An artificial intelligence-based water resource dynamic quantity determination method and system

Abstract

The application provides a kind of water resource dynamic nuclear amount method and system based on artificial intelligence, method includes: obtaining the multi-source heterogeneous data set corresponding to the region to be quantified;The region to be quantified is divided into several grid units, and the multi-source heterogeneous data set is mapped to the grid unit to obtain remote sensing unit data, node unit data and reported unit data;Based on real-time confidence, the three data are fused into unit fusion data;A spatial topology structure diagram corresponding to the region to be quantified is constructed, and the initial characteristics of the nodes to be quantified are obtained through the unit fusion data;Through the time series model, the initial characteristics and the graph neural network, the time series prediction quota and the correction amount are obtained, and then the dynamic nuclear quantification is obtained.Through the above-mentioned mode, the logical conflict problem caused by the inconsistency of multi-source data in space-time reference is solved;And the upgrading of the nuclear amount reference from static quota to AI dynamic quota is realized, so that the nuclear amount result can adapt to the water mode drift, and the accuracy of the nuclear amount is improved.

Description

Technical Field

[0001] This invention relates to the field of data prediction technology, and in particular to a method and system for dynamic measurement of water resources based on artificial intelligence. Background Technology

[0002] Water resources are a fundamental natural resource and a strategic economic resource.

[0003] Water resource verification is the foundation and key support for water abstraction permit supervision and water conservation management. Traditional verification work mainly relies on direct measurement using physical metering facilities (such as water meters, electromagnetic flow meters, and ultrasonic flow meters) or indirect estimation using empirical coefficients such as "electricity-to-water conversion" or "acre-to-water conversion." However, as modern water management shifts towards refinement and intelligence, higher demands are placed on the accuracy, reliability, and dynamic response capabilities of water abstraction and water consumption measurement.

[0004] To overcome the limitations of single-method approaches, multi-source data integration technology has been introduced, attempting to construct a comprehensive measurement system covering the entire region by integrating satellite remote sensing inversion, IoT sensor direct measurement, and user-reported data. However, this multi-source heterogeneous data fusion faces integration barriers in practical applications: firstly, there is a lack of effective data fusion methods, and simply superimposing multi-source data can easily lead to data logic conflicts due to inconsistencies in spatiotemporal benchmarks; secondly, the measurement benchmarks cannot adapt to dynamic changes in water use patterns such as climate, production, and water-saving renovations, resulting in continuous deviations between measurement results and actual consumption. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a dynamic water resource assessment method and system based on artificial intelligence. This method addresses the technical problem that existing technologies for water resource assessment using multi-source heterogeneous data fusion lack effective data fusion methods and cannot adapt to dynamic changes in water use patterns, resulting in significant errors in the assessment results.

[0006] To achieve the above objectives, in a first aspect, embodiments of this application provide an artificial intelligence-based dynamic water resource assessment method, comprising the following steps: Obtain a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data, and household-level reported water volume data. The region to be measured is divided into several grid cells, and the multi-source heterogeneous dataset is mapped to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data. Based on real-time confidence, the remote sensing unit data, the node unit data, and the reporting unit data are fused into unit fused data. Construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and obtain the initial features of the nodes to be measured through the unit fusion data; The time-series prediction quota corresponding to the node to be measured is obtained by using the time-series model and the initial features, and the correction amount is obtained by using the graph neural network and the initial features. The dynamic verification amount is obtained based on the time-series prediction quota and the correction amount.

[0007] Furthermore, the remote sensing inversion water volume data includes several pixel values ​​corresponding to remote sensing pixels, the IoT node flow data includes several node values ​​corresponding to IoT nodes, and the household-level reported water volume data includes several water intake values ​​corresponding to water intake points. The step of mapping the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data, and reporting unit data includes: Obtain the grid resolution of the grid cell and the remote sensing resolution of the remote sensing inverted water volume data, and compare the grid resolution with the remote sensing resolution; If the remote sensing resolution is greater than the grid resolution, then the pixel values ​​of several remote sensing pixels corresponding to the grid unit are aggregated to obtain remote sensing unit data. If the remote sensing resolution is less than the grid resolution, then the remote sensing unit data is obtained based on the grid unit and the overlapping area of ​​the remote sensing pixel corresponding to the grid unit. Based on the physical coordinates of IoT nodes and water intakes, node values ​​and water intake values ​​are mapped to corresponding grid cells to obtain node cell data and report cell data.

[0008] Furthermore, the real-time confidence level includes remote sensing confidence level, node confidence level, and reporting confidence level. The step of fusing the remote sensing unit data, the node unit data, and the reporting unit data into unit fusion data based on the real-time confidence level includes: Obtain the data source health, spatial adaptability, and temporal trend degree corresponding to the grid cell, and obtain the remote sensing data weight, node data weight, and reported data weight based on the data source health, spatial adaptability, and temporal trend degree. The weight sum is obtained based on the remote sensing data weight, the node data weight, and the reported data weight. The remote sensing confidence, the node confidence, and the reported confidence are obtained through the remote sensing data weight, the node data weight, the reported data weight, and the weight sum. The remote sensing unit data, the node unit data, and the reported unit data are weighted and fused into unit fused data using the remote sensing confidence level, the node confidence level, and the reporting confidence level.

[0009] Furthermore, the step of constructing a spatial topology diagram corresponding to the region to be measured, including a plurality of nodes to be measured, includes: All pipeline nodes, weir nodes, water source nodes, and water intake node nodes within the area to be measured are mapped to nodes to be measured. Based on the pipe segment connectivity and hydraulic transmission direction, the connection edges between the nodes to be measured are constructed to form a spatial topology diagram.

[0010] Furthermore, the step of obtaining the initial features of the node to be measured through the unit fusion data includes: Select several cells from a number of grid cells that correspond to the nodes to be measured; Obtain the Euclidean distance between the candidate cell and the node to be measured, and determine the cell weight of the candidate cell based on the Euclidean distance; The unit fusion data of several units to be used are weighted and summed using the unit weights to obtain the merging features; Obtain the original node features of the node to be measured, and concatenate the original node features and the merged features into an initial feature.

[0011] Furthermore, the formula for obtaining the unit weight is: , in, This represents the cell weight of the i-th cell corresponding to the j-th node to be measured. This represents the Euclidean distance between the i-th cell to be used and the j-th node to be measured. Represents the power exponent. This represents the Euclidean distance between the k-th cell to be used and the j-th node to be measured. This represents the total number of units to be used corresponding to the j-th node.

[0012] Furthermore, the graph neural network includes a first convolutional layer and a second convolutional layer, and the step of obtaining the correction amount through the graph neural network and the initial features includes: Obtain the adjacency matrix corresponding to the spatial topology graph, and obtain the normalized matrix based on the adjacency matrix; The initial features and the normalized matrix are input into the first convolutional layer to obtain stage features; The stage features and the normalization matrix are input into the second convolutional layer to obtain the correction amount.

[0013] Furthermore, the formula for obtaining the stage features is: , in, Indicates stage characteristics, Represents a normalized matrix. Let represent the initial characteristics at time t. This represents the first trainable weight matrix. Represents a nonlinear activation function; The formula for obtaining the correction amount is: , in, Indicates the correction amount. This represents the second trainable weight matrix.

[0014] Furthermore, the step of obtaining the dynamic verification amount based on the time-series prediction quota and the correction amount includes: Obtain the time period embedding amount, and concatenate the time period embedding amount and the initial feature into a prediction vector; The prediction vector is input into a multilayer perceptron to obtain fusion weights, and the time-series prediction quota and the correction amount are fused into a dynamic kernel amount based on the fusion weights.

[0015] Secondly, embodiments of this application provide an artificial intelligence-based dynamic water resource assessment system, applied to the artificial intelligence-based dynamic water resource assessment method described in the first aspect above, the system comprising: The acquisition module is used to acquire a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data, and household-level reported water volume data. The conversion module is used to divide the area to be measured into several grid cells and map the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data. The stitching module is used to fuse the remote sensing unit data, the node unit data, and the reporting unit data into unit fused data based on real-time confidence. The construction module is used to construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and to obtain the initial features of the nodes to be measured through the unit fusion data; The execution module is used to obtain the time-series prediction quota corresponding to the node to be measured through the time-series model and the initial features, obtain the correction amount through the graph neural network and the initial features, and obtain the dynamic verification amount based on the time-series prediction quota and the correction amount.

[0016] Thirdly, embodiments of this application provide a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the artificial intelligence-based dynamic water resource assessment method as described in the first aspect above.

[0017] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the artificial intelligence-based dynamic water resource assessment method as described in the first aspect above.

[0018] Compared with existing technologies, the beneficial effects of this invention are as follows: By mapping multi-source heterogeneous datasets to unified grid cells and performing weighted fusion of remote sensing unit data, node unit data, and reporting unit data based on real-time confidence, the logical conflict problem caused by the inconsistency of spatiotemporal benchmarks in multi-source data is solved, and the effective fusion of multi-source heterogeneous data is achieved; by constructing a spatial topology diagram and aggregating grid-level fused data into the initial features of the nodes to be measured, the spatial dependency relationship of the regional water use system is established; by generating time-series prediction quotas based on time-series patterns through a time-series model, and by extracting the mutual influence between the nodes to be measured from the spatial topology through a graph neural network to generate correction quantities, the time-series prediction quotas and correction quantities are fused into a dynamic verification quantity, realizing the upgrade of the verification benchmark from static quotas to AI dynamic quotas, enabling the verification results to adapt to water use pattern drift and improving the accuracy of verification. Attached Figure Description

[0019] Figure 1 This is a flowchart of the dynamic water resource assessment method based on artificial intelligence in the first embodiment of the present invention; Figure 2 This is a structural block diagram of the AI-based dynamic water resource measurement system in the second embodiment of the present invention. The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0020] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0021] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0022] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0023] Please see Figure 1 The first embodiment of the present invention provides a dynamic water resource assessment method based on artificial intelligence, which includes the following steps: S10: Obtain a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data, and household-level reported water volume data. In this embodiment, the remote sensing inversion water volume data originates from Sentinel-1 / 2 or GRACE series satellites, the IoT node flow data originates from electromagnetic flow meters, ultrasonic flow meters in the pipeline network, or NB-IoT water meters at the user end, and the user-reported water volume data originates from water collection records reported by the users themselves. It should be noted that the units of the remote sensing inversion water volume data and the user-reported water volume data are both cubic meters, while the electromagnetic flow meters, ultrasonic flow meters, or NB-IoT water meters at the user end measure instantaneous flow in cubic meters per hour. This can be converted (aligned sampling period) according to the sampling period of the remote sensing inversion water volume data and the user-reported water volume data to unify the dimensions of the IoT node flow data with those of the remote sensing inversion water volume data and the user-reported water volume data.

[0024] S20: Divide the area to be measured into several grid cells, and map the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data; Understandably, based on the required business accuracy (e.g., 100 meters, 500 meters), the grid resolution of the area to be measured is set to form a grid to be measured. In the grid to be measured, horizontal and vertical dividing lines are set at equal intervals along the x-axis and y-axis to divide the grid to be measured into several grid units. A unique ID is assigned to each grid unit, and its geographical range is recorded.

[0025] The remote sensing inversion water volume data includes several pixel values ​​corresponding to remote sensing pixels, the IoT node flow data includes several node values ​​corresponding to IoT nodes, and the household-level reported water volume data includes several water intake values ​​corresponding to water intake points.

[0026] Step S20 includes: S210: Obtain the grid resolution of the grid cell and the remote sensing resolution of the remote sensing inverted water volume data, and compare the grid resolution with the remote sensing resolution; S220: If the remote sensing resolution is greater than the grid resolution, then the pixel values ​​of several remote sensing pixels corresponding to the grid unit are aggregated to obtain remote sensing unit data; When the remote sensing resolution is greater than the grid resolution, one grid cell will cover multiple remote sensing pixels. In this case, the pixel values ​​of the remote sensing pixels it covers are summed to form remote sensing cell data.

[0027] S230: If the remote sensing resolution is less than the grid resolution, then obtain the remote sensing unit data based on the grid unit and the overlapping area of ​​the remote sensing pixel corresponding to the grid unit; When the remote sensing resolution is less than the grid resolution, one remote sensing cell will cover multiple grid cells. In this case, there are two scenarios: For the grid cells in the central region that are covered, they are completely covered by one remote sensing cell. In this case, the overlap area between the grid cell and the remote sensing cell is 100%, and the pixel value of this remote sensing cell is selected as the remote sensing cell data. For the grid cells in the edge region that are covered, they are only partially covered, and are also partially covered by other remote sensing cells. In this case, the overlap areas between the grid cell and its corresponding remote sensing cells are obtained, and the sum of the areas of these overlap areas is obtained. The overlap ratio is obtained by dividing the overlap area and the sum of the areas. The pixel values ​​of the remote sensing cells are then weighted and summed using the overlap ratio to obtain the remote sensing cell data.

[0028] S240: Based on the physical coordinates of IoT nodes and water intakes, the node values ​​and water intake values ​​are mapped to the corresponding grid cells to obtain node cell data and report cell data; Understandably, by comparing the physical coordinates of the IoT node and the water intake with the geographical range of the grid cell, the grid cell corresponding to the IoT node and the water intake can be determined. It should be noted that if there are multiple node values ​​within a grid cell, the multiple node values ​​are summed to obtain the node cell data. The method for obtaining the reported cell data is the same as that for obtaining the node cell data, and will not be elaborated here.

[0029] S30: Based on real-time confidence, the remote sensing unit data, the node unit data, and the reporting unit data are fused into unit fused data; The real-time confidence level includes remote sensing confidence, node confidence, and reporting confidence. Specifically, Step S30 includes: S310: Obtain the health, spatial adaptability, and temporal trend of the data source corresponding to the grid cell, and obtain the remote sensing data weight, node data weight, and reported data weight based on the health, spatial adaptability, and temporal trend of the data source; It should be noted that the remote sensing unit data, node unit data, and reporting unit data within the grid unit correspond to the data source health, spatial adaptability, and temporal trend, respectively.

[0030] For data source health: 1. Remote sensing unit data: Obtain the cloud mask percentage (0~1) corresponding to the remote sensing inversion water volume data, and subtract the cloud mask percentage from 1 to form the data health score; 2. Node unit data: Conduct a comprehensive evaluation based on the signal strength and remaining power in the IoT device self-inspection report (this can be based on a pre-trained neural network model, using the self-inspection values ​​from previous self-inspection reports as input and the preset health score as output for training; or using a prior knowledge base to preset mapping rules) to obtain a single health score (0~1). If there are multiple IoT nodes within a grid unit, then... 1. Average the health score of a single water intake to obtain the data health score. If there is only one IoT node in the grid cell, the single health score is directly selected as the data health score. 2. Report unit data: Determine whether the water intake value is reported at the preset time point. If it is reported at the preset time point, the data health score is set to 1. If it is not reported at the preset time point, obtain the overdue days, multiply the overdue days by 0.1 to form a deduction item, and obtain the data health score by subtracting the deduction item from 1. It can be understood that the minimum data health score is 0. If there are multiple water intakes in the grid cell, average the health score.

[0031] For spatial adaptability: 1. Remote sensing unit data: Obtain the water area ratio within the grid unit and determine the spatial adaptability based on the water area ratio; 2. Node unit data and reporting unit data: Determine whether there is an IoT node and water intake within the grid unit. If they exist, the spatial adaptability is 1; if they do not exist, the spatial adaptability is 0. For the temporal trend: Based on the sliding window data from remote sensing unit data, node unit data, and reporting unit data (which can be understood as including frame data from several consecutive time frames), remote sensing window data, node unit data, and reporting unit data are extracted respectively. The mean and standard deviation of the remote sensing data of the remote sensing window data are obtained, and the temporal trend is obtained based on the following formula: , in, This represents the temporal trend degree corresponding to the remote sensing unit data at time t. Represents an exponential function. This represents the frame data at time t in the remote sensing unit data. This represents the mean value of remote sensing data within time window T. This represents the standard deviation of remote sensing data within time window T, where t∈T. The temporal trend of node window data and reporting window data is similar and will not be elaborated upon here.

[0032] Furthermore, the formula for obtaining the remote sensing data weights is as follows: , in, This represents the weight of the remote sensing data at time t. This represents the health status of the data source corresponding to the remote sensing unit data at time t. This represents the spatial fit between the data at time t and the remote sensing unit data. , , Both represent adjustment coefficients, and + + =1.

[0033] S320: Obtain a weight sum based on the remote sensing data weight, the node data weight, and the reported data weight; obtain the remote sensing confidence level, the node confidence level, and the reported confidence level through the remote sensing data weight, the node data weight, the reported data weight, and the weight sum. The remote sensing confidence score, the node confidence score, and the reporting confidence score are respectively formed by dividing the product of the remote sensing data weight, the node data weight, the reported data weight, and the sum of the weights.

[0034] S330: The remote sensing unit data, the node unit data, and the reporting unit data are weighted and fused into unit fused data using the remote sensing confidence level, the node confidence level, and the reporting confidence level. By introducing three dimensions—data source health (reflecting device status and data quality), spatial adaptability (reflecting the degree of matching between the data source and the grid), and temporal trend (reflecting the degree of deviation between current data and historical patterns)—real-time confidence is comprehensively calculated. This enables data fusion to dynamically respond to anomalies such as device drift, cloud coverage, and overdue reporting, avoiding fusion bias caused by simple averaging or fixed weights, and improving the robustness and reliability of unit fusion data.

[0035] S40: Construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and obtain the initial features of the nodes to be measured through the unit fusion data; Step S40 includes: S410: Map the pipeline nodes, weir nodes, water source nodes, and water intake node within the area to be measured to nodes to be measured; S420: Based on the pipe segment connectivity and hydraulic transmission direction, construct the connection edges between the nodes to be measured to form a spatial topology diagram; The pipeline nodes exhibit a water transmission relationship between upstream and downstream (upstream water flow affects downstream flow), and pipeline leaks and pressure changes propagate between adjacent nodes. The weir / sluice gate nodes regulate flow, affecting multiple downstream nodes. The water source nodes supply water to multiple users, and their outflow is influenced by user demand. The user nodes exhibit similar water usage behaviors among adjacent users. By constructing this spatial topology, the spatial dependencies between nodes can be captured, allowing information from adjacent nodes to be measured to be used to correct the core parameters of individual nodes.

[0036] S430: Select several cells from a number of grid cells that correspond to the nodes to be measured; In this embodiment, the Thiessen polygon algorithm is used to generate several polygon regions covering the area to be measured. Each node to be measured corresponds to a polygon region. It should be noted that the distance from any point within the polygon region to the corresponding node to be measured is closer than the distance from that point to other nodes to be measured. After obtaining the polygon region, it is determined whether the center of the grid cell is located within the polygon region. If it is, the grid cell is selected as the unit to be used corresponding to the node to be measured.

[0037] S440: Obtain the Euclidean distance between the candidate cell and the node to be measured, and determine the cell weight of the candidate cell based on the Euclidean distance; Obtain the geographic coordinates of the center point of the unit to be used and the geographic coordinates of the node to be measured. Calculate the Euclidean distance based on the two geographic coordinates.

[0038] The formula for obtaining the unit weight is: , in, This represents the cell weight of the i-th cell corresponding to the j-th node to be measured. This represents the Euclidean distance between the i-th cell to be used and the j-th node to be measured. In this embodiment, the exponent is represented as a power. The value of is 2. This represents the Euclidean distance between the k-th cell to be used and the j-th node to be measured. This represents the total number of units to be used corresponding to the j-th node.

[0039] S450: The unit fusion data of several units to be used are weighted and summed using the unit weights to obtain the merging features; S460: Obtain the original node features of the node to be measured, and concatenate the original node features and the merged features into an initial feature; The initial features of the nodes include hydraulic features, historical quota features, and meteorological features. Hydraulic features: For pipeline nodes, readings from pressure sensors installed at the node are directly read. If no sensor is available, spatial interpolation is performed using pressure values ​​from upstream and downstream nodes to estimate the pressure, which is then converted into an equivalent head. For weir / sluice gate nodes, level gauge data from upstream and downstream of the gate is read. For water source nodes, level gauge data is read. For water intake nodes, there is usually no direct pressure sensor, so the value is set to 0. Historical quota features: The ratio of this year's water consumption to the water consumption of the node in question compared to the same period last year. Meteorological features: Based on meteorological station data mapped to the corresponding nodes in question. These initial features include regional background water volume information and the node's own state information, providing rich input dimensions for subsequent time-series models and graph neural networks.

[0040] S50: Obtain the time-series prediction quota corresponding to the node to be measured through the time-series model and the initial features, and obtain the correction amount through the graph neural network and the initial features. Obtain the dynamic kernel amount based on the time-series prediction quota and the correction amount. In this embodiment, the temporal model is a two-layer LSTM + Attention network. The first LSTM layer reads the input features and captures short-term temporal dependencies. The second LSTM layer, based on the output of the first layer, further extracts long-term trends and complex temporal patterns, outputting a hidden state sequence. The Attention mechanism performs a weighted summation on the hidden state sequence, identifying the key time points in the historical sequence most relevant to predicting future quotas, and outputting a context vector. This context vector is then converted into a temporal prediction quota through a fully connected layer and an activation function. Furthermore, during the training process of the temporal model, the initial historical feature sequence is obtained in the same way as described above. This initial historical feature sequence is then divided into an earlier training input feature sequence and a later training output feature sequence, which are used as input and output values ​​respectively to train the model. After training is completed, the temporal prediction quota is obtained by using the initial features as input values.

[0041] Step S50 includes: S510: Obtain the adjacency matrix corresponding to the spatial topology diagram, and obtain the normalized matrix based on the adjacency matrix; The formula for obtaining the normalized matrix is: , in, Represents a normalized matrix. This represents the diagonal matrix resulting from adding a self-loop to the adjacency matrix. ,in, This represents the degree matrix element of the a-th node to be measured in the diagonal matrix. This represents the adjacency matrix element between the a-th node and the b-th node in the adjacency matrix. By adding a self-loop, it can be ensured that the node retains its own information. This represents the adjacency matrix.

[0042] S520: Input the initial features and the normalized matrix into the first convolutional layer to obtain stage features; The formula for obtaining the stage features is: , in, Indicates stage characteristics, Represents a normalized matrix. Let represent the initial characteristics at time t. This represents the first trainable weight matrix. This represents a non-linear activation function.

[0043] S530: Input the stage features and the normalization matrix into the second convolutional layer to obtain the correction amount; The formula for obtaining the correction amount is: , in, Indicates the correction amount. This represents the second trainable weight matrix. The correction amount characterizes the kernel quantity that the node to be trained should possess based solely on spatial topological relationships. The first convolutional layer performs spatial message passing and nonlinear transformation on the initial features to extract low-order spatial features from the local neighborhood; the second convolutional layer further aggregates high-order neighborhood information based on the output of the first convolutional layer to capture spatial dependencies at greater distances; the introduction of the normalization matrix ensures that the numerical scale of nodes with different degrees is stable during message aggregation, avoids feature inflation of nodes with high degrees, and improves the convergence and generalization ability of the graph neural network training.

[0044] S540: Obtain the time period embedding amount, and concatenate the time period embedding amount and the initial feature into a prediction vector; The time period embedding is used to characterize the red light attribute at the current time point. It is a feature label that is defined by the user or learned by the model, including: month, whether it is the rainy season, whether it is a holiday, etc.

[0045] S550: Input the prediction vector into a multilayer perceptron to obtain fusion weights, and fuse the time-series prediction quota and the correction amount into a dynamic kernel quantity based on the fusion weights; When the predicted vector is input into the multilayer perceptron, it passes through a 64-dimensional hidden layer and a 32-dimensional hidden layer for computation (i.e., matrix multiplication and activation function). Finally, the output value is limited to 0~1 by the Sigmoid activation function, which is the fusion weight. After reading the predicted vector, the multilayer perceptron understands how the state of the node to be measured affects its credibility at a specific moment (such as a Monday in the rainy season). It can dynamically obtain the fusion weight. If it is a rainy day, it will prompt the network to increase the fusion weight, making the model more trusting of the time series prediction value based on historical rainy days. If it is a normal weekday, it may decrease the fusion weight and rely more on spatial topology correction. By introducing time-period embedding, the calculation of fusion weights can perceive the macroscopic time background of the current moment. The multilayer perceptron performs nonlinear mapping on the spliced ​​prediction vector and outputs fusion weights, realizing adaptive weighting of time-series prediction quotas and spatial corrections. When the weather is stable and historical regularity is strong, time-series predictions are trusted more, and when spatial disturbances are significant, spatial corrections are trusted more. Thus, the kernel results are automatically optimized under different working conditions, improving the robustness and adaptability of dynamic kernels.

[0046] The formula for obtaining the dynamic verification quantity is: , in, Indicates dynamic quantitative quantification. Indicates the time series forecast quota. This indicates the fusion weight.

[0047] By mapping multi-source heterogeneous datasets to unified grid cells and performing weighted fusion of remote sensing unit data, node unit data, and reporting unit data based on real-time confidence, the logical conflict problem caused by the inconsistency of spatiotemporal benchmarks in multi-source data was resolved, achieving effective fusion of multi-source heterogeneous data. By constructing a spatial topology map and aggregating grid-level fused data into the initial features of nodes to be measured, the spatial dependency relationship of the regional water use system was established. Time-series models were used to generate time-series prediction quotas based on time-series patterns, and graph neural networks were used to extract the mutual influence between nodes to be measured from the spatial topology to generate correction quantities. Then, the time-series prediction quotas and correction quantities were fused into dynamic verification quantities, realizing the upgrade of the verification benchmark from static quotas to AI dynamic quotas, enabling the verification results to adapt to water use pattern drift and improving the accuracy of verification.

[0048] Please see Figure 2The second embodiment of the present invention provides an artificial intelligence-based dynamic water resource assessment system. This system is applied to the artificial intelligence-based dynamic water resource assessment method described in the above embodiments, and will not be repeated hereafter. As used below, the terms "module," "unit," "subunit," etc., can refer to a combination of software and / or hardware that performs a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0049] The system includes: The acquisition module 10 is used to acquire a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data and household-level reported water volume data. The conversion module 20 is used to divide the area to be measured into several grid cells and map the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data. The conversion module 20 includes: The first unit is used to obtain the grid resolution of the grid cell and the remote sensing resolution of the remote sensing inversion water volume data, and to compare the grid resolution with the remote sensing resolution. The second unit is used to aggregate the pixel values ​​of several remote sensing pixels corresponding to the grid unit to obtain remote sensing unit data if the remote sensing resolution is greater than the grid resolution. The third unit is used to obtain remote sensing unit data based on the grid unit and the overlapping area of ​​the remote sensing pixels corresponding to the grid unit if the remote sensing resolution is less than the grid resolution. The fourth unit is used to map node values ​​and water intake values ​​to corresponding grid cells based on the physical coordinates of IoT nodes and water intakes, so as to obtain node cell data and report unit data. The splicing module 30 is used to fuse the remote sensing unit data, the node unit data and the reporting unit data into unit fused data based on real-time confidence. The splicing module 30 includes: The fifth unit is used to obtain the health, spatial adaptability and temporal trend of the data source corresponding to the grid unit, and to obtain the remote sensing data weight, node data weight and reported data weight based on the health, spatial adaptability and temporal trend of the data source. The sixth unit is used to obtain a weight sum based on the remote sensing data weight, the node data weight, and the reported data weight, and to obtain the remote sensing confidence, the node confidence, and the reported confidence through the remote sensing data weight, the node data weight, the reported data weight, and the weight sum; The seventh unit is used to weight and fuse the remote sensing unit data, the node unit data, and the reporting unit data into unit fused data by means of the remote sensing confidence level, the node confidence level, and the reporting confidence level. The construction module 40 is used to construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and to obtain the initial features of the nodes to be measured through the unit fusion data. The construction module 40 includes: The eighth unit is used to map the pipeline nodes, weir nodes, water source nodes and water intake nodes in the area to be measured as nodes to be measured. The ninth unit is used to construct the connection edges between the nodes to be measured based on the pipe segment connectivity and hydraulic transmission direction, so as to form a spatial topology diagram. The tenth unit is used to select several units corresponding to the nodes to be measured from several grid units; The eleventh unit is used to obtain the Euclidean distance between the candidate unit and the node to be measured, and to determine the unit weight of the candidate unit based on the Euclidean distance. The twelfth unit is used to perform a weighted summation of the unit fusion data of several units to be used through the unit weights, so as to obtain the merging features; The thirteenth unit is used to obtain the original node features of the node to be measured, and to concatenate the original node features and the merged features into an initial feature; Execution module 50 is used to obtain the time-series prediction quota corresponding to the node to be measured through the time-series model and the initial features, obtain the correction amount through the graph neural network and the initial features, and obtain the dynamic kernel amount based on the time-series prediction quota and the correction amount. The execution module 50 includes: The fourteenth unit is used to obtain the adjacency matrix corresponding to the spatial topology diagram, and to obtain the normalized matrix based on the adjacency matrix; The fifteenth unit is used to input the initial features and the normalized matrix into the first convolutional layer to obtain stage features; The sixteenth unit is used to input the stage features and the normalization matrix into the second convolutional layer to obtain the correction amount; The seventeenth unit is used to obtain the time period embedding amount and concatenate the time period embedding amount and the initial feature into a prediction vector. The eighteenth unit is used to input the prediction vector into a multilayer perceptron to obtain fusion weights, and to fuse the time-series prediction quota and the correction amount into a dynamic kernel quantity based on the fusion weights.

[0050] The present invention also provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the artificial intelligence-based dynamic water resource assessment method as described in the above technical solutions.

[0051] The present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the artificial intelligence-based dynamic water resource assessment method as described in the above technical solution.

[0052] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0053] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A dynamic water resource assessment method based on artificial intelligence, characterized in that, Includes the following steps: Obtain a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data, and household-level reported water volume data. The region to be measured is divided into several grid cells, and the multi-source heterogeneous dataset is mapped to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data. Based on real-time confidence, the remote sensing unit data, the node unit data, and the reporting unit data are fused into unit fused data. Construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and obtain the initial features of the nodes to be measured through the unit fusion data; The time-series prediction quota corresponding to the node to be measured is obtained by using the time-series model and the initial features, and the correction amount is obtained by using the graph neural network and the initial features. The dynamic verification amount is obtained based on the time-series prediction quota and the correction amount.

2. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The remote sensing inversion water volume data includes several pixel values ​​corresponding to remote sensing pixels, the IoT node flow data includes several node values ​​corresponding to IoT nodes, and the household-level reported water volume data includes several water intake values ​​corresponding to water intake points. The step of mapping the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data, and reporting unit data includes: Obtain the grid resolution of the grid cell and the remote sensing resolution of the remote sensing inverted water volume data, and compare the grid resolution with the remote sensing resolution; If the remote sensing resolution is greater than the grid resolution, then the pixel values ​​of several remote sensing pixels corresponding to the grid unit are aggregated to obtain remote sensing unit data. If the remote sensing resolution is less than the grid resolution, then the remote sensing unit data is obtained based on the grid unit and the overlapping area of ​​the remote sensing pixel corresponding to the grid unit. Based on the physical coordinates of IoT nodes and water intakes, node values ​​and water intake values ​​are mapped to corresponding grid cells to obtain node cell data and report cell data.

3. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The real-time confidence level includes remote sensing confidence level, node confidence level, and reporting confidence level. The step of fusing the remote sensing unit data, the node unit data, and the reporting unit data into unit fusion data based on the real-time confidence level includes: Obtain the data source health, spatial adaptability, and temporal trend degree corresponding to the grid cell, and obtain the remote sensing data weight, node data weight, and reported data weight based on the data source health, spatial adaptability, and temporal trend degree. The weight sum is obtained based on the remote sensing data weight, the node data weight, and the reported data weight. The remote sensing confidence, the node confidence, and the reported confidence are obtained through the remote sensing data weight, the node data weight, the reported data weight, and the weight sum. The remote sensing unit data, the node unit data, and the reported unit data are weighted and fused into unit fused data using the remote sensing confidence level, the node confidence level, and the reporting confidence level.

4. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The step of constructing a spatial topology diagram corresponding to the region to be measured, including a plurality of nodes to be measured, includes: All pipeline nodes, weir nodes, water source nodes, and water intake node nodes within the area to be measured are mapped to nodes to be measured. Based on the pipe segment connectivity and hydraulic transmission direction, the connection edges between the nodes to be measured are constructed to form a spatial topology diagram.

5. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the initial features of the node to be measured through the unit fusion data includes: Select several cells from a number of grid cells that correspond to the nodes to be measured; Obtain the Euclidean distance between the candidate cell and the node to be measured, and determine the cell weight of the candidate cell based on the Euclidean distance; The unit fusion data of several units to be used are weighted and summed using the unit weights to obtain the merging features; Obtain the original node features of the node to be measured, and concatenate the original node features and the merged features into an initial feature.

6. The method for dynamic water resource assessment based on artificial intelligence according to claim 5, characterized in that, The formula for obtaining the unit weight is: ， in, This represents the cell weight of the i-th cell corresponding to the j-th cell to be used. This represents the Euclidean distance between the i-th cell to be used and the j-th node to be measured. Indicates the power exponent. This represents the Euclidean distance between the k-th cell to be used and the j-th node to be measured. This represents the total number of units to be used corresponding to the j-th node.

7. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The graph neural network includes a first convolutional layer and a second convolutional layer. The step of obtaining the correction amount through the graph neural network and the initial features includes: Obtain the adjacency matrix corresponding to the spatial topology graph, and obtain the normalized matrix based on the adjacency matrix; The initial features and the normalized matrix are input into the first convolutional layer to obtain stage features; The stage features and the normalization matrix are input into the second convolutional layer to obtain the correction amount.

8. The method for dynamic water resource assessment based on artificial intelligence according to claim 7, characterized in that, The formula for obtaining the stage features is: ， in, Indicates stage characteristics, Represents a normalized matrix. Let represent the initial characteristics at time t. This represents the first trainable weight matrix. Represents a non-linear activation function; The formula for obtaining the correction amount is: ， in, Indicates the correction amount. This represents the second trainable weight matrix.

9. The method for dynamic water resource assessment based on artificial intelligence according to claim 1, characterized in that, The step of obtaining the dynamic verification amount based on the time-series prediction quota and the correction amount includes: Obtain the time period embedding amount, and concatenate the time period embedding amount and the initial feature into a prediction vector; The prediction vector is input into a multilayer perceptron to obtain fusion weights, and the time-series prediction quota and the correction amount are fused into a dynamic kernel amount based on the fusion weights.

10. An artificial intelligence-based dynamic water resource assessment system, applied to the artificial intelligence-based dynamic water resource assessment method as described in any one of claims 1 to 9, characterized in that, The system includes: The acquisition module is used to acquire a multi-source heterogeneous dataset corresponding to the area to be measured. The multi-source heterogeneous dataset includes remote sensing inversion water volume data, IoT node flow data, and household-level reported water volume data. The conversion module is used to divide the area to be measured into several grid cells and map the multi-source heterogeneous dataset to the grid cells to obtain remote sensing unit data, node unit data and reporting unit data. The stitching module is used to fuse the remote sensing unit data, the node unit data, and the reporting unit data into unit fused data based on real-time confidence. The construction module is used to construct a spatial topology diagram corresponding to the region to be measured, including several nodes to be measured, and to obtain the initial features of the nodes to be measured through the unit fusion data; The execution module is used to obtain the time-series prediction quota corresponding to the node to be measured through the time-series model and the initial features, obtain the correction amount through the graph neural network and the initial features, and obtain the dynamic verification amount based on the time-series prediction quota and the correction amount.

Images