Intelligent prediction system and method for groundwater storage change based on spatiotemporal sequence analysis

Abstract

The application provides a groundwater storage change intelligent prediction system and method based on space-time sequence analysis, relates to the technical field of groundwater storage prediction, and comprises the following steps: a data acquisition module is used for collecting and analyzing the data of the groundwater storage of a target region; the groundwater storage change data set is subjected to time sequence analysis and space analysis; a space-time mapping module is used for mapping the groundwater storage change trend characteristics and the groundwater storage change sensitive factors; a change prediction module is used for modeling according to the space-time law map, synchronizing the groundwater storage change data set to a deep prediction model for prediction, and generating groundwater storage change prediction data. Through the application, the technical problem of low accuracy of groundwater storage change prediction caused by the difficulty in simultaneously capturing the time dynamics and spatial distribution characteristics of groundwater storage change in the prior art can be solved, and the accuracy of groundwater storage change prediction can be improved by integrating time trends and space sensitive factors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of groundwater reserves prediction, and particularly relates to an intelligent groundwater reserves change prediction system and method based on spatiotemporal sequence analysis. BACKGROUND

[0002] Groundwater reserves refer to the total amount of water present in underground aquifers, which is usually determined by multiple factors such as precipitation, evaporation, groundwater flow, and extraction. The change in groundwater reserves can be divided into short-term fluctuations and long-term trends. Short-term fluctuations are usually caused by precipitation, evaporation, and human activities such as pumping, irrigation, etc. While long-term trends may be influenced by factors such as climate change, regional hydrogeological characteristics, and land use changes. Groundwater reserves change is a typical spatiotemporal coupling problem, which is affected by both seasonal changes and groundwater flow rules and regional differences. Existing prediction methods mostly focus on one of the time or spatial dimensions, and it is difficult to capture the interaction and complex relationship between the two. In the spatial scale, groundwater reserves change usually has significant spatial heterogeneity, and the hydrogeological conditions of different regions differ greatly, making it difficult for the model to generalize in multiple regions. In the time scale, groundwater reserves change is often a long-term process, while many prediction methods tend to be short-term, making it difficult to capture long-term trends. Relying solely on time series or spatial characteristics modeling cannot fully understand and predict the comprehensive dynamic behavior of the groundwater system, resulting in reduced prediction accuracy.

[0003] In summary, the existing technology has the technical problem of low accuracy in predicting groundwater reserves change due to the difficulty in simultaneously capturing the temporal dynamics and spatial distribution characteristics of groundwater reserves change. SUMMARY

[0004] The purpose of the present application is to provide an intelligent groundwater reserves change prediction system and method based on spatiotemporal sequence analysis, to solve the technical problem of low accuracy in predicting groundwater reserves change due to the difficulty in simultaneously capturing the temporal dynamics and spatial distribution characteristics of groundwater reserves change in the existing technology.

[0005] In view of the above problems, the present application provides an intelligent groundwater reserves change prediction system and method based on spatiotemporal sequence analysis.

[0006] In a first aspect, the application provides an intelligent groundwater storage change prediction system based on spatio-temporal sequence analysis, wherein the intelligent groundwater storage change prediction system based on spatio-temporal sequence analysis comprises: a data acquisition module, the data acquisition module is used for data acquisition and analysis of groundwater storage of a target region through multiple data sources, and obtains a groundwater storage change dataset; a time series analysis module, the time series analysis module is used for time series analysis based on the groundwater storage change dataset, and generates groundwater storage change trend characteristics; a spatial analysis module, the spatial analysis module is used for spatial analysis based on the groundwater storage change dataset, and generates groundwater storage change sensitive factors; a spatio-temporal mapping module, the spatio-temporal mapping module is used for spatio-temporal mapping of the groundwater storage change trend characteristics and the groundwater storage change sensitive factors, and constructs a spatio-temporal law map; a change prediction module, the change prediction module is used for modeling according to the spatio-temporal law map, constructing a deep prediction model, synchronizing the groundwater storage change dataset to the deep prediction model for prediction, and generating groundwater storage change prediction data.

[0007] In a second aspect, the application further provides an intelligent groundwater storage change prediction method based on spatio-temporal sequence analysis, wherein the intelligent groundwater storage change prediction method based on spatio-temporal sequence analysis comprises: data acquisition and analysis of groundwater storage of a target region through multiple data sources, and obtaining a groundwater storage change dataset; time series analysis based on the groundwater storage change dataset, and generating groundwater storage change trend characteristics; spatial analysis based on the groundwater storage change dataset, and generating groundwater storage change sensitive factors; spatio-temporal mapping of the groundwater storage change trend characteristics and the groundwater storage change sensitive factors, and constructing a spatio-temporal law map; modeling according to the spatio-temporal law map, constructing a deep prediction model, synchronizing the groundwater storage change dataset to the deep prediction model for prediction, and generating groundwater storage change prediction data.

[0008] One or more technical solutions provided in the application have at least the following technical effects or advantages:

[0009] The data acquisition module is used for data acquisition and analysis on the groundwater reserves of a target region through multiple data sources, to obtain a groundwater reserve change data set; the time series analysis module is used for time series analysis based on the groundwater reserve change data set, to generate a groundwater reserve change trend feature; the spatial analysis module is used for spatial analysis based on the groundwater reserve change data set, to generate a groundwater reserve change sensitive factor; the space-time mapping module is used for space-time mapping of the groundwater reserve change trend feature and the groundwater reserve change sensitive factor, to construct a space-time law map; and the change prediction module is used for modeling according to the space-time law map, to construct a deep prediction model, to synchronize the groundwater reserve change data set to the deep prediction model for prediction, and to generate groundwater reserve change prediction data. That is, through the acquisition of multiple data, time series analysis and spatial analysis are performed on the groundwater reserve change data, the time trend feature of the groundwater reserve change is mapped with the spatial sensitive factor, a space-time law map reflecting the groundwater reserve change law of the target region is constructed, and a deep prediction model is constructed accordingly, to predict the groundwater reserve change and improve the accuracy of the prediction of the groundwater reserve change.

[0010] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented in accordance with the content of the description, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only exemplary, and other drawings can be obtained by the provided drawings without creative labor for those skilled in the art.

[0012] Figure 1 Structure diagram of the groundwater reserve change intelligent prediction system based on space-time sequence analysis of the present application;

[0013] Figure 2 Flowchart of the groundwater reserve change intelligent prediction method based on space-time sequence analysis of the present application.

[0014] Explanation of reference signs: data acquisition module 11, time series analysis module 12, spatial analysis module 13, time-space mapping module 14, change prediction module 15. DETAILED DESCRIPTION

[0015] The present application provides an intelligent groundwater reserve change prediction system and method based on time-space sequence analysis, which solves the technical problem of low accuracy of groundwater reserve change prediction in the prior art due to the difficulty in simultaneously capturing the time dynamics and spatial distribution characteristics of groundwater reserve change. By collecting multi-party data, the time series analysis and spatial analysis of the groundwater reserve change data are performed, the time trend characteristics of the groundwater reserve change are time-space mapped with the spatial sensitive factors, the time-space rule map reflecting the change rule of the groundwater reserve of the target region is constructed, and a deep prediction model is constructed accordingly to predict the change of the groundwater reserve, thereby improving the accuracy of the prediction of the change of the groundwater reserve.

[0016] Hereinafter, the technical solutions in the present application will be described clearly and completely with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. It should be understood that the present application is not limited by the example embodiments described herein. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application. In addition, it should be noted that, for the convenience of description, only parts related to the present application are shown in the drawings, not all.

[0017] Embodiment one, please refer to the accompanying drawings Figure 1 The present application provides an intelligent groundwater reserve change prediction system based on time-space sequence analysis, wherein the intelligent groundwater reserve change prediction system based on time-space sequence analysis is used to implement the steps of the intelligent groundwater reserve change prediction method based on time-space sequence analysis, and the intelligent groundwater reserve change prediction system based on time-space sequence analysis comprises:

[0018] The data acquisition module 11 is used to collect and analyze the data of the groundwater reserve of the target region through multi-party data sources, and obtain a groundwater reserve change data set.

[0019] Specifically, the groundwater reserves of the target area are obtained through various data collection methods and technical means. Multiple data sources usually include a remote sensing device group, a ground sensing device group, and other information sources, which provide different dimensions of data about changes in groundwater reserves. The remote sensing device group and the sensing device group collect the groundwater reserves of the target area respectively to obtain multi-source remote sensing data sets and ground observation data sets. The multi-source remote sensing data sets and the ground observation data sets from the remote sensing device group and the sensing device group are integrated for learning, and multiple decision trees are trained through integrated learning. Each decision tree learns the relationship between changes in groundwater reserves and other factors such as meteorological data, vegetation coverage, and soil moisture. By combining the results of multiple decision trees, integrated learning can greatly improve the prediction accuracy of the model.

[0020] The first resolution of the target data set is set according to actual needs, and the low-resolution remote sensing data is combined with the ground observation data through the downscaling inversion model to generate higher spatial resolution groundwater reserves change prediction data, and the groundwater reserves change data set of the target area according to the first resolution is obtained. The specific process is described in detail in the subsequent corresponding refinement steps, and will not be described here. By combining remote sensing data with ground observation data, the model can more comprehensively capture the spatial and temporal dynamics of changes in groundwater reserves, making the prediction results more reliable and comprehensive. Integrated learning can integrate information from multiple data sources, avoiding bias from a single data source. The model can adapt to data from different regions and different time periods, enhancing the generalization ability of the prediction results.

[0021] The time series analysis module 12 is configured to perform time series analysis based on the groundwater reserves change data set to generate groundwater reserves change trend features.

[0022] Specifically, the time series analysis is performed on the groundwater storage change dataset to identify the regularity, trend, seasonality, periodicity and other characteristics in the data. According to the time variation of the groundwater storage, the groundwater storage change dataset is divided in chronological order to construct a time series dataset. The long short-term memory network is used to train the time series dataset to capture the long-term and short-term trends of the groundwater storage change, and to observe the overall change pattern and short-term fluctuation state of the groundwater storage over time. The long-term trend data is subjected to periodic analysis to identify whether there is periodic change in the data, such as fixed periodic fluctuations in the groundwater storage between years. The short-term trend data is subjected to fluctuation analysis to calculate the rate of change of the groundwater storage, such as the growth rate or reduction rate, and the rate of change of these rates. The obtained periodic characteristics and rate characteristics of the groundwater storage change are added to the groundwater storage change trend characteristics to form a more comprehensive feature set, and the groundwater storage change trend characteristics are obtained. Through time series analysis, the trend of the groundwater storage over time is revealed, which helps to identify the long-term water level change trend, and provides a more accurate basis for long-term and short-term prediction of the groundwater storage by identifying long-term trends and periodic fluctuations, thereby improving the accuracy of the prediction.

[0023] The spatial analysis module 13 is used for spatial analysis based on the groundwater storage change dataset to generate groundwater storage change sensitive factors.

[0024] Specifically, the spatial analysis is performed on the groundwater storage change dataset. First, the groundwater storage change dataset is divided into a plurality of equal-sized grid cells according to the geographical information of the target area. Each grid cell represents a basic unit of spatial analysis. Each grid cell is traversed, and the relationship between the groundwater storage change and other variables (such as geological structure, land use type, hydrological characteristics, etc.) within each grid cell is analyzed by combining the groundwater storage change dataset, i.e. the correlation coefficient between the groundwater storage change and these variables within each grid cell is calculated to obtain a plurality of correlation coefficients. Using the geographic weighted regression method, the correlation coefficients generated above are combined to perform weighted regression analysis on each grid cell.

[0025] The spatial influence weight values of each grid unit are obtained through the geographic weighted regression analysis, and the weight values reflect the influence degree of different variables on the change of the groundwater reserves. The obtained multiple spatial influence weight values are arranged in descending order to generate a spatial influence weight sequence. Based on the spatial influence weight sequence, the target region is identified to generate multiple identified regions. According to the spatial weight distribution data generated according to the identified regions, the sensitivity analysis is performed on the groundwater reserves change data set to evaluate the influence degree of different geographic information on the change of the groundwater reserves, and finally the groundwater reserves change sensitive factor is generated. That is, through the sensitivity analysis, the variables with the greatest influence on the change of the groundwater reserves are identified, and these variables are the groundwater reserves change sensitive factor. Through the geographic weighted regression, the accuracy of the spatial analysis is improved, and the analysis result is more in line with the actual situation.

[0026] The spatio-temporal mapping module 14 is configured to perform spatio-temporal mapping of the groundwater reserves change trend feature and the groundwater reserves change sensitive factor, and construct a spatio-temporal law map.

[0027] Specifically, the spatio-temporal mapping relationship between the groundwater reserves change trend feature and the groundwater reserves change sensitive factor is constructed, and the groundwater change law of different positions and time periods is presented in the form of a visual map. First, the groundwater reserves change trend feature and the groundwater reserves change sensitive factor are aligned in time to ensure that they match at the same time scale. Then, the groundwater reserves change trend feature and the groundwater reserves change sensitive factor are aligned in space, and the spatial region grid corresponding to the groundwater reserves change sensitive factor is matched with the geographic position of the groundwater reserves change trend feature to ensure that the groundwater reserves data at the same geographic position and its related sensitive factors (such as meteorological data, land use, etc.) can be matched. According to the first alignment parameter obtained by time alignment and the second alignment parameter obtained by space alignment, the groundwater reserves change trend feature and the sensitive factor are spatio-temporally aligned to construct a spatio-temporal mapping relationship.

[0028] According to the spatio-temporal mapping relationship, the target region is divided into a spatio-temporal grid, and each grid unit contains data of a specific time period and spatial position. Based on the spatio-temporal grid data, dynamic mapping calculation is performed to obtain the dynamic mapping value of each grid unit at different time points, which reflects the spatio-temporal law of the change of the groundwater reserves. According to the result of the spatio-temporal mapping model, the groundwater reserves change map of different time periods and spatial positions is generated, which clearly shows the change trend of the groundwater reserves in different regions at different time periods. The dynamic mapping value is synchronized to the spatio-temporal law map to visually display the groundwater change law of different positions and time periods. Through the spatio-temporal law map, the hot regions of the change of the groundwater reserves can be accurately identified, especially the regions where the groundwater resources are short, more accurate prediction of the groundwater reserves is provided, uncertainty is reduced, and the reliability of the prediction is improved.

[0029] a change prediction module 15 for modeling according to the spatiotemporal regularity map, constructing a deep prediction model, synchronizing the groundwater storage change dataset to the deep prediction model for prediction, and generating groundwater storage change prediction data.

[0030] Specifically, the spatiotemporal regularity map is modeled and a deep prediction model is constructed. The spatiotemporal regularity map is analyzed in time according to dynamic mapping values, and the features of the time series are extracted. Each spatial grid cell of the spatiotemporal regularity map is traversed, and the dynamic mapping values of each grid cell are analyzed to extract spatial features. The time feature set and the spatial feature set are normalized to eliminate the dimensional influence between different features and ensure that the data is on the same scale. According to the normalization result, a plurality of training data and a plurality of validation data sets are constructed to ensure the representativeness of the data sets. Generally, the proportion of training data is higher than that of validation data, which can be 7:3 or 6:4, depending on the degree of influence of time and space. Gradient descent algorithm is used to learn the plurality of training data to generate initial training parameters. The initial training parameters are evaluated according to the plurality of validation data to generate a training score. If the training score is less than the expected score, the training data is adjusted, which can include feature selection, data enhancement, parameter tuning, etc. The training process is repeated iteratively until the training score is greater than or equal to the expected score. When the training score meets the expectation, the iteration is stopped, and the model at this time is the deep prediction model.

[0031] The collected groundwater storage change dataset is input into the deep prediction model for training, so that the model can learn the change regularity in the data. The historical groundwater storage data, meteorological factors, etc. are combined with the spatiotemporal regularity map, and through preprocessing and standardization steps, the data format is converted into an input format that can be accepted by the deep learning model. The input data is trained in the deep prediction model, and the model parameters are optimized during the training process to ensure that the model can accurately capture the spatiotemporal change regularity in the data. Through the trained deep prediction model, the groundwater storage change in the future period is predicted, and the prediction result includes the numerical value and change trend of the groundwater storage in the future period, and the prediction result is usually presented in the form of time series. The groundwater storage change prediction data is the future groundwater storage change data generated by the prediction model, which is used to predict the dynamic change of groundwater in the future period. Through the combination of spatiotemporal regularity map and deep learning model, the change trend of future groundwater storage can be accurately predicted, not only considering the time series data, but also comprehensively reflecting the spatial features of groundwater storage change.

[0032] Further, the data collection module 11 in the groundwater storage change intelligent prediction system based on spatiotemporal sequence analysis is also used for:

[0033] The remote sensing data collection unit is configured to collect data on the groundwater reserves of the target area based on a remote sensing device group to obtain a multi-source remote sensing data set. The sensing data collection unit is configured to collect sensing data on the groundwater reserves of the target area based on a sensing device group to obtain a ground observation data set. The integrated learning unit is configured to perform integrated learning on the multi-source remote sensing data set and the ground observation data set to construct a downscaling inversion model. The change analysis unit is configured to set a first resolution and perform change analysis according to the first resolution by using the downscaling inversion model to obtain a groundwater reserve change data set. The groundwater reserve change data set has a corresponding relationship with the first resolution.

[0034] Specifically, the remote sensing device group (such as satellites, aerial cameras, radars, etc.) is used to collect data on the groundwater reserves of the target area to obtain a multi-source remote sensing data set. The multi-source remote sensing data set refers to data obtained from multiple different types of remote sensing devices or sensors, including data of different resolutions, different wavebands, or different platforms (such as satellite remote sensing, aerial remote sensing, etc.), such as satellite images, thermal infrared images, radar data, etc. Remote sensing devices are devices that use electromagnetic waves (such as optical, infrared, microwave, etc.) to sense ground objects and collect data. The remote sensing device group refers to multiple remote sensing devices working together to collect groundwater-related data in a large area, including satellite remote sensing and unmanned aerial vehicle remote sensing. For example, the GRACE (Gravity Recovery and Climate Experiment) satellite measures the changes in the Earth's gravitational field to indirectly calculate the changes in groundwater reserves.

[0035] The sensing device group is used to monitor the groundwater reserves of the target area in real time to obtain a ground observation data set, including air temperature, precipitation, humidity, etc. The sensing device group can be a visual sensor or other sensors combined with meteorological data or hydrogeological data, etc. for collection, and is usually installed at different locations in the target area to supplement the deficiencies of remote sensing data.

[0036] The multi-source data collected by the remote sensing device group and the ground observation data collected by the sensing device group are integrated and learned, and a plurality of weak learners (such as decision trees) are trained through gradient boosting decision trees to gradually improve the prediction accuracy of the model for the change of groundwater reserves. Ensemble learning is a machine learning method that combines multiple learners (i.e., multiple models) to improve prediction performance. Essentially, it combines multiple weak models to form a stronger model to improve accuracy and robustness. The downscaling inversion model is used to convert low-resolution remote sensing data into high-resolution data, and the existing data is used to calculate more detailed data. It is usually used to improve the spatial resolution of coarse resolution remote sensing data to a higher spatial resolution for more detailed analysis. The downscaling inversion model captures the spatial characteristics of the change in groundwater reserves through integrated analysis of remote sensing data and ground observation data.

[0037] The first resolution is set, for example, the first resolution is usually selected as 1 kilometer, which means that the groundwater reserve change data output by the model will be at a spatial scale of 1 kilometer. This is because in water resource management and groundwater monitoring, a spatial scale of 1 kilometer can provide sufficient precision and ensure data processing efficiency. Through the downscaling inversion model, the change in groundwater reserves in the target area is analyzed according to the first resolution, that is, the change in groundwater reserves in the entire target area is analyzed according to a resolution of 1 kilometer, and a groundwater reserve change data set corresponding to the first resolution is output, indicating the change in groundwater reserves at different geographic locations. During the entire downscaling inversion process, the generated groundwater reserve change data set maintains a strict correspondence with the set first resolution, meaning that each grid has a specific groundwater reserve change value, and these values are closely related to the spatial characteristics of the grid.

[0038] Through integrated learning, the model can fully utilize the information from various data sources, reduce errors caused by a single data source, and significantly improve the accuracy of groundwater reserve change prediction. By setting the first resolution and performing change analysis, the model can adapt to different data resolution requirements and provide multi-level groundwater reserve prediction from a macro to a micro perspective, enhancing the adaptability and accuracy of the model at different spatial scales.

[0039] Further, the data acquisition module 11 in the intelligent prediction system for the change in groundwater reserves based on spatio-temporal sequence analysis is further used for:

[0040] The meteorological factor extraction subunit is configured to extract spatial features of a target region based on the multi-source remote sensing dataset and extract meteorological factors of the target region based on the ground observation dataset. The decision tree construction subunit is configured to map and match the spatial features and the meteorological factors, generate a mapping connection network, and construct a gradient boosting decision tree according to the mapping connection network. The cross-validation subunit is configured to perform cross-validation at a coarse resolution based on the gradient boosting decision tree, prune the gradient boosting decision tree according to a verification result, and construct the downscaling inversion model.

[0041] Specifically, spatial features of a target region are extracted from a multi-source remote sensing dataset, and terrain features (such as elevation, slope, etc.), vegetation coverage, land use types, etc. are extracted from remote sensing images. Meteorological factors such as precipitation, air temperature, humidity, and evaporation are extracted from ground observation datasets. Meteorological factors refer to meteorological factors that affect changes in groundwater storage, such as air temperature, precipitation, humidity, and evaporation. These factors have a direct impact on groundwater recharge and consumption. The spatial features and the meteorological factors are mapped and matched to generate a mapping connection network.

[0042] Mapping and matching refers to establishing a mathematical or statistical relationship between spatial features and meteorological factors, such as correlation analysis or regression analysis, correlating data from different sources, and representing the relationship between spatial features and meteorological factors in a graph or network structure. Each node represents a feature or factor, and the edges between nodes represent their correlation. For example, the correlation coefficient between spatial features and meteorological factors is calculated by Pearson correlation coefficient, and the linear or nonlinear relationship between them is found. The Pearson correlation coefficient measures the linear relationship between two variables. Its value ranges from -1 to 1, where 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no linear relationship. If the correlation coefficient is close to 1 or -1, it indicates that there is a strong linear relationship between the two variables. If the correlation coefficient is close to 0, it indicates that there is no obvious linear relationship between the two variables.

[0043] On the basis of mapping the connection network, the mapping relationship between the input space features and the meteorological factors is input, and gradient boosting decision trees are used for modeling. In the construction process, each decision tree is trained based on the previous decision tree, and the prediction performance of the model is gradually optimized. Gradient boosting decision trees are an ensemble learning method that improves prediction accuracy by building a series of weak learners (usually decision trees). Each new tree is based on the previous tree, optimizing and reducing model error. Each tree has weak predictive ability, but by integrating multiple decision trees, a powerful prediction model is formed. By combining multiple models into a strong model, the strengths of each model are fully utilized, improving prediction accuracy. Gradient boosting decision trees are one of the commonly used ensemble learning methods, which build gradient boosting decision trees through multiple iterations, enabling the final model to more accurately capture the complex relationship between features and targets.

[0044] Cross-validation of gradient boosting decision trees using coarse resolution data divides the dataset into several subsets, each of which is used for training and testing to evaluate the model's generalization ability. During cross-validation, relevant hyperparameters of the decision tree are adjusted, such as tree depth (determining the maximum depth of the tree) and leaf node number (the minimum number of samples in each leaf node of the tree). Tree depth controls the maximum depth of the decision tree, and a larger tree depth may cause overfitting, while a smaller tree depth may cause underfitting. Leaf node number determines the minimum number of samples required for each leaf node, which controls the minimum number of samples required for each leaf node to avoid excessive small sample nodes. Coarse resolution refers to the low spatial resolution of the model or data, indicating that each pixel or data unit within a given spatial range represents less information. Cross-validation is a method for verifying model performance by dividing the dataset into multiple subsets and training and testing each subset to evaluate the model's generalization ability.

[0045] Based on the results of cross-validation, the pruning strategy of the model is determined to remove nodes that contribute less to the model's prediction and reduce the complexity of the model. Pruning includes pre-pruning and post-pruning. Pre-pruning refers to stopping the expansion of the tree early in the tree growth process (such as limiting the maximum depth or the minimum number of samples per node), while post-pruning refers to removing unnecessary branches based on the importance of the tree's nodes and their contribution to error after the tree has grown. The leaf node number and node sample minimum number can be set to prune the decision tree and remove redundant nodes. Pruning is an optimization method during the training of decision trees, used to remove nodes that contribute less to the model's prediction. Pruning can prevent overfitting of decision trees and improve the model's generalization ability. For example, suppose the gradient boosting decision tree has a maximum depth of 6 and a leaf node number of 15 after cross-validation. In the post-pruning phase, it is found that some nodes have little impact (such as very small error reduction), and these nodes will be removed, resulting in a simplified tree.

[0046] Based on the optimized gradient boosting decision tree model, a downscaling technique is used to convert low-resolution data into higher-resolution data. Through the inversion model, detailed prediction data of the groundwater reserves in the target area are obtained. The downscaling inversion model is a method for converting coarse-resolution data into high-resolution data, which maintains the overall trend of the data while improving the spatial resolution of the data for more detailed analysis and prediction. Through gradient boosting decision tree optimization and pruning, the model can effectively improve the accuracy of groundwater reserve change prediction. Cross-validation ensures that the model does not overfit and can better adapt to data in different regions.

[0047] Further, the time series analysis module 12 in the intelligent groundwater reserve change prediction system based on spatio-temporal sequence analysis is also used for:

[0048] A time series division unit is used to divide the groundwater reserve change data set according to the change time sequence and construct a time series data set. A data capture unit is used to capture data from the time series data set using a long short-term memory network to obtain a plurality of trend data, including long-term trend data and short-term trend data. A cycle analysis unit is used to analyze the long-term trend data to obtain groundwater reserve change cycle characteristics. A fluctuation analysis unit is used to analyze the short-term trend data to obtain groundwater reserve change rate characteristics. A feature integration unit is used to add the groundwater reserve change cycle characteristics and the groundwater reserve change rate characteristics to the groundwater reserve change trend characteristics.

[0049] Specifically, the groundwater reserve change data set is divided according to time sequence to construct a time series data set, that is, the groundwater reserve change data set is divided into consecutive time periods according to time, and the data in each time period is represented as the state of a time point. Ensure that the data format of each time window is consistent, usually each time point corresponds to a data record, including groundwater reserve value and other related characteristics.

[0050] The time series data set is converted into a format suitable for long short-term memory network input, which is usually a three-dimensional array (sample number, time step, feature number). Long short-term memory network (LSTM) is a special recurrent neural network that can capture long-term dependencies in time series data and effectively avoid the problem of gradient disappearance, so it has a significant advantage in time series prediction. The long short-term memory network is used to capture data from the time series data set to capture long-term dependencies in the time series data.

[0051] During the training process, LSTM gradually adjusts the weights and biases to minimize the error between the model's predicted values and the actual values. Through multiple iterations of training, LSTM is able to accurately capture both long-term and short-term trends in groundwater storage changes. After training, the LSTM model can output long-term trend data, which describes the long-term trend of groundwater storage changes. The long-term trend data is usually smooth, reflecting the gradual changes in groundwater storage. Short-term trend data captures fluctuations in the data over shorter time spans, often seasonal or periodic. Long-term trend data refers to the trend of changes in a time series over a long period of time, usually used to describe the overall direction of change, such as the long-term upward or downward trend of groundwater storage. Short-term trend data refers to the trend of changes in a time series over a short period of time, usually used to describe rapid fluctuations or seasonal changes. For example, fluctuations in groundwater storage over a quarter or a year.

[0052] Based on the long-term trend data generated by LSTM, periodic analysis is performed to identify the periodic characteristics of groundwater storage changes, such as seasonal changes (such as water level fluctuations in spring, summer, autumn and winter) or multi-year cycle changes (such as long-term water level changes caused by climate change or geological changes). Periodic analysis is a process of analyzing the periodic characteristics of time series data, aiming to identify the periodic patterns in the data, such as seasonal changes or long-term cycle changes. The goal of periodic analysis is to identify the periodic change patterns in groundwater storage in the long-term trend, such as seasonal changes, climate cycle changes, land use changes and other factors. Using the long-term trend data obtained by training the LSTM model as input for periodic analysis, the overall trend of groundwater storage changes can be reflected, and longer cycle change patterns can be extracted. Periodic analysis methods can be analyzed by Fourier transform and wavelet transform. Fourier transform is a commonly used periodic analysis tool that converts time series into frequency domain to identify periodic characteristics of different frequency components; wavelet transform can analyze signals in time-frequency domain simultaneously, which helps to identify periodic characteristics in nonlinear and non-stationary data.

[0053] Based on the short-term trend data generated by LSTM, volatility analysis is performed to identify rapid changes or sharp fluctuations in groundwater reserves. For example, groundwater reserves may change rapidly after certain seasons or meteorological events, such as extreme precipitation. Volatility analysis refers to analyzing the rapidly changing parts of the data in order to identify sharp changes in the short term, focusing on the short-term rate of change of groundwater reserves, such as seasonal fluctuations or rapid changes caused by weather events. Volatility analysis can be performed through methods such as difference method, standard deviation analysis, and rate of change analysis. The difference method analyzes the volatility of time series data by calculating the change in the data at each time point, that is, the difference between the current data and the previous data. Standard deviation can measure the degree of data fluctuation, with a larger standard deviation indicating greater volatility, allowing for the evaluation of the severity of water level changes. Rate of change analysis evaluates the speed of water level changes by calculating the rate of change at different time points. For example, the instantaneous rate of change (i.e., the rate of change of water level per unit time) can be used to quantify the rate of change of groundwater reserves.

[0054] The periodical characteristics of groundwater reserves extracted from long-term trend data and the rate of change characteristics of groundwater reserves extracted from short-term trend data are integrated to form a complete set of groundwater reserves change trend characteristics, including the periodicity of long-term trends, the rate of short-term fluctuations, and the overall trend of change patterns. Through the training of the LSTM model, the long-term trends and short-term fluctuations of groundwater reserves changes are accurately captured. Periodic analysis can effectively identify the periodicity of groundwater reserves changes, and volatility analysis can identify sudden fluctuations or abnormal situations in groundwater reserves changes, providing more comprehensive information for the prediction of groundwater reserves changes and helping to improve the accuracy and reliability of the prediction of groundwater reserves changes.

[0055] Further, the spatial analysis module 13 in the intelligent prediction system for groundwater reserves changes based on spatio-temporal sequence analysis is also used for:

[0056] The grid processing unit is configured to perform grid processing on the groundwater storage change dataset based on geographical information of the target region to generate a plurality of grid cells; the correlation analysis unit is configured to traverse the plurality of grid cells and perform correlation analysis on the groundwater storage change dataset to generate a plurality of correlation coefficients; the weighted regression unit is configured to perform geographical weighted regression on the plurality of grid cells based on the plurality of correlation coefficients to generate a plurality of spatial influence weight values; the region identification unit is configured to arrange the plurality of spatial influence weight values in descending order to generate a spatial influence weight sequence, identify the target region based on the spatial influence weight sequence, and generate a plurality of identified regions; and the sensitivity analysis unit is configured to determine spatial weight distribution data based on the plurality of identified regions, perform data sensitivity analysis on the groundwater storage change dataset based on the spatial weight distribution data, and generate the groundwater storage change sensitivity factor.

[0057] Specifically, the groundwater storage change dataset is processed based on geographical information of the target region, such as geological structure (specific location), land use type (e.g., farmland, forest land, city, wetland), hydrological characteristics (e.g., river, lake, distribution of groundwater reservoir), etc. The groundwater storage change dataset is processed by grid processing, and the region is divided into a plurality of grid cells, each grid cell representing a small area of the region. Grid processing refers to dividing a geographical region into a plurality of small grid cells. Each grid cell represents a part of the region, and the groundwater storage change is analyzed grid by grid based on geographical information data.

[0058] The correlation analysis is performed on each grid cell by traversing the plurality of grid cells, combining the corresponding groundwater storage change dataset (including groundwater level, precipitation, land use type, etc.) of each grid cell. The correlation coefficient between different variables in each grid cell is calculated, for example, the correlation coefficient between groundwater storage change and land use type. Correlation analysis is a statistical analysis method used to study the relationship between two or more variables. The correlation coefficient can quantify the degree of correlation between different variables, such as the relationship between groundwater storage change and geological structure, land use type, hydrological characteristics, etc. The closer the correlation coefficient is to 1 or -1, the stronger the relationship between the two variables; the closer to 0, the less linear relationship between the variables.

[0059] According to the calculated correlation coefficient, a geographically weighted regression is performed on the grid cells, the spatial position of each grid cell is considered, and different regression models are generated according to the different positions, so as to obtain the spatial influence weight value of each region. Through the geographically weighted regression, the spatial regression coefficient corresponding to each grid cell is obtained, which represents the influence of the local characteristics of each region on the change of groundwater reserves. A spatial influence weight value is assigned to each grid cell. The greater the value, the greater the influence of the region on the change of groundwater reserves. Through the spatial influence weight value, the influence intensity of each grid cell in the target region on the change of groundwater reserves is evaluated.

[0060] The generated multiple spatial influence weight values are arranged in descending order to generate a spatial influence weight sequence. The grid cell with a higher weight has a greater influence on the change of groundwater reserves. According to the spatial influence weight sequence, the target region is identified to generate multiple identified regions. According to the height of the spatial weight value, some regions can be identified as high-impact regions or low-impact regions. The grid cell with greater influence is identified as a high-impact region, and the grid cell with smaller influence is identified as a low-impact region.

[0061] According to the multiple identified regions, the spatial weight distribution data of the entire target region is determined, reflecting the spatial influence intensity of each grid cell. According to the spatial weight distribution data, data sensitivity analysis is performed, and by perturbing the input data (such as meteorological data, land use data, etc. in the groundwater reserves change data set), the change of these data is observed to affect the prediction result of the change of groundwater reserves. Generally, local sensitivity analysis or global sensitivity analysis is used to analyze the sensitivity of the data. Local sensitivity analysis observes the change of the prediction result by changing a single input variable. Global sensitivity analysis usually changes multiple input variables at the same time, and the comprehensive influence of multiple input variables on the change of groundwater reserves is simulated and calculated. According to the result of data sensitivity analysis, a groundwater reserves change sensitivity factor is generated, reflecting the influence degree of a factor (such as air temperature, precipitation, land use type, etc.) on the change of groundwater reserves. The greater the sensitivity factor, the stronger the influence of the factor on the change of groundwater reserves.

[0062] Through the grid processing and the geographically weighted regression, the spatial characteristics of different geographical units in the target region are extracted, the relationship between geographical factors and the change of groundwater reserves is revealed, the influence degree of different geographical factors on the change of groundwater reserves is understood, and the factors with greater influence on the change of groundwater reserves are identified according to the sensitivity analysis, which is helpful to understand and capture the spatial characteristics of the change of groundwater reserves.

[0063] Further, the spatio-temporal mapping module 14 in the intelligent prediction system for the change of groundwater reserves based on spatio-temporal sequence analysis is further used for:

[0064] a time alignment unit configured to time-align the groundwater storage change trend feature and the groundwater storage change sensitive factor, obtaining a first alignment parameter; a space alignment unit configured to space-align the groundwater storage change trend feature and the groundwater storage change sensitive factor, obtaining a second alignment parameter; a space-time alignment unit configured to space-time-align the groundwater storage change trend feature and the groundwater storage change sensitive factor according to the first alignment parameter and the second alignment parameter, constructing a space-time mapping relationship; a grid division unit configured to divide a target region according to the space-time mapping relationship, constructing space-time grid data; and a mapping calculation unit configured to perform dynamic mapping calculation based on the space-time grid data, obtaining a dynamic mapping value, and synchronizing the dynamic mapping value to the space-time regularity map.

[0065] Specifically, the groundwater storage change trend feature obtained through time series analysis and the groundwater storage change sensitive factor obtained through spatial analysis are aligned in time to obtain a first alignment parameter. Through timestamp matching, the groundwater storage change trend feature and the sensitive factor are aligned in time, involving time synchronization of data, which may require interpolation or resampling to ensure data consistency. For example, assuming that the time span of groundwater storage change is 2010-2020, and meteorological data is provided daily, daily data can be converted to annual data for alignment by averaging or other interpolation methods. If the time spans of the two are different, weight the data of certain time periods to ensure that they occupy an appropriate proportion in the space-time model. Time alignment refers to matching data of different time scales or time points. For the groundwater storage change trend feature and the sensitive factor, time alignment means aligning the two data sets in chronological order to ensure that the change features at the same time point can correspond.

[0066] The groundwater storage change trend feature obtained through time series analysis and the groundwater storage change sensitive factor obtained through spatial analysis are aligned in space to obtain a second alignment parameter. That is, the spatial region grid corresponding to the groundwater storage change sensitive factor is matched with the geographic location of the groundwater storage change trend feature to ensure that the data of each spatial location can be corresponded. Spatial alignment refers to matching data of different spatial locations. For the groundwater storage change trend feature and the sensitive factor, spatial alignment means matching the spatial region represented by the sensitive factor (such as a grid) with the geographic location of the trend feature to ensure that these features and factors correspond at the same geographic location.

[0067] The groundwater storage change trend characteristics are comprehensively spatio-temporally aligned with the sensitive factors through the first alignment parameter (temporal alignment) and the second alignment parameter (spatial alignment), and a spatio-temporal mapping relationship is constructed. That is, the groundwater storage change trend of each time point and spatial unit is combined with its corresponding sensitive factor to construct a comprehensive spatio-temporal dataset. The spatio-temporal mapping relationship is the relationship between the datasets after temporal alignment and spatial alignment, which describes the mutual influence of the groundwater storage change trend and its sensitive factors at different times and spaces. Spatio-temporal alignment is the combination of time and space information, which aligns the groundwater storage change trend characteristics and the sensitive factors at the same time. Through temporal alignment and spatial alignment, it is ensured that the groundwater storage change trend and its influencing factors are matched at a specific time point and a specific location, so as to obtain a more accurate spatio-temporal mapping relationship.

[0068] According to the spatio-temporal mapping relationship, the target area is divided into multiple spatio-temporal grid cells, and the spatial and temporal characteristics of the groundwater storage change are generated for each cell. Each grid cell contains data of a specific time period and spatial location. Through the spatio-temporal mapping relationship, the groundwater storage change data and the corresponding sensitive factors (such as meteorological data, land use, etc.) are mapped into each grid cell. Each grid cell not only contains spatial location information, but also combines data in the time dimension (such as groundwater storage change and related sensitive factors in a certain year or season).

[0069] According to the spatio-temporal grid data, dynamic mapping calculation is performed based on the spatio-temporal grid data, and dynamic prediction and calculation are performed through the spatio-temporal mapping relationship, so as to generate dynamic mapping values of the groundwater storage change, which represent the change trend of the groundwater storage at different times and spaces and the degree of influence by the sensitive factors. Dynamic mapping calculation is performed through spatio-temporal grid data and spatio-temporal mapping relationship. Using the historical data of these grid cells, combined with related sensitive factors, the groundwater storage change is predicted and calculated. Based on historical data and sensitive factors, a suitable model is selected for dynamic mapping calculation, including regression models (such as linear regression, ridge regression), machine learning models (such as random forest, support vector machine) and deep learning models (such as long short-term memory network LSTM). The historical spatio-temporal grid data is used for model training, and the training process is to minimize the prediction error and learn the relationship between the input data and the groundwater storage change. The trained model is used to predict the future groundwater storage change at a certain time or a certain time period, and the dynamic mapping value of each grid cell is obtained. Assuming that a linear regression model is selected, the regression equation is calculated through historical data, and the sensitive factors are used as input to predict the future groundwater storage change.

[0070] The dynamic mapping value is combined with the spatiotemporal regularity map, and finally the groundwater change regularity in different time and space segments is presented on the map. Through the spatiotemporal regularity map, the distribution regularity of the change of the groundwater reserves in time and space can be directly seen, helping to predict the future change trend of the groundwater reserves. Through spatiotemporal alignment and spatiotemporal mapping calculation, the change regularity of the groundwater reserves can be more accurately captured, and the accuracy of the prediction model can be improved.

[0071] Further, the change prediction module 15 in the intelligent prediction system for groundwater reserves change based on spatiotemporal sequence analysis is also used for:

[0072] a time analysis unit for performing time analysis on the spatiotemporal regularity map according to the dynamic mapping value to obtain a time feature set; a space analysis unit for performing space analysis on the spatiotemporal regularity map according to the dynamic mapping value to obtain a space feature set; a normalization processing unit for performing normalization processing on the time feature set and the space feature set, and determining data proportion information according to a processing result; a gradient descent unit for constructing a plurality of training data and a plurality of verification data according to the data proportion information, performing gradient descent based on the plurality of training data, and generating an initial training parameter; and a model training unit for evaluating the initial training parameter based on the plurality of verification data, generating a training score, adjusting the plurality of training data when the training score is less than an expected score, and iterating until the training score is greater than or equal to the expected score, thereby obtaining the deep prediction model.

[0073] Specifically, the spatiotemporal regularity map is traversed, the dynamic mapping value is analyzed in time, the change trend of the groundwater reserves in different time segments is analyzed, and a time feature set such as trend, period, seasonality, etc. is extracted. The time feature set is feature data related to the time dimension extracted from the spatiotemporal regularity map. For example, it may include information such as the trend of the groundwater reserves in different time segments, periodic fluctuations, seasonal changes, etc. According to the dynamic mapping value in the spatiotemporal regularity map, the spatial dimension is analyzed, and spatial distribution features such as spatial autocorrelation, spatial heterogeneity, etc. are extracted. The space feature set is feature data related to the spatial dimension extracted from the spatiotemporal regularity map. For example, it may include information such as the change pattern of the groundwater reserves in different geographic locations, spatial distribution, hot spot areas, etc.

[0074] The time feature set and the space feature set are normalized to eliminate the scale difference between the features, map the numerical values of all features to the range of [0, 1], and make the influence of each feature the same, thereby avoiding the dominance of certain features due to large numerical values. According to the processing results, determine the data proportion information, and the proportion of training data should be higher than that of validation data, which can be 7:3 or 6:4, depending on the degree of influence of time and space. According to the data proportion information, construct multiple training data and validation data. Training data refers to selecting a part (e.g., 80%) from the normalized data set as training data for model training, and the input data for training the model. Through the training data, the model can learn the relationship between the features and the results; validation data refers to selecting the remaining part (e.g., 20%) as validation data for evaluating the training effect and generalization ability of the model, and the test data set for evaluating the performance of the model during training, which helps to judge whether the model is overfitting or underfitting and adjust the training process.

[0075] The gradient descent algorithm is used to train the training data set to optimize the initial parameters of the deep prediction model. Gradient descent is an optimization algorithm used to minimize the loss function. In deep learning, the loss function measures the difference between the model's predicted values and the actual values. Gradient descent calculates the derivative (gradient) of the loss function with respect to the model parameters (such as neural network weights and biases), then adjusts the model parameters according to the gradient information, gradually reduces the value of the loss function, and converges to the optimal solution. Training data is divided into input features (such as time features, space features, etc.) and target outputs (such as groundwater storage change values). These input features and output values will be input into the model as training data. During training, the model will generate predicted values based on input features and calculate the difference between predicted values and actual values. This difference is measured by the loss function. Using the gradient descent algorithm, the model's parameters are updated by calculating the gradient of the loss function with respect to the model parameters. Each time the model updates, it adjusts the parameters in the negative direction of the gradient, thereby gradually reducing the value of the loss function.

[0076] During initial training, model parameters are generally randomly initialized. As training progresses, the model adjusts these parameters through each gradient descent iteration to reduce prediction error. The initial training parameters are the first batch of relatively reasonable parameters obtained by the model when optimizing the training data after multiple iterations. Through multiple iterations of optimization, the final parameters that can effectively predict groundwater storage changes are obtained. The gradient descent algorithm continuously adjusts the parameters to gradually reduce the loss function. In each iteration, the model parameters are updated according to the calculated gradient, gradually approaching the optimal solution. The learning rate of gradient descent determines the step size of each update, and a too large learning rate may cause the parameters to be updated too much, and a too small learning rate may cause the convergence speed to be too slow.

[0077] The initial training parameters are evaluated by multiple validation data, the initial training parameters are applied to the validation data, the model is used for prediction, and the error between the prediction result and the true value is calculated. According to the prediction error of the model, the training score is calculated, and the common score standards include mean square error (MSE) and mean absolute error (MAE). The mean square error quantifies the prediction accuracy of the model by calculating the average of the squared differences between the predicted value and the true value. The smaller the mean square error, the closer the prediction result of the model to the true value, and the better the performance of the model.

[0078] The calculated training score is compared with the expected score. If the training score is less than the expected score, the model performance is not good enough and needs to be adjusted. The expected score is a performance standard of the model, which is usually determined in the model design stage and set according to actual needs and business targets. Adjust the training data, adjust the training data set or parameter optimization algorithm, so that the model performance is gradually improved until the training score reaches or exceeds the expected score. According to the performance of the current model, adjust the training data or model parameters, for example, increase more sample data, or generate new training data through data enhancement technology, or adjust the hyperparameters of the model (such as learning rate, neural network layer number, tree depth, etc.), so that the model better fits the training data. Use the adjusted data or parameters for retraining, update the model parameters, and evaluate again using the validation data. Through iterative training and evaluation, the expected score is gradually approached.

[0079] After several iterations of adjustment, the training score of the model has reached the expected score. At this time, the model is accurate enough to effectively predict the change of groundwater reserves, and the model obtained at this time can be used as a deep prediction model to predict the future change of groundwater reserves and applied to practical problems. Through iterative training and adjustment, the deep prediction model can more accurately predict the change of groundwater reserves, and the normalization and feature extraction process enhances the robustness of the model, enabling it to handle complex spatio-temporal data.

[0080] In summary, the intelligent prediction system for groundwater reserve change based on spatio-temporal sequence analysis provided in the present application has the following technical effects:

[0081] The data acquisition module is used for data acquisition and analysis on the groundwater reserves of the target region through multiple data sources, and a groundwater reserve change data set is obtained; the time series analysis module is used for time series analysis based on the groundwater reserve change data set, and a groundwater reserve change trend feature is generated; the spatial analysis module is used for spatial analysis based on the groundwater reserve change data set, and a groundwater reserve change sensitive factor is generated; the space-time mapping module is used for space-time mapping of the groundwater reserve change trend feature and the groundwater reserve change sensitive factor, and a space-time law map is constructed; and the change prediction module is used for modeling according to the space-time law map, constructing a deep prediction model, synchronizing the groundwater reserve change data set to the deep prediction model for prediction, and generating groundwater reserve change prediction data. That is, through the collection of multiple data, time series analysis and spatial analysis are performed on the groundwater reserve change data, the time trend feature of the groundwater reserve change and the spatial sensitive factor are mapped in space-time, a space-time law map reflecting the groundwater reserve change law of the target region is constructed, and a deep prediction model is constructed accordingly to predict the groundwater reserve change, thereby improving the accuracy of the groundwater reserve change prediction.

[0082] In the second embodiment, based on the same inventive concept as the groundwater reserve change intelligent prediction system based on space-time sequence analysis in the first embodiment, the present application also provides a groundwater reserve change intelligent prediction method based on space-time sequence analysis. Please refer to the accompanying drawings Figure 2 The groundwater reserve change intelligent prediction method based on space-time sequence analysis comprises:

[0083] S100: data acquisition and analysis on the groundwater reserves of the target region through multiple data sources, and a groundwater reserve change data set is obtained; S200: time series analysis based on the groundwater reserve change data set, and a groundwater reserve change trend feature is generated; S300: spatial analysis based on the groundwater reserve change data set, and a groundwater reserve change sensitive factor is generated; S400: space-time mapping of the groundwater reserve change trend feature and the groundwater reserve change sensitive factor, and a space-time law map is constructed; and S500: modeling according to the space-time law map, constructing a deep prediction model, synchronizing the groundwater reserve change data set to the deep prediction model for prediction, and generating groundwater reserve change prediction data.

[0084] Further, the data acquisition and analysis on the groundwater reserves of the target region through multiple data sources, and the obtaining of the groundwater reserve change data set, comprises:

[0085] S110: data collection on the groundwater reserves of the target region based on the remote sensing device group, to obtain a multi-source remote sensing dataset; S120: data sensing on the groundwater reserves of the target region based on the sensing device group, to obtain a ground observation dataset; S130: integrated learning on the multi-source remote sensing dataset and the ground observation dataset, to construct a downscaling inversion model; S140: setting a first resolution, and performing change analysis according to the first resolution through the downscaling inversion model, to obtain a groundwater reserve change dataset, wherein the groundwater reserve change dataset and the first resolution have a corresponding relationship.

[0086] Further, the integrated learning on the multi-source remote sensing dataset and the ground observation dataset, and the construction of the downscaling inversion model, comprise:

[0087] S131: extracting spatial features of the target region based on the multi-source remote sensing dataset, and extracting meteorological factors of the target region based on the ground observation dataset; S132: mapping and matching the spatial features and the meteorological factors, to generate a mapping connection network, and constructing a gradient boosting decision tree according to the mapping connection network; S133: cross-validation according to a coarse resolution based on the gradient boosting decision tree, pruning the gradient boosting decision tree according to the validation result, and constructing the downscaling inversion model.

[0088] Further, the time series analysis based on the groundwater reserve change dataset, to generate groundwater reserve change trend features, comprises:

[0089] S210: dividing according to change time series based on the groundwater reserve change dataset, to construct a time series dataset; S220: data capturing on the time series dataset by using a long short-term memory network, to obtain a plurality of trend data, wherein the plurality of trend data comprises long-term trend data and short-term trend data; S230: period analysis according to the long-term trend data, to obtain groundwater reserve change period features; S240: fluctuation analysis according to the short-term trend data, to obtain groundwater reserve change rate features; S250: adding the groundwater reserve change period features and the groundwater reserve change rate features to the groundwater reserve change trend features.

[0090] Further, the spatial analysis based on the groundwater reserve change dataset, to generate groundwater reserve change sensitive factors, comprises:

[0091] S310: grid processing is performed on the groundwater reserve change dataset according to geographical information of a target region to generate a plurality of grid cells; S320: correlation analysis is performed on the plurality of grid cells in combination with the groundwater reserve change dataset to generate a plurality of correlation coefficients; S330: geographical weighted regression is performed on the plurality of grid cells according to the plurality of correlation coefficients to generate a plurality of spatial influence weight values; S340: the plurality of spatial influence weight values are arranged in descending order to generate a spatial influence weight sequence, and the target region is identified based on the spatial influence weight sequence to generate a plurality of identified regions; and S350: spatial weight distribution data is determined according to the plurality of identified regions, data sensitivity analysis is performed on the groundwater reserve change dataset based on the spatial weight distribution data, and the groundwater reserve change sensitive factor is generated.

[0092] Further, the spatiotemporal mapping of the groundwater reserve change trend feature and the groundwater reserve change sensitive factor to construct a spatiotemporal law map comprises:

[0093] S410: time alignment is performed on the groundwater reserve change trend feature and the groundwater reserve change sensitive factor to obtain a first alignment parameter; S420: space alignment is performed on the groundwater reserve change trend feature and the groundwater reserve change sensitive factor to obtain a second alignment parameter; S430: spatiotemporal alignment is performed on the groundwater reserve change trend feature and the groundwater reserve change sensitive factor according to the first alignment parameter and the second alignment parameter to construct a spatiotemporal mapping relationship; S440: a target region is divided according to the spatiotemporal mapping relationship to construct spatiotemporal grid data; and S450: dynamic mapping calculation is performed based on the spatiotemporal grid data to obtain a dynamic mapping value, and the dynamic mapping value is synchronized to the spatiotemporal law map.

[0094] Further, the modeling according to the spatiotemporal law map to construct a deep prediction model comprises:

[0095] S510: traversing the spatiotemporal regularity map to perform time analysis according to the dynamic mapping value to obtain a time feature set; S520: traversing the spatiotemporal regularity map to perform space analysis according to the dynamic mapping value to obtain a space feature set; S530: normalizing the time feature set and the space feature set, and determining data proportion information according to a processing result; S540: constructing multiple training data and multiple verification data according to the data proportion information, performing gradient descent based on the multiple training data to generate an initial training parameter; S550: evaluating the initial training parameter based on the multiple verification data to generate a training score, and adjusting the multiple training data when the training score is less than an expected score, thereby iterating until the training score is greater than or equal to the expected score to obtain the deep prediction model.

[0096] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The foregoing Figure 1 The intelligent groundwater reserve change prediction system based on spatiotemporal sequence analysis in Embodiment One and the specific examples are also applicable to the intelligent groundwater reserve change prediction method based on spatiotemporal sequence analysis in the present embodiment. Those skilled in the art can clearly understand the intelligent groundwater reserve change prediction method based on spatiotemporal sequence analysis in the present embodiment through the foregoing detailed description of the intelligent groundwater reserve change prediction system based on spatiotemporal sequence analysis. Therefore, for the sake of brevity of the specification, no further detailed description is given here. For the method disclosed in the embodiments, since it corresponds to the system disclosed in the embodiments, the description is relatively simple, and the relevant part can be referred to the system part description.

[0097] The foregoing description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

[0098] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the present application and its equivalents, the present application also intends to include these modifications and variations.

Claims

1. An intelligent prediction system for groundwater storage changes based on spatiotemporal series analysis, characterized in that, include: The data acquisition module is used to collect and analyze groundwater storage data in the target area through multiple data sources to obtain a dataset of groundwater storage changes. The time series analysis module is used to perform time series analysis based on the groundwater storage change dataset to generate groundwater storage change trend characteristics. A spatial analysis module is used to perform spatial analysis based on the groundwater storage change dataset and generate groundwater storage change sensitivity factors. The spatiotemporal mapping module is used to perform spatiotemporal mapping between the characteristics of the groundwater storage change trend and the sensitive factors of the groundwater storage change, and to construct a spatiotemporal pattern map. The change prediction module is used to model according to the spatiotemporal pattern map, construct a deep prediction model, synchronize the groundwater storage change dataset to the deep prediction model for prediction, and generate groundwater storage change prediction data. The time series analysis module includes: The time-series partitioning unit is used to divide the groundwater storage change dataset according to the change time sequence to construct a time-series dataset; The data capture unit is used to capture data from the time series dataset using a long short-term memory network to obtain multiple trend data, which include long-term trend data and short-term trend data. The periodic analysis unit is used to perform periodic analysis based on the long-term trend data to obtain the periodic characteristics of groundwater storage changes. The fluctuation analysis unit is used to perform fluctuation analysis based on the short-term trend data to obtain the characteristics of the rate of change of groundwater storage. The feature integration unit is used to add the groundwater storage change period feature and the groundwater storage change rate feature to the groundwater storage change trend feature; The spatial analysis module includes: A grid processing unit is used to perform gridding processing on the groundwater storage change dataset according to the geographic information of the target area, and generate multiple grid cells. The correlation analysis unit is used to traverse the multiple grid cells and perform correlation analysis in conjunction with the groundwater storage change dataset to generate multiple correlation coefficients. A weighted regression unit is used to perform geographic weighted regression on the multiple grid units according to the multiple correlation coefficients to generate multiple spatial influence weight values; The region identification unit is used to sort the multiple spatial influence weight values ​​in descending order to generate a spatial influence weight sequence, and to identify the target region based on the spatial influence weight sequence to generate multiple identified regions; The sensitivity analysis unit is used to determine spatial weight distribution data based on the multiple identified areas, perform data sensitivity analysis on the groundwater storage change dataset based on the spatial weight distribution data, and generate the groundwater storage change sensitivity factor. The spatiotemporal mapping module includes: A time alignment unit is used to align the groundwater storage change trend characteristics with the groundwater storage change sensitivity factor over time to obtain a first alignment parameter; A spatial alignment unit is used to spatially align the groundwater storage change trend characteristics with the groundwater storage change sensitivity factor to obtain a second alignment parameter. The spatiotemporal alignment unit is used to spatiotemporally align the groundwater storage change trend characteristics with the groundwater storage change sensitivity factor according to the first alignment parameter and the second alignment parameter, and construct a spatiotemporal mapping relationship. A grid partitioning unit is used to divide the target region according to the spatiotemporal mapping relationship and construct spatiotemporal grid data; The mapping calculation unit is used to perform dynamic mapping calculations based on the spatiotemporal grid data, obtain dynamic mapping values, and synchronize the dynamic mapping values ​​to the spatiotemporal pattern map.

2. The intelligent prediction system for groundwater storage changes based on spatiotemporal series analysis as described in claim 1, characterized in that, The data acquisition module includes: The remote sensing data acquisition unit is used to collect data on the groundwater storage of the target area based on the remote sensing equipment group, and obtain a multi-source remote sensing dataset; The sensor data acquisition unit is used to sense the groundwater storage in the target area based on the sensor equipment group and obtain the ground observation dataset. An integrated learning unit is used to integrate the multi-source remote sensing dataset and the ground observation dataset to construct a downscaling inversion model; The change analysis unit is used to set a first resolution and perform change analysis according to the first resolution using the downscaling inversion model to obtain a groundwater storage change dataset, wherein the groundwater storage change dataset corresponds to the first resolution.

3. The intelligent prediction system for groundwater storage changes based on spatiotemporal sequence analysis as described in claim 2, characterized in that, The integrated learning unit includes: The meteorological factor extraction subunit is used to extract spatial features of the target area based on the multi-source remote sensing dataset and to extract meteorological factors of the target area based on the ground observation dataset. A decision tree construction subunit is used to map and match the spatial features with the meteorological factors, generate a mapping connection network, and construct a gradient boosting decision tree according to the mapping connection network; The cross-validation subunit is used to perform cross-validation based on the gradient boosting decision tree at a coarse resolution, prune the gradient boosting decision tree according to the validation results, and construct the downscaling inversion model.

4. The intelligent prediction system for groundwater storage changes based on spatiotemporal series analysis as described in claim 1, characterized in that, The change prediction module is specifically used for: The time analysis unit is used to traverse the spatiotemporal pattern map and perform time analysis according to the dynamic mapping value to obtain a time feature set; The spatial analysis unit is used to traverse the spatiotemporal pattern map and perform spatial analysis according to the dynamic mapping value to obtain a spatial feature set. The normalization processing unit is used to normalize the time feature set and the spatial feature set, and determine the data ratio information based on the processing result; The gradient descent unit is used to construct multiple training data and multiple validation data according to the data ratio information, and perform gradient descent based on the multiple training data to generate initial training parameters. The model training unit is used to evaluate the initial training parameters based on the multiple validation data, generate a training score, and adjust the multiple training data when the training score is less than the expected score. This process is repeated until the training score is greater than or equal to the expected score, thereby obtaining the deep prediction model.

5. A method for intelligent prediction of groundwater storage changes based on spatiotemporal series analysis, characterized in that, The method for intelligent prediction of groundwater storage changes based on spatiotemporal sequence analysis, as described in any one of claims 1 to 4, is executed by the intelligent prediction system for groundwater storage changes based on spatiotemporal sequence analysis, wherein the method comprises: By collecting and analyzing groundwater reserves in the target area from multiple data sources, a dataset of groundwater reserve changes can be obtained. A time-series analysis is performed based on the aforementioned groundwater storage change dataset to generate groundwater storage change trend characteristics; Spatial analysis is performed based on the aforementioned groundwater storage change dataset to generate groundwater storage change sensitivity factors; The characteristics of groundwater storage change trends are spatiotemporally mapped with the sensitive factors of groundwater storage change to construct a spatiotemporal pattern map. Modeling is performed based on the spatiotemporal pattern map, a depth prediction model is constructed, and the groundwater storage change dataset is synchronized to the depth prediction model for prediction, generating groundwater storage change prediction data.

Images