CN111539155B - Phenomenon-oriented time-space correlation mode analysis and visualization method - Google Patents
Phenomenon-oriented time-space correlation mode analysis and visualization method Download PDFInfo
- Publication number
- CN111539155B CN111539155B CN202010314044.7A CN202010314044A CN111539155B CN 111539155 B CN111539155 B CN 111539155B CN 202010314044 A CN202010314044 A CN 202010314044A CN 111539155 B CN111539155 B CN 111539155B
- Authority
- CN
- China
- Prior art keywords
- formula
- matrix
- time
- spatial position
- visualization
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/10—Numerical modelling
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- General Engineering & Computer Science (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A phenomenon-oriented time-space correlation mode analysis and visualization method comprises the steps of firstly dividing a time-space sequence into a plurality of time regions along the time direction, extracting characteristics of each time region, then utilizing the extracted characteristics and corresponding phenomenon data to carry out model solution by adopting a linear support vector machine, and finally realizing micro visualization and macro visualization according to the solved model; by the method, the invention can explore the mode or knowledge related to the specific phenomenon in the spatio-temporal data and present the mode or knowledge in an intuitive and easily understood form by a graphical method.
Description
Technical Field
The invention belongs to the technical field of machine learning and pattern recognition, and particularly relates to a phenomenon-oriented spatiotemporal correlation pattern analysis and visualization method.
Technical Field
Spatio-temporal data generally includes three elements, spatial location information, temporal information, and variables describing real-world phenomena. To explore the regularity of spatio-temporal data, meaningful knowledge or patterns need to be obtained from the measured data. In many cases, data visualization can be an effective means of exploring knowledge or patterns from highly dynamic and continuous spatiotemporal data.
Exploiting the knowledge implicit in spatiotemporal data is beneficial from the point of view. From this perspective, data visualization may take two forms: and (4) representing and exploring. The data is displayed in an intuitive way based on the visualization of the representation, and the answer "what" can be answered. For example, what is the rainfall variation in different regions, months or years? The answer "what" is typically the first step in analyzing the data. The next step is to answer "why", which is what the exploration-based visualization can answer. Visual analytics has become an important method of understanding and insights into large complex data sets over the past decade. Visual analytics focus on analytical reasoning through an interactive visual interface. This is a dynamic, iterative process in which the answer to "what" and "why" behind the data can be found. The visualization method is widely applied to space-time data analysis.
The first law in Tobler geography holds that "everything is related, except that nearby things are more closely related". Researchers comprehensively consider time and space position factors and carry out deeper research on the time and space data analysis and visualization method. However, sometimes we need to mine, explore, etc. patterns or knowledge implied in spatio-temporal data related to a particular phenomenon. The definition of "phenomenon" here depends on the field of application. For example, what weather conditions in a given area affect the rainfall in the given area over a given period of time, where "phenomenon" refers to rainfall in the given area; what is the relationship between the occurrence of tornadoes in one area and the weather in other areas? Here, the "phenomenon" refers to tornado. Therefore, when analyzing the spatio-temporal data sequence, the method is a problem worthy of exploration by combining with a specific phenomenon to research a phenomenon-oriented spatio-temporal data correlation mode analysis and visualization method.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a phenomenon-oriented space-time correlation mode analysis and visualization method. By the method, the mode or knowledge related to the specific phenomenon in the spatio-temporal data can be explored and presented in an intuitive and easily understood form through a graphical method.
In order to achieve the purpose, the technical scheme of the invention is as follows:
a phenomenon-oriented spatio-temporal correlation mode analysis and visualization method comprises the following steps:
1. to analyze the length T, spatial position { s } i Multivariate spatio-temporal sequence of i =1,2, \8230;, M }And visualizing the pattern or knowledge associated with the particular phenomenon, using a sliding window of length N to->Sequence of divisions K time zones along the time direction into a sequence of->
In formula (1):is t j Time of day, spatial position s i The measured data, Q is the number of measured variables;
in formula (3):for the kth time region t j Time of day at spatial position s i The measured data of (a);
3. For the time region K (K =1,2, \ 8230;, K), the spatial position s i (i =1,2, \8230;, M) construct a vertical (4) regression model:
in formula (4): x is the input of the model;
4. selecting a regression model of the formula (4) as a linear regression, wherein the regression model is shown as the formula (5);
in formula (5): x is the input to the model and,is a coefficient of the linear model>p = M (M + 1)/2, b is a constant,<·>represents->Dot product in space;
5. for input x k And an outputSolving the vector of the model formula (5) to obtain the vector of the formula (8) by adopting a linear support vector machine
in formula (8): alpha is alpha i,j Andis a spatial position s i Lagrange multipliers of the site model;
6. microscopic visualization: to spatial position s i Time region k, vectorConversion to mxm symmetric matrix The upper triangle of (a) is the vector->The components are arranged line by line to obtain a set>To a matrix>M rows and n columns of elements, by spatial position s m And spatial position s n The connecting lines between the lines represent that the depth of the line color represents the size of a value, and the larger the value is, the darker the line color is, so that microscopic visualization is realized;
7. macroscopic visualization: symmetric matrix for time region K (K =1,2, \8230;, K)According to the formula (9), the average value is->Then according to equation (10), make a pair->Vector based on the row sum shown in formula (11)>According to>The macroscopic visualization of the time region k can be realized by drawing the contour line;
in the formula (10), the compound represented by the formula (10),is a matrix->M rows and n columns of elements;
in the step two, in the characteristic extraction,the calculation adopts the following specific steps:
first, using the distance metric g (-) a matrix is calculated according to equation (12)M rows n columns elements>Get MxM matrix>
In formula (12): the distance metric g (-) takes the euclidean distance;
secondly, utilizing a regularized graph Laplacian method, and performing the following steps according to equations (13) and (14)Conversion into a symmetrical positive decision matrix->
In the formula (14), the regularization parameter lambda is more than 0, and I is an identity matrix;
thirdly, symmetrical positive definite matrixThe space is a Riemann manifold which is based on equation (15)>Manifold from the location Riemann>Mapping to £ er>Inner reference point->Is located in the cutting plane->Inner point->
In formula (15): the logarithm operation of the matrix is to logarithm each element of the matrix;
the fourth step is to divide the M × M symmetric matrix according to equations (16) and (17)Conversion into a vector>p=M(M+1)/2;/>
In formula (17): b i,j (i =1,2, \8230;, M; j =1,2, \8230;, M) is the i row and j column elements of an M by M symmetric matrix b.
The invention provides a visual analysis method for exploring space-time data, finding out relevant modes or knowledge between a specific spatial position of a specific phenomenon and other spatial positions, and presenting the relevant modes or knowledge in an intuitive and easily understood form. In the invention, the mining of relevant patterns in data is focused, and more importantly, the patterns related to specific phenomena are focused.
Drawings
Fig. 1 is a riemann manifold and tangent plane.
FIG. 2 is a visualization of spatiotemporal correlation patterns oriented to temperature phenomena.
FIG. 3 is a global mode visualization under temperature phenomena.
Detailed Description
The present invention will be described in detail with reference to the accompanying drawings.
A phenomenon-oriented spatio-temporal correlation mode analysis and visualization method comprises the following steps:
1. to analyze the spatial position { s } of length T i Multivariate spatio-temporal sequence of i =1,2, \8230;, M }And visualizing the pattern associated with the particular phenomenon, using a sliding window of length N to->Sequence of divisions K time zones along the time direction into a sequence of->
In formula (1):is t j Time of day, spatial position s i The measured data, Q is the number of measured variables;
in formula (3):for the kth time region t j Time of day at spatial position s i The measured data of (a);
In the step two, in the characteristic extraction,the calculation adopts the following specific steps:
first, a matrix is calculated according to equation (12) using the distance metric g (·)M rows n columns elements>Get the MxM matrix->/>
In formula (12): the distance metric g (-) takes the euclidean distance;
secondly, utilizing a regularized graph Laplacian method, and performing the following steps according to equations (13) and (14)Conversion into a symmetrical positive decision matrix>
In formula (14): the regularization parameter lambda is greater than 0, and I is an identity matrix;
thirdly, a symmetric positive definite matrixThe space is a Riemann manifold which is based on equation (15)>Manifold from the location Riemann>Mapping to £ er>Internal reference point>Is located in the cutting plane->Inner point->
In formula (15): the logarithm operation of the matrix is to logarithm each element of the matrix.And &>The corresponding relationship between them is shown in fig. 1. In the figure, a dotted line γ is a geodesic distance, and a solid line is a euclidean distance;
fourthly, the M multiplied by M symmetric matrix is formed according to the formulas (16) and (17)Conversion into a vector> p=M(M+1)/2;
In formula (17): b i,j (i =1,2, \8230;, M; j =1,2, \8230;, M) is the i row and j column elements of the M symmetric matrix b.
3. For the time region K (K =1,2, \ 8230;, K), the spatial position s i (i =1,2, \8230;, M) construct a vertical (4) regression model:
in formula (4): x is the input of the model;
4. selecting the regression model of formula (4) as a linear regression, as shown in formula (5):
in formula (5): x is the input of the model, vectorIs a coefficient of the linear model> p = M (M + 1)/2, b is a constant term,<·>represents->Dot product in space;
5. for input x k And an outputSolving the coefficient vector of the model formula (5) to the coefficient vector of the formula (8) by adopting a linear support vector machine/>
in formula (8): alpha is alpha i,j Andis a spatial position s i Lagrange multipliers of the site model;
6. microscopic visualization: for spatial position s i Time region k, vectorConversion to mxm symmetric matrix The upper triangle of (a) is the vector->The components are arranged line by line to obtain a set>To a matrix>M rows and n columns of elements, by spatial position s m And spatial position s n The connecting lines between the lines represent that the depth of the line color represents the size of a value, and the larger the value is, the darker the line color is, so that microscopic visualization is realized;
fig. 2 is a microscopic visualization of hourly meteorological data for 102 observatory stations in the united states in 2016. Selected characteristics are average temperature, total precipitation, total solar energy, average infrared surfaceTemperature and average relative humidity. Each observation station has 8768 sampling points, divided into 4 time zones. 4 observation points were selected for analysis. The phenomenon is the average temperature of the observation points. FIG. 2 contains 16 subgraphs, corresponding toExcept alaska and hawaii, each sub-graph against a map of the united states shows how the climate in the region of a circle affects the temperature of a square point in a given place. The shade of the spatial position connecting line indicates the matrix->The dark color indicates that high temperatures are observed and the light color indicates low temperatures. In FIG. 2, for ease of illustration, the @isomitted>The smaller absolute value of the component. Some important patterns can be found in fig. 2. For example, the high temperatures in most regions are highly correlated with some regions in the southwest; the temperature of some regions is associated with remote regions rather than adjacent regions (see the last two rows of fig. 2). These phenomena provide important information that can be used to help and guide us in revealing hidden rules behind the phenomena;
7. macroscopic visualization: symmetric matrix for time region K (K =1,2, \8230;, K)Evaluation of the mean value according to formula (9)>Then, according to equation (10), the matrix is +>The sum by row results in the vector ≥ according to equation (11)>According to>Drawing contours enables macroscopic visualization of the time region K (K =1,2, \8230;, K);
in the formula (10), the compound represented by the formula (10),is a matrix->M rows and n columns of elements.
FIG. 3 is a macroscopic visualization, the graph being based onDrawing a contour line. The year round data serves as a time region. In the figure, the black spot is the meteorological observation point, and is based on the weather>Corresponding to the i-th observation point, the color gray scale represents->Different values of (a). Different regions show different types of climate. Comparison with the U.S. Koppen climate type map shown in Wikipedia shows that under some typical climate conditions, such as hot desert climate in the southwest region, oceanic climate in the northwest pacific region, and wet subtropical climate in the southeast region, there is a high degree of unityCausing the disease. This consistency verifies the effectiveness of the method to some extent. Further, it is believed that more climate observers can obtain more detailed and meaningful results. />
Claims (2)
1. A phenomenon-oriented spatio-temporal correlation mode analysis and visualization method is characterized by comprising the following steps:
1. to analyze the length T, spatial position { s } i Multivariate spatio-temporal sequence of i =1,2, \8230;, M }And visualizing the pattern or knowledge associated with the particular phenomenon, using a sliding window of length N to->Sequence of divisions K time zones along the time direction into a sequence of->
In formula (1):is t j Time of day, spatial position s i The measured data, Q is the number of measured variables;
in formula (3):for the kth time region t j Time of day at spatial position s j The measured data of (a);
3. For the time region K (K =1,2, \ 8230;, K), the spatial position s i (i =1,2, \8230;, M) construct a vertical (4) regression model:
4. selecting a regression model of the formula (4) as a linear regression, wherein the regression model is shown as the formula (5);
in formula (5): x is the input to the model and,is a coefficient of the linear model>p = M (M + 1)/2, b is a constant,<·>represents->A dot product in space;
5. for inputOutput->Solving the vector of the model formula (5) to obtain the vector of the formula (8) by adopting a linear support vector machine
in formula (8): alpha is alpha i,j Andis a spatial position s i Lagrange multipliers of the site model; />
6. Microscopic visualization: for spatial position s i Time region k, vectorSymmetrical matrix which is converted into MxM->The upper triangle of (a) is the vector->The components are arranged line by line to obtain a set>To the matrix->M rows and n columns of elements, by spatial position s m And spatial position s n The connecting lines between the lines represent that the depth of the line color represents the size of a value, and the larger the value is, the darker the line color is, so that microscopic visualization is realized;
7. macroscopic visualization: symmetric matrix for time region K (K =1,2, \8230;, K)Evaluation of the mean value according to formula (9)>Then according to equation (10), make a pair->The sum by row results in the vector ≥ according to equation (11)>According to>The macroscopic visualization of the time region k can be realized by drawing the contour line;
in the formula (10), the compound represented by the formula (10),is a matrix->M rows and n columns of elements;
2. the method of claim 1, wherein the spatiotemporal correlation pattern analysis and visualization is a spatiotemporal correlation model,
in the step two, in the characteristic extraction,the calculation adopts the following specific steps:
first, a matrix is calculated according to equation (12) using the distance metric g (·)M rows and n columns of elements->Get the MxM matrix->
In formula (12): the distance metric g (-) takes the euclidean distance;
secondly, utilizing a regularized graph Laplacian method to convert the data into a normalized graph according to the formulas (13) and (14)Conversion into symmetrical positive definite matrix
In the formula (14), the regularization parameter lambda is more than 0, and I is an identity matrix;
thirdly, symmetrical positive definite matrixIs located in a space which is a Riemann manifold which combines +>Manifold from the location Riemann>Mapping to £ er>Inner reference point->Is located in the cutting plane->Inner point->/>
In formula (15): the logarithm operation of the matrix is to logarithm each element of the matrix;
fourthly, the M multiplied by M symmetric matrix is formed according to the formulas (16) and (17)Conversion into a vector> p=M(M+1)/2;
In formula (17): b i,j (i =1,2, \8230;, M; j =1,2, \8230;, M) is the i row and j column elements of a symmetric M by M matrix b.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010314044.7A CN111539155B (en) | 2020-04-20 | 2020-04-20 | Phenomenon-oriented time-space correlation mode analysis and visualization method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010314044.7A CN111539155B (en) | 2020-04-20 | 2020-04-20 | Phenomenon-oriented time-space correlation mode analysis and visualization method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111539155A CN111539155A (en) | 2020-08-14 |
CN111539155B true CN111539155B (en) | 2023-03-28 |
Family
ID=71953508
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010314044.7A Active CN111539155B (en) | 2020-04-20 | 2020-04-20 | Phenomenon-oriented time-space correlation mode analysis and visualization method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111539155B (en) |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7430313B2 (en) * | 2004-05-04 | 2008-09-30 | Zbilut Joseph P | Methods using recurrence quantification analysis to analyze and generate images |
US8849728B2 (en) * | 2007-10-01 | 2014-09-30 | Purdue Research Foundation | Visual analytics law enforcement tools |
CN110059147A (en) * | 2019-04-21 | 2019-07-26 | 黎慧斌 | The map visualization system and method for knowledge excavation is carried out based on space big data |
-
2020
- 2020-04-20 CN CN202010314044.7A patent/CN111539155B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN111539155A (en) | 2020-08-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Wolanin et al. | Estimating and understanding crop yields with explainable deep learning in the Indian Wheat Belt | |
Ailliot et al. | Stochastic weather generators: an overview of weather type models | |
CN110163472B (en) | Large-range extreme drought emergency monitoring and influence evaluation method and system | |
CN111738066B (en) | Grid late rice sheath blight disease habitat evaluation method integrating multisource remote sensing information | |
Gibson et al. | Evaluating synoptic systems in the CMIP5 climate models over the Australian region | |
US10943173B2 (en) | Computer-implemented methods, computer readable medium and systems for generating a satellite data model for a precision agriculture platform | |
CN113553697B (en) | Long-time-sequence multi-source data-based vegetation disturbance analysis method for coal mining | |
Prasad et al. | Cloud affected solar UV prediction with three-phase wavelet hybrid convolutional long short-term memory network multi-step forecast system | |
CN104952070A (en) | Near-rectangle guide based remote-sensing cornfield image segmentation method | |
CN116580318A (en) | Soil nutrient inversion method, device, equipment and medium for crop coverage area | |
CN114005048A (en) | Multi-temporal data-based land cover change and thermal environment influence research method | |
CN111460733B (en) | Mountain fire early warning method based on deep learning | |
Shi et al. | SA-JSTN: Self-attention joint spatiotemporal network for temperature forecasting | |
Bhattacharjee et al. | Spatio-temporal change modeling of LULC: a semantic kriging approach | |
Camacho et al. | An analysis of spectral variability in hyperspectral imagery: a case study of stressed oil palm detection in Colombia | |
Deo et al. | Forecasting solar photosynthetic photon flux density under cloud cover effects: novel predictive model using convolutional neural network integrated with long short-term memory network | |
CN111539155B (en) | Phenomenon-oriented time-space correlation mode analysis and visualization method | |
Sun et al. | Monitoring rice lodging grade via Sentinel-2A images based on change vector analysis | |
CN116227758B (en) | Agricultural product maturity prediction method and system based on remote sensing technology and deep learning | |
Ye et al. | GeoUNet: A novel AI model for high-resolution mapping of ecological footprint | |
CN114782835B (en) | Crop lodging area proportion detection method and device | |
Arunarani et al. | Crop yield prediction using spatio temporal CNN and multimodal remote sensing | |
CN112668615B (en) | Satellite cloud picture prediction method based on depth cross-scale extrapolation fusion | |
Arai et al. | Customer Profiling Method with Big Data based on BDT and Clustering for Sales Prediction | |
Kobayashi et al. | Colored mosaic matrix: Visualization technique for high-dimensional data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |