentation by SVM, the m spatial regions of water quality data have high internal homogeneity of regional water quality characteristics; the n spatial regions of water ecological data have high internal consistency of regional water ecological characteristics.

[0128] Granular alignment of spatially differentiated multi-source data Construct a nonlinear mapping relationship model between hydrodynamics and water quality to solve the problem of water quality data deduction in complex environments; provide a water environment modeling technology route that adapts to diverse regional characteristics; and achieve upscaling prediction of regional water quality data from local observations to large-scale deductions.

[0129] Hydrodynamics → Water Quality: Using aligned hydrodynamic data and water quality data as samples, in the segmented m water quality spaces, in each SVM segmentation space, a multi-layer perceptron (MLP) algorithm is applied to construct the hydrodynamics → water quality spatial response relationship, 80% of the data is used as a training set, and 20% of the data is used as a test set. The hydrodynamics-water quality model constructed with MLP is used to achieve water quality data upscaling.

[0130] (1) Constructing the MLP hydrodynamics → water quality spatial response relationship in m water quality spaces In the m water quality spaces segmented by support vector machine (SVM), a response model was constructed using multi-layer perceptron (MLP) to reveal the nonlinear impact mechanism of hydrodynamic changes on water quality. Through the spatial segmentation of water quality characteristic data by SVM, complex waters are divided into multiple homogeneous subspaces. The response law of water quality parameters (Chl-a, DO, COD, TN, etc.) in each subspace to hydrodynamic conditions has spatial characteristics, and there are spatial differences between different spaces.

[0131] MLP is further applied in each SVM subspace to construct a nonlinear response model of hydrodynamics and water quality. MLP captures complex nonlinear relationships through the powerful expressive power of the neural network structure. In the model, hydrodynamic parameters (velocity, flow, tidal amplitude and water depth) are used as input layer variables, and key features are extracted through nonlinear mapping of the hidden layer, and finally the predicted values ​​of water quality indicators are output. For the design of the hidden layer, it is necessary to select the appropriate number of neurons and activation functions according to the complexity of the water quality characteristics. For water quality parameters that show strong nonlinear relationships, the ReLU activation function is used to better capture the boundary effect. For smoother water quality indicators, linear activation functions are applied. In addition, during model training, the network weights are iteratively adjusted through the back-propagation algorithm to gradually minimize the prediction error and ensure the fitting effect of the model. The constructed model can accurately predict water quality parameters, and the influence of hydrodynamics on various water quality indicators is quantified by analyzing the network weights and bias parameters.

[0132] In each SVM segmented space, MLP is used to construct the hydrodynamic and water quality response model, and the formula is as follows: Input layer to hidden layer: ; For the hidden layer The output of a neuron; is the activation function (ReLU or Sigmoid); hydrodynamic parameters include velocity (V, m / s), flow (Q, m³ / s), tidal amplitude (T, m), and water depth (H, m); is the hydrodynamic characteristic vector, including velocity (V, m / s), flow (Q, m³ / s), tidal amplitude (T, m), and water depth (H, m); is the weight matrix from the input layer to the hidden layer; For the hidden layer The bias of a neuron.

[0133] Hidden layer to output layer: ;in: is the predicted value of water quality parameter; is the activation function of the output layer (linear or Softmax); is the weight from the hidden layer to the output layer; is the bias for the output layer.

[0134] After the modeling is completed, the output results include the predicted values ​​of water quality indicators and model weights. The spatially differentiated distribution characteristics of water quality are quantitatively described. At the same time, the model output quantifies the sensitivity of hydrodynamic variables to water quality indicators, thereby identifying key driving factors. For example, in the differentiated space, flow in some areas may be the dominant factor affecting COD, while in other areas, tidal amplitude may have a greater impact on TN.

[0135] (2) Scaling formula Using the trained MLP model, low-resolution water quality data are mapped to a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity. The core idea of ​​upscaling is to use the spatial distribution characteristics of hydrodynamic parameters to predict the water quality parameters of each grid point through the MLP model, thereby making up for the lack of space in the original monitoring data. The input of upscaling is a high-resolution hydrodynamic parameter grid, including information such as flow velocity, flow, tidal amplitude and water depth. The MLP model uses these parameters as input variables and outputs the water quality parameter values ​​(DO, TN, TP, etc.) of the corresponding grid points. During the deduction process, a subspace-by-subspace deduction strategy is adopted. By inputting the hydrodynamic parameters of the unobserved area, the MLP model is used to calculate the corresponding water quality parameter output values ​​one by one. Thereby improving the spatial resolution of water quality data and ensuring the physical consistency and environmental rationality of the data. By generating a continuous water quality data set in the entire study area, a high-precision data basis is provided for the analysis and prediction of dynamic changes in water quality. The result of water quality upscaling is a set of high-resolution water quality data sets with strong continuity and accuracy in space. In order to verify the reliability of the deduction results, the study compared some observation data with the deduction results and calculated error indicators (such as MSE and R^2). When the deduction results are highly consistent with the actual observations, it indicates that the model's predictions in the unobserved areas are highly credible.

[0136] Scaling formula: ;in: : The spatial coordinates of the grid points. : Hydrodynamic parameters of the grid points. :Predicted water quality parameters[ ].

[0137] In order to evaluate the performance of the model, the data was divided into a training set (80%) and a test set (20%), and cross-validation was used to evaluate the generalization ability of the model. The regularization method was used to prevent overfitting. The final optimized MLP model was able to accurately fit the hydrodynamic-water ecology relationship in the subspace and had good generalization ability, which could be used to predict water ecological parameters in unobserved areas. The upscaling results were analyzed by the determination coefficient ( ) and mean square error (MSE) evaluation: ;in: is the mean of the actual observed values.

[0138] Hydrodynamics → Water Ecology: Using aligned hydrodynamic data and water quality data as samples, in the n segmented water ecological spaces, in each SVM segmentation space, the multi-layer perceptron (MLP) algorithm is applied to construct the hydrodynamics → water ecological spatial response relationship, 80% of the data is used as a training set, and 20% of the data is used as a test set. The hydrodynamics-water ecology model constructed by MLP is used to achieve water ecological data upscaling.

[0139] (1) Constructing the MLP hydrodynamics → hydroecological space response relationship in n water ecological spaces Through the joint application of support vector machine (SVM) and multi-layer perceptron (MLP), the impact path of hydrodynamics on aquatic ecosystems is identified and its nonlinear mechanism is revealed. Aquatic ecosystems are a multidimensional system composed of indicators such as phytoplankton (PP), zooplankton (ZP), benthic organisms (BM) and fish (FM), and their health status is highly dependent on hydrodynamic conditions. Different hydrodynamic parameters, such as velocity (V), flow (Q), tidal amplitude (T) and water depth (H), have significant temporal and spatial differences in the impact on ecological indicators. Changes in variables may directly affect the growth rate of phytoplankton, the predation efficiency of zooplankton and the diversity characteristics of the ecosystem.

[0140] The MLP algorithm is further applied in each SVM subspace to construct a response model of hydrodynamics and water ecology. The input layer of MLP includes hydrodynamic parameters such as velocity, flow, tidal amplitude and water depth, and the output layer predicts water ecological parameters (such as BM, ZP, PP, FM). The hidden layer captures nonlinear relationships through activation functions, and the number of neurons needs to be determined according to the complexity of ecological characteristics. In some cases, changes in flow velocity may have a significant impact on phytoplankton density (PP), while tidal amplitude may more directly affect zooplankton density (ZP). MLP gradually approaches the response law of the real ecosystem by optimizing learning weights and bias parameters layer by layer. Through the model output, it can be clarified which hydrodynamic parameters are the key drivers of ecological changes.

[0141] In each ecological segmentation space, a response model of hydrodynamics and water ecology is established, and the formula is as follows: Input layer to hidden layer: ;in: For the hidden layer The output of a neuron; is the hydrodynamic characteristic vector (velocity, flow, tidal amplitude, water depth).

[0142] Hidden layer to output layer: ;in: is the predicted value of water ecological parameters; is the output layer activation function; is the weight from the hidden layer to the output layer; is the bias for the output layer.

[0143] The output results include the response model, predicted values ​​and related sensitivity analysis of hydrodynamics and water ecology. The prediction results of the model can provide a quantitative basis for regional water ecological management. For example, in a specific environmental field, if the model shows that the ecological index is the highest within a certain flow rate range, managers can optimize the regional flow rate to maintain ecological health. Model sensitivity analysis can help identify the indicators in the ecosystem that are most susceptible to hydrodynamic changes and provide direction for ecological restoration and management. For example, in a specific environmental field, if it is identified that tidal amplitude is a limiting factor for phytoplankton growth, tidal control measures can be used to improve ecological conditions.

[0144] (2) Scaling formula Using the trained MLP model, the hydrodynamic parameters of the unobserved area are mapped to the water ecological parameters, achieving the goal of deducing water ecological indicators from local observations to large-scale. In water environment monitoring, observation data often have the problem of uneven spatial distribution. It is impossible to fully understand the ecological status of the entire region based on limited local observations. By upscaling and deducing, the study can make high-precision predictions of water ecological indicators in the entire region, providing a scientific basis for regional ecological management.

[0145] During the simulation, the input data are the hydrodynamic parameters (V, Q, T, H) of the unobserved area. The MLP model uses its learned weights and bias parameters to map these hydrodynamic data into water ecological parameters ( , , , The prediction results were verified by comparing with some observed data, and the mean square error (MSE) and determination coefficient ( ) to evaluate the applicability and accuracy of the model in unobserved areas. The study found that when the subspace segmentation is reasonable and the MLP model is fully trained, the prediction results of the upscaling are highly consistent with the actual observation data, with small errors and good model stability.

[0146] In addition, an important feature of upscaling is that it can take into account the spatial distribution characteristics of ecological parameters, so that the prediction results are not only numerically accurate, but also reflect the trend of regional ecological function changes. Through upscaling, the efficient expansion from small-scale local observation data to large-scale regional water ecological status is achieved, solving the problem of insufficient spatial coverage of traditional monitoring methods.

[0147] The upscaling simulation is based on the trained MLP model, and the hydrodynamic parameters of the unobserved area are input into the model to predict its water ecological indicators: Scaling formula: ;in: : spatial coordinates of the grid points. : Hydrodynamic parameters of the grid points. : Predicted water quality parameters .

[0148] In order to evaluate the performance of the model, the data was divided into a training set (80%) and a test set (20%), and cross-validation was used to evaluate the generalization ability of the model. The regularization method was used to prevent overfitting. The final optimized MLP model was able to accurately fit the hydrodynamic-water ecology relationship in the subspace and had good generalization ability, which could be used to predict water ecological parameters in unobserved areas. The upscaling results were analyzed by the determination coefficient ( ) and mean square error (MSE) evaluation: in: is the mean of the actual observed values. 2.3 Multi-granularity data fusion to build a water quality-water ecology chain feedback reaction model: Water quality-water ecology, water ecology-water quality: reversely verify the rationality of water quality and water ecology data after upscaling. Use water quality data and water ecology data derived from different regions for cross-validation. Determine the accuracy of the water quality and water ecology chain feedback reaction model.

[0149] Bidirectional chain feedback verification: Chain feedback verification tests the consistency of water quality and water ecology models in two-way predictions and the rationality of the mutual feedback mechanism. This process uses water quality to drive water ecology prediction, and then uses water ecology to reverse the water quality in a two-way chain process, ensuring that the model can not only accurately predict unidirectional changes, but also reflect the circular feedback relationship in complex ecosystems. Through chain feedback verification, the stability and applicability of the model in different regional subspaces can be verified.

[0150] In the water quality driven aquatic ecological succession model, water quality parameters drive the growth of phytoplankton through nutrient concentrations (such as TN and TP) ( ), dissolved oxygen ( ) supports benthic organisms by providing oxygen ( ) and fish ( ). The nonlinear structure of MLP captures these complex relationships, including As a quantitative indicator of phytoplankton, it directly reflects the driving effect of TN and TP. The dilution effect is reflected through the weighted calculation of the hidden layer.

[0151] In the water ecology-driven water quality change model, the water ecosystem reacts to water quality parameters through metabolism, including phytoplankton ( ) photosynthesis increases dissolved oxygen ( ). Benthic organisms ( ) activities may increase the organic matter content in the water body, thereby affecting COD and BOD. The model extracts the nonlinear pattern of water ecological characteristics through the hidden layer and deduces the details of water quality changes.

[0152] ;Water quality drives water ecology model, output The predicted value of water ecological parameters is . Water ecology drives water quality model, output The predicted value of water quality parameter is .

[0153] Chain feedback error: The error of the chain feedback reflects the consistency of the model in the bidirectional prediction. If the error is small, it means that the water quality and water ecology models maintain logical consistency in the prediction and back-inference process; if the error is large, it may indicate that some input features are not important enough or the model weights need to be re-optimized. The results of chain feedback verification can also be used to identify key driving factors in the system. For example, if the prediction error of phytoplankton density (PP) is large, it may indicate that the model is not sensitive enough to total phosphorus (TP) or flow velocity (V). By analyzing the source of deviation in the chain feedback process, the feature selection and training weights of the model can be further optimized. By identifying key parameters, more precise intervention measures can be formulated for regional water quality management, such as controlling total phosphorus emissions or optimizing hydrodynamic conditions.

[0154] If the predicted water ecological parameters (are small, it means that the water quality and water ecological models maintain logical consistency in the prediction and back-inference process; if the error is large, it may indicate that some input features are not important enough or the weights of the model need to be re-optimized. The results of chain feedback verification can also be used to identify key driving factors in the system. For example, if the prediction error of phytoplankton density (PP) is large, it may indicate that the model is not sensitive enough to total phosphorus (TP) or flow velocity (V). By analyzing the source of deviation in the chain feedback process, the feature selection and training weights of the model can be further optimized. By identifying key parameters, more precise intervention measures can be formulated for regional water quality governance, such as controlling total phosphorus emissions or optimizing hydrodynamic conditions.

[0155] The predicted water ecological parameters ( ) as input to infer water quality parameters ( ), compare the initial water quality input and the inverse water quality output, and calculate the feedback error: in: is the original input water quality parameter; is the water quality prediction value after reverse calculation; is the sample size.

[0156] Please refer to Figure 2 , Figure 2 An aquatic habitat multi-granularity data alignment device 110 based on machine learning provided in an embodiment of the present invention includes: The acquisition module 1101 is used to acquire the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area; based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the feature space of the target area is segmented by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, where m and n are both positive integers; based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed by a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; The analysis module 1102 is used to perform multi-granularity data fusion based on the water quality data and water ecological data that have completed data upscaling, and obtain a multi-granularity data alignment result of the water habitat.

[0157] It should be noted that the implementation principle of the aforementioned aquatic habitat multi-granularity data alignment device 110 based on machine learning can refer to the implementation principle of the aforementioned aquatic habitat multi-granularity data alignment method based on machine learning, which will not be repeated here. It should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into a physical entity, or physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the aquatic habitat multi-granularity data alignment device 110 based on machine learning can be a separately established processing element, or it can be integrated in a chip of the above device for implementation. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The function of the aquatic habitat multi-granularity data alignment device 110 based on machine learning. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in a processor element or an instruction in software form.

[0158] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more microprocessors (digital signal processors, DSP), or one or more field programmable gate arrays (FPGA), etc. For another example, when a module above is implemented in the form of a processing element scheduling program code, the processing element may be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call program code. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0159] The embodiment of the present invention provides a computer device 100, which includes a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned machine learning-based aquatic habitat multi-granularity data alignment device 110. Figure 3 As shown, Figure 3 The computer device 100 provided in the embodiment of the present invention is a structural block diagram. The computer device 100 includes a machine learning-based aquatic habitat multi-granularity data alignment device 110, a memory 111, a processor 112 and a communication unit 113.

[0160] In order to realize data transmission or interaction, the memory 111, the processor 112 and the communication unit 113 are electrically connected to each other directly or indirectly. For example, the electrical connection between these elements can be realized through one or more communication buses or signal lines. The machine learning-based aquatic habitat multi-granularity data alignment device 110 includes at least one software function module that can be stored in the memory 111 in the form of software or firmware or solidified in the operating system (OS) of the computer device 100. The processor 112 is used to execute the machine learning-based aquatic habitat multi-granularity data alignment device 110 stored in the memory 111, such as the software function modules and computer programs included in the machine learning-based aquatic habitat multi-granularity data alignment device 110.

[0161] An embodiment of the present invention provides a readable storage medium, which includes a computer program. When the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the aforementioned machine learning-based aquatic habitat multi-granularity data alignment device 110.

[0162] For illustrative purposes, the foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the present disclosure to the precise form disclosed. Numerous modifications and variations are possible in accordance with the above teachings. These embodiments are selected and described in order to best illustrate the principles of the present disclosure and its practical application, so that those skilled in the art can best utilize the present disclosure and utilize various embodiments with different modifications to suit the intended specific application.

Claims

1. A method for aligning multi-granularity data of aquatic habitats based on machine learning, characterized in that: include: Obtain the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area; Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, where m and n are both positive integers; Based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed through a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; Based on the water quality data and water ecological data that have completed data upscaling, multi-granularity data fusion is performed to obtain the multi-granularity data alignment results of the water habitat.

2. The method according to claim 1, characterized in that The step of obtaining the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area includes: Obtaining water flow velocity, water flow direction and water depth in the target area as original hydrodynamic parameters; Obtaining the pH value, dissolved oxygen, conductivity and turbidity of the target area as original water quality parameters; Obtaining the abundance of benthic animals, zooplankton, phytoplankton and fish in the target area as original water ecological parameters; Based on the PCA algorithm, data features are extracted from the hydrodynamic original parameters, water quality original parameters, and water ecological original parameters, redundant variables are eliminated, key features are retained, and decorrelated high-dimensional feature data are obtained; The hydrodynamic parameters, water quality parameters and water ecological parameters of the target area are obtained based on the decorrelated high-dimensional feature data.

3. The method according to claim 1, characterized in that The feature space segmentation of the target area is performed by a support vector machine based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, including: Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented in feature space by a support vector machine, and a hyperplane that maximizes the classification boundary is determined in the feature space: ;in, : hyperplane weight vector; : bias; : Slack variable, used to deal with classification errors; : Penalty factor, used to control model complexity and balance classification accuracy; according to the determined hyperplane that maximizes the classification boundary, the water quality data divided into m water quality spaces and the water ecological data divided into n water ecological spaces are obtained.

4. The method according to claim 1, characterized in that: Based on the water quality data of the m water quality spaces, a hydrodynamic water quality model is constructed through a multi-layer perceptron to achieve data upscaling for the water quality data, including: The input layer to the hidden layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the activation function; the hydrodynamic parameters include velocity V, flow Q, tidal amplitude T, and water depth H; is the hydrodynamic eigenvector; is the weight matrix from the input layer to the hidden layer; For the hidden layer The bias of each neuron; The hidden layer to output layer of the hydrodynamic water quality model constructed by the multi-layer perceptron is: ; in, is the predicted value of water quality parameter; is the activation function of the output layer; is the weight from the hidden layer to the output layer; is the output layer bias; The trained hydrodynamic water quality model is used to map low-resolution water quality data onto a high-resolution hydrodynamic data grid to generate a high-resolution water quality data set with spatiotemporal continuity, so as to achieve data upscaling for the water quality data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, The predicted water quality parameters.

5. The method according to claim 1, characterized in that Based on the water ecological data of the n water ecological spaces, a hydrodynamic water ecological model is constructed through a multi-layer perceptron to achieve data upscaling for the water ecological data, including: The input layer to the hidden layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, For the hidden layer The output of a neuron; is the hydrodynamic characteristic vector, V is the velocity, Q is the flow, T is the tidal amplitude, and H is the water depth; The hidden layer to output layer of the hydrodynamic water ecology model constructed by the multi-layer perceptron is: ; in, is the predicted value of water ecological parameters; is the output layer activation function; is the weight from the hidden layer to the output layer; is the output layer bias; The trained hydrodynamic and water ecological model is used to map the hydrodynamic parameters of the unobserved area to water ecological parameters to achieve data upscaling for the water ecological data, wherein the upscaling formula is: , : spatial coordinates of the grid points, : hydrodynamic parameters of the grid points, For the predicted water ecological parameters.

6. The method according to claim 1, characterized in that The multi-granularity data fusion is performed on the water quality data and the water ecological data according to the completed data upscaling to obtain the multi-granularity data alignment result of the water habitat, including: A two-way chain feedback verification is carried out based on the water quality data and water ecological data that have completed data upscaling, and the multi-granularity data alignment results of the water habitat are determined in combination with the chain feedback error.

7. The method according to claim 6, characterized in that The chain feedback error is given by the formula: Calculated, among which, is the chain feedback error, is the original input water quality parameter; is the water quality prediction value after reverse calculation; is the sample size.

8. A device for aligning multi-granularity data of aquatic habitats based on machine learning, characterized in that: include: An acquisition module is used to obtain the hydrodynamic parameters, water quality parameters and water ecological parameters of the target area; Based on the hydrodynamic parameters, the water quality parameters and the water ecological parameters, the target area is segmented into feature spaces by a support vector machine to obtain water quality data divided into m water quality spaces and water ecological data divided into n water ecological spaces, where m and n are both positive integers; based on the water quality data of the m water quality spaces and the water ecological data of the n water ecological spaces, a hydrodynamic water quality model and a hydrodynamic water ecological model are respectively constructed by a multi-layer perceptron to achieve data upscaling for the water quality data and the water ecological data; The analysis module is used to perform multi-granularity data fusion based on the water quality data and water ecological data that have completed data upscaling, and obtain the multi-granularity data alignment results of the water habitat.

9. A computer device, characterized in that: The computer device comprises a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device executes the method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium includes a computer program, and when the computer program is executed, the computer device where the readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.

Images