Acoustic rainfall enhancement decision method, system, equipment and medium for reservoir storage
By using prediction models in the field of reservoir accumulation to predict rainfall and runoff, and calculating decision coefficients and trend curves, the problem of intelligent and automated decision-making in reservoir accumulation is solved, and the efficiency and accuracy of operations are improved , optimized water resource allocation.
Patent Information
- Application Number
- CN202411298587.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-09-18
AI Technical Summary
In the prior art, the application of sound wave rain increase technology in the field of reservoir storage has not yet achieved intelligent and automated decision-making, and it is difficult to accurately meet the reservoir's water storage needs.
By obtaining digital elevation images and meteorological data of the upstream flow area, pre-trained prediction models are used to predict rainfall and runoff, and the decision coefficient and trend curve are calculated to determine whether the sonic rain increase operation is started and continued.
It improves the efficiency and accuracy of sound wave rain-enhancing operations, optimizes water resource allocation, enhances the pertinence and flexibility of operations, and provides clear operation guidance.
Smart Images

Figure CN118822222B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a sound wave rainfall enhancement decision method, system, equipment and medium for reservoir storage, and belongs to the technical field of sound wave rainfall enhancement operations. Background Art
[0002] Against the backdrop of global warming, droughts are becoming more frequent, and key water storage facilities such as reservoirs are often faced with water shortages. As a result, the regional water supply security and the power generation efficiency of hydropower companies cannot be effectively guaranteed. Artificial rainfall is one of the important ways to solve this problem. Unlike traditional rainfall enhancement methods such as catalyst rainfall enhancement, acoustic rainfall enhancement technology emits sound waves of a specific frequency to change the cloud structure and promote the condensation of water vapor in the clouds into raindrops, achieving rainfall enhancement results that are controllable in time, location and effect, and accurately meet the water storage needs of reservoirs.
[0003] However, the research on applying acoustic rainfall enhancement technology to reservoir storage is still in the demonstration and verification stage. The decision on whether to perform the acoustic rainfall enhancement operation can only be made by the operators based on experience and other conditions such as the reservoir water level. There is no relevant research on how to combine this technology with reservoir management to realize intelligent and automated rainfall enhancement operation decision-making. Summary of the invention
[0004] In order to solve the above problems existing in the prior art, the present invention proposes an acoustic rainfall enhancement decision method, system, equipment and medium for reservoir storage.
[0005] The technical solution of the present invention is as follows:
[0006] On the one hand, the present invention proposes a method for making a decision on acoustic rainfall enhancement for increasing reservoir storage, comprising the following steps:
[0007] According to the location of the reservoir, the upstream flow-generating area A that affects the reservoir runoff and the target area B affected by the acoustic rainfall enhancement operation are selected, and area A includes area B;
[0008] A digital elevation image of area A is obtained, and the entire upstream runoff generating area is divided into a plurality of grids of the same size according to the digital elevation image of area A;
[0009] Obtain meteorological data corresponding to each grid in area A at time t; input the corresponding meteorological data into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t;
[0010] Obtain the historical runoff sequence of area A at time t, input the historical runoff sequence, the rainfall prediction results of each grid and the digital elevation image into the pre-trained second prediction model, obtain the predicted runoff sequence, and obtain the reservoir inflow runoff prediction results based on the predicted runoff sequence ;
[0011] According to the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Trend curves over time;
[0012] According to the decision coefficient Determine whether to start the sonic rain enhancement operation; and after the sonic rain enhancement operation starts, determine whether to continue the operation based on the trend curve.
[0013] As a preferred implementation, the pre-training step of the first prediction model is specifically as follows:
[0014] Obtain high-resolution meteorological data of region A at a historical moment and put it into the first training sample set, where the high-resolution meteorological data includes different types of meteorological physical variables; the number of initial field images of each meteorological physical variable is n×m;
[0015] For each training sample, a graph structure is constructed with each meteorological physical variable as a node, consisting of nodes and edges connecting node pairs;
[0016] The data initial field image of each training sample is input into the encoder module. After k layers of convolutional layers and pooling layers, different levels of spatial features are obtained for each meteorological physical variable, and the n×m data initial field image is compressed into a height of , the width is , a vector array with 128 channels, connecting the vector arrays of each meteorological physical variable to form a vector matrix ;
[0017] The vector matrix Input attention module for enhancement to form the second matrix , the second matrix The input coupling aggregation module performs graph convolution and outputs the third matrix after filtering. ;
[0018] The decoder module is used to decode the third matrix Decode to obtain the row vector corresponding to rainfall, and use super-resolution technology to increase the dimension of the row vector corresponding to rainfall to restore the predicted field rainfall data of n×m;
[0019] The loss function is calculated based on the rainfall data of the forecast field, and the model parameters of the first forecast model are dynamically optimized by combining the loss function and the Adam algorithm, and the model training is performed iteratively.
[0020] As a preferred implementation, the spatial resolution of each meteorological physical variable does not exceed the resolution of the grid division, and n=N / longitude resolution, m=M / latitude resolution; N and M are the longitude span and latitude span of region A respectively;
[0021] The attention module adopts a hybrid attention network architecture;
[0022] The coupled aggregation module adopts either the ChebNet graph convolution model or the GCN graph convolution model.
[0023] As a preferred implementation, the pre-training step of the second prediction model is specifically as follows:
[0024] The digital elevation image of area A is preprocessed to obtain the slope, flow direction, catchment area and river network information of area A. The convergence process of surface precipitation runoff is qualitatively analyzed according to the D8 unidirectional flow algorithm to form a tree-shaped flow structure.
[0025] Construct a flow graph based on the tree-like flow structure, with each grid as a node, and edges between nodes with flow relationships;
[0026] The corresponding rainfall prediction sequence is obtained based on the rainfall prediction results of the grid, and the rainfall prediction sequence and flow direction are combined. Figure 1 As with the input rainfall driving module, the rainfall prediction sequence of each grid is transformed from the time domain to the frequency domain through Hartley transform. After capturing the spatiotemporal relationship of the rainfall prediction sequence in the frequency domain, it is converted back to the time domain through inverse Hartley transform.
[0027] The rainfall time domain sequence based on the flow direction graph is convolved to obtain the spatiotemporal feature vector U of the rainfall on the flow direction graph st ;
[0028] Collect runoff series data of hydrological stations in area A during the same period as rainfall, and use the forward filling method to fill in the missing data of each station to obtain the runoff series data set;
[0029] Perform one-dimensional vector convolution on each sample in the runoff sequence data set to extract the time series feature vector U of the runoff sequence s ;
[0030] The space-time feature vector U st and the time series feature vector U s After being spliced together, they are input into the fully connected layer and the predicted runoff sequence is obtained through nonlinear transformation;
[0031] Based on the predicted runoff sequence, the loss is calculated, the combined loss and the Adam algorithm dynamically optimize the model parameters of the second prediction model, and the model training is iteratively performed.
[0032] As a preferred implementation, the graph convolution adopts any one of the ChebNet graph convolution model or the GCN graph convolution model;
[0033] The activation function of the fully connected layer uses the ReLU function.
[0034] As a preferred embodiment, the rainfall prediction results of each grid and the reservoir runoff prediction results are and reservoir operation data to calculate decision coefficients The calculation formula is:
[0035] ;
[0036] in, It is the preset acoustic rainfall enhancement ratio; is the area of a single river channel in region B; is the rainfall prediction result for each grid in area B, The storage capacity corresponding to the safe water level Z of the reservoir in the current period; is the reservoir capacity corresponding to the current water level at time t; It is the discharge capacity of the water discharge facility during the reservoir operation period; is the reservoir inflow prediction result, The calculation period.
[0037] As a preferred implementation, the step of determining whether to continue the operation according to the trend curve after the acoustic rain enhancement operation begins is specifically:
[0038] According to the trend change of the trend curve, when the decision coefficient When there is a trend towards 0, keep working. If there is a trend towards 1, consider stopping the operation.
[0039] On the other hand, the present invention also proposes an acoustic rainfall enhancement decision system for reservoir storage, comprising:
[0040] The area selection module selects the upstream flow-generating area A that affects the reservoir runoff and the target area B affected by the acoustic rainfall enhancement operation according to the location of the reservoir. Area A includes area B.
[0041] A grid division module is used to obtain a digital elevation image of area A, and divide the entire upstream runoff-generating area into a plurality of grids of the same size according to the digital elevation image of area A;
[0042] A rainfall prediction module is used to obtain the meteorological data corresponding to each grid in area A at time t; input the corresponding meteorological data into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t;
[0043] The inflow runoff prediction module is used to obtain the historical runoff sequence of area A at time t, input the historical runoff sequence, the rainfall prediction results of each grid and the digital elevation image into the pre-trained second prediction model, obtain the predicted runoff sequence, and obtain the reservoir inflow runoff prediction results based on the predicted runoff sequence ;
[0044] Auxiliary decision parameter calculation module, used to calculate the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Trend curves over time;
[0045] Decision module, used to Determine whether to start the sonic rain enhancement operation; and after the sonic rain enhancement operation starts, determine whether to continue the operation based on the trend curve.
[0046] On the other hand, the present invention further proposes an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements a method for making decisions on acoustic wave rainfall enhancement for reservoir storage as described in any embodiment of the present invention.
[0047] On the other hand, the present invention further proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements an acoustic wave rainfall enhancement decision-making method for reservoir storage as described in any embodiment of the present invention.
[0048] The present invention has the following beneficial effects:
[0049] The present invention calculates the decision coefficient and its related trend curve by comprehensively calculating rainfall prediction and runoff prediction. The decision coefficient characterizes the carrying capacity of the reservoir capacity for the runoff generated by the original rainfall and the net rainfall increase caused by acoustic rainfall enhancement under the condition of the safe operating water level of the reservoir. The decision coefficient and trend curve provide clear operation guidance such as the feasibility of the operation and the decision on the timing of the operation for the operators. This data-driven decision-making method not only improves the efficiency and accuracy of the acoustic rainfall enhancement operation, but also optimizes the allocation of water resources and enhances the pertinence and flexibility of the operation. BRIEF DESCRIPTION OF THE DRAWINGS
[0050] Figure 1 This is a schematic diagram of the method steps of Embodiment 1 of the present invention;
[0051] Figure 2 This is a schematic diagram of a specific process of Embodiment 1 of the present invention;
[0052] Figure 3An example diagram of a digital elevation image provided by an embodiment of the present invention;
[0053] Figure 4 An example diagram of grid division and flow direction provided in an embodiment of the present invention;
[0054] Figure 5 This is an example diagram of the meteorological physical variable graph structure provided in an embodiment of the present invention;
[0055] Figure 6 Schematic diagram of the process of optimizing a gated recurrent neural network using the Grey Wolf Algorithm in an embodiment of the present invention. DETAILED DESCRIPTION
[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0057] It should be understood that the step numbers used in this document are only for convenience of description and are not intended to limit the order in which the steps are executed.
[0058] It should be understood that the terms used in the present specification are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0059] The terms “include” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.
[0060] The term "and / or" means and includes any and all possible combinations of one or more of the associated listed items.
[0061] Embodiment 1:
[0062] In existing acoustic wave operations, operators often make decisions on operation conditions based on water level data and personal experience. However, it is difficult for operators to fully consider the reservoir's rainfall capacity in different seasons, and it is impossible to combine acoustic wave rainfall enhancement technology with reservoir management to achieve intelligent and automated rainfall enhancement operation decisions.
[0063] See also Figure 1 and Figure 2In order to solve the above problems, this embodiment proposes a sound wave rainfall enhancement decision method for reservoir storage, which specifically includes the following steps:
[0064] S100, selecting an upstream flow-generating area A that affects the reservoir runoff and a target area B affected by the acoustic rainfall enhancement operation according to the location of the reservoir, wherein area A includes area B;
[0065] S200, obtaining a digital elevation image (DEM image) of area A. The digital elevation image is as follows: Figure 3 According to the digital elevation image of area A, the entire upstream runoff area is divided into multiple grids of the same size, as shown in Figure 4 As shown, Figure 4 Point G in the middle is the reservoir inlet point.
[0066] S300, obtaining meteorological data corresponding to each grid in region A at time t, including evaporation, surface temperature, atmospheric pressure, humidity, atmospheric precipitable water, wind speed and wind direction, etc.; inputting the above meteorological data into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t;
[0067] S400, obtaining the historical runoff sequence of region A at time t, inputting the historical runoff sequence, the rainfall prediction result of each grid and the digital elevation image into the pre-trained second prediction model, obtaining the predicted runoff sequence, and obtaining the reservoir inflow according to the predicted runoff sequence, that is, Figure 4 Runoff prediction results for the middle g section ;
[0068] S500, based on the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Varying over time Trend curve;
[0069] In this embodiment, the decision coefficient The specific calculation formula is:
[0070] ;
[0071] in, It is the preset acoustic wave rainfall enhancement ratio, which can be selected according to the specific situation; is the area of a single river channel in region B (m 2 ); is the rainfall prediction result for each grid of each river in region B (m), is the storage capacity corresponding to the safe water level Z of the reservoir in the current period (m 3), where Z can be the flood limit water level, design water level or normal high water level, but cannot be the design flood level, verification flood level or dam crest elevation water level; is the corresponding reservoir capacity of the current water level at time t (m 3 ); is the discharge capacity of the discharge facility during reservoir operation (m 3 / s); is the reservoir runoff prediction result (m 3 / s), is the calculation period (s).
[0072] S600, according to the decision coefficient Determine whether to start the sonic rain enhancement operation; and after the sonic rain enhancement operation starts, determine whether to continue the operation based on the trend curve.
[0073] This embodiment calculates the decision coefficient of acoustic rainfall enhancement operation based on the meteorological data of the runoff-producing area of the basin, the digital elevation information of the basin, the historical runoff sequence and the reservoir operation data. The decision coefficient represents the carrying capacity of the reservoir capacity for the runoff generated by the original rainfall and the net rainfall increase by acoustic rainfall enhancement under the condition of the safe operating water level of the reservoir; the decision coefficient that can support the start of acoustic rainfall enhancement operation It should be in the range of 0 to 1. The closer it is to 0, the stronger the reservoir's current carrying capacity for the increased rainfall, and the greater the necessity of acoustic rainfall enhancement operations. Conversely, the closer it is to 1, the weaker the reservoir's carrying capacity for the increased rainfall, and the less necessary the acoustic rainfall enhancement operations.
[0074] When the acoustic precipitation enhancement operation begins, the decision coefficient can be updated every fixed period of time. curve; according to The curve trend makes a decision on whether to continue the operation. If there is a trend towards 0, you can consider continuing the operation. If there is a trend towards 1, the operation should be stopped. The proposed method can provide operators with an accurate basis for decision-making during rainfall enhancement operations.
[0075] Since this embodiment needs to calculate the decision coefficient more accurately, it is necessary to obtain a more accurate rainfall forecast within the influence range of the acoustic rain enhancement technology within a certain visible period, and due to the limitation of resolution and the difficulty of characterizing convection and cloud microphysics, numerical weather forecast models often find it difficult to provide high-resolution precipitation forecasts. This embodiment uses a data-driven method to obtain a sophisticated and spatially consistent rainfall prediction model from high-resolution meteorological data.
[0076] In one implementation of this embodiment, the pre-training step of the first prediction model is specifically:
[0077] S301, obtaining high-resolution meteorological data of region A at a historical moment, and putting it into a first training sample set, wherein the high-resolution meteorological data includes different types of meteorological physical variables, and the different types of meteorological physical variables include rainfall, evaporation, surface temperature, atmospheric pressure, humidity, atmospheric precipitable water, wind speed and wind direction; according to the spatial resolution interpolation and the longitude (N) and latitude (M) spanned by region A and the spatial resolution, the number of initial field pictures of data for each meteorological physical variable is n×m;
[0078] S302. Divide the first training sample set into a training set, a validation set and a test set in a ratio of 7:1:2, wherein the training set is used for training the first prediction model, the validation set is used for testing and iteratively optimizing the effect of the model during the training process, and the test set is used for evaluating the prediction effect.
[0079] S303. For each training sample, a graph structure D is constructed with eight physical variables as nodes and nodes and edges connecting node pairs. Specifically, Figure 5 As shown, each node in the graph expresses a meteorological physical variable of a data type, and the edges connecting the nodes reflect a fuzzy relationship extracted from the laws of physics.
[0080] S304, input the data initial field image of each training sample into the encoder module, obtain different levels of spatial features for each meteorological physical variable after passing through k layers of convolution layers and pooling layers, and compress the n×m data initial field image into a height of , the width is , a vector array with 128 channels, connecting the vector arrays of 8 meteorological physical variables to form an 8×( × ×128) vector matrix , each row represents a meteorological physical variable.
[0081] S305: transform the vector matrix Input attention module for enhancement to form the second matrix , the second matrix The input coupling aggregation module performs graph convolution. The object of graph convolution is a graph structure D consisting of eight physical variables as nodes and nodes and edges connecting node pairs. The relationship characteristics of each meteorological physical variable are introduced through the attention module and graph convolution, and the third matrix is output after filtering. Among them, the role of the attention module is to strengthen the weight of the edges between nodes in the graph structure. The weight of the edges is determined by the physical relationship between the meteorological physical variables of different nodes; the role of the coupling aggregation module is to map the vector from the graph (spatial domain) to the spectrum domain, and the role of the filter is to aggregate the neighbor features of the node and dynamically update the representation of the node in the iteration.
[0082] S306, finally, the third matrix is decoded by the decoder module Decode to obtain the row vector corresponding to rainfall, and use the super-resolution technology PixelShuffle to increase the dimension of the row vector corresponding to rainfall, and restore it to the predicted field rainfall data of n×m;
[0083] S307. Calculate the loss function based on the rainfall data of the forecast field. The loss function can use the Dice loss function, the combined loss function and the Adam algorithm to dynamically optimize the model parameters of the first forecast model. Perform model training iteratively. Verify the model through the model verification set during the iteration process. End the iteration when the calculated loss is the smallest or the maximum number of iterations is reached.
[0084] In step S301, the spatial resolution of each meteorological physical variable does not exceed, for example, Figure 3 The grid resolution shown is n=N / longitude resolution, m=M / latitude resolution.
[0085] In step S304, the convolution layer of the encoder uses the Conv2D function, and the pooling layer uses the MaxPooling2D function.
[0086] In step S305, the attention module adopts the Transformer (HAT) hybrid attention network architecture.
[0087] In step S305, the coupled aggregation module can use the ChebNet graph convolution model to approximate the graph convolution kernel in the frequency domain through Chebyshev expansion, thereby realizing the convolution operation on the graph structure data. The process can be expressed as:
[0088] , where k is the order of the convolution kernel and K is the maximum order of the convolution kernel. is the scaled Laplacian matrix, is a Chebyshev polynomial, are model parameters, is the feature vector of the node.
[0089] In step S305, the coupled aggregation module can also adopt the GCN graph convolution model, and only use the first-order Chebyshev polynomial to approximate the graph convolution, which can also implement the convolution operation on the graph structure data. The process can be expressed as:
[0090] ,in is the adjacency matrix with self-connection added, yes The degree matrix of is the feature matrix of the node, is a trainable weight matrix, is the output feature matrix after convolution.
[0091] In one implementation of this embodiment, the pre-training step of the second prediction model is specifically:
[0092] S401. Preprocess the digital elevation image of area A to obtain information such as the slope, flow direction, catchment area, and river network of area A. Assuming that rain falls on a grid in the terrain, the water in the grid will flow to the grid with the lowest terrain among the surrounding 8 grids. According to the D8 unidirectional flow algorithm, the convergence process of surface precipitation runoff is qualitatively analyzed to form a tree-shaped flow structure pointing to the outlet grid g, such as Figure 4 shown.
[0093] S402, constructing a flow graph according to the tree-shaped flow structure, with each grid as a node, and edges between nodes with flow relationships. By default, the flow rate of each edge is the same, that is, the weight of the edge is 1.
[0094] S403, obtain the corresponding rainfall prediction sequence according to the rainfall prediction result of the grid, and combine the rainfall prediction sequence and the flow direction Figure 1 As with the input rainfall driving module, the rainfall prediction sequence of each grid is transformed from the time domain to the frequency domain through the Hartley transform. After capturing the spatiotemporal relationship of the rainfall prediction sequence in the frequency domain, it is converted back to the time domain through the inverse Hartley transform.
[0095] S404: Perform graph convolution on the rainfall time-domain sequence based on the flow direction graph to obtain the spatiotemporal feature vector U of the rainfall on the flow direction graph. st ;
[0096] S405. Collect runoff sequence data of hydrological stations in area A during the same period as the rainfall, and use the forward filling method to fill in the missing data of each station to obtain a runoff sequence data set; divide the runoff sequence data set as a whole into a training set, a validation set, and a test set according to a ratio of 7:1:2, wherein the training set is used for training the first prediction model, the validation set is used for testing and iteratively optimizing the effect of the model during the training process, and the test set is used to evaluate the prediction effect.
[0097] S406: Perform one-dimensional vector convolution on each sample in the runoff sequence data set to extract the time series feature vector U of the runoff sequence. s .
[0098] S407, the spatiotemporal feature vector U st and the time series feature vector U s After being spliced together, they are input into the fully connected layer and the predicted runoff sequence is obtained through nonlinear transformation;
[0099] S408. Calculate the loss based on the predicted runoff sequence. The loss function may use the MSE loss function, the combined loss function, and the Adam algorithm to dynamically optimize the model parameters of the second prediction model. Perform model training iteratively. Verify the model with the model verification set during the iteration process. End the iteration when the calculated loss is the minimum or the maximum number of iterations is reached.
[0100] In steps SS403 and S406, the acquisition of spatiotemporal feature vectors and time series feature vectors is achieved through the gated recurrent unit GWO-GRU neural network optimized by the Grey Wolf Algorithm. The gated recurrent neural network (GRU) is an effective feature extraction tool that solves the gradient vanishing and gradient exploding problems in traditional recurrent neural networks (RNNs) by introducing a gating mechanism, thereby effectively capturing long-term dependencies in sequence data. GRU achieves feature extraction of sequence data through two key components: the update gate and the reset gate. The update gate controls whether the current state should be retained or updated, while the reset gate determines whether the previous information should be discarded from the current state. This mechanism enables GRU to maintain the integrity of information when processing long sequences, avoiding gradient problems that may occur in traditional RNNs.
[0101] like Figure 5 As shown in the figure, the optimal network hyperparameter values of GRU in the two feature vector acquisition modules are optimized and designed respectively by the Gray Wolf Optimization Algorithm, including the initial learning rate, the number of hidden layer neurons, the batch size and the number of training iterations, so as to improve the nonlinear fitting ability of GRU and suppress the overfitting phenomenon.
[0102] In step S404, the graph convolution adopts the ChebNet graph convolution model, and approximates the graph convolution kernel in the frequency domain through Chebyshev expansion, thereby realizing the convolution operation on the graph structure data. The process can be expressed as:
[0103] , where k is the order of the convolution kernel and K is the maximum order of the convolution kernel. is the scaled Laplacian matrix, are Chebyshev polynomials, are model parameters.
[0104] In step S404, the graph convolution can also adopt the GCN graph convolution model, and only use the first-order Chebyshev polynomial to approximate the graph convolution, which can also implement the convolution operation on the graph structure data. The process can be expressed as:
[0105] ,in is the adjacency matrix with self-connection added, yes The degree matrix of is the feature matrix of the node, is a trainable weight matrix.
[0106] In step S407, the activation function of the fully connected layer adopts the ReLU function.
[0107] Embodiment 2:
[0108] This embodiment provides an acoustic wave rainfall enhancement decision system for increasing reservoir storage, including:
[0109] The area selection module selects the upstream flow generation area A that affects the reservoir runoff and the target area B affected by the acoustic rainfall enhancement operation according to the location of the reservoir, and the area A includes the area B; this module is used to implement the function of step S100 in the first embodiment, which will not be repeated here;
[0110] A grid division module is used to obtain a digital elevation image of area A, and divide the entire upstream runoff generating area into a plurality of grids of the same size according to the digital elevation image of area A; this module is used to implement the function of step S200 in the first embodiment, and will not be described in detail here;
[0111] A rainfall prediction module is used to obtain the meteorological data corresponding to each grid in area A at time t; the corresponding meteorological data is input into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t; this module is used to implement the function of step S300 in embodiment 1, which will not be repeated here;
[0112] The inflow runoff prediction module is used to obtain the historical runoff sequence of area A at time t, input the historical runoff sequence, the rainfall prediction results of each grid and the digital elevation image into the pre-trained second prediction model, obtain the predicted runoff sequence, and obtain the reservoir inflow runoff prediction results based on the predicted runoff sequence ; This module is used to implement the function of step S400 in the first embodiment, which will not be described in detail here;
[0113] Auxiliary decision parameter calculation module, used to calculate the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Trend curve changing over time; this module is used to implement the function of step S500 in the first embodiment, which will not be described in detail here;
[0114] Decision module, used to Determine whether to start the acoustic rain enhancement operation; and after the acoustic rain enhancement operation starts, determine whether to continue the operation based on the trend curve; this module is used to implement the function of step S600 in embodiment 1, which will not be repeated here.
[0115] Embodiment three:
[0116] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, an acoustic wave rainfall enhancement decision method for reservoir storage enhancement as described in any embodiment of the present invention is implemented.
[0117] Embodiment 4:
[0118] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for making decisions on acoustic rainfall enhancement for reservoir storage as described in any embodiment of the present invention is implemented.
[0119] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.
[0120] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0121] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0122] In several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), disk or optical disk, and other media that can store program codes.
[0123] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A method for making decisions on acoustic rainfall enhancement for increasing reservoir storage, characterized in that: The following steps are involved: Based on the location of the reservoir, the upstream runoff area A that affects the reservoir runoff and the target area B affected by the acoustic rainfall enhancement operation are selected. Area A includes area B. Obtain a digital elevation image of area A, and divide the entire upstream runoff generating area into multiple grids of equal size based on the digital elevation image of area A; Obtain meteorological data corresponding to each grid in area A at time t; input the corresponding meteorological data into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t; Obtain the historical runoff sequence of area A at time t, input the historical runoff sequence, the rainfall prediction results of each grid and the digital elevation image into the pre-trained second prediction model, obtain the predicted runoff sequence, and obtain the reservoir inflow runoff prediction results based on the predicted runoff sequence ; According to the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Trend curves over time; According to the decision coefficient Determine whether to start the acoustic precipitation enhancement operation; and determine whether to continue the operation based on the trend curve after the acoustic precipitation enhancement operation begins; The pre-training steps of the first prediction model are specifically as follows: Obtain high-resolution meteorological data of region A at a historical moment and put it into the first training sample set. The high-resolution meteorological data includes different types of meteorological physical variables. The number of initial field images of each meteorological physical variable is n×m. For each training sample, a graph structure is constructed with each meteorological physical variable as a node, consisting of nodes and edges connecting node pairs; The data initial field image of each training sample is input into the encoder module, and each meteorological physical variable is passed through k layers of convolution and pooling layers to obtain spatial features of different levels, and the data initial field image of n×m is compressed into a height of , with a width of , a vector array with 128 channels, connecting the vector arrays of each meteorological physical variable to form a vector matrix ; The vector matrix Input attention module for enhancement to form the second matrix , the second matrix The input coupling aggregation module performs graph convolution and outputs the third matrix after filtering. ; The third matrix is processed by the decoder module Decode to obtain the row vector corresponding to rainfall, and use super-resolution technology to increase the dimension of the row vector corresponding to rainfall to restore the predicted field rainfall data of n×m; The loss function is calculated based on the rainfall data of the forecast field. The model parameters of the first forecast model are dynamically optimized by combining the loss function and the Adam algorithm, and the model training is performed iteratively.
2. The acoustic wave rainfall enhancement decision-making method for reservoir storage according to claim 1, characterized in that: The spatial resolution of each meteorological physical variable shall not exceed the resolution of the grid division, and n=N / longitude resolution, m=M / latitude resolution; N and M are the longitude span and latitude span of area A respectively; The attention module adopts a hybrid attention network architecture; The coupled aggregation module adopts either the ChebNet graph convolution model or the GCN graph convolution model.
3. The acoustic wave rainfall enhancement decision-making method for reservoir storage according to claim 1 is characterized in that: The pre-training steps of the second prediction model are specifically as follows: The digital elevation image of area A was preprocessed to obtain information on its slope, flow direction, catchment area, and river network. The D8 unidirectional flow algorithm was used to qualitatively analyze the convergence of surface precipitation and runoff, forming a tree-like flow structure. Construct a flow graph based on the tree-like flow structure, with each grid as a node and edges between nodes with flow relationships; The corresponding rainfall prediction sequence is obtained based on the rainfall amount prediction results of the grid. The rainfall prediction sequence and the flow direction map are input into the rainfall driving module. The rainfall prediction sequence of each grid is transformed from the time domain to the frequency domain through Hartley transform. After capturing the spatiotemporal relationship of the rainfall prediction sequence in the frequency domain, it is converted back to the time domain through inverse Hartley transform. The rainfall time domain sequence based on the flow direction graph is convolved to obtain the spatiotemporal feature vector U of the rainfall amount on the flow direction graph. st ; Collect runoff series data from hydrological stations in region A during the same period as rainfall, and use the forward filling method to fill in the missing data at each station to obtain a runoff series dataset; Perform one-dimensional vector convolution on each sample in the runoff sequence dataset to extract the time series feature vector U of the runoff sequence s ; The spatiotemporal feature vector U st and the time series feature vector U s After being spliced together, they are input into the fully connected layer and subjected to nonlinear transformation to obtain the predicted runoff sequence; Based on the loss calculation of the predicted runoff sequence, the combined loss and Adam algorithm dynamically optimize the model parameters of the second prediction model, and iteratively train the model.
4. The acoustic wave rainfall enhancement decision-making method for reservoir storage according to claim 3 is characterized by: The graph convolution adopts either the ChebNet graph convolution model or the GCN graph convolution model; The activation function of the fully connected layer uses the ReLU function.
5. The acoustic wave rainfall enhancement decision-making method for reservoir storage according to claim 1 is characterized in that: The rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients The calculation formula is: ; in, It is the preset acoustic rainfall enhancement ratio; is the area of a single river channel in region B; is the rainfall prediction result for each grid in region B, The storage capacity corresponding to the safe water level Z of the reservoir in the current period; is the reservoir capacity corresponding to the current water level at time t; The discharge capacity of the water discharge facilities during the reservoir operation period; is the reservoir inflow prediction result, The calculation period.
6. The acoustic wave rainfall enhancement decision-making method for reservoir storage according to claim 5, characterized in that: The steps of determining whether to continue the operation according to the trend curve after the acoustic rain enhancement operation begins are specifically: According to the trend change of the trend curve, when the decision coefficient When there is a trend towards 0, keep working. If there is a trend towards 1, consider stopping the operation.
7. An acoustic rainfall enhancement decision system for reservoir storage, characterized in that: include: The region selection module selects the upstream runoff region A that affects the reservoir runoff and the target region B affected by the acoustic rainfall enhancement operation based on the location of the reservoir. Region A includes region B. A grid division module is used to obtain a digital elevation image of area A and divide the entire upstream runoff generating area into multiple grids of the same size based on the digital elevation image of area A; The rainfall prediction module is used to obtain the meteorological data corresponding to each grid in area A at time t; input the corresponding meteorological data into the pre-trained first prediction model to obtain the rainfall prediction result of each grid at time t; The inflow runoff prediction module is used to obtain the historical runoff sequence of area A at time t, input the historical runoff sequence, the rainfall prediction results of each grid and the digital elevation image into the pre-trained second prediction model, obtain the predicted runoff sequence, and obtain the reservoir inflow runoff prediction results based on the predicted runoff sequence ; Auxiliary decision parameter calculation module, used to calculate the rainfall prediction results of each grid and the reservoir runoff prediction results and reservoir operation data to calculate decision coefficients and the decision coefficient Trend curves over time; Decision module, used to determine the Determine whether to start the acoustic precipitation enhancement operation; and determine whether to continue the operation based on the trend curve after the acoustic precipitation enhancement operation begins; The pre-training steps of the first prediction model are specifically as follows: Obtain high-resolution meteorological data of region A at a historical moment and put it into the first training sample set. The high-resolution meteorological data includes different types of meteorological physical variables. The number of initial field images of each meteorological physical variable is n×m. For each training sample, a graph structure is constructed with each meteorological physical variable as a node, consisting of nodes and edges connecting node pairs; The data initial field image of each training sample is input into the encoder module, and each meteorological physical variable is passed through k layers of convolution and pooling layers to obtain spatial features of different levels, and the data initial field image of n×m is compressed into a height of , with a width of , a vector array with 128 channels, connecting the vector arrays of each meteorological physical variable to form a vector matrix ; The vector matrix Input attention module for enhancement to form the second matrix , the second matrix The input coupling aggregation module performs graph convolution and outputs the third matrix after filtering. ; The third matrix is processed by the decoder module Decode to obtain the row vector corresponding to rainfall, and use super-resolution technology to increase the dimension of the row vector corresponding to rainfall to restore the predicted field rainfall data of n×m; The loss function is calculated based on the rainfall data of the forecast field. The model parameters of the first forecast model are dynamically optimized by combining the loss function and the Adam algorithm, and the model training is performed iteratively.
8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method for making decisions on acoustic rainfall enhancement for reservoir storage as described in any one of claims 1 to 6 is implemented.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, an acoustic wave rainfall enhancement decision method for reservoir storage is implemented as described in any one of claims 1 to 6.
