Nothing Special   »   [go: up one dir, main page]

CN117634554A - Space-time deep learning runoff prediction method and system based on spatial information fusion - Google Patents

Space-time deep learning runoff prediction method and system based on spatial information fusion Download PDF

Info

Publication number
CN117634554A
CN117634554A CN202311721815.4A CN202311721815A CN117634554A CN 117634554 A CN117634554 A CN 117634554A CN 202311721815 A CN202311721815 A CN 202311721815A CN 117634554 A CN117634554 A CN 117634554A
Authority
CN
China
Prior art keywords
data
runoff
space
spatial
graph
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311721815.4A
Other languages
Chinese (zh)
Inventor
邓梁堃
张翔
刘浩源
陶士勇
赵烨
刘书颖
吴可怡
李宜伦
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuhan University WHU
Original Assignee
Wuhan University WHU
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuhan University WHU filed Critical Wuhan University WHU
Priority to CN202311721815.4A priority Critical patent/CN117634554A/en
Publication of CN117634554A publication Critical patent/CN117634554A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Business, Economics & Management (AREA)
  • Tourism & Hospitality (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Economics (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a space-time deep learning runoff prediction method and a space-time deep learning runoff prediction system based on spatial information fusion, wherein the space-time deep learning runoff prediction method comprises the following steps: acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted; extracting European spatial characteristic information in grid point precipitation data and potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity and historical runoff monitoring data of each station by adopting a graph neural network, and splicing to form spatial correlation data; processing the space correlation data by adopting a long-term and short-term memory network to obtain space-time correlation data; and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data. The method can effectively improve the accuracy of multi-step prediction of multi-station runoffs in the river basin, and provides basic water situation prediction data for relieving flood disasters, treating water ecological water environment and developing and utilizing water resources.

Description

Space-time deep learning runoff prediction method and system based on spatial information fusion
Technical Field
The invention relates to the technical field of hydrologic prediction, in particular to a space-time deep learning runoff prediction method and system based on spatial information fusion.
Background
The accurate river runoff prediction method can provide effective technical support for relieving flood disasters, treating water ecological water environment and developing and utilizing water resources. The river flow is the result of the combined action of natural factors and human factors, and is very easily influenced by natural condition changes and human activities. Reliable radial flow prediction has long been a difficult business requirement due to its high complexity, strong nonlinearity and non-stationarity of complex features.
Many methods have been proposed in the industry to make accurate radial flow predictions. These methods can be largely divided into two categories, physical process-based and data-driven-based. Hydrologic models based on physical processes, including lumped and distributed models, are being used in the field of river runoff prediction. However, hydrologic models based on physical processes still have certain limitations in practical business applications, subject to errors due to their stringent data input requirements and simplifying assumptions. Emerging models based on data driving, particularly deep learning techniques, have been introduced in the field of river runoff prediction in recent years as additions or alternatives to traditional hydrologic models. They establish an empirical relationship between input and output without having to consider the physical process of hydrologic cycle. Deep learning techniques have so far shown a powerful capability in river runoff prediction. In particular, long-short term memory network model (LSTM) based predictive effects have proven to be superior to physical process based hydrologic models on the regional spatial and time-of-day scales.
Although deep learning-based models have made much progress in the field of river runoff prediction, there are still some shortcomings. First, the prior deep learning model tends to use basin-averaged meteorological hydrologic elements that ignore spatial features as inputs. However, hydrologic runoffs are affected by weather and geophysical spatial correlations, exhibiting complex spatial variability. There are still few methods to explore spatial variability using deep learning models, as compared to the substantial progress made by distributed hydrologic models based on physical processes. Today, spatially distributed data sets, such as satellite weather products and re-analysis products, have been widely used to describe spatial heterogeneity, providing a more potentially valuable learning sample for deep learning radial flow prediction models, which helps achieve higher radial flow prediction accuracy. Therefore, the gravity center is of great significance in the development of a space-time deep learning runoff prediction method. Secondly, the prior space-time deep learning radial flow prediction method only considers one type of spatial information, namely European or non-European spatial information. European spatial information generally comes from site interpolation, satellite remote sensing and reanalysis products, which have regular lattice point structures (namely, data are distributed on longitude and latitude coordinates with fixed intervals), and can directly describe the spatial variability of meteorological and hydrologic characteristics. non-European spatial information, such as the topological connectivity of a river network, differs from regular European spatial information in that its structure depends on the actual physical river network structure and appears as an irregular data form. It is another dominant factor affecting river runoff variation in addition to meteorological factors (e.g., precipitation, air temperature, etc.). The method describes the relativity of upstream and downstream runoff monitoring stations, and can quantitatively describe the influence of the transfer effect of the runoff of the upstream monitoring stations on the runoff of the downstream monitoring stations through rivers. However, the prior space-time deep learning runoff prediction method only considers one spatial information, and omits the potential of fusion of multiple types of spatial information on improvement of river runoff prediction effect.
Disclosure of Invention
The invention provides a space-time deep learning runoff prediction method and a space-time deep learning runoff prediction system based on spatial information fusion, which are used for solving the defects in the prior art.
In a first aspect, the present invention provides a space-time deep learning runoff prediction method based on spatial information fusion, including:
acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted;
extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data;
processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data;
and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
According to the space information fusion-based space-time deep learning runoff prediction method provided by the invention, grid point precipitation data, potential evaporation meteorological data, topological connectivity of each site and historical runoff monitoring data of river network runoff to be predicted are obtained, and the space information fusion-based space-time deep learning runoff prediction method comprises the following steps:
collecting an original data set of all runoff monitoring stations in the river network runoff to be predicted;
and carrying out data cleaning, data normalization and sliding window preprocessing on the original data set to obtain the grid point precipitation data, the potential evaporation meteorological data, the topological connectivity of each site and the historical runoff monitoring data.
According to the space information fusion-based space-time deep learning runoff prediction method provided by the invention, the parallel convolutional neural network is adopted to extract European space characteristic information in the grid point precipitation data and the potential evaporation meteorological data, and the method comprises the following steps:
determining that a single convolutional neural network comprises a two-dimensional convolutional layer, a pooling layer, a modified linear unit ReLU activation layer and a convolutional attention mechanism module CBAM which are sequentially connected;
the single convolutional neural network extracting the European spatial feature information of the grid point precipitation data or the potential evaporation meteorological data comprises the following steps:
where i, j and l represent the ith input neuron, the jth output neuron and the ith block; m represents the number of input channels;representing the characteristics of the input; />Representing a convolution kernel; * Representing a convolution operation; />A learnable bias coefficient representing a j-th output in the i-th block; />Representing the j-th output result in the l-th block after a series of operations.
According to the space-time deep learning runoff prediction method based on spatial information fusion, provided by the invention, the topological connectivity of each site and the non-European spatial characteristic information of the historical runoff monitoring data are extracted by adopting a graph neural network, and the space-time deep learning runoff prediction method comprises the following steps:
determining a Graph neural network comprising a Graph generation layer and a Graph roll-up layer, outputting a static Graph and an adaptive Graph by the Graph generation layer, the static Graph and the adaptive Graph each having a structure graph= (V, E, a), wherein V represents a set of vertices, i.e. all runoff monitoring stations in the river network, E represents a set of directed edges, i.e. representing spatial topological connectivity between upstream to downstream stations,the method comprises the steps that the method is an adjacency matrix, static or self-adaptive adjacency weights of connecting vertexes along directed edges are represented, N represents the number of runoff monitoring stations, V and E of a static graph and a self-adaptive graph are consistent, and A is different;
and inputting the static diagram and the self-adaptive diagram to the diagram volume lamination layer to obtain the non-European space characteristic information.
According to the space-time deep learning radial flow prediction method based on spatial information fusion, the adjacency matrix of the static diagram is A st ＝(a st ) N×N Wherein:
according to the space-time deep learning runoff prediction method based on spatial information fusion, the self-adaptive graph is connected with a graph annotation meaning network GAT, and the GAT comprises a sharing operator
Wherein e ij Is the attention coefficient, characterizes the importance of the feature in site i to site j,b is the input feature number, B' is the output feature number,>representing the characteristics of the input. T Represents the operation of the transposition, the i represents a merge operation;
attention coefficient e using softmax function ij And (3) performing standardization:
obtaining an adaptive weighted adjacency matrix A sa ＝(α dyij ) N×N 。
According to the space-time deep learning runoff prediction method based on spatial information fusion, the static diagram and the self-adaptive diagram are input into the diagram volume lamination layer, and the non-European space characteristic information is obtained, and the method comprises the following steps:
wherein L is k Is the normalized adjacency matrix in the kth diffusion of the graph roll stack, l=a/rowsum (a), W k The parameter to be learned in the kth diffusion is Z, and the Z is non-European space characteristic information which is output after being learned by a graph convolution layer.
According to the space information fusion-based space-time deep learning runoff prediction method provided by the invention, the space correlation data is processed by adopting a long-term and short-term memory network, and the space-time correlation data is obtained, and the space-time deep learning runoff prediction method comprises the following steps:
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
c′ t ＝tanh(W c ·[h t-1 ,x t ]+b c )
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
c t ＝f t ⊙c t-1 +i t ⊙c′ t
o t ＝σ(W o ·[h t-1 ,x t ]+b o )
h t ＝o t ⊙tanh(c t )
wherein i is t ,f t And o t For input door, forget door and output door, c' t Representing potential update units, c t Representing a memory cell, h t Indicating the hidden state, σ is a sigmoid activation function, as well as the matrix point multiplication operation, W and b respectively indicate the corresponding learnable weight matrix and bias term.
In a second aspect, the present invention further provides a space-time deep learning runoff prediction system based on spatial information fusion, including:
the acquisition module is used for acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted;
the space extraction module is used for extracting European space characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European space characteristic information of the topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European space characteristic information and the non-European space characteristic information to form space correlation data;
the time extraction module is used for processing the space correlation data by adopting a long-term and short-term memory network to acquire space-time correlation data;
and the output module is used for outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer.
In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements any one of the spatial information fusion-based spatio-temporal deep learning runoff prediction methods described above when the processor executes the program. .
The space information fusion-based space-time deep learning runoff prediction method and system can effectively improve the accuracy of multi-step prediction of multi-station runoffs in the river basin, and provide basic water situation prediction data for relieving flood disasters, treating water ecological water environment and developing and utilizing water resources.
Drawings
In order to more clearly illustrate the invention or the technical solutions of the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow diagram of a spatial information fusion-based spatial-temporal deep learning runoff prediction method provided by the invention;
FIG. 2 is a model framework diagram of a spatial information fusion-based space-time deep learning runoff prediction method provided by the invention;
FIG. 3 is a schematic structural diagram of a spatial information fusion-based spatial-temporal deep learning runoff prediction system provided by the invention;
fig. 4 is a schematic structural diagram of an electronic device provided by the present invention.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Aiming at a plurality of limitations existing in the prior art, the invention creatively provides a space-time deep learning runoff prediction method based on European and non-European spatial information fusion, so as to improve the accuracy of river runoff prediction by fusing European and non-European spatial information. The model framework mainly comprises four modules, namely a model input module, a space information extraction module, a time feature extraction module and a model result output module.
Fig. 1 is a schematic flow chart of a spatial information fusion-based space-time deep learning runoff prediction method, as shown in fig. 1, including:
step 100: acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted;
step 200: extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data;
step 300: processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data;
step 400: and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
Specifically, in the model framework shown in fig. 2, the model input data includes euclidean space information (grid-point precipitation and potential evaporation meteorological data) and non-euclidean space information (topological connectivity of river network runoff monitoring sites and historical runoff monitoring data of each site), and the input data is subjected to data cleaning, data normalization and preprocessing of sliding windows.
The spatial information extraction module comprises a Convolutional Neural Network (CNN) and a Graph Neural Network (GNN) which are respectively used for learning the spatial characteristics of European spatial information and non-European spatial information.
For European spatial information, i.e. grid-point precipitation and potential evaporation input data, the module employs two parallel CNN structures to automatically learn their spatial features. For each CNN, it is formed by sequentially connecting a plurality of structures. In each block, it contains in sequence a two-dimensional convolution layer, a pooling layer, a ReLU nonlinear activation layer, and a attention mechanism (CBAM) suitable for extraction of critical features of the european spatial information. The operation rule of one CNN module is as follows:
where i, j and l represent the ith input neuron, the jth output neuron and the ith block; m represents the number of input channels;representing the characteristics of the input; />Representing a convolution kernel; * Representing a convolution operation; />A learnable bias coefficient representing a j-th output in the i-th block; />Representing the j-th output result in the l-th block after a series of operations.
For non-European spatial information, namely the topological connectivity of river network runoff monitoring sites and the corresponding historical runoff monitoring data of each site, the module adopts GNN to learn the spatial characteristics of the non-European spatial information. The traffic at the downstream site receives traffic from the upstream site, both of which create a tight spatial topological connectivity along the river channel. The GNN realizes the characterization of the topological connectivity of the sites and the quantification of the mutual influence of the runoffs of the sites by constructing a graph structure. The GNN structure is composed of a graph generation layer and a graph roll stack layer. In the graph generation layer, the GNN module generates static and adaptive graphs to describe the static and adaptive nature of the topology connectivity. They all have a similar structure graph= (V, E, a), where V is a set of vertices corresponding to runoff monitoring stations in the river network, E is a set of directed edges, revealingSpatial topological connectivity between upstream to downstream stations,(N represents the number of runoff monitoring stations) adjacency matrix representing static or adaptive adjacency weights of connected vertices along directed edges. For both constructed plots, their V and E are uniform, but with different a.
For static diagrams, it is predefined based on the actual relative position of the runoff monitoring station along the river channel. Since the locations of the stations are relatively stable over a long period of time, they exhibit stable characteristics that hardly change over time. In this case, the corresponding graph structure is referred to as a "static graph". Its adjacency matrix is denoted as A st ＝(a st ) N×N ：
For an adaptive graph, its construction is to take into account that due to the complexity of hydrologic loops, a static graph formed from static connectivity alone may not be sufficient to fully capture potential spatial correlations, and thus attention mechanism graph attention networks (GATs) suitable for non-european spatial information key feature extraction are introduced to adaptively capture potential changes. GAT is defined by a shared operatorThe composition is as follows:
in e ij Is the attention coefficient, characterizes the importance of the feature in site i to site j,b is the input feature number, B' is the output feature number,>representing the characteristics of the input. T Represents the operation of the transposition, the i indicates a merge operation. To normalize the attention coefficients of all selected sites i, the softmax function is used:
then an adaptive weighted adjacency matrix A is obtained sa ＝(α dyij ) N×N 。
After two constructed static and adaptive graphs are obtained by the graph generation layer, they are input into the graph convolution layer, the two graph structures are convolved and fused into a whole by the graph convolution operation, and the graph convolution layer is described as a diffusion process with K steps:
wherein L is k Is the normalized adjacency matrix in the kth diffusion, l=a/rowsum (a); w (W) k Is the parameter to be learned in the kth diffusion; z is the output after the learning of the GNN module.
After European spatial information and non-European spatial information are simultaneously input into the spatial information extraction module, three characteristic outputs with spatial representation are output, and finally the three characteristic outputs with European and non-European spatial representation are combined into one characteristic output with European and non-European spatial representation through matrix splicing operation, namely, the spatial correlation of different types of spatial data is automatically learned through the deep learning spatial information extraction module.
In the time feature extraction module, an LSTM network structure is adopted, the output of the space feature extraction module is taken as input, and the time correlation of the space feature extraction module is further acquired in the LSTM. After the operation of the space information extraction module and the time feature extraction module, the method can acquire the space-time correlation of the input information. The operational formula of LSTM is as follows:
i t ＝σ(W i ·[h t-1 ,x t ]+b i )
c′ t ＝tanh(W c ·[h t-1 ,x t ]+b c )
f t ＝σ(W f ·[h t-1 ,x t ]+b f )
c t ＝f t ⊙c t-1 +i t ⊙c′ t
o t ＝σ(Q O ·[h t-1 ,x t ]+b o )
h t ＝o t ⊙tanh(c t )
wherein i is t ,f t And o t The input door, the forget door and the output door; c' t Representing potential update units; c t Representing a memory unit; h is a t Representing a hidden state; sigma is a sigmoid activation function; the "" indicates a matrix dot product operation; w and b represent the respective learnable weight matrix and bias term.
And the model result output module is used for finally outputting multi-step flow prediction results of all runoff monitoring stations through the ReLU activation function and the full connection layer by receiving the output of the time feature extraction module.
Based on the above embodiment, taking a certain basin as an example, 8 runoff monitoring sites with representativeness on the basin are selected as prediction objects. Example time ranges were selected from 1 month, 1 day 2001 to 31 days, 12 months, 2015, the entire time period being according to 6:2: the 2 scale is divided into training period, verification period and verification period. The input data comprise grid point precipitation and potential evaporation meteorological data in a downstream range in a 30-day-old basin, spatial topological relations of 8 runoff monitoring stations and historical monitoring runoffs thereof, and output results are runoff prediction processes of 8 runoff monitoring stations in the future 5 days.
The application effect of the space-time prediction model SCGDL which fuses the European and non-European spatial information is compared with the prediction results of the traditional prediction model LSTM, the prediction model SCDL which only considers the European spatial information and the prediction model SGDL which only considers the non-European spatial information. The model super parameters are all found to be the optimal combination through trial and error. The model prediction accuracy comparison uses three evaluation indexes including Nash efficiency coefficient (NSE), correlation Coefficient (CC) and Kelin-ancient tower efficiency coefficient (KGE). The closer these three indices are to 1, the better the prediction effect of the model.
The method and the device can improve the prediction precision of multi-station multi-step runoffs of the river channel by simultaneously fusing European and non-European spatial information. Table 1 shows the comparison results of the proposed space-time prediction model SCGDL and the comparison model in the test period, and the displayed evaluation index is the average value of the multi-station multi-step runoff prediction evaluation index. The result shows that after the European and non-European spatial information is fused, the SCGDL exceeds all the comparison prediction models in terms of three evaluation indexes. Taking NSE evaluation index as an example, SCGDL is improved by 17.75%, 12.67% and 5.96% relative to LSTM, SCDL and SGDL respectively. Compared with a deep learning runoff prediction model which does not consider spatial information and only considers one spatial information alone, the space-time deep learning runoff prediction method based on European and non-European spatial information fusion provided by the invention has the advantage that the runoff prediction precision is greatly improved.
TABLE 1
Note that: the best prediction results in table 1 are bolded.
The model in the invention can extract the spatial correlation of European and non-European information at the same time, performs multi-step space-time runoff prediction on multiple sites of a river channel in a river basin, and provides basic water situation prediction data for relieving flood disasters, treating water ecological water environment and developing and utilizing water resources.
The space information fusion-based space-time deep learning runoff prediction system provided by the invention is described below, and the space information fusion-based space-time deep learning runoff prediction system described below and the space information fusion-based space-time deep learning runoff prediction method described above can be correspondingly referred to each other.
Fig. 3 is a schematic structural diagram of a spatial information fusion-based spatial-temporal deep learning radial-flow prediction system according to an embodiment of the present invention, as shown in fig. 3, including: an acquisition module 31, a space extraction module 32, a time extraction module 33 and an output module 34, wherein:
the obtaining module 31 is configured to obtain grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station, and historical runoff monitoring data of river network runoff to be predicted; the space extraction module 32 is configured to extract euclidean space feature information in the grid-point precipitation data and the potential evaporative meteorological data by using a parallel convolutional neural network, extract non-euclidean space feature information of the topological connectivity of each site and the historical runoff monitoring data by using a graph neural network, and splice the euclidean space feature information and the non-euclidean space feature information to form space correlation data; the time extraction module 33 is configured to process the spatial correlation data by using a long-short-term memory network, and obtain space-time correlation data; the output module 34 is configured to output the multi-step flow prediction results of all the runoff monitoring stations through the activation function and the full connection layer.
Fig. 4 illustrates a physical schematic diagram of an electronic device, as shown in fig. 4, which may include: processor 410, communication interface (Communications Interface) 420, memory 430 and communication bus 440, wherein processor 410, communication interface 420 and memory 430 communicate with each other via communication bus 440. The processor 410 may invoke logic instructions in the memory 430 to perform a spatio-temporal deep learning radial flow prediction method based on spatial information fusion, the method comprising: acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted; extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data; processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data; and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
Further, the logic instructions in the memory 430 described above may be implemented in the form of software functional units and may be stored in a computer-readable storage medium when sold or used as a stand-alone product. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.
In another aspect, the present invention also provides a computer program product, where the computer program product includes a computer program, where the computer program can be stored on a non-transitory computer readable storage medium, and when the computer program is executed by a processor, the computer can execute the spatial information fusion-based spatial-deep learning runoff prediction method provided by the above methods, and the method includes: acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted; extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data; processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data; and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
In yet another aspect, the present invention further provides a non-transitory computer readable storage medium having stored thereon a computer program which, when executed by a processor, is implemented to perform the spatial information fusion-based spatio-temporal deep learning runoff prediction method provided by the above methods, the method comprising: acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted; extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data; processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data; and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
The apparatus embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.
From the above description of the embodiments, it will be apparent to those skilled in the art that the embodiments may be implemented by means of software plus necessary general hardware platforms, or of course may be implemented by means of hardware. Based on this understanding, the foregoing technical solution may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a computer readable storage medium, such as ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the respective embodiments or some parts of the embodiments.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and are not limiting; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. The space-time deep learning runoff prediction method based on spatial information fusion is characterized by comprising the following steps of:
acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted;
extracting European spatial characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European spatial characteristic information of topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European spatial characteristic information and the non-European spatial characteristic information to form spatial correlation data;
processing the spatial correlation data by adopting a long-short-term memory network to obtain space-time correlation data;
and outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer by using the space-time correlation data.
2. The spatial information fusion-based space-time deep learning runoff prediction method of claim 1, wherein obtaining grid point precipitation data, potential evaporation meteorological data, topological connectivity of each site and historical runoff monitoring data of river network runoff to be predicted comprises:
collecting an original data set of all runoff monitoring stations in the river network runoff to be predicted;
and carrying out data cleaning, data normalization and sliding window preprocessing on the original data set to obtain the grid point precipitation data, the potential evaporation meteorological data, the topological connectivity of each site and the historical runoff monitoring data.
3. The spatial information fusion-based spatio-temporal deep learning runoff prediction method of claim 1, wherein extracting the euro-spatial feature information in the grid-point precipitation data and the potential evaporative meteorological data by adopting a parallel convolutional neural network includes:
determining that a single convolutional neural network comprises a two-dimensional convolutional layer, a pooling layer, a modified linear unit ReLU activation layer and a convolutional attention mechanism module CBAM which are sequentially connected;
the single convolutional neural network extracting the European spatial feature information of the grid point precipitation data or the potential evaporation meteorological data comprises the following steps:
where i, j and l represent the ith input neuron, the jth output neuron and the ith block; m represents the number of input channels;representing the characteristics of the input; />Representing a convolution kernel; * Representing a convolution operation; />Represents the first blockA leavable coefficient of deviation of the inner jth output; />Representing the j-th output result in the l-th block after a series of operations.
4. The spatial information fusion-based spatio-temporal deep learning runoff prediction method of claim 1, wherein extracting non-european spatial feature information of the topological connectivity of each site and the historical runoff monitoring data using a graph neural network comprises:
determining a Graph neural network comprising a Graph generation layer and a Graph roll-up layer, outputting a static Graph and an adaptive Graph by the Graph generation layer, the static Graph and the adaptive Graph each having a structure graph= (V, E, a), wherein V represents a set of vertices, i.e. all runoff monitoring stations in the river network, E represents a set of directed edges, i.e. representing spatial topological connectivity between upstream to downstream stations,the method comprises the steps that the method is an adjacency matrix, static or self-adaptive adjacency weights of connecting vertexes along directed edges are represented, N represents the number of runoff monitoring stations, V and E of a static graph and a self-adaptive graph are consistent, and A is different;
and inputting the static diagram and the self-adaptive diagram to the diagram volume lamination layer to obtain the non-European space characteristic information.
5. The spatial information fusion-based spatio-temporal deep learning radial-flow prediction method of claim 4, wherein the adjacency matrix of said static graph is a st ＝(a st ) N×N Wherein:
6. the substrate according to claim 4The space-time deep learning runoff prediction method based on spatial information fusion is characterized in that the self-adaptive graph is connected with a graph annotation meaning network GAT, and the GAT comprises a sharing operator
Wherein e ij Is the attention coefficient, characterizes the importance of the feature in site i to site j,b is the input feature number, B' is the output feature number,>representing the characteristics of the input. T Represents the operation of the transposition, the i represents a merge operation;
attention coefficient e using softmax function ij And (3) performing standardization:
obtaining an adaptive weighted adjacency matrix A sa ＝(α dyij ) N×N 。
7. The spatial information fusion-based spatio-temporal deep learning runoff prediction method of claim 4, wherein inputting said static map and said adaptive map to said map convolution layer to obtain said non-european spatial feature information comprises:
wherein L is k Is the normalized adjacency matrix in the kth diffusion of the graph roll stack, l=a/rowsum (a), W k The parameter to be learned in the kth diffusion is Z, and the Z is non-European space characteristic information which is output after being learned by a graph convolution layer.
8. The spatial information fusion-based spatio-temporal deep learning runoff prediction method of claim 1, wherein said spatial correlation data is processed by a long-short term memory network to obtain spatio-temporal correlation data, comprising:
i t ＝σ(W i ·[h t-1 ，x t ]+b i )
c′ t ＝tanh(W c ·[h t-1 ，x t ]+b c )
f t ＝σ(W f ·[h t-1 ，x t ]+b f )
c t ＝f t ⊙c t-1 +i t ⊙c′ t
o t ＝σ(W o ·[h t-1 ，x t ]+b o )
h t ＝o t ⊙tanh(c t )
wherein i is t ，f t And o t For input door, forget door and output door, c' t Representing potential update units, c t Representing a memory cell, h t Indicating the hidden state, σ is a sigmoid activation function, as well as the matrix point multiplication operation, W and b respectively indicate the corresponding learnable weight matrix and bias term.
9. A space-time deep learning runoff prediction system based on spatial information fusion, comprising:
the acquisition module is used for acquiring grid point precipitation data, potential evaporation meteorological data, topological connectivity of each station and historical runoff monitoring data of river network runoff to be predicted;
the space extraction module is used for extracting European space characteristic information in the grid point precipitation data and the potential evaporation meteorological data by adopting a parallel convolutional neural network, extracting non-European space characteristic information of the topological connectivity of each station and the historical runoff monitoring data by adopting a graph neural network, and splicing the European space characteristic information and the non-European space characteristic information to form space correlation data;
the time extraction module is used for processing the space correlation data by adopting a long-term and short-term memory network to acquire space-time correlation data;
and the output module is used for outputting multi-step flow prediction results of all runoff monitoring stations through the activation function and the full connection layer.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the spatial information fusion based spatio-temporal deep learning runoff prediction method of any one of claims 1 to 8 when executing the program.
CN202311721815.4A 2023-12-13 2023-12-13 Space-time deep learning runoff prediction method and system based on spatial information fusion Pending CN117634554A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311721815.4A CN117634554A (en) 2023-12-13 2023-12-13 Space-time deep learning runoff prediction method and system based on spatial information fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311721815.4A CN117634554A (en) 2023-12-13 2023-12-13 Space-time deep learning runoff prediction method and system based on spatial information fusion

Publications (1)

Publication Number Publication Date
CN117634554A true CN117634554A (en) 2024-03-01

Family

ID=90021439

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311721815.4A Pending CN117634554A (en) 2023-12-13 2023-12-13 Space-time deep learning runoff prediction method and system based on spatial information fusion

Country Status (1)

Country Link
CN (1) CN117634554A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118364292A (en) * 2024-05-06 2024-07-19 上海勘测设计研究院有限公司 Offshore wind power cluster short-term wind power prediction method, system and terminal

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118364292A (en) * 2024-05-06 2024-07-19 上海勘测设计研究院有限公司 Offshore wind power cluster short-term wind power prediction method, system and terminal

Similar Documents

Publication Publication Date Title
Shabri et al. Streamflow forecasting using least-squares support vector machines
CN112651665B (en) Surface water quality index prediction method and device based on graph neural network
CN109583565A (en) Forecasting Flood method based on the long memory network in short-term of attention model
CN111027686B (en) Landslide displacement prediction method, device and equipment
CN114220271A (en) Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network
CN113194493B (en) Wireless network data missing attribute recovery method and device based on graph neural network
CN112529296B (en) Water quality prediction method, device and server
Nevo et al. ML for flood forecasting at scale
CN113190688A (en) Complex network link prediction method and system based on logical reasoning and graph convolution
CN112766600B (en) Urban area crowd flow prediction method and system
Li et al. A novel multichannel long short-term memory method with time series for soil temperature modeling
Chen et al. WSN sampling optimization for signal reconstruction using spatiotemporal autoencoder
CN111310995A (en) Service prediction method and device based on convolutional neural network
CN114881286A (en) Short-time rainfall prediction method based on deep learning
CN117634554A (en) Space-time deep learning runoff prediction method and system based on spatial information fusion
CN112241802A (en) Interval prediction method for wind power
CN116562414A (en) Land water reserve change prediction method based on multisource hydrographic meteorological data
CN111402028A (en) Information processing method, device and equipment
CN114169646A (en) Water bloom prediction method, device, electronic equipment and computer readable storage medium
CN113537354B (en) Aquifer structure staged stochastic inversion identification method based on deep learning
Shan et al. ACGL-TR: A deep learning model for spatio-temporal short-term irradiance forecast
Das et al. A Bayesian sparse generalized linear model with an application to multiscale covariate discovery for observed rainfall extremes over the United States
CN115829163A (en) Multi-mode integration-based runoff prediction method and system for middle and lower reaches of Yangtze river
He et al. A TMA-Seq2seq Network for Multi-Factor Time Series Sea Surface Temperature Prediction.
CN113011657A (en) Method and device for predicting typhoon water level, electronic equipment and storage medium

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination