Hydrological trend prediction method based on big data analysis

Abstract

The invention provides a hydrological trend prediction method based on big data analysis, and the method comprises the steps: building an initial graph structure which takes a monitoring station as a node and geographic distance and water system connectivity as an edge weight, carrying out the iterative aggregation of node features through a message passing mechanism, and capturing the space interaction of hydrological variables between stations; dynamically adjusting a water system connectivity parameter and an edge weight based on rainfall change, and generating an adaptive graph structure; in combination with the graph neural network, the dynamic Bayesian network and the space-time collaborative Kriging interpolation algorithm, hour-level hydrological dynamic transmission features are extracted, and a space-time coupling prediction model is formed; model parameter optimization and message passing mechanism adjustment are triggered through a prediction deviation threshold value, and a self-adaptive prediction process of'dynamic modeling-feature fusion-closed loop optimization 'is realized. According to the method, hydrological element space-time correlation can be accurately described, the analysis precision of a complex hydrological process is improved, and the adaptability to sudden hydrological events and basin environment changes is enhanced.

Description

Technical Field

[0001] The present invention relates to the field of hydrological data analysis, and in particular to a hydrological trend prediction method based on big data analysis. Background Art

[0002] Time series analysis of hydrological data is a key area in water resources management and flood prevention and disaster reduction. By dynamically modeling variables such as rainfall, flow, and soil moisture, it can reveal the spatiotemporal patterns of hydrological processes in a watershed, providing important support for scientific decision-making. In recent years, neural networks and deep learning technologies have been widely used in hydrological time series forecasting due to their powerful nonlinear modeling capabilities. However, existing methods often focus on forecasting a single site or the entire watershed, ignoring the spatial correlation between monitoring sites and the dynamic transmission process of hydrological variables. This leads to insufficient characterization of the spatial heterogeneity of hydrological processes within the watershed, especially in the dynamic adjustment of connectivity caused by changes in rainfall intensity, which significantly reduces the accuracy and adaptability of the model.

[0003] First, the transmission of hydrological variables between stations is constrained by both geographic distance and river system connectivity, making it difficult for traditional models to integrate these complex spatial relationships. Second, the dynamic adjustment of edge weights due to changes in rainfall intensity increases the complexity of time series modeling, and existing methods lack a real-time response mechanism to this dynamic nature. Furthermore, the multivariate interactions of node characteristics such as rainfall, flow, and soil moisture further exacerbate modeling difficulties. Unresolved dynamic spatial correlation modeling issues limit the accuracy of basin hydrological forecasts. Summary of the Invention

[0004] The embodiments of the present invention provide a hydrological trend prediction method based on big data analysis, which aims to solve the problem that the existing technology is difficult to capture the spatial correlation between sites in a basin and its dynamic evolution process, and the hydrological trend analysis model has low accuracy.

[0005] To achieve the above objectives, the present invention provides a hydrological trend prediction method based on big data analysis, the method comprising:

[0006] Obtaining geographic coordinates, rainfall, flow, and soil moisture data for monitoring stations within the basin, and constructing an initial graph structure with the stations as nodes. The node feature vectors include the rainfall, flow, and soil moisture, and edge weights are calculated based on the geographic distance between stations and water system connectivity.

[0007] A message passing mechanism is used to aggregate node features of the initial graph structure, and feature vectors of the nodes are updated through iterative calculation to capture the mutual influence of hydrological variables between sites, including rainfall, flow, and soil moisture.

[0008] If the change in rainfall in the node feature exceeds a preset threshold, the water system connectivity parameter is adjusted and the edge weight is recalculated to generate a dynamically adjusted graph structure;

[0009] The dynamically adjusted graph structure is input into the graph neural network prediction layer, and the rainfall, flow and soil moisture prediction values of each station are calculated through the fully connected layer mapping;

[0010] Performing time series analysis on the predicted values, extracting hourly hydrological dynamic transmission characteristics, and constructing a coupling model based on a dynamic Bayesian network and a spatiotemporal co-kriging interpolation algorithm;

[0011] According to the spatial distribution characteristics of the coupling model output, the parameters of the graph neural network are updated and the message passing mechanism is optimized;

[0012] If the prediction value deviation of the optimized model output is lower than the preset threshold, the final hydrological variable prediction result is output; otherwise, the dynamic adjustment and parameter optimization process is repeated.

[0013] Furthermore, the edge weight is calculated as follows:

[0014] W ij =D ij (1+α(1-C ij )),

[0015] Where W ij is the edge weight between site i and site j; D ij is the geographical distance between site i and site j; α is the adjustment parameter; C ij is the connectivity factor; if there is a water system connection between site i and site j, then C ij =1, otherwise C ij =0.

[0016] Furthermore, the message passing mechanism is used to aggregate node features of the initial graph structure, and the node feature vectors are updated through iterative calculation to capture the mutual influence of hydrological variables between sites. The hydrological variables include rainfall, flow and soil moisture, including:

[0017] Through the message passing mechanism, the neighbor node information of each node is obtained from the initial graph structure, and the influence of the neighbor node's feature vector on the current node is calculated;

[0018] According to the message vector of each node, the aggregation function is used to fuse the message vectors;

[0019] From the aggregated node feature vector, determine whether the convergence condition is met. If so, stop the iteration and determine the node feature vector; if not, continue the next round of iteration to obtain the updated node feature vector;

[0020] According to the updated node feature vector, calculate the feature change trend of each node;

[0021] The K-means algorithm is used to group nodes based on feature change trends;

[0022] From the node grouping results, obtain the central feature vector of each group, calculate the feature similarity between groups, and obtain the similarity matrix between groups;

[0023] Based on the similarity matrix between groups, the interaction intensity of hydrological variables between groups is determined, and the group relationship reflecting the dynamic interaction of hydrological variables is obtained.

[0024] Furthermore, if the rainfall intensity in the node feature changes by more than a preset threshold, the water system connectivity parameter is adjusted and the edge weight is recalculated to generate a dynamically adjusted graph structure, including:

[0025] If the rainfall intensity change in the node feature exceeds the preset threshold, the water system connectivity parameters are adjusted and the edge weights are updated according to the rainfall intensity change value;

[0026] According to the dynamically adjusted graph structure, the neighbor node set of each node is obtained, the influence of the edge weight of the neighbor node on the current node is calculated, and the edge weight influence value of each node is obtained;

[0027] If the edge weight influence value exceeds the preset influence threshold, the edge weight influence value is normalized using standardization to obtain the normalized edge weight influence value;

[0028] According to the normalized edge weight influence value, the hierarchical clustering algorithm is used to group the nodes to obtain the node grouping results;

[0029] From the node grouping results, obtain the center edge weight influence value of each group, calculate the edge weight similarity between groups, and obtain the similarity matrix between groups;

[0030] If an element in the similarity matrix between groups exceeds the preset similarity threshold, it is determined that the groups have strong connectivity, and the water system connectivity parameters are updated to obtain the group relationship reflecting the dynamic connectivity of the water system;

[0031] According to the grouping relationship reflecting the dynamic connectivity of the water system, the weighted average method is used to calculate the comprehensive connectivity strength of each group to obtain the comprehensive connectivity strength of each group.

[0032] Furthermore, the dynamically adjusted graph structure is input into the graph neural network prediction layer, and the rainfall, flow and soil moisture prediction values of each station are calculated through the fully connected layer mapping, including:

[0033] The dynamic graph structure is processed by graph neural network, and the message passing mechanism is adopted to update the characteristics of each site to obtain the updated site characteristics;

[0034] Based on the updated site features, the fully connected layer mapping is used to calculate the initial prediction value of each site;

[0035] If the deviation between the initial prediction value and the preset threshold exceeds the specified range, the weight matrix and bias parameters are optimized by the gradient descent algorithm to obtain the optimized weight matrix and bias parameters;

[0036] Based on the optimized weight matrix and bias parameters, re-map through the fully connected layer to update the predicted value of each site;

[0037] If there is a significant deviation between the updated forecast value and the historical data, the forecast value will be adjusted through time series analysis;

[0038] Based on the adjusted forecast values, the weighted average method is used to integrate the rainfall, flow and soil moisture forecast values to obtain the comprehensive forecast value of each station;

[0039] Through cluster analysis, the comprehensive prediction values are grouped to obtain the grouping results of the site prediction values.

[0040] Furthermore, the predicted values are subjected to time series analysis to extract hourly hydrological dynamic transmission characteristics, and a coupling model based on a dynamic Bayesian network and a spatiotemporal collaborative Kriging interpolation algorithm is constructed, including:

[0041] Through time series analysis, hourly dynamic features are extracted from hydrological data and the sliding window method is used to obtain hourly dynamic features;

[0042] Based on hourly dynamic features, a dynamic Bayesian network is constructed to define the node state transition probability and obtain the node state probability distribution;

[0043] The spatiotemporal collaborative kriging interpolation algorithm is used to calculate spatial correlation based on the node state probability distribution to obtain spatial interpolation results.

[0044] If the deviation between the spatial interpolation result and the preset threshold exceeds the specified range, the weight coefficient is optimized by gradient descent;

[0045] Recalculate the spatial interpolation results based on the optimized interpolation weights;

[0046] Through data fusion method, the node state probability distribution and the updated spatial interpolation results are combined to obtain the comprehensive prediction value;

[0047] If there is a significant deviation between the comprehensive forecast value and the historical data, the model parameters will be adjusted through a dynamic update mechanism.

[0048] Furthermore, updating the parameters of the graph neural network and optimizing the message passing mechanism according to the spatial distribution characteristics of the coupling model output includes:

[0049] Obtain the interaction strength between nodes through spatial distribution characteristics;

[0050] Determine the set of neighbor nodes based on the interaction strength between nodes;

[0051] Obtain node feature vector through neighbor node set;

[0052] According to the node feature vector, the node message is calculated using the message passing mechanism;

[0053] Calculate the loss function through node messages;

[0054] If the loss function value is greater than the preset threshold, the weight matrix and bias are optimized using gradient descent;

[0055] Update the graph neural network parameters based on the optimized weight matrix and bias.

[0056] Furthermore, if the prediction value deviation of the optimized model output is lower than a preset threshold, the final hydrological variable prediction result is output; otherwise, the dynamic adjustment and parameter optimization process is repeated, including:

[0057] The hydrological variable characteristics are obtained by inputting data, and the characteristic vector is calculated using the feature extraction function;

[0058] Based on the hydrological variable feature vector, a machine learning model is used to generate initial prediction results;

[0059] Calculate the prediction value deviation based on the initial prediction results and the true value data;

[0060] If the predicted value deviation is lower than the preset threshold, the prediction result is output;

[0061] If the deviation is higher than the preset threshold, dynamic adjustment is performed;

[0062] According to the deviation of the predicted value, the model structure is dynamically adjusted and the feature weights are redistributed using the structure update function;

[0063] Through the adjusted model structure, the gradient descent algorithm is used to optimize the parameters;

[0064] Regenerate the prediction results based on the optimized parameters;

[0065] Based on the new prediction results, the deviation calculation and threshold judgment are repeated, and the cycle is executed until the deviation is lower than the preset threshold, and the final hydrological variable prediction results are output.

[0066] The above technical solution has the following technical effects:

[0067] (1) The initial graph structure is constructed through geographical distance and water system connectivity, and the edge weights are dynamically adjusted according to the rainfall change threshold, so that the model can respond in real time to sudden changes in water system connectivity caused by extreme weather, accurately capture the dynamic correlation between factors such as rainfall, flow, and soil moisture in the basin, and significantly improve the dynamic modeling capability of sudden hydrological events.

[0068] (2) The features of neighborhood nodes are iteratively aggregated through the message passing mechanism to explicitly model the physical transmission process of hydrological variables between sites. At the same time, the graph neural network and dynamic Bayesian network + spatiotemporal kriging interpolation are combined to achieve the dual feature fusion of spatial topology and time series, and to analyze the hourly dynamic transmission characteristics of hydrological variables.

[0069] (3) By setting a prediction value deviation threshold, a closed-loop process is triggered to dynamically adjust the graph structure edge weights, update the graph neural network parameters, and optimize the message transmission mechanism, forming a "prediction-feedback-optimization" adaptive mechanism. When the hydrological conditions in the basin change, the model can automatically calibrate the parameters through iterative optimization, continuously reducing the prediction deviation and ensuring high prediction accuracy and stability under different environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 Schematic diagram of the flow of a hydrological trend prediction method based on big data analysis in one embodiment of the present invention;

[0071] Figure 2 A schematic diagram of a process for dynamically adjusting a model structure according to a model prediction result in one embodiment of the present invention;

[0072] Figure 3 Schematic diagram of a process for updating graph neural network parameters based on the spatial distribution characteristics of the coupling model output in one embodiment of the present invention. DETAILED DESCRIPTION

[0073] To further illustrate various embodiments, the present invention is provided with accompanying drawings. These drawings form part of the present disclosure and are primarily used to illustrate the embodiments and, in conjunction with the relevant description in the specification, to explain the operating principles of the embodiments. By referring to these drawings, one of ordinary skill in the art will understand other possible embodiments and the advantages of the present invention. The components in the figures are not drawn to scale, and similar reference numerals are generally used to represent similar components.

[0074] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0075] like Figure 1-3 In this embodiment, a hydrological data mining and trend prediction method based on big data may specifically include:

[0076] Step S101, obtain the geographical coordinates, rainfall, flow and soil moisture data of the monitoring stations in the basin, and construct an initial graph structure, wherein the stations are used as nodes, the node feature vectors contain the rainfall, flow and soil moisture, and the edge weights are calculated based on the geographical distance between the stations and the water system connectivity.

[0077] The geographic coordinates, rainfall, flow, and soil moisture data are obtained from the monitoring stations in the basin. The original monitoring data set is obtained through database query and sensor data collection. Based on the original monitoring data set, the geographic coordinates of each station are extracted, and the geographic distance between stations is calculated using the Euclidean distance formula to obtain the geographic distance matrix. The distance between stations is obtained from the geographic distance matrix. Combined with the pre-established water system connectivity topology map, if there is a water system connection between stations i and j, the connectivity factor C ij =1, otherwise C ij =0, using formula W ij =D ij (1+α(1-C ij )) Calculate the edge weights, where Wij is the edge weight and α is the adjustment parameter, and obtain the edge weight matrix. According to the original monitoring data set, the rainfall, flow and soil moisture of each station are extracted, and the feature vectors are formed after normalization to obtain the node feature matrix. Generate nodes according to the stations, combine the edge weight matrix and the node feature matrix to construct the initial graph structure, node association feature vectors, edge association edge weights, and use the adjacency matrix to represent the graph structure to obtain the initial graph structure data. For the initial graph structure data, the graph convolutional network (GCN) algorithm is used for feature extraction to obtain an enhanced node feature matrix. Extract features from the enhanced node feature matrix, combine with the edge weight matrix, and use the K-means algorithm to cluster the stations to obtain the station clustering results.

[0078] For example, the geographical coordinates, rainfall, flow and soil moisture data of monitoring sites in the basin can be collected in real time through database query combined with sensors.

[0079] For example, within a river basin, there are five monitoring stations located upstream, midstream, and downstream. Sensors record data hourly, and the database stores the station's latitude and longitude, such as Station A (116.5, 39.8) and Station B (116.7, 39.9). Rainfall, flow, and soil moisture are expressed in millimeters, cubic meters per second, and percentages, respectively. This approach ensures comprehensive and real-time data, providing a reliable foundation for subsequent analysis.

[0080] In a possible implementation, the Euclidean distance formula is used to calculate the geographical distance between sites.

[0081] For example, if we substitute the longitude and latitude of sites A and B into the formula, we get a distance of approximately 22 kilometers. After calculating the distances for all sites pairwise, we form a geographic distance matrix. This matrix intuitively reflects the spatial relationships between sites and provides a basis for subsequent weight calculations.

[0082] It should be noted that the Euclidean distance ignores terrain factors, but is simple to calculate and suitable for quickly constructing a distance matrix.

[0083] Specifically, the water system connectivity topology map can be generated through the basin hydrological map.

[0084] For example, sites A and B are located on the same tributary, and C ij =1; site C is located in another tributary, C ij = 0. Edge weight formula W ij In , α is set to 0.5, and if there is no water system connection, the weight increases to reflect the impact of hydrological isolation.

[0085] For example, W from A to B ij 22, A to C W ij is 33. This weight design balances the geographical distance and water system relationship, which helps to accurately model the characteristics of the watershed.

[0086] Preferably, the node feature matrix is generated by normalizing rainfall, flow, and soil moisture.

[0087] For example, if site A has 50 mm rainfall, 100 m³ / s flow, and 60% soil moisture, and its maximum values are 100 mm, 200 m³ / s, and 80%, respectively, the normalized feature vector is [0.5, 0.5, 0.75]. This normalization eliminates dimensional differences, enhances feature comparability, and provides a unified scale for graph structure modeling.

[0088] In one embodiment, the initial graph structure uses sites as nodes, edge weight matrices as edges, and node feature matrices as attributes.

[0089] For example, 5 sites generate a 5-node graph, and the adjacency matrix records W ij , node-associated feature vectors. This graph structure integrates spatial, hydrological, and environmental information, providing input for the graph convolutional network. The advantage of the graph structure is that it can capture both topological and feature relationships.

[0090] It can be understood that graph convolutional networks enhance node features through multi-layer feature aggregation.

[0091] For example, after the feature vector of station A is processed by GCN, it is integrated with the rainfall and flow information of neighboring stations to generate a more global feature. This enhanced feature reflects the overall hydrological dynamics of the basin and helps to discover hidden patterns.

[0092] For example, K-means clustering divides the sites into three categories based on the enhanced feature matrix.

[0093] For example, upstream stations could be grouped into one category due to high rainfall, while midstream stations could be grouped into another category due to flow characteristics. This clustering can be used for zoning management, such as optimizing flood control measures for high-rainfall areas. This clustering improves the targeted and efficient management of watersheds.

[0094] In one implementation, the overall solution forms a complete chain from data collection to clustering, centered on graph construction and GCN feature extraction. Extended solutions can incorporate time series features to enrich dynamic analysis. The integration of multiple aspects, such as distance, connectivity, and features, supports efficient watershed zoning, significantly improving water resource management and disaster prevention capabilities.

[0095] Step S102 : A message passing mechanism is used to aggregate node features of the initial graph structure, and the node feature vectors are updated through iterative calculation to capture the mutual influence of hydrological variables between sites. The hydrological variables include rainfall, flow, and soil moisture.

[0096] Through the message passing mechanism, the neighbor node information of each node is obtained from the initial graph structure, and the influence of the feature vector of the neighbor node on the current node is calculated. The formula is:

[0097]

[0098] Among them is The message vector of node i in the t+1th iteration, N is the set of neighbor nodes of node i, is the feature vector of neighbor node j at the tth iteration, W ij is the edge weight between nodes i and j, and the message vector of each node is obtained. According to the message vector of each node, the aggregation function is used to fuse the message vectors. The formula is:

[0099]

[0100] in is the aggregated feature vector of node i at the t+1th iteration; is the feature vector of node i at the tth iteration, k is the activation function, and the aggregated node feature vector is obtained.

[0101] From the aggregated node feature vector, determine whether the convergence condition is met. If so, stop the iteration and determine the node feature vector as If not, the next iteration proceeds to obtain an updated node feature vector. Based on the updated node feature vector, the feature change trend of each node is calculated. The nodes are then grouped using the K-means algorithm based on this trend. From the node grouping results, the central feature vector of each group is obtained, and the feature similarity between groups is calculated (using cosine similarity) to obtain a similarity matrix between groups. Based on the similarity matrix between groups, the interaction strength of hydrological variables between groups is determined, resulting in a grouping relationship that reflects the dynamic interaction of hydrological variables.

[0102] For example, the message passing mechanism updates feature vectors through feature interactions between nodes. In a watershed monitoring scenario, assume there are four monitoring stations within the basin, recording rainfall, flow, and soil moisture, respectively. In the initial graph structure, nodes are stations, and edge weights are based on geographic distance and water connectivity. The core of message passing is that each node collects information from its neighbors.

[0103] For example, the neighbors of station A are B and C. The rainfall at station B is 40 mm and the flow is 80 cubic meters per second. The rainfall at station C is 60 mm and the flow is 100 cubic meters per second. The message vector is generated by weighted aggregation of neighbor features, with weight W AB and W AC Reflecting distance and water system relationships. This mechanism ensures that node features incorporate surrounding environment information.

[0104] In one possible implementation, the aggregation function combines the message vector with the node's own features. For example, the initial feature vector for site A is [0.5, 0.4, 0.6]. After the message vector aggregates the features of B and C, an activation function such as ReLU is applied to generate an aggregated feature vector. The aggregation process emphasizes the influence of neighbors on the current node.

[0105] For example, the rainfall feature of A is enhanced by the high rainfall of B and C. This aggregation method makes the node features more global.

[0106] It should be noted that the convergence condition is determined based on the change in the eigenvector. Assuming the threshold ε is set to 0.01, if the difference between the aggregated eigenvectors of site A between two iterations is less than 0.01, the iteration is stopped and the final eigenvector is determined.

[0107] For example, if the feature vector of A is updated from [0.5, 0.4, 0.6] to [0.52, 0.42, 0.61], the difference satisfies the condition and the iteration is terminated. This method ensures the stability of the feature.

[0108] Specifically, the characteristic change trend is calculated by comparing the initial and final characteristic vectors. For site A, the initial characteristic vector is [0.5, 0.4, 0.6], the final characteristic vector is [0.52, 0.42, 0.61], and the trend is [0.02, 0.02, 0.01]. The trend reflects the dynamic evolution of the node in the hydrological environment.

[0109] For example, the rainfall and flow characteristics of A increased slightly, indicating that it was affected by the high rainfall in its neighbor.

[0110] Preferably, the K-means algorithm groups sites based on their characteristic trends. For example, suppose four sites are divided into two groups: upstream sites A and B are grouped together due to similar rainfall trends, while downstream sites C and D are grouped together. This grouping reflects the similarity of the sites' hydrological characteristics.

[0111] For example, high rainfall trends in A and B indicate active upstream hydrology.

[0112] In one embodiment, the similarity of the central feature vectors of the groups is calculated by cosine similarity. Assume that the central feature of group 1 is [0.03, 0.02, 0.01] and that of group 2 is [0.01, 0.01, 0.02]. The similarity S 12 A higher value indicates that the hydrological characteristics of the two groups are similar. This similarity matrix quantifies the relationship between the groups.

[0113] For example, the judgment of the interaction intensity of hydrological variables is based on the similarity threshold θ. If θ is set to 0.8, S 12 If the value exceeds the threshold, it indicates that there is a strong interaction between group 1 and group 2.

[0114] For example, high rainfall in upstream group 1 may affect flow in downstream group 2. Grouping relationships reveal interconnected hydrological dynamics within the basin. This approach, from message passing to interactive analysis, forms a complete solution. Extending this solution can introduce a temporal dimension, enriching dynamic interactive analysis.

[0115] Step S103: If the change in rainfall intensity in the node feature exceeds a preset threshold, the water system connectivity parameter is adjusted and the edge weight is recalculated to generate a dynamically adjusted graph structure.

[0116] If the change in rainfall intensity in the node feature exceeds the preset threshold, the water system connectivity parameters are adjusted according to the rainfall intensity change value, the edge weights are updated, and the dynamically adjusted graph structure is obtained. According to the dynamically adjusted graph structure, the set of neighbor nodes of each node is obtained, and the influence of the edge weights of the neighbor nodes on the current node is calculated to obtain the edge weight influence value of each node. If the edge weight influence value exceeds the preset influence threshold, the edge weight influence value is normalized by standardization. According to the normalized edge weight influence value, the nodes are grouped using a hierarchical clustering algorithm. From the node grouping results, the central edge weight influence value of each group is obtained, and the edge weight similarity between groups is calculated (calculated using cosine similarity) to obtain the similarity matrix between groups. If an element S in the similarity matrix between groups kl If the similarity exceeds the preset threshold, group k and group l are determined to have strong connectivity. The water system connectivity parameters are updated to obtain a group relationship that reflects the dynamic connectivity of the water system. Based on the group relationship that reflects the dynamic connectivity of the water system, the weighted average method is used to calculate the comprehensive connectivity strength of each group.

[0117] Specifically, in the watershed hydrological monitoring scenario, dynamically adjusting the graph structure is the key to optimizing water system connectivity analysis.

[0118] For example, changes in rainfall intensity can trigger edge weight updates. Taking a river basin as an example, assuming there are four monitoring stations A, B, C, and D, the initial edge weights are based on geographical distances, and the edge weight W between stations A and B is AB is 0.6. Rainfall intensity change ΔR A is 0.2, the adjustment coefficient α is 0.5, and the updated formula W AB '=W AB (1+αΔR A ) After calculation, W AB ' changes to 0.66. This adjustment reflects the dynamic impact of rainfall on water system connectivity and enhances the environmental adaptability of edge weights.

[0119] Specifically, the calculation of edge weight influence value focuses on the contribution of neighboring nodes. The neighbors of site A are B and C, and the updated edge weight W AB ' is 0.66, W AC ' is 0.72. Calculate the edge weight influence value E of A A = 0.66 + 0.72 = 1.38. This method quantifies the comprehensive impact of neighbors on the node's hydrological environment and provides a data basis for subsequent analysis.

[0120] It should be noted that if the edge weight influence value is too high, standardization processing ensures data comparability, facilitates cross-site comparison, and highlights connectivity differences.

[0121] In one possible implementation, hierarchical clustering is performed based on normalized edge weight influence values. For example, suppose sites A and B have edge weight influence values of 1 and 0.8, respectively, indicating a close proximity, and are grouped together. Sites C and D, with edge weight influence values of 0.4 and 0, are grouped together. This grouping reflects the similarity of sites in hydrological connectivity and facilitates identification of distinct regional characteristics within a watershed.

[0122] Preferably, the edge weight similarity between groups is calculated by the center edge weight influence value. The center C1 of group 1 is 0.9, and the center C2 of group 2 is 0.2. The cosine similarity formula is used to quantify the relationship between the two groups. If S 12 The value is 0.85, which exceeds the threshold of 0.8, indicating that groups 1 and 2 are strongly connected and that the water system parameters need to be updated. This analysis reveals the intensity of hydrological interactions within the basin.

[0123] This approach, understandably, progresses from dynamic adjustments to connectivity analysis, ensuring a comprehensive capture of the basin's hydrological characteristics. Expanding this approach could include time series analysis to analyze long-term trends in rainfall variability and further optimize water system management.

[0124] In step S104, the dynamically adjusted graph structure is input into the graph neural network prediction layer, and the rainfall, flow and soil moisture prediction values of each station are calculated through full connection layer mapping.

[0125] A graph neural network processes the dynamic graph structure, employing a message passing mechanism to update the characteristics of each site. Based on the updated site characteristics, a fully connected layer mapping is used to calculate the initial prediction value for each site. If the initial prediction value deviates from the preset threshold by more than a specified range, the weight matrix and bias parameters are optimized using a gradient descent algorithm. Based on the optimized weight matrix and bias parameters, the prediction value for each site is updated again through the fully connected layer mapping. If the updated prediction value deviates significantly from the historical data, the prediction value is adjusted through time series analysis. Based on the adjusted prediction value, a weighted average method is used to integrate the rainfall, flow, and soil moisture prediction values to obtain a comprehensive prediction value for each site. Cluster analysis is used to group the comprehensive prediction values.

[0126] Specifically, in the watershed hydrological monitoring scenario, the graph neural network updates site features through a message passing mechanism, which can effectively capture the dynamic interaction relationship between sites.

[0127] For example, the core of the message passing mechanism is to use the characteristics of neighboring sites to update the characteristics of the target site. Suppose there are sites A, B, and C in a watershed, where A's neighbors are B and C. The rainfall characteristic of site B is 10 mm, and that of site C is 15 mm. The message passing weight matrix is set based on hydrological correlation. Through message passing, the characteristics of site A are integrated with the rainfall information of sites B and C, generating a more comprehensive feature representation. This approach enhances the adaptability of site characteristics to the watershed environment and provides a reliable foundation for subsequent predictions.

[0128] Specifically, the fully connected layer maps the updated features into the initial predicted values.

[0129] In one embodiment, the updated features for station A include rainfall and flow information, which are mapped through a fully connected layer to generate a rainfall forecast for A, for example, 12 mm. This forecast reflects the conversion of station features into specific hydrological variables, facilitating subsequent analysis.

[0130] It should be noted that if the predicted value deviates greatly from the actual observed value, for example, the predicted value is 12 mm but the actual value is 15 mm, the weight matrix and bias parameters need to be optimized.

[0131] Preferably, the gradient descent algorithm reduces the prediction deviation by iteratively adjusting the weights.

[0132] For example, the adjusted weights make the predicted values closer to the actual observed values, improving the model accuracy.

[0133] In one possible implementation, the optimized weights and biases are used for remapping to update the predicted values.

[0134] For example, the rainfall forecast for station A was adjusted from 12 mm to 14 mm, which is closer to the actual value. This update ensures the dynamic adaptability of the forecast results.

[0135] It is understandable that if the forecast value deviates significantly from the historical data, time series analysis can be used for further correction.

[0136] For example, if the historical rainfall average for A is 13 mm and the adjustment factor is 0.5, the forecast value can be generated by weighted integration of historical data to produce a more stable result. This method effectively reduces the interference of short-term fluctuations.

[0137] For example, the generation of a composite prediction value is achieved by fusing multivariate prediction values through weighted averaging.

[0138] In one example, the rainfall forecast for site A is 14 mm, the flow forecast is 200 cubic meters per second, and the soil moisture forecast is 1.2 meters, with weights of 0.4, 0.3, and 0.3, respectively. The combined forecast value reflects multidimensional hydrological characteristics. This fusion improves the comprehensiveness of the forecast and facilitates watershed management.

[0139] Preferably, the cluster analysis groups the samples according to their comprehensive prediction values.

[0140] For example, sites A and B have similar composite prediction values and are grouped together, while sites C and D are grouped together. This grouping reveals the hydrological similarities among sites within the basin and facilitates targeted management.

[0141] In one possible implementation, an extended approach to time series analysis can further optimize forecasts.

[0142] For example, analyzing rainfall trends at site A over the past week and combining them with current forecasts can generate a more accurate long-term forecast. This approach captures the dynamics of hydrological variables and improves the robustness of the forecast.

[0143] It is understandable that the extended application of cluster grouping can be combined with geographical location to refine the watershed zoning management.

[0144] For example, if A and B are located upstream, and C and D are located downstream, the grouping results guide water resource allocation in different areas. This layered analysis logic, from feature updating to comprehensive prediction, fully supports watershed hydrological management.

[0145] Step S105 , performing time series analysis on the predicted values, extracting hourly hydrological dynamic transmission characteristics, and constructing a coupling model based on a dynamic Bayesian network and a spatiotemporal collaborative Kriging interpolation algorithm.

[0146] Through time series analysis, hourly dynamic features are extracted from hydrological data using a sliding window approach. Based on these hourly dynamic features, a dynamic Bayesian network is constructed, node state transition probabilities are defined, and a node state probability distribution is obtained. A spatiotemporal co-kriging interpolation algorithm is used to calculate spatial correlations based on the node state probability distribution to obtain spatial interpolation results. If the spatial interpolation result deviates from the preset threshold by more than a certain range, the weight coefficients are optimized using gradient descent to obtain optimized interpolation weights. Based on the optimized interpolation weights, the spatial interpolation result is recalculated to obtain an updated spatial interpolation result. Using a data fusion method, the node state probability distribution and the updated spatial interpolation result are combined and weighted averaged to obtain a comprehensive prediction value. If the comprehensive prediction value deviates significantly from the historical data, the model parameters are adjusted through a dynamic update mechanism to obtain the optimized model parameters.

[0147] Specifically, in the watershed hydrological monitoring scenario, time series analysis extracts hourly dynamic features through the sliding window method, which can capture the short-term fluctuation patterns of variables such as rainfall and flow.

[0148] For example, the sliding window method takes the rainfall data of the past 6 hours as input and combines it with the weight matrix to generate the feature representation of the current hour.

[0149] For example, if the rainfall at a station over the previous six hours was 8 mm, 10 mm, 12 mm, 15 mm, 14 mm, and 13 mm, respectively, weighted fusion can be used to generate dynamic features for the current hour, reflecting the rainfall trend. This method facilitates capturing rapidly changing hydrological patterns and provides a foundation for subsequent analysis.

[0150] Specifically, a dynamic Bayesian network is constructed based on hourly dynamic features to describe the dependencies between site states.

[0151] In one embodiment, assume that there are stations A and B in a watershed, and the rainfall state of A is affected by the flow state of B. The Bayesian network calculates the probability of A being in a high rainfall state based on dynamic characteristics.

[0152] For example, if the rainfall characteristic of A is 12 mm and the flow characteristic of B is 180 cubic meters per second, the network infers that the probability of high rainfall in A is 0.7. This probability distribution provides a state basis for spatial analysis.

[0153] Preferably, the spatiotemporal co-kriging interpolation algorithm uses node state probabilities to calculate the hydrological values of unmonitored points in the watershed.

[0154] For example, the observed rainfall values at sites A and B are 14 mm and 16 mm, respectively, but there is no observational data for target site C. By calculating the spatial correlation between sites A and B and C, an interpolated rainfall estimate of 15 mm is generated for site C. This method effectively fills in data gaps and improves watershed coverage.

[0155] It should be noted that if the interpolation result deviates greatly from the actual value, for example, the actual rainfall of C is 17 mm, the weight coefficient is optimized by gradient descent.

[0156] In one possible implementation, initial weights are set based on distance, which are then optimized to more accurately reflect the topographic influences between sites. After reinterpolation, the estimated value of C is adjusted to 16.5 mm, a significantly reduced deviation.

[0157] In one embodiment, the data fusion method integrates the Bayesian network prediction and interpolation results to generate a composite prediction value.

[0158] For example, the Bayesian prediction for rainfall in A is 13 mm, and the interpolated result is 14 mm. The weighting coefficients are 0.6 and 0.4, respectively, and the combined prediction value is 13.4 mm. This fusion balances the advantages of probabilistic inference and spatial estimation, enhancing the stability of the forecast.

[0159] It is understandable that if the composite prediction value deviates significantly from the historical data, for example, the historical mean is 15 mm, the dynamic update mechanism adjusts the model parameters.

[0160] For example, by weighting historical data, the forecast value was adjusted to 14.2 mm, which is closer to the long-term trend. This dynamic correction reduces the impact of short-term anomalies and improves the robustness of the model.

[0161] For example, in an extended scenario, time series analysis is combined with long-term trends to optimize forecasts.

[0162] For example, analyzing rainfall fluctuations at site A over the past 24 hours and combining them with the current composite forecast yields a more stable hourly forecast. This approach captures the cyclical changes in hydrological variables and provides a reliable basis for watershed management.

[0163] Step S106: Update the parameters of the graph neural network and optimize the message passing mechanism based on the spatial distribution characteristics of the coupling model output.

[0164] The spatial distribution characteristics are used to determine the interaction strength between nodes. Based on the interaction strength between nodes, a set of neighboring nodes is determined. The node feature vector is obtained from the neighboring node set. Based on the node feature vector, a message passing mechanism is used to calculate node messages. The loss function is calculated based on the node messages. If the loss function value is greater than a preset threshold, gradient descent is used to optimize the weight matrix and bias. Based on the optimized weight matrix and bias, the graph neural network parameters are updated to obtain the optimized graph neural network parameters.

[0165] Specifically, in the watershed hydrological monitoring scenario, the analysis of spatial distribution characteristics and the calculation of interaction intensity between nodes are the core of building a hydrological prediction model.

[0166] For example, spatial distribution characteristics can be understood as the geographical location of each monitoring station within the basin, the terrain height, or the distribution pattern of hydrological variables such as rainfall and discharge. By analyzing these characteristics, the mutual influence between stations can be quantified.

[0167] For example, site A is located on a hillside where rainfall easily forms runoff, while site B is located downstream and the flow is affected by site A. The interaction intensity calculation reflects the degree of association between sites through spatial distance and terrain differences.

[0168] In one possible implementation, assuming that the distance between A and B is 5 kilometers and the terrain height difference is 200 meters, the interaction intensity is weighted based on the distance and height difference to generate a numerical value representing the hydrological impact of A on B.

[0169] Specifically, the neighbor node set is determined based on interaction strength screening.

[0170] Preferably, a threshold is set to filter out weakly connected nodes.

[0171] For example, there are sites A, B, C, and D in the basin. If the interaction strength between A and B and C is higher than the threshold, and D is lower than the threshold, then B and C are neighbors of A. This screening ensures that only strongly connected nodes are considered, reducing computational redundancy.

[0172] In one embodiment, A's neighbor set includes B and C, which provide data of rainfall of 12 mm and flow of 150 cubic meters per second, respectively, for subsequent feature extraction. The acquisition of node feature vectors is the integration of spatial distribution data and neighbor information.

[0173] It can be understood that the feature vector is not only the hydrological data of a single station, but also incorporates the dynamics of neighboring stations.

[0174] For example, if rainfall at station A is 10 mm, the flow at neighbor B is 160 cubic meters per second, and rainfall at station C is 13 mm, weighted fusion generates a feature vector for station A, reflecting its hydrological status within the watershed network. The message passing mechanism further utilizes feature vectors to transmit dynamic information between stations.

[0175] In one possible implementation, the traffic characteristics of site B are converted into messages through a weight matrix and transmitted to A, reflecting the impact of B's traffic changes on A's rainfall forecast.

[0176] For example, an increase in flow at B may indicate increased rainfall upstream, and A adjusts its forecast after receiving this information. The loss function is used to evaluate the deviation between the model prediction and the actual value.

[0177] For example, the predicted rainfall for A is 11 mm, but the actual rainfall is 12 mm. The loss function quantifies the error and guides parameter optimization.

[0178] In one embodiment, a loss value above a threshold triggers gradient descent to optimize the weight matrix and bias.

[0179] For example, initial weights are set based on distance, and after optimization, they more accurately reflect topographic and hydrological correlations. After optimizing the weight matrix and bias, the graph neural network parameters are updated to improve prediction accuracy.

[0180] Specifically, the updated parameters better capture the spatial dependencies among sites.

[0181] For example, the predicted rainfall for Area A was adjusted from 11 mm to 11.8 mm, which is closer to the actual value. This dynamic adjustment enhances the model's adaptability to complex watershed environments.

[0182] For example, the extended solution optimizes forecasts by incorporating long-term spatial trends. It analyzes spatial fluctuations in rainfall over the past 24 hours in A and integrates them with current messaging results to generate more stable hourly forecasts. This approach effectively balances short-term fluctuations with long-term trends, providing a reliable basis for watershed hydrological management.

[0183] Step S107: If the deviation of the predicted value output by the optimized model is lower than a preset threshold, the final hydrological variable prediction result is output; otherwise, the dynamic adjustment and parameter optimization process is repeated.

[0184] Hydrological variable characteristics are obtained from input data, and a feature extraction function is used to calculate the hydrological variable feature vector. Based on the hydrological variable feature vector, a machine learning model is used to generate an initial prediction result. The prediction value deviation is calculated using the initial prediction result and the actual value data. If the prediction value deviation is lower than the preset threshold, the prediction result is output to obtain the final hydrological variable prediction result. If the deviation is higher than the preset threshold, dynamic adjustment is performed. Based on the prediction value deviation, the model structure is dynamically adjusted, and the feature weights are redistributed using the structure update function to obtain the adjusted model structure. Using the adjusted model structure, the parameters are optimized using the gradient descent algorithm. Based on the optimized parameters, the prediction result is regenerated and calculated using the updated model to obtain a new prediction result. Based on the new prediction result, the deviation calculation and threshold judgment are repeated, and the cycle is repeated until the deviation falls below the preset threshold, at which point the final hydrological variable prediction result is output.

[0185] Specifically, in the watershed hydrological prediction scenario, the extraction of hydrological variable characteristics is the basis for building a prediction model.

[0186] For example, the hydrological variables include rainfall, flow, evaporation, etc. The function of the feature extraction function is to convert these raw data into feature vectors that can reflect the hydrological status of the site.

[0187] Specifically, a monitoring station receives 15 mm of rainfall and 200 cubic meters per second of flow. The feature extraction function weights and integrates this data to generate a multidimensional vector containing the station's hydrological characteristics. This method effectively captures the station's dynamic characteristics within the watershed, providing reliable input for subsequent forecasts. Generating initial forecasts using a machine learning model is a key step.

[0188] In one possible implementation, the model uses the feature vectors to calculate the predicted values through a weight matrix and a bias.

[0189] For example, the feature vector of a station contains rainfall and flow information. The model predicts that the rainfall in the next hour will be 12 mm based on the weight matrix trained with historical data.

[0190] It should be noted that the initial values of the weight matrix are based on the basin's topography and hydrological context, while the bias is used to correct for model deviations. This approach ensures that the predictions are closely aligned with the actual hydrological environment. The calculation of the deviations from the predicted values is used to evaluate model performance.

[0191] Preferably, the deviation calculation function quantifies the error by comparing the predicted value with the true value.

[0192] For example, if the predicted rainfall is 12 mm and the actual rainfall is 13 mm, the deviation calculation function generates an error value for subsequent optimization. This deviation assessment can intuitively reflect the accuracy of the model prediction and provide a basis for dynamic adjustment. When the predicted value deviation exceeds the preset threshold, dynamic adjustment of the model structure becomes necessary.

[0193] In one embodiment, the structure update function redistributes feature weights based on the deviations.

[0194] For example, if the flow characteristics of a particular station have a significant impact on the forecast, the update function will increase its weight. This adjustment can enhance the model's sensitivity to key hydrological variables and improve the adaptability of the forecast. A gradient descent algorithm is used to optimize the model parameters.

[0195] It can be understood that the optimization process gradually adjusts the parameters by calculating the gradient of the bias with respect to the weights and biases.

[0196] For example, the initial weight matrix is set based on distance and topographic differences, and after optimization, it more accurately reflects the hydrological connections between stations. This approach ensures that the model parameters can dynamically adapt to changes in the watershed environment. Regenerating the prediction results after optimization is a key step.

[0197] For example, the optimized model predicted rainfall at a certain site, adjusted from 12 mm to 12.8 mm, closer to the actual value. This iterative prediction method continuously approximates the actual hydrological conditions. The deviation calculation and threshold determination are repeated repeatedly until the deviation falls below the threshold, ultimately outputting the prediction result.

[0198] For example, the final forecast results indicate that a station will receive 12.5 mm of rainfall in the next hour, providing reliable information for hydrological management. This cyclical optimization mechanism effectively improves the model's forecast accuracy. In the expanded solution, the forecast is further optimized by incorporating long-term hydrological trends.

[0199] In one implementation, rainfall fluctuations at a site over the past 48 hours are analyzed and integrated with current feature vectors to generate more stable forecasts. This approach balances short-term dynamics with long-term patterns, providing more comprehensive support for basin water resource scheduling.

[0200] Although the present invention has been particularly shown and described in conjunction with preferred embodiments, it will be understood by those skilled in the art that various changes in form and details may be made to the present invention without departing from the spirit and scope of the invention as defined in the appended claims, and all such changes are within the scope of protection of the present invention.

Claims

1. A hydrological trend prediction method based on big data analysis, characterized in that: The method comprises: Obtaining geographic coordinates, rainfall, flow, and soil moisture data for monitoring stations within the basin, and constructing an initial graph structure with the stations as nodes. The node feature vectors include the rainfall, flow, and soil moisture, and edge weights are calculated based on the geographic distance between stations and water system connectivity. A message passing mechanism is used to aggregate node features of the initial graph structure, and feature vectors of the nodes are updated through iterative calculation to capture the mutual influence of hydrological variables between sites, including rainfall, flow, and soil moisture. If the change in rainfall in the node feature exceeds a preset threshold, the water system connectivity parameter is adjusted and the edge weight is recalculated to generate a dynamically adjusted graph structure; The dynamically adjusted graph structure is input into the graph neural network prediction layer, and the rainfall, flow and soil moisture prediction values of each station are calculated through the fully connected layer mapping; Performing time series analysis on the predicted values, extracting hourly hydrological dynamic transmission characteristics, and constructing a coupling model based on a dynamic Bayesian network and a spatiotemporal co-kriging interpolation algorithm; According to the spatial distribution characteristics of the coupling model output, the parameters of the graph neural network are updated and the message passing mechanism is optimized; If the prediction value deviation of the optimized model output is lower than the preset threshold, the final hydrological variable prediction result is output; otherwise, the dynamic adjustment and parameter optimization process is repeated.

2. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: The edge weight is calculated as follows: W ij =D ij (1+α(1-C ij )), Where W ij is the edge weight between site i and site j; D ij is the geographical distance between site i and site j; α is the adjustment parameter; C ij is the connectivity factor; if there is a water system connection between site i and site j, then C ij =1, otherwise C ij =0.

3. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: The message passing mechanism is used to aggregate node features of the initial graph structure, and the node feature vectors are updated through iterative calculation to capture the mutual influence of hydrological variables between sites. The hydrological variables include rainfall, flow and soil moisture, including: Through the message passing mechanism, the neighbor node information of each node is obtained from the initial graph structure, and the influence of the neighbor node's feature vector on the current node is calculated; According to the message vector of each node, the aggregation function is used to fuse the message vectors; From the aggregated node feature vector, determine whether the convergence condition is met. If so, stop the iteration and determine the node feature vector; if not, continue the next round of iteration to obtain the updated node feature vector; According to the updated node feature vector, calculate the feature change trend of each node; The K-means algorithm is used to group nodes based on feature change trends; From the node grouping results, obtain the central feature vector of each group, calculate the feature similarity between groups, and obtain the similarity matrix between groups; Based on the similarity matrix between groups, the interaction intensity of hydrological variables between groups is determined, and the group relationship reflecting the dynamic interaction of hydrological variables is obtained.

4. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: If the rainfall intensity in the node feature changes by more than a preset threshold, the water system connectivity parameter is adjusted and the edge weight is recalculated to generate a dynamically adjusted graph structure, including: If the rainfall intensity change in the node feature exceeds the preset threshold, the water system connectivity parameters are adjusted and the edge weights are updated according to the rainfall intensity change value; According to the dynamically adjusted graph structure, the neighbor node set of each node is obtained, the influence of the edge weight of the neighbor node on the current node is calculated, and the edge weight influence value of each node is obtained; If the edge weight influence value exceeds the preset influence threshold, the edge weight influence value is normalized using standardization to obtain the normalized edge weight influence value; According to the normalized edge weight influence value, the hierarchical clustering algorithm is used to group the nodes to obtain the node grouping results; From the node grouping results, obtain the center edge weight influence value of each group, calculate the edge weight similarity between groups, and obtain the similarity matrix between groups; If an element in the similarity matrix between groups exceeds the preset similarity threshold, it is determined that the groups have strong connectivity, and the water system connectivity parameters are updated to obtain the group relationship reflecting the dynamic connectivity of the water system; According to the grouping relationship reflecting the dynamic connectivity of the water system, the weighted average method is used to calculate the comprehensive connectivity strength of each group to obtain the comprehensive connectivity strength of each group.

5. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: The dynamically adjusted graph structure is input into the graph neural network prediction layer, and the rainfall, flow and soil moisture prediction values of each station are calculated through the fully connected layer mapping, including: The dynamic graph structure is processed by graph neural network, and the message passing mechanism is adopted to update the characteristics of each site to obtain the updated site characteristics; Based on the updated site features, the fully connected layer mapping is used to calculate the initial prediction value of each site; If the deviation between the initial prediction value and the preset threshold exceeds the specified range, the weight matrix and bias parameters are optimized by the gradient descent algorithm to obtain the optimized weight matrix and bias parameters; Based on the optimized weight matrix and bias parameters, re-map through the fully connected layer to update the predicted value of each site; If there is a significant deviation between the updated forecast value and the historical data, the forecast value will be adjusted through time series analysis; Based on the adjusted forecast values, the weighted average method is used to integrate the rainfall, flow and soil moisture forecast values to obtain the comprehensive forecast value of each station; Through cluster analysis, the comprehensive prediction values are grouped to obtain the grouping results of the site prediction values.

6. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: The time series analysis of the predicted values is performed to extract hourly hydrological dynamic transmission characteristics, and a coupling model based on a dynamic Bayesian network and a spatiotemporal collaborative Kriging interpolation algorithm is constructed, including: Through time series analysis, hourly dynamic features are extracted from hydrological data and the sliding window method is used to obtain hourly dynamic features; Based on hourly dynamic features, a dynamic Bayesian network is constructed to define the node state transition probability and obtain the node state probability distribution; The spatiotemporal collaborative kriging interpolation algorithm is used to calculate spatial correlation based on the node state probability distribution to obtain spatial interpolation results. If the deviation between the spatial interpolation result and the preset threshold exceeds the specified range, the weight coefficient is optimized by gradient descent; Recalculate the spatial interpolation results based on the optimized interpolation weights; Through data fusion method, the node state probability distribution and the updated spatial interpolation results are combined to obtain the comprehensive prediction value; If there is a significant deviation between the comprehensive forecast value and the historical data, the model parameters will be adjusted through a dynamic update mechanism.

7. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: The updating of the parameters of the graph neural network and the optimization of the message passing mechanism according to the spatial distribution characteristics of the coupling model output include: Obtain the interaction strength between nodes through spatial distribution characteristics; Determine the set of neighbor nodes based on the interaction strength between nodes; Obtain node feature vector through neighbor node set; According to the node feature vector, the node message is calculated using the message passing mechanism; Calculate the loss function through node messages; If the loss function value is greater than the preset threshold, the weight matrix and bias are optimized using gradient descent; Update the graph neural network parameters based on the optimized weight matrix and bias.

8. The hydrological trend prediction method based on big data analysis according to claim 1 is characterized in that: If the deviation of the predicted value output by the optimized model is lower than the preset threshold, the final hydrological variable prediction result is output; otherwise, the dynamic adjustment and parameter optimization process is repeated, including: The hydrological variable characteristics are obtained by inputting data, and the characteristic vector is calculated using the feature extraction function; Based on the hydrological variable feature vector, a machine learning model is used to generate initial prediction results; Calculate the prediction value deviation based on the initial prediction results and the true value data; If the predicted value deviation is lower than the preset threshold, the prediction result is output; If the deviation is higher than the preset threshold, dynamic adjustment is performed; According to the deviation of the predicted value, the model structure is dynamically adjusted and the feature weights are redistributed using the structure update function; Through the adjusted model structure, the gradient descent algorithm is used to optimize the parameters; Regenerate the prediction results based on the optimized parameters; Based on the new prediction results, the deviation calculation and threshold judgment are repeated, and the cycle is executed until the deviation is lower than the preset threshold, and the final hydrological variable prediction results are output.

Images