CN115168327A - Large-scale data space-time prediction method based on multilayer tree long-short term memory network - Google Patents

Large-scale data space-time prediction method based on multilayer tree long-short term memory network Download PDF

Info

Publication number
CN115168327A
CN115168327A CN202210687301.0A CN202210687301A CN115168327A CN 115168327 A CN115168327 A CN 115168327A CN 202210687301 A CN202210687301 A CN 202210687301A CN 115168327 A CN115168327 A CN 115168327A
Authority
CN
China
Prior art keywords
data
prediction
time
scale
correlation
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
CN202210687301.0A
Other languages
Chinese (zh)
Inventor
朱弘恣
李淳钦
楼紫阳
过敏意
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.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
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 Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN202210687301.0A priority Critical patent/CN115168327A/en
Publication of CN115168327A publication Critical patent/CN115168327A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21—Design, administration or maintenance of databases
    • G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • 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)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Pure & Applied Mathematics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Mathematical Optimization (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Analysis (AREA)
  • Computational Mathematics (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Computing Systems (AREA)
  • Operations Research (AREA)
  • Artificial Intelligence (AREA)
  • General Health & Medical Sciences (AREA)
  • Algebra (AREA)
  • Health & Medical Sciences (AREA)
  • Fuzzy Systems (AREA)
  • Remote Sensing (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A large-scale data space-time prediction method based on a multilayer tree-like long-term and short-term memory network comprises the steps of firstly, carrying out normalization on original city scale perception data and carrying out data preprocessing for filling missing by using a linear interpolation method; then, through analyzing the information entropy of the data at the same position, the mutual information of the data at different positions and the redundancy data, the time and space correlation of the sensing data is obtained, so that the time periodicity of the data at each place and the spatial correlation of the data at any place and other position data are determined; reconstructing short-term data by using a multi-channel singular spectrum analysis algorithm to generate a training set for training a prediction model based on deep learning; in the on-line prediction stage, the prediction is realized by using a trained prediction model for the preprocessed large-scale city perception data. The invention solves the problems of large data scale and low quality, and greatly improves the prediction accuracy.

Description

Large-scale data space-time prediction method based on multilayer tree-like long-term and short-term memory network
Technical Field
The invention relates to a technology in the field of neural network application, in particular to a space-time prediction method for urban mass traffic and classified garbage data based on a multilayer tree-like long-term and short-term memory network.
Background
The development of smart cities and the technology of the Internet of things promotes a series of applications such as intelligent traffic systems and urban environment monitoring. The applications extract features from large-scale historical data, predict the future space-time trend of the data and have important significance on the development of cities. The existing space-time prediction technology can be mainly divided into a method based on mathematical analysis and a method based on deep learning. The method based on mathematical analysis is characterized by small calculated amount and low operation cost, but can not effectively process large-scale data of city level; in addition, the method cannot capture complex nonlinear space-time correlation, so that the prediction accuracy is low. The method based on deep learning has strict requirements on the quality of data, and the application of the deep learning method is limited by short data time, noise and missing phenomena in urban scale data; and the method generally cannot extract the temporal and spatial features of the data at the same time, so that the prediction result is inaccurate, and the calculation of large-scale data also increases the training and running cost of the method.
Disclosure of Invention
The invention provides a large-scale data space-time prediction method based on a multilayer tree-like long and short term memory network, aiming at the defects and shortcomings of small data processing scale, high requirement on data quality and incapability of fully extracting the time-space correlation of data in the prior art.
The invention is realized by the following technical scheme:
the invention relates to a large-scale data space-time prediction method based on a multilayer tree-shaped long-short term memory network, which comprises the steps of firstly carrying out normalization and linear interpolation on original city scale perception data to fill missing data preprocessing; then, through analyzing the information entropy of the data at the same position, the mutual information of the data at different positions and the redundancy data, the time and space correlation of the sensing data is obtained, so that the time periodicity of the data at each place and the spatial correlation of the data at any place and other position data are determined; reconstructing short-term data by using a multi-channel singular spectrum analysis (MSSA) algorithm to generate a training set for training a prediction model based on deep learning; and in the on-line prediction stage, the prediction is realized on the preprocessed large-scale city perception data by using a trained prediction model.
The original city scale perception data is as follows: time series data over multiple locations generated in smart city applications, such as average traffic speed per half hour per road segment, traffic density per city, amount of dry, wet, recoverable, hazardous waste per day per waste site per city, city noise monitoring data, city air pollution monitoring data. The original sensing data mainly comprises corresponding sensing data time sequences acquired at different positions according to a certain acquisition frequency, and the length of the sequences can be shorter according to the acquisition starting time of the system.
The linear interpolation is to fill the missing data at a certain moment with the average of the data at the previous moment and the data at the next moment.
The time and space correlation refers to: for sequence data X = (v) 0 ，v 1 ，...v N-1 ) Data Y n time units earlier n ＝(v 0-n ，v 1 -n，...v N-1-n ) The time correlation between the two can be calculated by conditional entropy: h (X | Y) n )＝ H(Y n X) -H (X), wherein: for data XAnd Y n The formed sequence pair is discretized, and then P (y) n X) is a discretized sequence W in (y) n And x) the number of occurrences of the compound,
Figure BDA0003700105100000021
Figure BDA0003700105100000022
for data (v) 0 ，v 1 ，...v N-1 ) V. will be i Is discretized into Q disjoint subintervals, so that the original data is identical to (k) 0 ，k 1 ，...k N-1 )，k i ∈[0，Q-1]，s j Represents the number of occurrences of j in a discrete value, j ∈ [0, Q-1]The probability of j being X is
Figure BDA0003700105100000023
The time period of the data can be obtained by observing the change of the conditional entropy along with the n, namely the n corresponding to the minimum value of the conditional entropy. For spatial correlation, let
Figure BDA0003700105100000024
And
Figure BDA0003700105100000025
are respectively shown in position l 1 And l 2 Can calculate
Figure BDA0003700105100000026
And
Figure BDA0003700105100000027
of mutual information
Figure BDA0003700105100000028
Wherein:
Figure BDA0003700105100000029
and
Figure BDA00037001051000000210
redundancy of (2) measures spatial correlation:
Figure BDA00037001051000000211
for each position data, calculating the redundancy with the data of all the rest positions, wherein the higher the redundancy is, the higher the spatial correlation of the data of the two positions is.
The multichannel singular spectrum analysis algorithm is as follows: by using
Figure BDA00037001051000000212
Is shown in position l i And data measured at time t
Figure BDA00037001051000000216
Is a value of (a), i ∈ {1, 2., L }, t ∈ {1, 2., N }, L is the total number of positions, N is the total number of measurement data. All data is lagged and embedded in a matrix with a window length of M:
Figure BDA00037001051000000214
wherein: x is a matrix with rows L M and columns N-M +1, the autocovariance matrix C X ＝XX T Is a block Toeplitz matrix of L × M:
Figure BDA00037001051000000215
wherein: matrix C ij Represents from l i And l i A lag covariance matrix between the data collected at the locations. Then calculate C X Characteristic value λ of K And a feature vector P K The feature vector corresponding to the larger feature value reflects the main trend of the original data, and the feature vector corresponding to the smaller feature value reflects noise and can be removed; t-th column X of matrix X t At P K The orthogonal projection coefficients above are:
Figure BDA0003700105100000031
wherein: t is more than or equal to 1 and less than or equal to T-M +1,
Figure BDA0003700105100000032
indicating the k-th feature vector at position l i Has a time lag of j, reflecting the spatial and temporal variation of the data, P k Is a time-space empirical orthogonal function (ST-EOF). a is t，k To represent
Figure BDA0003700105100000033
At X t The weight above, is called the kth temporal spatial principal component (ST-PC). Reconstruction of the data requires the co-participation of ST-EOF and ST-PC. When the k characteristic vector is reconstructed, the k characteristic vector is obtained
Figure BDA0003700105100000034
To remove noise in the original data, the first k principal components reconstructed data are selected:
Figure BDA0003700105100000035
in the data reconstruction process, for each position, selecting a corresponding time period of the data as a window length M, and only selecting the position with high redundancy with the current position to form a matrix X.
The prediction model based on deep learning comprises: a multi-layered tree-like long-short term memory network and a three-layered fully-connected network, wherein: the multi-layer tree-like long-term and short-term memory network extracts and processes the time-space correlation of input data, and the full-connection network predicts according to the output of the multi-layer tree-like long-term and short-term memory network.
The training uses the mean square error as a loss function, i.e. the square of the difference between the prediction and the label.
Technical effects
The invention integrally solves the defects of insufficient calculated amount, lack of training data and low accuracy in the space-time prediction problem of large-scale data in the prior art. Compared with the prior art, the method can reconstruct a large amount of data sets from a small amount of original data for model training, and ensures that the model can more fully extract data characteristics in the prediction process by using the time-space correlation and the multi-layer tree-shaped long-short-term memory neural network, thereby improving the final prediction accuracy.
Drawings
FIG. 1 is a diagram of an embodiment system architecture;
FIG. 2 is a diagram of a multi-layer tree-like long/short term memory network;
fig. 3 and 4 are schematic diagrams illustrating effects of the embodiment.
Detailed Description
As shown in fig. 1, the present embodiment relates to a large-scale data spatio-temporal prediction system based on a multi-layer tree-like long-short term memory network, which includes: the device comprises a data preprocessing module, a time-space correlation analysis module, a data reconstruction module and a model prediction module, wherein: the data preprocessing module is used for carrying out normalization processing on data according to original data to be predicted and filling missing data by adopting a linear interpolation method to obtain preprocessed data; the spatio-temporal correlation analysis module calculates information entropy, mutual information and redundancy among the data according to the preprocessed data, so that the spatio-temporal correlation of the data, namely the time period of the data of each place and the spatial similarity of the data among different places, is obtained; the data reconstruction module reconstructs data by using MSSA according to preprocessed data and the spatio-temporal correlation of the data, the model prediction module uses a data set as the input of model training according to a reconstructed data set and spatio-temporal correlation information, uses the temporal correlation as the length of the data, uses the spatial correlation to adjust the connection structure of the model, calculates the prediction result and the label of the data through a loss function, and finally obtains a trained prediction model.
In the reconstruction, the period of the data is used as a time window of the MSSA, similar data among different places are mutually channels of the MSSA, and a reconstructed data set is obtained through calculation.
The model prediction module comprises: multilayer tree-like long short-term memory neural network, full connection neural network, wherein: the input of the multilayer tree-like long-short term memory neural network is a reconstructed data set, the spatial correlation of the data determines the connection mode of the data of different places, the output of the underlying network is used as the input of the upper network, and finally the multilayer tree-like long-short term memory neural network obtains the intermediate representation of the data after feature extraction and space-time correlation analysis; the fully-connected neural network takes the intermediate representation output by the multi-layer tree long-short term memory neural network as input, and obtains the final prediction result of each place through three-layer linear calculation and activation function calculation.
As shown in fig. 2, the three-layer linear calculation specifically includes: in the first layer, at spatial position l i The data of (2) corresponds to the tree-shaped long and short term memory network unit
Figure BDA0003700105100000041
T is the time length of the data. Thus, input at time t
Figure BDA0003700105100000042
Data of
Figure BDA0003700105100000043
In order to take into account the spatial correlation,
Figure BDA0003700105100000044
is not only inputted with
Figure BDA0003700105100000045
Including also other network elements at the previous instant j-1
Figure BDA0003700105100000046
To output of (c). If data
Figure BDA0003700105100000047
And
Figure BDA0003700105100000048
is high, two positions are considered to have spatial correlation, then will be
Figure BDA0003700105100000049
Is connected to an output of
Figure BDA00037001051000000410
It is marked as
Figure BDA00037001051000000411
The p-th layer unit is connected in the same way as the topology of the first layer, and each network unit also has as one of its inputs the corresponding output of the p-1 th layer. For example, a unit
Figure BDA00037001051000000412
Will also
Figure BDA00037001051000000413
As an input. Finally, the full connection layer takes the output of the last layer of tree-shaped long and short term memory network as input to obtain the final prediction results of all positions. On the parameter design of the model, for the input length T of the tree-shaped long and short term memory network, the time period length of the data is used; for the number P of the layers of the tree-shaped long-term and short-term memory network, the structure of two layers is selected because the network learning capability of the multiple layers is stronger; for the number of the remaining positions of each position that are most spatially correlated, the highest 5 positions are selected according to the redundancy calculation.
Through specific practical experiments, a model code is built on a computer running an Ubuntu16.04LTS 64bit operating system by using a PyTorch deep learning framework. The method selects two data sets for training, namely a traffic data set PeMS of the state of California of America and a rubbish data set of the Shanghai Xuhui region. In the training process, each data set is trained for 200 rounds, the initial model learning rate is set to be 0.01, and the learning rate is reduced by 10% after each 10 rounds of training.
The state of california traffic data set PeMS in the united states refers to: the U.S. transportation bureau records the number of passing vehicles every 30 seconds from near 44681 independent probe sets collected in real time across highway systems in all major metropolitan areas of california. The data set used in the training of the method is traffic flow data of two months, i.e. 1 month 8 in 2018 to 30 months 9 in 2018, which are selected from 3860 sensors in the urban area of los angeles.
The Shanghai xuhui area garbage data set refers to: data sets consisting of daily garbage production of 4061 garbage sites in the Shanghai Xuhui district since 1 month 2021 collected by the Shanghai City environmental institute. The model is authorized to use data from 2021, month 1, to 2021, month 6, and day 30.
As shown in fig. 3 and fig. 4, under the above environment, the prediction accuracy rates of the method on the PeMS data set and the xu-hui garbage data set are 90.75% and 91.19%, respectively, which are greatly improved compared with the prior art. Wherein FIG. 3 is the predicted result accuracy on the PeMS dataset for the present method and the prior art, and FIG. 4 is the predicted result accuracy on the xu-hui region garbage dataset for the present method and the prior art; the prior art in the figures includes: a Support Vector Regression (SVR), a model for regression by using a support vector machine; the multi-layer perceptron MLP comprises a plurality of basic neural networks of hidden layers; the space-time k neighbor KNN-ST is a weighted combination of the space k neighbor and the time k neighbor; seer, a model for prediction based on multi-channel singular spectrum analysis; long-short term memory neural network LSTM, a recurrent neural network commonly used for time series prediction; a convolution long-short term memory neural network ConvLSTM, a model combining convolution operation and LSTM; GPTE, a neural network that uses a graph neural network for prediction; ST-Tran, combines spatio-temporal correlation with a Transformer prediction model.
The foregoing embodiments may be modified in many different ways by those skilled in the art without departing from the spirit and the principle of the invention, and the scope of the invention is not limited by the foregoing embodiments, but rather by the claims, and each implementation within the scope is limited by the invention.

Claims (8)

1. A large-scale data space-time prediction method based on a multilayer tree-like long-term and short-term memory network is characterized in that firstly, normalization and linear interpolation methods are carried out on original city scale perception data to fill missing data preprocessing; then, through analyzing the information entropy of the data at the same position, the mutual information of the data at different positions and the redundancy data, the time and space correlation of the sensing data is obtained, so that the time periodicity of the data at each place and the spatial correlation of the data at any place and other position data are determined; reconstructing short-term data by using a multi-channel singular spectrum analysis algorithm to generate a training set for training a prediction model based on deep learning; and in the on-line prediction stage, the prediction is realized on the preprocessed large-scale city perception data by using a trained prediction model.
2. The large-scale data spatiotemporal prediction method based on the multi-layer tree-like long short-term memory network as claimed in claim 1, wherein the raw city scale perceptual data is: time series data over multiple locations generated in a smart city application.
3. The large-scale data spatiotemporal prediction method based on the multi-layered tree-like long short-term memory network as claimed in claim 1, wherein the temporal and spatial correlations are: for sequence data X = (v) 0 ，v 1 ，...v N-1 ) Data Y n time units before n ＝(v 0-n ，v 1 -n，...v N-1-n ) And the time correlation conditional entropy calculation between the two is as follows: h (X | Y) n )＝H(Y n X) -H (X), wherein: for data X and Y n The formed sequence pair is discretized, and P (y) n X) is a discretized sequence W in (y) n X) the number of occurrences of the compound,
Figure FDA0003700105090000011
Figure FDA0003700105090000012
for data (v) 0 ，v 1 ，...v N-1 ) V is to be i Is discretized into Q disjoint sub-intervals, so the original data is identical to (k) 0 ，k 1 ，…k N-1 )，k i ∈[0，Q-1]，s j Is the number of occurrences of j in a discrete value, j ∈ [0, Q-1]The probability of j being X is
Figure FDA0003700105090000013
4. The method as claimed in claim 1, wherein the spatial correlation is determined by the order of the prediction of the spatiotemporal data
Figure FDA0003700105090000014
And
Figure FDA0003700105090000015
are each in position l 1 And l 2 Data of, calculate
Figure FDA0003700105090000016
And
Figure FDA0003700105090000017
of mutual information
Figure FDA0003700105090000018
Wherein:
Figure FDA0003700105090000019
and
Figure FDA00037001050900000110
redundancy of (2) measures spatial correlation:
Figure FDA00037001050900000111
for the data of each position, the redundancy with the data of all the rest positions is calculated, and the higher the redundancy is, the higher the spatial correlation of the data of the two positions is.
5. The large-scale data spatio-temporal prediction method based on the multi-layer tree-like long-short term memory network as claimed in claim 1, wherein the multi-channel singular spectrum analysis algorithm is as follows: by using
Figure FDA0003700105090000021
To be at position l i And data measured at time t
Figure FDA0003700105090000022
Is in the range of 1,2,.., L, is in the range of 1,2,.., N, L is the total number of positions, N is the total number of measurement data; all data is lagged and embedded in a matrix with a window length of M:
Figure FDA0003700105090000023
wherein: x is a matrix with rows L M and columns N-M +1, and its autocovariance matrix C X ＝XX T Is a block Toeplitz matrix of L × M:
Figure FDA0003700105090000024
wherein: matrix C ij Is derived from i And l j A lag covariance matrix between the position-collected data; then calculate C X Characteristic value λ of K And a feature vector P K Eliminating the characteristic vector reflecting noise corresponding to smaller characteristic value, and the t-th column X of the matrix X t At P K The orthogonal projection coefficients of (a) are:
Figure FDA0003700105090000025
Figure FDA0003700105090000026
wherein: t is more than or equal to 1 and less than or equal to T-M +1,
Figure FDA0003700105090000027
for the k-th feature vector at position l i A time lag of j, reflecting the spatial and temporal variations of the data, P k Is a time-space empirical orthogonal function (ST-EOF); a is a t，k Is composed of
Figure FDA0003700105090000028
At X t The weight above, referred to as the kth temporal spatial principal component (ST-PC); the reconstruction of data requires the joint participation of ST-EOF and ST-PC; when the k-th feature vector is reconstructed, the noise in the original data is removed
Figure FDA0003700105090000029
Figure FDA00037001050900000210
The first k principal components are selected to reconstruct data:
Figure FDA00037001050900000211
in the data reconstruction process, for each position, selecting a corresponding time period of the data as a window length M, and only selecting the position with high redundancy with the current position to form a matrix X.
6. The large-scale data spatiotemporal prediction method based on the multi-layer tree-like long short-term memory network as claimed in claim 1, wherein the prediction model based on deep learning comprises: a multi-layered tree-like long-short term memory network and a three-layered fully-connected network, wherein: the multi-layer tree-shaped long and short term memory network extracts and processes the time-space correlation of input data, and the full-connection network predicts according to the output of the multi-layer tree-shaped long and short term memory network.
7. The large-scale data spatio-temporal prediction method based on the multi-layer tree-like long short-term memory network as claimed in claim 1, wherein the training adopts the mean square error as the loss function, namely the square of the difference between the prediction result and the label.
8. A large-scale data spatiotemporal prediction system based on a multilayer tree-like long short-term memory network for realizing the method of any one of claims 1 to 7, which is characterized by comprising the following steps: the device comprises a data preprocessing module, a time-space correlation analysis module, a data reconstruction module and a model prediction module, wherein: the data preprocessing module is used for carrying out normalization processing on data according to original data to be predicted and filling missing data by adopting a linear interpolation method to obtain preprocessed data; the spatiotemporal correlation analysis module calculates information entropy, mutual information and redundancy among the data according to the preprocessed data, so as to obtain spatiotemporal correlation of the data, namely the time period of the data of each place and the spatial similarity of the data among different places; the data reconstruction module reconstructs data by using MSSA according to preprocessed data and the space-time correlation of the data, the model prediction module reconstructs the data according to a reconstructed data set and space-time correlation information by using the data set as the input of model training and using the time correlation as the length of the data, the connection structure of a model is adjusted by using the space correlation, and the prediction result and the label of the data are calculated by a loss function to finally obtain a trained prediction model.
CN202210687301.0A 2022-06-17 2022-06-17 Large-scale data space-time prediction method based on multilayer tree long-short term memory network Pending CN115168327A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210687301.0A CN115168327A (en) 2022-06-17 2022-06-17 Large-scale data space-time prediction method based on multilayer tree long-short term memory network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210687301.0A CN115168327A (en) 2022-06-17 2022-06-17 Large-scale data space-time prediction method based on multilayer tree long-short term memory network

Publications (1)

Publication Number Publication Date
CN115168327A true CN115168327A (en) 2022-10-11

Family

ID=83485109

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210687301.0A Pending CN115168327A (en) 2022-06-17 2022-06-17 Large-scale data space-time prediction method based on multilayer tree long-short term memory network

Country Status (1)

Country Link
CN (1) CN115168327A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116740949A (en) * 2023-08-16 2023-09-12 北京航空航天大学 Urban traffic data prediction method based on continuous learning space-time causal prediction

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116740949A (en) * 2023-08-16 2023-09-12 北京航空航天大学 Urban traffic data prediction method based on continuous learning space-time causal prediction
CN116740949B (en) * 2023-08-16 2023-10-24 北京航空航天大学 Urban traffic data prediction method based on continuous learning space-time causal prediction

Similar Documents

Publication Publication Date Title
CN111832814B (en) Air pollutant concentration prediction method based on graph attention mechanism
CN111223301B (en) Traffic flow prediction method based on graph attention convolution network
JP7004364B1 (en) Multi-source timing data failure diagnosis method and medium based on graph neural network
CN114299723B (en) Traffic flow prediction method
CN111158068B (en) Short-term prediction method and system based on simple convolution cyclic neural network
CN115578851A (en) Traffic prediction method based on MGCN
CN113762338B (en) Traffic flow prediction method, equipment and medium based on multiple graph attention mechanism
CN114936691A (en) Temperature forecasting method integrating relevance weighting and space-time attention
CN116596151B (en) Traffic flow prediction method and computing device based on time-space diagram attention
CN114287023A (en) Multi-sensor learning system for traffic prediction
CN115935796A (en) Time-space heterogeneous and synchronous graph convolution network traffic flow prediction method
CN115168327A (en) Large-scale data space-time prediction method based on multilayer tree long-short term memory network
CN116307152A (en) Traffic prediction method for space-time interactive dynamic graph attention network
CN115587335A (en) Training method of abnormal value detection model, abnormal value detection method and system
CN113610329B (en) Short-time rainfall approaching forecasting method of double-current convolution long-short term memory network
CN117290706A (en) Traffic flow prediction method based on space-time convolution fusion probability sparse attention mechanism
CN117131991A (en) Urban rainfall prediction method and platform based on hybrid neural network
CN112528557A (en) Flood flow prediction system and method based on deep learning
CN116434569A (en) Traffic flow prediction method and system based on STNR model
CN117077843A (en) Space-time attention fine granularity PM2.5 concentration prediction method based on CBAM-CNN-converter
CN115293399A (en) Traffic flow prediction method based on space-time graph convolutional network
Nagineni et al. An Efficient and Robust Deep Learning Approach to Predict Air Pollution by Employing Long Short-Term Memory
CN114611764A (en) Monitoring and early warning method for enterprise industrial water abnormity in specific area
El Bakali et al. Data-Based Solar Radiation Forecasting with Pre-Processing Using Variational Mode Decomposition
CN117456736B (en) Traffic flow prediction method based on multi-scale space-time dynamic interaction network

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