CN116401291A - Time sequence causal discovery method, device, medium and equipment for digital agricultural information - Google Patents

Time sequence causal discovery method, device, medium and equipment for digital agricultural information Download PDF

Info

Publication number
CN116401291A
CN116401291A CN202310378471.5A CN202310378471A CN116401291A CN 116401291 A CN116401291 A CN 116401291A CN 202310378471 A CN202310378471 A CN 202310378471A CN 116401291 A CN116401291 A CN 116401291A
Authority
CN
China
Prior art keywords
time
causal
data
window
module
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
CN202310378471.5A
Other languages
Chinese (zh)
Inventor
姜京池
沈汝佳
王勃然
刘劼
关毅
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.)
Harbin Institute of Technology
Original Assignee
Harbin Institute of Technology
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 Harbin Institute of Technology filed Critical Harbin Institute of Technology
Priority to CN202310378471.5A priority Critical patent/CN116401291A/en
Publication of CN116401291A publication Critical patent/CN116401291A/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
    • G06F16/24—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
    • 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
    • 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
    • G06Q50/02—Agriculture; Fishing; Forestry; Mining
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Business, Economics & Management (AREA)
  • Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Mining & Mineral Resources (AREA)
  • Biomedical Technology (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Probability & Statistics with Applications (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Agronomy & Crop Science (AREA)
  • Animal Husbandry (AREA)
  • Marine Sciences & Fisheries (AREA)
  • Artificial Intelligence (AREA)
  • Fuzzy Systems (AREA)
  • Economics (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention provides a time sequence cause and effect discovery method, a device, a medium and equipment of digital agricultural information, belonging to the technical field of agricultural informatization, wherein the device comprises: an acquisition module for acquiring a dataset comprising a plurality of observable time series of digital agricultural information; a decomposition module for expanding the dataset into a plurality of window representations and decomposing a time-sequential causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module; the first calculation module is used for outputting a causal numerical conversion relation through the mechanism invariance module; the second calculation module is used for outputting a window causal graph through the time invariance module; and the training module is used for acquiring time sequence causal relations of different variables in the digital agricultural information according to the causal numerical conversion relation and the window causal graph. The invention solves the problems of noise diversification and smaller data quantity of the time sequence data of the digital agricultural information, and has good application prospect in the aspect of agricultural mechanism research.

Description

Time sequence causal discovery method, device, medium and equipment for digital agricultural information
Technical Field
The invention relates to the technical field of agricultural informatization, in particular to a time sequence cause and effect discovery method, a device, a medium and equipment for digital agricultural information.
Background
Causal discovery of time series data aims at capturing causal relationships between time slice variables and between time slice variables in a time chain, such as the growth and development of crops, various states at the current moment of the crops have an influence on the state at the next moment, and causal effects are included. Causal discovery of time series data requires more data samples and is more time complex than causal discovery of non-time series data.
For the data of the digital agricultural information, the plant height data, the leaf nitrogen content data, the chlorophyll content data, the soil temperature data, the soil humidity data, the weather temperature data, the weather humidity data, the wind speed data, the carbon dioxide concentration data, the illumination intensity data, the effective illumination time number data and the like all show the causal relationship in the form of a normal differential equation, but the time sequence data noise in the digital agricultural information data is diversified, the data quantity is smaller, and the causal discovery method aiming at the time sequence data in the prior art is difficult to apply to the analysis of the digital agricultural information data and has great limitation.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a causal discovery device and equipment based on hierarchical supervised learning.
In a first aspect, the present invention provides a time-sequential causal discovery method for digital agricultural information, comprising:
obtaining a dataset comprising a plurality of observable time series of digital agricultural information, wherein each of the observable time series comprises samples in a same time range, variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
expanding the dataset into a plurality of window representations based on a convolutional neural network, and decomposing a time-sequential causal mechanism of the dataset into a summation form having a mechanism invariance module and a time invariance module;
outputting a causal numerical transformation relationship through the mechanism invariance module according to the data set;
outputting a window causal graph through the time invariance module according to the data set;
And according to the causal numerical value transformation relation and the window causal graph, performing time prediction through a square error training model, and obtaining time sequence causal relations of different variables in the digital agricultural information.
Optionally, the acquiring a dataset comprising a plurality of observable time series of digital agricultural information comprises:
acquiring a dataset X epsilon R containing an observable time series of d digital agricultural information d×T Each of the observable time series includes the same time range t: samples in t+T-1.
Optionally, the expanding the data set into a plurality of window representations based on the convolutional neural network includes:
by a length of
Figure BDA0004171260030000021
Is transformed into a matrix +.>
Figure BDA0004171260030000022
Each M of the matrix M ω Representing window observations, wherein +.>
Figure BDA0004171260030000023
τ represents the maximum hysteresis delay, +.>
Figure BDA0004171260030000024
Figure BDA0004171260030000025
ω represents the subscript of the window, ω e { 1..c }.
Optionally, the decomposing the time-based causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module comprises:
decomposing the time sequence causal mechanism into a sum form, wherein the sum form comprises causal weights representing time invariance and causal transformation relations representing mechanism invariance, and the causal transformation relations are represented by a formula I, and the formula I comprises the following components:
Figure BDA0004171260030000026
Wherein,,
Figure BDA0004171260030000027
observation data representing a variable i at time t, w representing said causal weights, f representing said causal transformation, i and j representing different time sequences, +.>
Figure BDA0004171260030000028
Weight representing causal influence of time series j on time series i, t-t' representing time delay,/->
Figure BDA0004171260030000029
Representing from->
Figure BDA00041712600300000210
To->
Figure BDA00041712600300000211
Is>
Figure BDA00041712600300000212
Observation data representing variable j at time t ', t' ∈ { t- τ,..once., t }, and }>
Figure BDA00041712600300000213
Representing noise.
Optionally, the outputting, by the mechanism invariance module, the causal numerical transformation relationship according to the dataset includes:
using transform kernels
Figure BDA0004171260030000031
For said window observations in each of said matrices M +.>
Figure BDA0004171260030000032
Figure BDA0004171260030000033
Executing the Hadamard product and then obtaining the output +.>
Figure BDA0004171260030000034
The expression is represented by a formula II, wherein the formula II comprises:
Figure BDA0004171260030000035
wherein,,
Figure BDA0004171260030000036
representing the window observations M corresponding to the window ω Transformed matrix, K m Representing transform kernels, M ω Indicating the window observations, +.;
repeating the above process by using different transformation cores, and repeating the above processes to obtain c window observed values M ω The transformed matrix obtained by transformation
Figure BDA0004171260030000037
Splicing to obtain three-dimensional matrix->
Figure BDA0004171260030000038
And then output as a transformation equation f (X) between the different variables in the dataset.
Optionally, the outputting, by the time invariance module, a window causal graph according to the dataset includes:
for any of the window observations
Figure BDA0004171260030000039
Aggregating similar information between time series data within the window using a formula III, the formula III comprising:
Figure BDA00041712600300000310
wherein P is ω Representation, K t Representation extraction of each M ω ω e { 1..the c }, the hadamard product between matrices, d the number of variables, τ the maximum hysteresis delay;
mapping the obtained time relationship P to a feedforward neural network
Figure BDA00041712600300000311
And when the time delay is zero, limiting the self-circularity of the window causal graph, and adopting a formula IV, wherein the formula IV comprises:
Figure BDA00041712600300000312
wherein,,
Figure BDA00041712600300000313
a weight representing the causal influence of a time sequence i on j with a time delay t e { 0..the τ }, p representing the effect of +.>
Figure BDA00041712600300000314
A sparse threshold;
and obtaining the window causal graph according to the results output by the formula III and the formula IV.
Optionally, the time prediction is performed through a square error training model according to the causal numerical transformation relationship and the window causal graph, and the obtaining of the time sequence causal relationship of different variables in the digital agricultural information includes:
According to the causal numerical transformation relation output by the mechanism invariance module and the window causal graph output by the time invariance module, parameter learning is carried out in an autoregressive mode, and a time sequence causal relation is obtained, wherein the time sequence causal relation is expressed by adopting a formula five, and the formula five comprises:
Figure BDA0004171260030000041
wherein,,
Figure BDA0004171260030000042
predicted data representing variable j at time t' +τ, +.>
Figure BDA0004171260030000043
Representing the original input matrix X t′:′τ One time slice after conversion by the mechanism invariance module,/for>
Figure BDA0004171260030000045
A weight matrix indicating all causal edges pointing to the time series j, d indicating the number of variables, τ indicating the maximum lag delay, and a hadamard product between the matrices;
prediction by optimization
Figure BDA0004171260030000046
And the loss of square error of the real value X between each time t' to find a window cause and effect matrix W that satisfies the condition, using the formula six comprising:
Figure BDA0004171260030000047
wherein X is t′ Observation data representing all variables at time t',
Figure BDA0004171260030000048
predictive data representing all variables at time t';
and acquiring time sequence causal relations of different variables in the digital agricultural information according to the window causal matrix.
In a second aspect, the present invention provides a time-series causal discovery apparatus for digital agricultural information, comprising:
An acquisition module for acquiring a dataset comprising a plurality of observable time series of digital agricultural information, wherein each of the observable time series contains samples in a same time range, variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
a decomposition module for expanding the dataset into a plurality of window representations based on a convolutional neural network and decomposing a time-sequential causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module;
the first calculation module is used for outputting a causal numerical conversion relation through the mechanism invariance module according to the data set;
a second calculation module for outputting a window causal graph through the time invariance module according to the data set;
and the training module is used for carrying out time prediction through a square error training model according to the causal numerical value transformation relationship and the window causal graph, and obtaining the time sequence causal relationship of different variables in the digital agricultural information.
In a third aspect, the present invention provides a computer readable storage medium storing a computer program for executing the time-series causal discovery method of digital agricultural information described above.
In a fourth aspect, the present invention provides a computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the processor executing the time-series causal discovery method of digital agricultural information described above.
According to the time sequence causal discovery method, device, medium and equipment for the digital agricultural information, the convolutional neural network is used as a basic network structure, an obtained data set of an observable time sequence of the digital agricultural information is expanded into a plurality of window representations, a time sequence causal mechanism in the data set is decomposed into a summation form with a mechanism invariance module and a time invariance module, a causal numerical transformation relation and a window causal chart between different variables in the data set are further obtained through the mechanism invariance module and the time invariance module respectively, then a result output by the two modules is predicted through a square error model, so that the time sequence causal relation between the different variables in the digital agricultural information is finally obtained, theoretical basis is provided for mechanism research of the digital agricultural information, the time invariance information and the time invariance information are fused in the causal discovery process, the time invariance and mechanism invariance characteristics of the effect relation in each observation window are fully utilized, the time and sample efficiency are improved, the causal numerical transformation relation and the window chart output by the mechanism invariance module in the time chain are recognized, the causal numerical transformation relation between time slice variables and the time slice relation in the time chain is better, the causal relation between the time slice variables and the time slice in the time chain is better, the causal relation between the time sequence variable and the time sequence variable has better quality in the aspect of the digital agricultural information is better, and the digital information is better applied to the aspect is better, and has better quality than the problem.
Drawings
FIG. 1 is an application environment diagram of a time-series causal discovery method for digital agricultural information in an embodiment of the present invention;
FIG. 2 is a flow chart of a method for time-series causal discovery of digital agricultural information according to an embodiment of the present invention;
FIG. 3 is a block diagram of a timing cause and effect discovery apparatus for digital agricultural information in accordance with an embodiment of the present invention;
fig. 4 is an internal structural diagram of a computer device in an embodiment of the present invention.
Detailed Description
In order that the above objects, features and advantages of the invention will be readily understood, a more particular description of the invention will be rendered by reference to specific embodiments thereof which are illustrated in the appended drawings. While the invention is susceptible of embodiment in the drawings, it is to be understood that the invention may be embodied in various forms and should not be construed as limited to the embodiments set forth herein, but rather are provided to provide a more thorough and complete understanding of the invention. It should be understood that the drawings and embodiments of the invention are for illustration purposes only and are not intended to limit the scope of the present invention.
It should be understood that the various steps recited in the method embodiments of the present invention may be performed in a different order and/or performed in parallel. Furthermore, method embodiments may include additional steps and/or omit performing the illustrated steps. The scope of the invention is not limited in this respect.
The term "including" and variations thereof as used herein are intended to be open-ended, i.e., including, but not limited to. The term "based on" is based at least in part on. The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "alternative embodiments". Related definitions of other terms will be given in the description below. It should be noted that the terms "first," "second," and the like herein are merely used for distinguishing between different devices, modules, or units and not for limiting the order or interdependence of the functions performed by such devices, modules, or units.
It should be noted that references to "one", "a plurality" and "a plurality" in this disclosure are intended to be illustrative rather than limiting, and those skilled in the art will appreciate that "one or more" is intended to be construed as "one or more" unless the context clearly indicates otherwise.
The names of messages or information interacted between the devices in the embodiments of the present invention are for illustrative purposes only and are not intended to limit the scope of such messages or information.
In the prior art, causal discovery methods for time series data generally utilize time intra-slice and inter-slice information, such as linear regression, conditional independence, and a smooth score function, and the like, and can be broadly classified into a glabellar causality-based method, a constraint-based method, and a score-based method.
Among them, the method based on the graininess causality uses a regression model to search the directed acyclic graph, however, such method is difficult to generalize to other types of data sets under the assumption of a stationary linear system and zero-mean independent random noise;
constraint-based methods test the conditional independence of all time sequences between two variables. For example if and only if
Figure BDA0004171260030000071
When they build up from +.>
Figure BDA0004171260030000072
To->
Figure BDA0004171260030000073
Of (3), wherein
Figure BDA0004171260030000074
However, as the number variable and time maximum lag delay increase, the solution space of the causal graph grows exponentially, and worse, in the case of limited data, such methods require powerful but impractical assumptions (e.g., markov conditions, loyalty and gaussian noise);
the score-based method uses one score function S and searches all solution spaces of the directed acyclic graph to maximize the score function S using the intra-time-slice directed graph G and the inter-time-slice directed graph a. Inspired by the smooth characterization, the score-based approach uses acyclic behavior to constrain G to learn the temporal causal structure. While loop-free regularization coefficients can optimize the score function more efficiently, they do not guarantee that the correct causal graph is found and often fall into a locally optimal solution.
Therefore, the causal discovery method for the time series data in the prior art has certain limitation, and in addition, the time series data noise of the digital agricultural information is diversified, the data volume is smaller, and the difficulty of applying the causal discovery method for the time series data in the prior art in the time series data analysis of the digital agricultural information is further increased.
In order to make the disclosure of the embodiments of the present invention clearer, technical terms and technical backgrounds used in the embodiments of the present invention are described before describing specific technical schemes.
1. Time causal discovery:
consider a data set X ε R containing d observable time series d×T Each time series comprises the same time range t: samples in t+T-1. Timing causal discovery aims at discovering observable time sequences (X 1 ，...，X d ) Causal relationship between the two. Such causal relationships include time lag delays and causal relationships within a time slice, and a weight matrix may be used
Figure BDA0004171260030000075
Representation of->
Figure BDA0004171260030000076
Is obtained by adding one to the maximum time delay tau>
Figure BDA0004171260030000077
Non-zero term in W->
Figure BDA0004171260030000078
Representing a time series X i For time series X j The time delay t e {0,... For the windowed cause and effect matrix, it may form a windowed cause and effect graph to describe cause and effect relationships within a time slice or between time slices. For example, for digital agricultural information data, plant height data, Blade nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, effective illumination time data and the like all show time sequence causal relationships in the form of a normal differential equation, wherein the maximum time lag delay tau is usually 1.
Wherein, the definition of the window causal graph is: let X be a multivariate data set comprising d observable time series. Window the time
Figure BDA0004171260030000081
The windowed causal graph of middle X is defined as a directed acyclic graph g= (V, E), where τ represents the maximum time lag delay, then +.>
Figure BDA0004171260030000082
Nodes in G represent data samples +.>
Figure BDA0004171260030000083
Figure BDA0004171260030000084
Edges represent causal relationships between nodes. Further, edge->
Figure BDA0004171260030000085
X represents i Resulting in X j The time delay is equal to or less than 0 and equal to or less than t' and equal to or less than tau at any time t. Note that when i=j, t' +.0.
The window causal graph described above is typically found using the means of a gland causality test. The graininess causality is based on numerical calculations, and the time causality is determined by calculating numerical fit losses and variances.
Specifically, glaring is causally defined as: let X be a multivariate dataset comprising d observable time series, V denote all nodes of G. If it is
Figure BDA0004171260030000086
Wherein->
Figure BDA0004171260030000087
Representing use of V prediction +.>
Figure BDA0004171260030000088
Variance of (X) j Resulting in X i And the time delay is t-t', which is defined by +.>
Figure BDA0004171260030000089
And (3) representing.
2. Causal invariance
In the time causal findings, invariance conditions include time invariance and mechanism invariance, which overall form causal invariance.
Wherein, the time invariance is defined as: let X epsilon R d×T Is a sequence of d observable times within a time T range. For any two variables i, j e {1,., d }, if
Figure BDA00041712600300000810
Then->
Figure BDA00041712600300000811
Figure BDA00041712600300000812
As long as t 1 -t 2 ＝t 3 -t 4 Where Pa (-) is the parent node of the vertex in the window causal graph.
Specifically, taking the time series data of digital agricultural information as an example, the above time invariance is expressed as: if there is a causal relationship between leaf nitrogen content and chlorophyll content, this causal relationship holds for any instant t. Furthermore, the invariance over time can be written as a function of time, with T representing the time index. Further, the mechanism invariance can be obtained from a given time series through the T value.
The mechanism invariance is defined as: for any variable V i And a set of variables S, a conditional distribution P (V i S) is constant at different values of TIs if and only if
Figure BDA0004171260030000091
The above mechanism invariance indicates that the conditional probability (mechanism of action) between nodes is constant for any time t, that is, the causal formula between leaf nitrogen content and chlorophyll content does not change with time.
Further, from the above two properties of the time series causal relationship, the following quotients can be further deduced:
window independence lemma: on a finite time series data set with a time maximum lag delay τ, including the variable X t -τ ，...，X t If there is in P (X)
Figure BDA0004171260030000092
And 0.ltoreq.t i Not more than τ, then
Figure BDA0004171260030000093
For all maxt i -τ≤t′≤mint i T' of (2).
The above quotients show that for an identifiable time-series causal graph, the independence of time shift is true for any time t, in particular, both causal edges and causal formulas between leaf nitrogen content and chlorophyll content are independent of time shift. Further, the observation data X can be converted into window observation data by this argument, and cause and effect discovery can be performed. In addition, the final window causal graph and the data conversion equation can be obtained by extracting the common information in each window.
3. Necessity of convolution
By a gram representation and a spectral representation of a covariance sequence, the graininess causal relationship indicates that the sequence can be decomposed into the sum or integral of uncorrelated components. Inspired by such representation and fourier transformation, for observed data X e R d×T ，
Figure BDA0004171260030000094
Considering such a function x=wf (X) +e, wf (X) can be decomposed into fourier integral form:
Figure BDA0004171260030000095
Where s denotes the projection of Wf (X) on the spatial domain, t denotes the projection of Wf (X) on the time domain,
Figure BDA0004171260030000098
representing the use of the transformation equation->
Figure BDA0004171260030000096
The coupling acting on both vectors.
Then, the time series data is decomposed using a multi-element fourier transform:
Figure BDA0004171260030000097
a number. Similar to the time-independent schrodinger equation, we assume that f (x, y) can be decomposed into spatial and temporal domains, i.e
Figure BDA0004171260030000101
According to the convolution theorem, F { h×g } = F { h·f { g }, where F { · } represents the fourier transform. Further, the convolution formula may be converted into:
Figure BDA0004171260030000102
the above formula shows that the observed value X can be obtained by convolving its convolution kernel with timing information with its window signal with structural details, i.e.
Figure BDA0004171260030000103
And sliding in the form of a window is one way to achieve the structural order. By the aboveThe transformation can prove that the causal discovery from the digital agricultural information can be decomposed into a spatial domain component and a time domain component respectively, and the two components are respectively convolved and then coupled, so that the corresponding causal information can be obtained, and further the causal discovery result is obtained.
Furthermore, the graininess causality indicates that if X i Provides the past value of X j Unique statistically significant information of future values, time series X i Graminium effect on X j . Thus, the causal discovery can be solved as a numerical fitting problem according to the definition of the gland causal test.
FIG. 1 is an application environment diagram of a time-series causal discovery method for digital agricultural information in one embodiment. Referring to fig. 1, the time-series causal discovery method of digital agricultural information is applied to a time-series causal discovery system of digital agricultural information. The time-series causal discovery system of digital agricultural information includes a terminal 110 and a server 120. The terminal 110 and the server 120 are connected through a network. The terminal 110 may be a desktop terminal or a mobile terminal, and the mobile terminal may be at least one of a mobile phone, a tablet computer, a notebook computer, and the like. The server 120 may be implemented as a stand-alone server or as a server cluster composed of a plurality of servers.
As shown in fig. 2, an embodiment of the present invention provides a time-sequence causal discovery method for digital agricultural information, including:
step 210, acquiring a data set comprising a plurality of observable time series of digital agricultural information, wherein each of the observable time series contains samples in the same time range, and variables in the data set include at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data and effective illumination time data;
Step 220, expanding the data set into a plurality of window representations based on a convolutional neural network, and decomposing a time-sequence causal mechanism of the data set into a summation form with a mechanism invariance module and a time invariance module;
step 230, outputting a causal numerical transformation relationship through the mechanism invariance module according to the data set;
step 240, outputting a window causal graph through the time invariance module according to the data set;
and step 250, performing time prediction through a square error training model according to the causal numerical transformation relationship and the window causal graph, and obtaining time sequence causal relationships of different variables in the digital agricultural information.
FIG. 2 is a flow chart of a method of time-series causal discovery of digital agricultural information in one embodiment. It should be understood that, although the steps in the flowchart of fig. 2 are shown in sequence as indicated by the arrows, the steps are not necessarily performed in sequence as indicated by the arrows. The steps are not strictly limited to the order of execution unless explicitly recited herein, and the steps may be executed in other orders. Moreover, at least some of the steps in fig. 2 may include multiple sub-steps or stages that are not necessarily performed at the same time, but may be performed at different times, or the order in which the sub-steps or stages are performed is not necessarily sequential, but may be performed in rotation or alternatively with at least a portion of the other steps or other steps.
In step 210, a dataset comprising a plurality of observable time series of digital agricultural information, each of the observable time series comprising samples in a same time range, is convenient for causal discovery of time series data between different variables, the variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data.
Wherein each variable in the dataset has a sample at each time in the time range, one sample for each variable.
Illustratively, the dataset is a dataset X ε R comprising an observable time series of d digital agricultural information d×T Each of the observable time series includes the same time range t: the sample in t+T-1, the data set variables include plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data.
In step 220, the expanding the data set into a plurality of window representations based on the convolutional neural network includes:
by a length of
Figure BDA0004171260030000121
Is transformed into a matrix +.>
Figure BDA00041712600300001219
Each M of the matrix M ω Representing window observations, wherein +.>
Figure BDA0004171260030000122
τ represents the maximum hysteresis delay, +.>
Figure BDA0004171260030000123
Figure BDA0004171260030000124
ω represents the subscript of the window, ω e { 1..c }.
That is, a length of
Figure BDA0004171260030000125
Will contain d sets of data X e R of observable time series of digital agricultural information d×T Conversion into matrix->
Figure BDA0004171260030000126
In matrix MEach M is a window representation, i.e. a plurality of window representations are included in the matrix M, correspondingly M ω Representing window observations.
The decomposing the time-sequential causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module, comprising:
decomposing the time sequence causal mechanism into a sum form, wherein the sum form comprises causal weights representing time invariance and causal transformation relations representing mechanism invariance, and the causal transformation relations are represented by a formula I, and the formula I comprises the following components:
Figure BDA0004171260030000127
wherein,,
Figure BDA0004171260030000128
observation data representing a variable i at time t, w representing said causal weights, f representing said causal transformation, i and j representing different time sequences, +. >
Figure BDA0004171260030000129
Weight representing causal influence of time series j on time series i, t-t' representing time delay,/->
Figure BDA00041712600300001210
Representing from->
Figure BDA00041712600300001211
To->
Figure BDA00041712600300001212
Is>
Figure BDA00041712600300001213
Observation data representing variable j at time t ', t' ∈ { t- τ,..once., t }, and }>
Figure BDA00041712600300001214
Representing noise。
Based on a basic network structure of a convolutional neural network, a data set is expanded into a three-step structure after window representation, and the data set has a mechanism invariance module B m And time invariance module B t Specifically, the causal relation in the time slice and between the time slices can be fully utilized in the time sequence causal discovery process of the digital agricultural information, the time invariance information and the mechanism invariance information are fused, and the time and sample efficiency can be improved.
In step 230, the outputting, by the mechanism invariance module, the causal numerical transformation relationship according to the dataset includes:
using transform kernels
Figure BDA00041712600300001220
Window observations in each of said matrices M>
Figure BDA00041712600300001215
Figure BDA00041712600300001216
Executing the Hadamard product and then obtaining the output +.>
Figure BDA00041712600300001217
The expression is represented by a formula II, wherein the formula II comprises:
Figure BDA00041712600300001218
wherein,,
Figure BDA0004171260030000131
representing the window observations M corresponding to the window ω Transformed matrix, K m Representing transform kernels, M ω Indicating the window observations, +.;
repeating the above process by using different transformation cores, and repeating the above processes to obtain c window observed values M ω The transformed matrix obtained by transformation
Figure BDA0004171260030000132
Splicing to obtain three-dimensional matrix->
Figure BDA0004171260030000133
And then output as a transformation equation f (X) between the different variables in the dataset.
Through the above steps, a unified transformation equation can be found for all window representations, including equation relationships between different variables (e.g., leaf nitrogen content data and chlorophyll content data) in the dataset of digital agricultural information temporal data.
It should be noted that the transformation equation obtained through the above steps is not completely a causal transformation process, and further screening by using a time invariance module is required.
In step 240, the time information within the window is fused by a causal convolutional network structure, specifically:
for any of the window observations
Figure BDA0004171260030000134
Aggregating similar information between time series data within the window using a formula III, the formula III comprising:
Figure BDA0004171260030000135
wherein P is ω Representation, K t Representation extraction of each M ω ω e { 1..the c }, the hadamard product between matrices, d the number of variables, τ the maximum hysteresis delay;
Using the same extraction kernel
Figure BDA0004171260030000136
After the hadamard product is performed,the output is capable of displaying similar features between time series data while K t As an invariance representation of time series data, future values of a target time series can be predicted by learning a specific repetitive pattern in an input sequence. Representing the correlation (and possibly also the causality) between the input sequence and the output sequence by these learned patterns is crucial for causal discovery.
Mapping the obtained time relationship P to a feedforward neural network
Figure BDA0004171260030000137
And when the time delay is zero, limiting the self-circularity of the window causal graph, and adopting a formula IV, wherein the formula IV comprises:
Figure BDA0004171260030000138
wherein,,
Figure BDA0004171260030000139
a weight representing the causal influence of a time sequence i on j with a time delay t e { 0..the τ }, p representing the effect of +.>
Figure BDA0004171260030000141
A sparse threshold;
and obtaining a window causal graph through the results output by the formula III and the formula IV.
In one embodiment, since the weight matrix in the windowed causal graph may display causal directions and causal weights, after the windowed causal graph is obtained, causal directions between different of the variables, including causal edges between leaf nitrogen content and chlorophyll content in the digital agricultural information timing data, can be further obtained.
In step 250, both the Cramer representation and the spectral representation of the covariance sequence indicate that the sequence can be decomposed into sums or integrals of uncorrelated components, so that the outputs of the mechanism invariance module and the time invariance module can be combined to calculate
Figure BDA0004171260030000142
Given +.>
Figure BDA0004171260030000143
The method is used for representing the prediction data of all variables from the t+tau moment to the T moment, namely, the parameter learning is carried out in an autoregressive mode, so that the accurate time sequence causal relationship is obtained.
Specifically, according to the causal numerical transformation relationship output by the mechanism invariance module and the window causal graph output by the time invariance module, parameter learning is performed in an autoregressive mode to obtain a time sequence causal relationship, and a formula five is adopted for expression, wherein the formula five comprises:
Figure BDA0004171260030000144
wherein,,
Figure BDA0004171260030000145
predicted data representing variable j at time t' +τ, +.>
Figure BDA0004171260030000146
Representing the original input matrix X t′：t′+τ One time slice after conversion by the mechanism invariance module,/for>
Figure BDA0004171260030000147
A weight matrix indicating all causal edges pointing to the time series j, d indicating the number of variables, τ indicating the maximum lag delay, and a hadamard product between the matrices;
Figure BDA0004171260030000148
including causal edges within and between time slices, have been calculated in W;
Prediction by optimization
Figure BDA0004171260030000149
And the loss of square error of the real value X between each time t' to find a window cause and effect matrix W that satisfies the condition, using the formula six comprising:
Figure BDA00041712600300001410
wherein X is t′ Observation data representing all variables at time t',
Figure BDA00041712600300001411
predictive data representing all variables at time t';
and acquiring time sequence causal relations of different variables in the digital agricultural information according to the window causal matrix.
In order to perform the steps in the foregoing embodiments and the various alternative embodiments, as shown in fig. 3, another embodiment of the present invention provides a time-series causal discovery apparatus for digital agricultural information, including:
an acquisition means 310 for acquiring a dataset comprising a plurality of observable time series of digital agricultural information, wherein each of said observable time series contains samples in the same time range, variables in said dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data and effective illumination time data;
A decomposition module 320 that expands the dataset into a plurality of window representations based on a convolutional neural network and decomposes a time-sequential causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module;
a first calculation module 330 that outputs a causal numerical transformation relationship from the dataset via the mechanism invariance module;
a second calculation module 340 outputting a window causal graph from the dataset through the time invariance module;
and the training module 350 is used for carrying out time prediction through a square error training model according to the causal numerical transformation relationship and the window causal graph, and obtaining the time sequence causal relationship of different variables in the digital agricultural information.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, performs the steps of:
obtaining a dataset comprising observable time series of a plurality of digital agricultural information, wherein each of the observable time series comprises samples in a same time range, variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
Expanding the dataset into a plurality of window representations based on a convolutional neural network, and decomposing a time-sequential causal mechanism of the dataset into a summation form having a mechanism invariance module and a time invariance module;
outputting a causal numerical transformation relationship through the mechanism invariance module according to the data set;
outputting a window causal graph through the time invariance module according to the data set;
and according to the causal numerical value transformation relation and the window causal graph, performing time prediction through a square error training model, and obtaining time sequence causal relations of different variables in the digital agricultural information.
In one embodiment, a computer device is provided comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of when executing the computer program:
obtaining a dataset comprising a plurality of observable time series of digital agricultural information, wherein each of the observable time series comprises samples in the same time range, and variables in the dataset comprise at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
Expanding the dataset into a plurality of window representations based on a convolutional neural network, and decomposing a time-sequential causal mechanism of the dataset into a summation form having a mechanism invariance module and a time invariance module;
outputting a causal numerical transformation relationship through the mechanism invariance module according to the data set;
outputting a window causal graph through the time invariance module according to the data set;
and according to the causal numerical value transformation relation and the window causal graph, performing time prediction through a square error training model, and obtaining time sequence causal relations of different variables in the digital agricultural information.
FIG. 4 illustrates an internal block diagram of a computer device in one embodiment. The computer device may be specifically the terminal 110 (or the server 120) in fig. 1. As shown in fig. 4, the computer device includes a processor, a memory, a network interface, an input device, and a display screen connected by a system bus. The memory includes a nonvolatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may also store a computer program which, when executed by a processor, causes the processor to implement a time-sequential causal discovery method of digital agricultural information. The internal memory may also have stored therein a computer program which, when executed by the processor, causes the processor to perform a time-series causal discovery method for digital agricultural information. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, the input device of the computer equipment can be a touch layer covered on the display screen, can also be keys, a track ball or a touch pad arranged on the shell of the computer equipment, and can also be an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that all or part of the processes in the methods of the above embodiments may be implemented by a computer program for instructing relevant hardware, where the program may be stored in a non-volatile computer readable storage medium, and where the program, when executed, may include processes in the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
The foregoing is only a specific embodiment of the invention to enable those skilled in the art to understand or practice the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (10)

1. A time-series causal discovery method for digital agricultural information, comprising:
obtaining a dataset comprising observable time series of a plurality of digital agricultural information, wherein each of the observable time series comprises samples in a same time range, variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
expanding the dataset into a plurality of window representations based on a convolutional neural network, and decomposing a time-sequential causal mechanism of the dataset into a summation form having a mechanism invariance module and a time invariance module;
Outputting a causal numerical transformation relationship through the mechanism invariance module according to the data set;
outputting a window causal graph through the time invariance module according to the data set;
and according to the causal numerical value transformation relation and the window causal graph, performing time prediction through a square error training model, and obtaining time sequence causal relations of different variables in the digital agricultural information.
2. The time-series causal discovery method of digital agricultural information of claim 1, wherein said acquiring a dataset comprising a plurality of observable time series of digital agricultural information comprises:
acquiring a dataset X epsilon R containing an observable time series of d digital agricultural information d×T Each of the observable time series includes samples in the same time range T: t+T-1.
3. The time-sequential causal discovery method of digital agricultural information of claim 2, wherein the convolutional neural network-based expanding the dataset into a plurality of window representations comprises:
by a length of
Figure FDA0004171260020000011
Is transformed into a matrix +.>
Figure FDA0004171260020000012
Each M of the matrix M ω Representing window observations, wherein +.>
Figure FDA0004171260020000013
τ represents the maximum hysteresis delay, c= = - >
Figure FDA0004171260020000014
ω denotes the subscript of the window, ω∈ {1, …, c }.
4. A method of time-based causal discovery of digital agricultural information according to claim 3, wherein said decomposing the time-based causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module, comprises:
decomposing the time sequence causal mechanism into a sum form, wherein the sum form comprises causal weights representing time invariance and causal transformation relations representing mechanism invariance, and the causal transformation relations are represented by a formula I, and the formula I comprises the following components:
Figure FDA0004171260020000021
wherein,,
Figure FDA0004171260020000022
observation data representing a variable i at time t, w representing said causal weights, f representing said causal transformation, i and j representing different time sequences, +.>
Figure FDA0004171260020000023
Weight representing causal influence of time series j on time series i, t-t' representing time delay,/->
Figure FDA0004171260020000024
Representing from->
Figure FDA0004171260020000025
To->
Figure FDA0004171260020000026
Is>
Figure FDA0004171260020000027
Observation data representing variable j at time t ', t' ∈ { t- τ,..once., t }, and }>
Figure FDA0004171260020000028
Representing noise.
5. A time-series causal discovery method of digital agricultural information according to claim 3, wherein said outputting, from said dataset, a causal numerical transformation relationship via said mechanism invariance module comprises:
using transform kernels
Figure FDA0004171260020000029
For said window observations in each of said matrices M +.>
Figure FDA00041712600200000210
Figure FDA00041712600200000211
Executing the Hadamard product and then obtaining the output +.>
Figure FDA00041712600200000212
The expression is represented by a formula II, wherein the formula II comprises:
Figure FDA00041712600200000213
wherein,,
Figure FDA00041712600200000214
representing the window observations M corresponding to the window ω Transformed matrix, K m Representing transform kernels, M ω Indicating the window observations, +.;
repeating the above process by using different transformation cores, and repeating the above processes to obtain c window observed values M ω The transformed matrix obtained by transformation
Figure FDA00041712600200000215
Splicing to obtain three-dimensional matrix->
Figure FDA00041712600200000216
And then output as a transformation equation f (X) between the different variables in the dataset.
6. The method of time-series causal discovery of digital agricultural information of claim 5, wherein said outputting, from said dataset, a window causal graph via said time invariance module, comprises:
for any of the window observations
Figure FDA00041712600200000217
Aggregating similar information between time series data within the window using a formula III, the formula III comprising:
Figure FDA00041712600200000218
wherein P is ω Representation, K t Representation extraction of each M ω ω e { 1..the c }, the hadamard product between matrices, d the number of variables, τ the maximum hysteresis delay;
Mapping the obtained time relationship P to a feedforward neural network
Figure FDA0004171260020000031
And when the time delay is zero, limiting the self-circularity of the window causal graph, and adopting a formula IV, wherein the formula IV comprises:
Figure FDA0004171260020000032
wherein,,
Figure FDA0004171260020000033
a weight representing the causal impact of time series i on j, the time delay is t e {0,., τ }, p denotes the enable
Figure FDA0004171260020000034
A sparse threshold;
and obtaining the window causal graph according to the results output by the formula III and the formula IV.
7. The method for time-series causal discovery of digital agricultural information according to claim 6, wherein said obtaining time-series causal relationships of different variables in digital agricultural information by performing time prediction through a square error training model according to the causal numerical transformation relationship and the window causal graph comprises:
according to the causal numerical transformation relation output by the mechanism invariance module and the window causal graph output by the time invariance module, parameter learning is carried out in an autoregressive mode, a time sequence causal relation is obtained, and a formula five is adopted, wherein the formula five comprises:
Figure FDA0004171260020000035
wherein,,
Figure FDA0004171260020000036
predicted data representing variable j at time t' +τ, +.>
Figure FDA0004171260020000037
Representing the original input matrix X t′:t′+τ One time slice after conversion by the mechanism invariance module,/for>
Figure FDA0004171260020000038
A weight matrix indicating all causal edges pointing to the time series j, d indicating the number of variables, τ indicating the maximum lag delay, and a hadamard product between the matrices;
prediction by optimization
Figure FDA0004171260020000039
And the loss of square error of the real value X between each time t' to find a window cause and effect matrix W that satisfies the condition, using the formula six comprising:
Figure FDA00041712600200000310
wherein X is t′ Observation data representing all variables at time t',
Figure FDA00041712600200000311
predictive data representing all variables at time t';
and acquiring time sequence causal relations of different variables in the digital agricultural information according to the window causal matrix.
8. A time-series causal discovery apparatus for digital agricultural information, comprising:
an acquisition module for acquiring a dataset comprising a plurality of observable time series of digital agricultural information, wherein each of the observable time series contains samples in a same time range, variables in the dataset comprising at least two of plant height data, leaf nitrogen content data, chlorophyll content data, soil temperature data, soil humidity data, weather temperature data, weather humidity data, wind speed data, carbon dioxide concentration data, illumination intensity data, and effective illumination time data;
A decomposition module for expanding the dataset into a plurality of window representations based on a convolutional neural network and decomposing a time-sequential causal mechanism of the dataset into a summed form having a mechanism invariance module and a time invariance module;
the first calculation module is used for outputting a causal numerical conversion relation through the mechanism invariance module according to the data set;
a second calculation module for outputting a window causal graph through the time invariance module according to the data set;
and the training module is used for carrying out time prediction through a square error training model according to the causal numerical value transformation relationship and the window causal graph, and obtaining the time sequence causal relationship of different variables in the digital agricultural information.
9. A computer readable storage medium, characterized in that the storage medium stores a computer program for executing the time-sequential causal discovery method of digital agricultural information according to any of the preceding claims 1-7.
10. A computer device comprising a processor, a memory and a computer program stored on the memory and executable on the processor, the processor implementing the time-series causal discovery method of digital agricultural information according to any of the preceding claims 1-7 when executing the computer program.
CN202310378471.5A 2023-04-10 2023-04-10 Time sequence causal discovery method, device, medium and equipment for digital agricultural information Pending CN116401291A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310378471.5A CN116401291A (en) 2023-04-10 2023-04-10 Time sequence causal discovery method, device, medium and equipment for digital agricultural information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310378471.5A CN116401291A (en) 2023-04-10 2023-04-10 Time sequence causal discovery method, device, medium and equipment for digital agricultural information

Publications (1)

Publication Number Publication Date
CN116401291A true CN116401291A (en) 2023-07-07

Family

ID=87008656

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310378471.5A Pending CN116401291A (en) 2023-04-10 2023-04-10 Time sequence causal discovery method, device, medium and equipment for digital agricultural information

Country Status (1)

Country Link
CN (1) CN116401291A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118333162A (en) * 2024-06-14 2024-07-12 安徽大学 Mixed local text entity causal structure learning method and device and electronic equipment

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118333162A (en) * 2024-06-14 2024-07-12 安徽大学 Mixed local text entity causal structure learning method and device and electronic equipment
CN118333162B (en) * 2024-06-14 2024-08-30 安徽大学 Mixed local text entity causal structure learning method and device and electronic equipment

Similar Documents

Publication Publication Date Title
US8645304B2 (en) Change point detection in causal modeling
Sangiorgio et al. Forecasting of noisy chaotic systems with deep neural networks
Munch et al. Recent developments in empirical dynamic modelling
Pérez et al. Data augmentation through multivariate scenario forecasting in Data Centers using Generative Adversarial Networks
US8341097B2 (en) Systems, methods and circuits for learning of relation-based networks
CN113420421B (en) QoS prediction method based on time sequence regularized tensor decomposition in mobile edge calculation
Lagos-Álvarez et al. A Kalman filter method for estimation and prediction of space–time data with an autoregressive structure
CN116401291A (en) Time sequence causal discovery method, device, medium and equipment for digital agricultural information
Qi et al. An efficient GAN-based predictive framework for multivariate time series anomaly prediction in cloud data centers
Beyaztas et al. Forecasting functional time series using weighted likelihood methodology
Wentz et al. Derivative-based SINDy (DSINDy): Addressing the challenge of discovering governing equations from noisy data
Hirata et al. Parsimonious description for predicting high-dimensional dynamics
Graafland et al. The probabilistic backbone of data-driven complex networks: an example in climate
Beyaztas et al. Functional linear models for interval-valued data
Gandy et al. Scoring predictions at extreme quantiles
Muschalik et al. isage: An incremental version of SAGE for online explanation on data streams
Wang et al. TANGO: A temporal spatial dynamic graph model for event prediction
Dudek Block bootstrap for periodic characteristics of periodically correlated time series
Amiri et al. Density estimation over spatio-temporal data streams
Muyskens et al. Partition-based nonstationary covariance estimation using the stochastic score approximation
Torres et al. PO-MOESP subspace identification of directed acyclic graphs with unknown topology
Sheth et al. Causal Discovery for Feature Selection in Physical Process-Based Hydrological Systems
Patel et al. Weather prediction using machine learning
Lin et al. PRNN: Piecewise recurrent neural networks for predicting the tendency of services invocation
Cinquemani Stochastic reaction networks with input processes: Analysis and application to gene expression inference

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