CN112232543A - Multi-site prediction method based on graph convolution network - Google Patents
Multi-site prediction method based on graph convolution network Download PDFInfo
- Publication number
- CN112232543A CN112232543A CN202010895383.9A CN202010895383A CN112232543A CN 112232543 A CN112232543 A CN 112232543A CN 202010895383 A CN202010895383 A CN 202010895383A CN 112232543 A CN112232543 A CN 112232543A
- Authority
- CN
- China
- Prior art keywords
- model
- data
- time
- prediction
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 230000007246 mechanism Effects 0.000 claims abstract description 17
- 238000002474 experimental method Methods 0.000 claims abstract description 12
- 238000012549 training Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 9
- 238000005259 measurement Methods 0.000 claims description 7
- 238000012795 verification Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000008569 process Effects 0.000 claims description 4
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 3
- 238000003672 processing method Methods 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 3
- 230000006872 improvement Effects 0.000 abstract description 3
- 238000004140 cleaning Methods 0.000 abstract description 2
- 238000012545 processing Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 9
- 238000011160 research Methods 0.000 description 5
- 230000002123 temporal effect Effects 0.000 description 5
- 230000009466 transformation Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000003915 air pollution Methods 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 238000012731 temporal analysis Methods 0.000 description 1
- 238000000700 time series analysis Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- 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
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- 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
- 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
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Business, Economics & Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a multi-site prediction method based on a graph convolution network, which is used for acquiring relevant atmospheric visibility data and cleaning the data after the data is acquired. And performing pretreatment. And based on a prediction model of the graph convolution network, performing prediction by using the same configuration in a contrast experiment, and finally comparing results. The method is improved based on the graph convolution network, and provides the reason for extracting the features in the space and time dimensions by utilizing the advantages of the graph convolution processing of non-European data and then introducing an attention mechanism to improve the model effect. The method provided by the invention has certain improvement in multi-station atmospheric visibility prediction compared with other models.
Description
Technical Field
The invention belongs to the technical field of machine learning, and particularly relates to a multi-station atmospheric visibility prediction task based on a graph convolution network and other related deep learning technologies.
Background
Atmospheric visibility is closely related to daily life of people, and the quality of the atmospheric visibility can reflect the quality of atmospheric environment in a region to a certain extent. In recent years, atmospheric visibility is reduced, air quality is poor, and more cities are often accompanied by haze weather, so that life and work of people are greatly affected, and the air quality problem is concerned by governments and people. The method can ensure traffic safety for accurate prediction of atmospheric visibility at multiple stations, particularly for prediction of low-visibility weather information, can make people control and prevent air pollution events in a targeted manner and reduce various losses caused by polluted weather, and has positive significance for traffic operation management departments, wide traveling citizens and maintenance of good air environment. In addition, with the reduction of data collection and storage cost and the rapid development of machine learning technology in recent years, various meteorological data of most regions can be accurately observed and stored, which provides a basis for abundant data and algorithm research for scientific research in the meteorological field.
Neural networks have evolved dramatically since 2012, with deep learning techniques making a number of breakthroughs in the research efforts in both computer vision and natural language processing areas. Compared with a series of steps that the features are required to be manually extracted when a traditional method is used for carrying out a picture classification task, and then the features are input into a classifier to finally obtain a classification result, a deep learning technology (such as a convolutional neural network) can directly input a picture into a model according to a fixed coding format, and finally a prediction label to which the picture belongs is directly output. Therefore, the two steps of feature extraction and classification can be combined into one, the tedious operation of manually extracting features is avoided, the end-to-end learning is adopted as the end-to-end learning, the features and the patterns with higher latitudes can be learned compared with the traditional method, and the labor consumption is reduced while the accuracy is improved.
Convolutional neural networks are limited in the type of data available, which requires that the input data must be confined to the euclidean domain. The most significant features of the european data are that the data have a regular spatial structure, for example, the image is a regular square grid, the voice is regular one-dimensional sequence data, and the data can be represented by a matrix. In the actual situations, such as electronic transactions, brain signals, recommendation systems and multi-site atmospheric visibility prediction problems in the research, most of the data in the problems do not have a regular spatial structure and are called non-European data. In these data structures, each node connection is different, i.e., the degree of each node may be different. In order to transplant the Convolution operation to such non-european data, a Graph Convolution Network (GCN) has appeared, and generally, the Graph Convolution Network has three steps, one is transmission, that is, each node transmits its own feature information to an adjacent node after transformation, which can achieve the purpose of extracting and transforming the feature information of each node, the second step is reception, each node aggregates the feature information of the adjacent nodes, thereby fusing the local structure information of the nodes, and the last step is transformation, that is, the previous information is aggregated and then non-linear transformation is performed, thereby achieving the purpose of increasing the expression capability of the model. In addition, the GCN model also has three properties of deep learning, namely hierarchical structure (characteristics are extracted layer by layer, one layer is more abstract and higher than one layer), nonlinear transformation (the expression capability of the model is increased) and end-to-end training, namely the model can learn by itself and feature information and structural information are fused by only giving a mark to a node of the graph without defining any rule. Certainly, the GCN also has the common property of the convolutional neural network, one is local parameter sharing, and the second is that the receptive field is proportional to the number of layers, when the first layer of calculation is started, each node contains the information of the respective directly adjacent node, and when the second layer of calculation is started, the information of the second-order neighbor can be also contained, so that the information participating in the calculation is more and more sufficient, that is to say, the more the number of layers, the wider the receptive field, and the more the information participating in the calculation.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a multi-site prediction method based on a graph convolution network, and the used data set is atmospheric visibility related data of 16 sites.
The technical scheme adopted by the invention is a multi-site prediction method based on a graph convolution network, which comprises the following steps:
And 2, preprocessing the data aiming at the conditions of data loss, unsmooth and the like in the atmospheric visibility data collected in the step one.
And 3, constructing a training set and a test set by using the experimental data subjected to the preprocessing step based on the prediction model of the graph convolution network, and optimizing the hyper-parameters by using gradient descent in the experimental process so as to obtain the optimal solution of the prediction model.
And 4, predicting by using the same configuration in a control experiment, and finally comparing results.
Preferably, step 2 specifically comprises the following steps:
step 2.1, filling the missing values in the data by using the average values of the previous and next moments;
2.2, aiming at the unstable time sequence, converting the unstable time sequence into a stable sequence by a first-order difference processing method;
2.3, constructing an adjacency matrix of the meteorological sites through the geographical position distance between every two meteorological sites based on the positions of the 16 meteorological sites, and using the adjacency matrix as the other input data of the graph convolution network;
preferably, step 3 specifically comprises the following steps:
and 3.1, dividing the initial data set into a training set, a verification set and a test set according to a data division method, wherein the proportion is 60%, 20% and 20% respectively. The training set is used for training parameters of the model, and then the verification set is used for testing the accuracy of the current model after each iteration;
step 3.2, the model is composed of two space-time convolution modules, each space-time convolution module is composed of two time convolution layers and a space convolution layer, time and space characteristics are extracted by performing convolution operation on input data in a time dimension and a space dimension respectively, the purpose of accurate prediction is achieved, and in addition, the purpose of endowing the input data with different weights is achieved by adding an attention mechanism;
step 3.3, defining a loss function, using an average absolute error MAE loss function and an Adam version with a random gradient descending, and additionally setting 12 hours as historical time steps respectively, wherein 3, 6, 9 and 12 hours are prediction time steps, namely predicting the atmospheric visibility in the future of 3, 6, 9 and 12 hours by using visibility data in the past 12 hours;
step 3.4, the model is suitable for 512 training periods with the batch size of 128, then the model is trained, and the hyper-parameter is continuously adjusted to obtain the optimal solution of the model;
preferably, step 4 specifically comprises the following steps:
step 4.1, using a model based on a seq2seq structure as a contrast experiment model, and respectively obtaining seq2seq-LSTM with an LSTM encoder and a LSTM decoder, seq2seq-LSTM-AM with an LSTM encoder and a stress-adding mechanism, seq2seq-GRU with a GRU encoder and a stress-adding mechanism, and seq2seq-GRU-AM with a GRU encoder and a stress-adding mechanism;
and 4.2, performing visibility prediction with the same parameter configuration by using a contrast experiment model to obtain a result and carrying out error measurement, wherein the measurement indexes are MAE, MSE and RMSE, and compared with a seq2seq model for respectively carrying out modeling prediction on each station, the visibility prediction method based on the seq2seq model has certain improvement in time and accuracy through the experiment results shown in the figures 4-7.
Compared with the prior art, the invention has the following obvious advantages:
the method is improved based on the graph convolution network, and provides the reason for extracting the features in the space and time dimensions by utilizing the advantages of the graph convolution processing of non-European data and then introducing an attention mechanism to improve the model effect. The method provided by the invention has certain improvement in multi-station atmospheric visibility prediction compared with other models.
Description of the drawings:
FIG. 1 is a structure of a time convolution layer in a model of the present invention;
FIG. 2 is a schematic representation of a spatio-temporal convolution layer in a model of the present invention;
FIG. 3 is a flow chart of a method to which the present invention relates;
FIGS. 4-7 are comparisons of experimental results of the present invention.
Detailed Description
The present invention will be described in further detail below with reference to specific embodiments and with reference to the attached drawings.
The hardware equipment used by the invention comprises 1 PC and 2 1080 video cards;
as shown in fig. 3, the present invention provides a multi-site prediction method based on graph convolution network, which specifically includes the following steps:
And 2, preprocessing the data aiming at the conditions of data loss, unsmooth and the like.
And 2.1, filling missing values in the data by using average values of the previous time and the next time to complete.
And 2.2, aiming at the unstable time sequence, converting the unstable time sequence into a stable sequence by a first-order difference processing method.
And 2.3, constructing an adjacency matrix of the meteorological sites through the distance based on the positions of the 16 meteorological sites.
And 3, building a model, constructing a training set and a testing set, and optimizing parameters to obtain the optimal solution of the prediction model.
Step 3.1, the initial data set is divided into a training set, a verification set and a test set according to a widely used conventional data division method, and the proportion is 60%, 20% and 20% respectively. The training set is used to train the parameters of the model, and then the validation set is used to test the accuracy of the current model after each iteration.
And 3.2, defining a model which consists of two space-time convolution modules, wherein each space-time convolution module consists of two time convolution layers and a space convolution layer, and an attention mechanism is added to realize the purpose of endowing different weights to input data.
And 3.3, defining a loss function, using a Mean Absolute Error (MAE) loss function and an efficient Adam version with a random gradient descending, and additionally setting 12 hours as historical time steps respectively, wherein 3, 6, 9 and 12 hours are prediction time steps, namely predicting the atmospheric visibility at 3, 6, 9 and 12 hours in the future by using visibility data at the last 12 hours.
And 3.4, the model is suitable for 512 training periods with the batch size of 128, then the model is trained, and the hyper-parameter is continuously adjusted to obtain the optimal solution of the model.
And 4, predicting by using the same configuration in a contrast experiment, and finally comparing results.
And 4.1, mainly using a model based on a seq2seq structure in the contrast experiment model, and respectively using a seq2seq-LSTM with LSTM both as an encoder and a decoder, a seq2seq-LSTM-AM with LSTM both as an encoder and a decoder and with an attention system added, a seq2seq-GRU with GRU both as an encoder and a decoder, and a seq2seq-GRU-AM with GRU both as an encoder and a decoder and with an attention system added.
And 4.2, performing visibility prediction with the same parameter configuration by using a contrast experiment model to obtain a result and carrying out error measurement, wherein the measurement indexes are MAE, MSE and RMSE, and compared with the experiment results of FIGS. 4-7, the model disclosed by the invention is improved to a certain extent in time and accuracy compared with the case that a seq2seq model is used for carrying out modeling prediction on each station respectively.
The weather station is essentially a graph structure, and the characteristics of each node can be regarded as signals on the graph. Therefore, in order to fully utilize the topological characteristics of the network, the signal is directly processed by adopting graph convolution based on spectrogram theory on each time slice, and the signal correlation on the network is mined in a space dimension. The spectrum method converts the graph into an algebraic form and analyzes topological properties of the graph, such as connectivity in the graph. In spectrogram analysis, the map can be prepared fromThe graph is represented by a laplacian matrix, and the properties of the graph structure can be obtained by analyzing the laplacian matrix and eigenvalues thereof. GCN is a convolution operation using diagonalized linear operators in the Fourier domain instead of classical convolution operators and based thereon using the kernel gθThe signal x on the graph G is filtered as shown in the following equation:
gθ*Gx＝gθ(L)x＝gθ(U∧UT)x＝∪gθ(∧)UTx
where G represents the graph convolution operation, the above formula can also be understood as the G values are each separately transformed by the graph Fourier transform since the graph signal convolution operation is equal to the product of the signals that have been transformed into the spectral domain by the graph Fourier transformθAnd x are subjected to Fourier transform, then the transform results are multiplied, and then the products are subjected to inverse Fourier transform, so that the final result of the convolution operation can be obtained. However, when there are many points in the graph and the data is large, it is time consuming to perform the eigen decomposition directly on the laplacian matrix. The present invention therefore employs Chebyshev polynomials to approximate and effectively solve this problem, as shown in the following formula:
wherein
Is a vector of coefficients of the polynomial,
the formula (5) is shown in (5-3):
wherein λmaxIs the maximum eigenvalue of the laplacian matrix, and the recursion of chebyshev polynomials defines equation (5-4):
Tk(x)＝2xTk-1(x)-Tk-2(x)
wherein T is0(x)＝1，T1(x) X. Solving this formula using approximate expansion of the Chebyshev polynomial is equivalent to passing through the convolution kernel gθThe distance near each node in the extraction graph is from 0 to (K-1)thNeighbor information of the range, eventually using the ReLU as the final activation function.
In the aspect of time feature extraction, some RNN-based models are common in time series analysis research, but a cyclic network model for multi-site atmospheric visibility requires a long time iteration, wherein a complex door mechanism cannot make timely adjustment well when dynamically changing the number of sites, and even needs to retrain the model. On the contrary, the graph convolution network model has certain advantages, can simplify a part of work and has no dependency on previous steps, the time dynamic characteristics are captured by utilizing a convolution structure on a time axis, and the specific design can form layered representation through a multilayer convolution structure to realize parallel training, so that the training time of the model is shortened. As shown in FIG. 1, the time convolution has a width k of 1-DtFollowed by the connection of a gated linear unit GLU to provide the non-linear transformation. For each node in the graph, the time convolution layer is k on the time axis for the input elementtThe neighbors perform a convolution operation to obtain the temporal signature. It has been described so far how to use GCN to extract the spatial and temporal features of the nodes in the graph, and finally, to splice these two modules to form a space-time convolution layer, as shown in fig. 2, so as to further extract the space-time correlation between the nodes in the graph, and to add a full-connection layer at the end of the model to ensure that the final output has the same size and shape as the predicted target.
In addition, in order to acquire dynamic time and space correlation on the meteorological site network, a space attention mechanism and a time attention mechanism are introduced to solve the problem. In the spatial dimension, meteorological sites at different positions are mutually influenced, the influence is dynamically changed, and an attention mechanism is used for adaptively learning the dynamic correlation among nodes in the spatial dimension, and the formula is as follows:
wherein
Is of the gammathInput to a space-time convolution module, Cr-1Is of the gammathNumber of channels of input data of a layer. v. ofs，bb，
Is a parameter to be learned, and σ is a sigmoid function used as an activation function therein. And dynamically updating an attention matrix s according to different inputs in each layer, wherein si，jThe value of (d) represents how semantically inordinate inodes and j nodes are related, and then the softmax function is used to ensure that the attention weight sum between the nodes is 1. In the process of graph convolution, the adjacency matrix A and the space attention moment matrix are utilized
The impact weights between the nodes are dynamically adjusted together.
In the time dimension, there is also a certain correlation between atmospheric visibility at different moments, and the correlation is different in different situations, and likewise, an attention mechanism is used to adaptively give different importance to different data, see the following formula:
wherein v ise，
Is a learnable parameter, the temporal attention matrix E is determined by varying inputs, the element E in Ei，jThe strength of the dependency between time i and time j is shown over time, and finally E is normalized using the softmax function. Applying the normalized temporal attention matrix directly to the input then yields:
the input is dynamically adjusted by incorporating the relevant information. By then combining the temporal attention module and the spatial attention module, the spatiotemporal convolution module can automatically place a higher degree of attention on the valuable information.
The model structure provided by the invention is introduced, the space characteristic and the time characteristic in the multi-station atmospheric visibility data are completed by mainly utilizing the space-time convolution layers, each space-time convolution layer comprises two time convolution layers and one space convolution layer, so that the model learns the time characteristic and the space characteristic in the data, and in addition, an attention mechanism is introduced in the process to ensure that the model gives higher weight to more important information in the input data, thereby achieving the effect of improving the accuracy of the model.
The above embodiments are only exemplary embodiments of the present invention, and are not intended to limit the present invention, and the scope of the present invention is defined by the claims. Various modifications and equivalents may be made by those skilled in the art within the spirit and scope of the present invention, and such modifications and equivalents should also be considered as falling within the scope of the present invention.
Claims (4)
1. A multi-site prediction method based on graph convolution network is characterized in that: the method comprises the following steps:
step 1, obtaining relevant atmospheric visibility data of a meteorological station, wherein the relevant atmospheric visibility data comprises seven characteristics of sensible temperature, wind power, humidity, precipitation, atmospheric pressure and atmospheric visibility;
step 2, preprocessing the data aiming at the situations of data loss and unsmooth in the atmospheric visibility data collected in the step 1;
step 3, constructing a training set and a test set by using the preprocessed experimental data based on a prediction model of the graph convolution network, and optimizing the hyper-parameters by using gradient descent in the experimental process to obtain the optimal solution of the prediction model;
and 4, predicting by using the same configuration in a control experiment, and finally comparing results.
2. The multi-site prediction method based on graph and volume network as claimed in claim 1, wherein: the step 2 specifically comprises the following steps:
step 2.1, filling the missing values in the data by using the average values of the previous and next moments;
2.2, aiming at the unstable time sequence, converting the unstable time sequence into a stable sequence by a first-order difference processing method;
and 2.3, constructing an adjacency matrix of the meteorological sites through the geographical position distance between every two meteorological sites based on the positions of the 16 meteorological sites, and using the adjacency matrix as the other input data of the graph convolution network.
3. The multi-site prediction method based on graph and volume network as claimed in claim 1, wherein: the step 3 specifically comprises the following steps:
step 3.1, dividing the initial data set into a training set, a verification set and a test set according to a data division method, wherein the proportion is 60%, 20% and 20% respectively; the training set is used for training parameters of the model, and then the verification set is used for testing the accuracy of the current model after each iteration;
step 3.2, the model is composed of two space-time convolution modules, each space-time convolution module is composed of two time convolution layers and a space convolution layer, time and space characteristics are extracted by performing convolution operation on input data in a time dimension and a space dimension respectively, and the purpose of giving different weights to the input data is realized by adding an attention mechanism;
step 3.3, defining a loss function, using an average absolute error MAE loss function and an Adam version with a random gradient descending, and additionally setting 12 hours as historical time steps respectively, wherein 3, 6, 9 and 12 hours are prediction time steps, namely predicting the atmospheric visibility in the future of 3, 6, 9 and 12 hours by using visibility data in the past 12 hours;
and 3.4, the model is suitable for 512 training periods with the batch size of 128, then the model is trained, and the hyper-parameter is continuously adjusted to obtain the optimal solution of the model.
4. The multi-site prediction method based on graph and volume network as claimed in claim 1, wherein: the step 4 specifically comprises the following steps:
step 4.1, using a model based on a seq2seq structure as a contrast experiment model, and respectively obtaining seq2seq-LSTM with an LSTM encoder and a LSTM decoder, seq2seq-LSTM-AM with an LSTM encoder and a stress-adding mechanism, seq2seq-GRU with a GRU encoder and a stress-adding mechanism, and seq2seq-GRU-AM with a GRU encoder and a stress-adding mechanism;
and 4.2, performing visibility prediction with the same parameter configuration by using a contrast experiment model to obtain a result and carrying out error measurement, wherein the measurement indexes are MAE, MSE and RMSE, and compared with a seq2seq model for respectively carrying out modeling prediction on each station, the model is improved to a certain extent in terms of time and accuracy.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010895383.9A CN112232543B (en) | 2020-08-31 | 2020-08-31 | Multi-station prediction method based on graph convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010895383.9A CN112232543B (en) | 2020-08-31 | 2020-08-31 | Multi-station prediction method based on graph convolution network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112232543A true CN112232543A (en) | 2021-01-15 |
| CN112232543B CN112232543B (en) | 2024-08-20 |
Family
ID=74115779
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010895383.9A Active CN112232543B (en) | 2020-08-31 | 2020-08-31 | Multi-station prediction method based on graph convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112232543B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112906941A (en) * | 2021-01-21 | 2021-06-04 | 哈尔滨工程大学 | Prediction method and system for dynamic correlation air quality time series |
| CN112926771A (en) * | 2021-02-22 | 2021-06-08 | 中国石油大学(华东) | Oil production prediction method and system based on improved space-time graph convolution network |
| CN113241117A (en) * | 2021-05-21 | 2021-08-10 | 北京工业大学 | Method for discriminating RNA-protein binding site based on residual error map convolution neural network |
| CN113240169A (en) * | 2021-05-10 | 2021-08-10 | 东南大学 | Short-term rainfall prediction method of GRU network based on multi-mode data and up-down sampling |
| CN113761806A (en) * | 2021-09-15 | 2021-12-07 | 中国人民解放军国防科技大学 | Atmospheric visibility prediction method based on neural network |
| CN115860286A (en) * | 2023-03-01 | 2023-03-28 | 江苏省生态环境监测监控有限公司 | Air quality prediction method and system based on time sequence door mechanism |
| CN118070843A (en) * | 2024-04-18 | 2024-05-24 | 成都信息工程大学 | Multi-station low-visibility prediction method based on graph recurrent neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929603A (en) * | 2019-11-09 | 2020-03-27 | 北京工业大学 | Weather image identification method based on lightweight convolutional neural network |
| CN111126704A (en) * | 2019-12-25 | 2020-05-08 | 河海大学常州校区 | Multi-region precipitation prediction model construction method based on multi-graph convolution and memory network |
| CN111563611A (en) * | 2020-04-13 | 2020-08-21 | 北京工业大学 | Cloud data center renewable energy space-time prediction method for orientation graph convolutional network |
-
2020
- 2020-08-31 CN CN202010895383.9A patent/CN112232543B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110929603A (en) * | 2019-11-09 | 2020-03-27 | 北京工业大学 | Weather image identification method based on lightweight convolutional neural network |
| CN111126704A (en) * | 2019-12-25 | 2020-05-08 | 河海大学常州校区 | Multi-region precipitation prediction model construction method based on multi-graph convolution and memory network |
| CN111563611A (en) * | 2020-04-13 | 2020-08-21 | 北京工业大学 | Cloud data center renewable energy space-time prediction method for orientation graph convolutional network |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112906941A (en) * | 2021-01-21 | 2021-06-04 | 哈尔滨工程大学 | Prediction method and system for dynamic correlation air quality time series |
| CN112926771A (en) * | 2021-02-22 | 2021-06-08 | 中国石油大学(华东) | Oil production prediction method and system based on improved space-time graph convolution network |
| CN112926771B (en) * | 2021-02-22 | 2022-10-25 | 中国石油大学(华东) | Oil yield prediction method and system based on improved space-time diagram convolution network |
| CN113240169A (en) * | 2021-05-10 | 2021-08-10 | 东南大学 | Short-term rainfall prediction method of GRU network based on multi-mode data and up-down sampling |
| CN113241117A (en) * | 2021-05-21 | 2021-08-10 | 北京工业大学 | Method for discriminating RNA-protein binding site based on residual error map convolution neural network |
| CN113241117B (en) * | 2021-05-21 | 2024-05-31 | 北京工业大学 | Residual map-based convolutional neural network RNA-protein binding site discrimination method |
| CN113761806A (en) * | 2021-09-15 | 2021-12-07 | 中国人民解放军国防科技大学 | Atmospheric visibility prediction method based on neural network |
| CN115860286A (en) * | 2023-03-01 | 2023-03-28 | 江苏省生态环境监测监控有限公司 | Air quality prediction method and system based on time sequence door mechanism |
| CN118070843A (en) * | 2024-04-18 | 2024-05-24 | 成都信息工程大学 | Multi-station low-visibility prediction method based on graph recurrent neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112232543B (en) | 2024-08-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112232543B (en) | Multi-station prediction method based on graph convolution network | |
| CN111798051B (en) | Air quality space-time prediction method based on long-term and short-term memory neural network | |
| CN113487066B (en) | Long-time-sequence freight volume prediction method based on multi-attribute enhanced graph convolution-Informer model | |
| CN113450568A (en) | Convolutional network traffic flow prediction method based on space-time attention mechanism | |
| CN111340288B (en) | Urban air quality time sequence prediction method considering time-space correlation | |
| CN113487061A (en) | Long-time-sequence traffic flow prediction method based on graph convolution-Informer model | |
| CN114240000A (en) | Air quality prediction method based on space-time graph convolution network | |
| CN111242395B (en) | Method and device for constructing prediction model for OD (origin-destination) data | |
| CN112183625A (en) | PM based on deep learning2.5High-precision time-space prediction method | |
| CN113516304B (en) | Regional pollutant space-time joint prediction method and device based on space-time diagram network | |
| CN112285376A (en) | Wind speed prediction method based on CNN-LSTM | |
| CN116720156A (en) | Weather element forecasting method based on graph neural network multi-mode weather data fusion | |
| CN116092294A (en) | Expressway traffic flow prediction method based on Transformer and graph meaning network | |
| CN115376317B (en) | Traffic flow prediction method based on dynamic graph convolution and time sequence convolution network | |
| CN109583588A (en) | A kind of short-term wind speed forecasting method and system | |
| CN115759461A (en) | Internet of things-oriented multivariate time sequence prediction method and system | |
| CN115629160A (en) | Air pollutant concentration prediction method and system based on space-time diagram | |
| CN114970946A (en) | PM2.5 pollution concentration long-term space prediction method based on deep learning model and empirical mode decomposition coupling | |
| CN116343046A (en) | Multi-source remote sensing urban river water quality monitoring method based on self-optimizing machine learning | |
| CN114444757A (en) | Combined prediction method for plateau mountain multi-model multi-scale new energy power station output | |
| Luo et al. | Stgin: A spatial temporal graph-informer network for long sequence traffic speed forecasting | |
| CN116992305A (en) | Weather forecast method and system based on big data | |
| Sabri et al. | Photovoltaic power forecasting with a long short-term memory autoencoder networks | |
| CN116525135B (en) | Method for predicting epidemic situation development situation by space-time model based on meteorological factors | |
| CN112668808A (en) | Hand-foot-and-mouth disease incidence trend prediction system based on graph convolution |
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 | ||
| GR01 | Patent grant |