CN109934917B - Parallelization point cloud generation DEM method based on machine learning prediction computation strength - Google Patents
Parallelization point cloud generation DEM method based on machine learning prediction computation strength Download PDFInfo
- Publication number
- CN109934917B CN109934917B CN201910151140.1A CN201910151140A CN109934917B CN 109934917 B CN109934917 B CN 109934917B CN 201910151140 A CN201910151140 A CN 201910151140A CN 109934917 B CN109934917 B CN 109934917B
- Authority
- CN
- China
- Prior art keywords
- dem
- point cloud
- tiles
- tile
- intensity
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Geophysics And Detection Of Objects (AREA)
- Investigation Of Foundation Soil And Reinforcement Of Foundation Soil By Compacting Or Drainage (AREA)
Abstract
The invention provides a parallelization point cloud generation DEM method based on machine learning prediction computation intensity, which is characterized by comprising the following steps of: training a CART model, including feature selection, sample data generation, model training and storage; and recursively dividing point cloud data to be processed by using a quadtree, predicting the calculation intensity of each tile by adopting a stored CART model, performing dimension reduction processing on the tiles in a two-dimensional space based on a Z-curve, and mapping the tiles to each process for interpolating the DEM in parallel based on the predicted tile calculation intensity and spatial coding. The technical scheme of the invention has high efficiency and feasibility. Compared with the traditional serial point cloud generation DEM, the method for generating the DEM by using the parallelized point cloud based on the machine learning prediction computation intensity has the advantages that the performance is improved, the breakthrough of the magnitude can be realized, the execution time is effectively saved, the method is particularly suitable for data processing of massive dense point clouds, and the real-time application of the geology is supported.
Description
Technical Field
The invention belongs to the technical field of application of network geographic information systems, and relates to a method for quickly and efficiently generating a DEM (Digital Elevation Model ) based on point clouds, in particular to a method for generating the DEM by the point clouds in a parallel mode.
Background
With the rapid development of data acquisition technology in recent years, dense point cloud acquisition has become more and more popular, however, how to efficiently and quickly process raw point cloud data and extract useful information therefrom, such as rapid DEM generation using point clouds, is still a challenge at present. High performance computing has been widely used in the GIS (geographic information system) field as a typical data-intensive and computation-intensive application, and point cloud generation DEM can achieve a considerable performance improvement through parallel computing. However, the distribution of the point cloud serving as spatial data generally has spatial heterogeneity, and the computation intensity of the point cloud interpolation DEM algorithm is not only dependent on the number of the point clouds, but also causes serious load imbalance phenomenon and reduces the parallelization efficiency if a conventional data division mode and an inaccurate computation intensity evaluation method are adopted in the parallelization process. Therefore, how to accurately evaluate the calculation strength of the algorithm, and to realize rapid and efficient DEM generation based on the point cloud becomes an urgent need at present.
Disclosure of Invention
In order to solve the problems, the invention provides a method for generating a DEM (digital elevation model) by parallelization point cloud based on machine learning prediction calculation intensity, which realizes accurate evaluation of algorithm calculation intensity and rapidly and efficiently generates the DEM based on the point cloud.
The technical scheme adopted by the invention comprises a parallelization point cloud generation DEM method based on machine learning prediction computation intensity, and CART model training is carried out, wherein the method comprises feature selection, sample data generation, model training and storage; and recursively dividing point cloud data to be processed by using a quadtree, predicting the calculation intensity of each tile by adopting a stored CART model, performing dimension reduction processing on the tiles in a two-dimensional space based on a Z-curve, and mapping the tiles to each process for interpolating the DEM in parallel based on the predicted tile calculation intensity and spatial coding.
Moreover, the method is realized by adopting the following steps,
step 1, selecting sample characteristics, wherein the sample characteristics comprise the number of the tile point clouds, the number of the neighborhood point clouds, the number of DEM grids in the tile, the point cloud density and the point cloud distribution variance;
step 2, generating sample data, which comprises the steps of generating point cloud data with different areas and different numbers, interpolating the DEM according to the uniform resolution, counting the characteristics of each sample, and taking the time of interpolating the DEM by the point cloud as a sample label;
step 3, splitting sample data into a training set and a verification set, constructing a CART model based on the training set, performing post pruning treatment on the CART model based on the verification set, and locally storing the constructed CART model; the CART represents a classification regression tree;
step 4, dividing the point cloud into tiles based on the quadtree and a preset initial recursion parameter, reading a CART model stored locally, and predicting the calculation intensity of each tile;
step 5, carrying out space coding on the divided tiles based on Z-curves to realize conversion from two-dimensional space to one-dimensional space, and sequencing the tiles according to the Z-curve codes;
step 6, mapping the tiles to each process according to the coding sequence, wherein the mapping strategy is to sum the calculation intensities of all the tiles, divide the calculation intensities by the number of processes to obtain the average calculation intensity contained by each process, and finally map the tiles to each process in sequence according to the coding sequence;
step 7, calculating a parameter time difference ratio reflecting the load balancing condition, if the parameter is larger than a preset corresponding time difference ratio threshold, continuously dividing tiles with calculation intensity larger than a preset corresponding calculation intensity threshold, generating multi-granularity tiles, predicting calculation intensity, and returning to the step 5; otherwise, jumping to the step 8;
and 8, interpolating the DEM by each process based on the mapped tiles and sending the result to a main process, wherein the main process merges the DEM.
Moreover, the point cloud interpolation DEM is implemented in an IDW manner, which represents inverse distance weighted interpolation.
And, construct CART model based on training set, generate complete decision tree, the implementation is as follows,
firstly, selecting a feature to perform data splitting based on an input training set so as to minimize the error of the split data set; and then dividing the data into two subsets according to the characteristics, judging whether the difference value between the current node error and the node split subset error is smaller than a preset corresponding error change threshold value, if so, stopping dividing, otherwise, continuously judging whether the number of samples contained in the two subsets is smaller than a preset corresponding number threshold value, if so, stopping dividing, otherwise, continuing dividing.
And performing post pruning processing on the CART model based on the verification set, wherein the implementation mode is that based on each subtree detected from the bottom up by the verification set, the verification errors of two leaf nodes under the subtree and the combined verification errors are calculated, and if the combined verification errors are smaller than the verification errors of the two leaf nodes, the two leaf nodes are combined to finish pruning processing.
Furthermore, the time difference ratio is calculated by dividing the difference between the maximum time consumption and the minimum time consumption in the process by the maximum time consumption.
In step 7, a backtracking parameter is set to control the granularity of tile division and additional division time, and prevent no solution.
Moreover, the prediction of each tile computational intensity is implemented as follows:
firstly, counting the number of tile point clouds, the number of neighborhood point clouds and the number of DEM grids in the tiles, wherein the number of the tile point clouds, the number of the neighborhood point clouds and the number of the DEM grids in the tiles are divided, and the distribution variance of the point clouds is calculated; reading a CART model stored locally, and sequentially judging each characteristic of the tile to a leaf node according to the split characteristic selected in the CART model; finally, the value of the leaf node is the calculation intensity of the tile.
Compared with the traditional serial point cloud generation DEM, the method for generating the DEM by using the parallelized point cloud based on the machine learning prediction calculation intensity has the advantages that the performance is improved, the breakthrough of the magnitude can be realized, the execution time of an algorithm is effectively saved, the method is particularly suitable for data processing of massive dense point clouds, the real-time application of the geochemistry is supported, and the method has important economic value.
Detailed Description
The present invention will be further described in detail below in conjunction with the following examples, for the purpose of facilitating understanding and practicing the present invention by those of ordinary skill in the art, it being understood that the examples described herein are for the purpose of illustration and explanation only and are not intended to limit the invention.
The invention provides a parallel point cloud generation DEM method based on machine learning prediction computation intensity, which realizes accurate evaluation of algorithm computation intensity and rapidly and efficiently generates the DEM based on the point cloud.
The technical scheme adopted by the invention is that the method for generating the DEM by using the parallelization point cloud based on the machine learning prediction computation strength is used for training the CART model and comprises the steps of feature selection, sample data generation, model training and storage; and recursively dividing point cloud data to be processed by using a quadtree, predicting the calculation intensity of each tile by adopting a stored CART model, performing dimension reduction processing on the tiles in a two-dimensional space based on a Z-curve, and then mapping the tiles to each process based on the predicted calculation intensity of the tiles and space coding.
Moreover, the method is realized by adopting the following steps,
step 1, selecting algorithm characteristics, adopting IDW (Inverse Distance Weighted, inverse distance weighted interpolation) to realize point cloud interpolation DEM in experiments, and taking factors affecting algorithm calculation intensity including the number of tile point clouds, the number of neighborhood point clouds, the number of DEM grids in the tiles, the density of the point clouds and the distribution variance of the point clouds as sample characteristics;
in the step, factors influencing the calculation intensity of the IDW algorithm are comprehensively considered, and the number of point clouds is not only used as an index for evaluating the calculation intensity. Where the neighborhood range is defined as a large rectangle that extends the tile space range by a specified distance, which is typically the maximum search radius in the IDW algorithm.
The IDW algorithm process is that an initial search radius is initialized, a search radius increases step distance and a maximum search radius are firstly increased, then for each DEM grid, whether the number of point clouds in the initial search radius is larger than the number needed by interpolation is judged, if so, the number of point clouds in the initial search radius increases step by step according to the increase step distance until the number of point clouds in the radius reaches the number needed by interpolation, the DEM is generated based on an IDW interpolation formula, and if the search radius is larger than the maximum search radius, the point clouds still cannot reach the number needed by interpolation, and then the interpolation is not carried out. In specific implementation, the IDW interpolation formula may be implemented by using the prior art, which is not described in detail in the present invention.
Step 2, generating sample data, generating point cloud data with different areas and different numbers, interpolating DEM (digital elevation model) according to uniform resolution to obtain sample data tiles, counting various characteristics (the number of tile point clouds, the number of neighborhood point clouds, the number of DEM grids in the tiles, the density of the point clouds and the distribution variance of the point clouds) of each sample, and taking the time of interpolating the DEM by the point clouds as a sample label;
step 3, splitting sample data into a training set and a verification set, constructing a CART (Classification and Regression Tree, classification regression tree) model based on the training set, performing post pruning treatment on the CART model based on the verification set, and locally storing the constructed CART model;
in the step, the situation that the trained CART model generates overfitting to the training set is comprehensively considered, and sample data are split into an independent training set and an independent verification set. The training set is used for model training, and the verification set performs tree pruning on the generated model.
The training set is used for model training to generate a complete decision tree, the complete decision tree is a CART model under specific parameters, two parameters are defined when the CART model is generated and are respectively used for controlling error change and the least sample number of segmentation features, and the meaning of error change is that if the difference value between the current node error and the subset error after node splitting is smaller than the corresponding error change threshold value, splitting is not performed any more; the meaning of the minimum number of samples of the segmentation feature is that if the number of samples contained in the node after the splitting is smaller than the minimum number of samples of the segmentation feature, the splitting is not performed any more. The complete decision tree is generated under the condition that the error change threshold value is 0 and the minimum sample number of the segmentation characteristics is 1, namely the complete decision tree is split as far as possible, and preparation is made for post pruning.
The training generation of the complete decision tree is specifically realized as follows:
firstly, based on an input training set, selecting a feature to perform data splitting so that the error of the split data set is minimum, in the embodiment, the error of the data set adopts the variance of a predicted value in the data set, traversing each value on each feature in a sample during splitting, then performing binary division on the sample according to the value, counting the error sum of the two divided samples, and finally selecting the feature value which makes the split error sum minimum to divide the sample;
and then dividing the data into two subsets according to the characteristics, judging whether the difference value between the current node error and the node split subset error is smaller than a preset corresponding error change threshold value, if so, stopping dividing, otherwise, continuously judging whether the number of samples contained in the two subsets is smaller than a preset corresponding number threshold value, if so, stopping dividing, otherwise, continuing dividing. In the embodiment, the subset error adopts the variance of the predicted value in the subset, and the subset error after node splitting is the error sum of the two divided subsets.
The verification set performs tree pruning on the generated model, and is specifically implemented as follows:
based on the detection of each subtree from the bottom to the top, calculating the verification errors of the two leaf nodes under the subtree and the combined verification errors, if the combined verification errors are smaller than the verification errors of the two leaf nodes, combining the two leaf nodes, and completing pruning. The verification error is the square of the difference between the predicted value and the actual value.
Step 4, dividing the point cloud into tiles based on the quadtree and a preset initial recursion parameter, reading a CART model stored locally, and predicting the calculation intensity of each tile;
in particular, the initial recursion parameters may be preset. The initial recursion parameter value principle is that the area of the recursively divided tiles falls in a sample space as much as possible, so that more accurate prediction is realized.
The invention adopts CART model to predict the calculation intensity, and realizes the accurate evaluation of the calculation intensity of the parallel tasks. The prediction implementation is as follows:
firstly, counting the number of tile point clouds of each divided tile, the number of neighborhood point clouds, the number of DEM grids in the tile, the density of the point clouds and the distribution variance of the point clouds; reading a CART model stored locally, and sequentially judging each characteristic of the tile to a leaf node according to the split characteristic selected in the CART model; finally, the value of the leaf node is the calculation intensity of the tile.
Step 5, carrying out space coding on the divided tiles based on Z-curves to realize conversion from two-dimensional space to one-dimensional space, and sequencing the tiles according to the Z-curve codes;
space coding based on Z-curve is the prior art, which is used to realize the conversion from two dimensions to one dimension and ensure a certain space adjacency, and the invention is not repeated.
Step 6, mapping the tiles to each process according to the coding sequence, wherein the mapping strategy is to sum the calculation intensities of all the tiles, divide the sum by the number of the processes to obtain the calculation intensity contained by each process, and sequentially map the tiles to each process according to the coding sequence;
in the step, the tiles are mapped to each process based on the predicted tile calculation intensity and the spatial coding, and the method is specifically realized by firstly predicting the calculation intensity of each tile according to the step 4, calculating the calculation intensity sum of all the tiles, and averaging the calculation intensity sum to each process to obtain the calculation intensity accommodated by each process; and then mapping the tiles to each process in turn according to the coding sequence, and counting the sum of the calculation intensities of the tiles of each process until the sum is larger than the calculation intensity which can be accommodated by the process. The mapping strategy ensures that the sum of the computational intensities of the tiles to be processed by each process is substantially equal.
Step 7, calculating a parameter time difference ratio reflecting the load balancing condition, if the parameter is larger than a preset corresponding time difference ratio threshold, continuously dividing tiles with calculation intensity larger than a preset corresponding calculation intensity threshold, generating multi-granularity tiles, predicting calculation intensity, and returning to the step 5; otherwise, go to step 8.
Furthermore, a backtracking parameter can be added to control the division granularity and the additional division time of the tiles, and meanwhile, the situation of no solution is prevented;
the dividing mode in the step is consistent with that in the step 4, and the tiles are continuously divided in a quadtree mode. It should be noted that the time difference ratio (the threshold is generally 0.1) is used to determine whether to continue dividing, and the condition is not satisfied, and the dividing is continued. Therefore, a backtracking parameter needs to be added to control the division granularity of the tiles and the additional division time so as to prevent the situation of no solution. The number of trace back steps depends on the additional dividing time that can be tolerated in practical applications, and it is also necessary to consider that the divided tiles should fall as far as possible in the sample space.
In this step, a time difference ratio parameter is used to evaluate the load balancing, where the parameter is the difference between the maximum time consumption and the minimum time consumption in the process divided by the maximum time consumption, and it is generally considered that less than 0.1 is preferable. In this step, if the parameter does not meet the given condition, the tiles are further divided by backtracking, so that a backtracking step number needs to be added to control the additional dividing time, and the situation that the time difference ratio cannot meet the given condition all the time is prevented.
And 8, interpolating the DEM by each process based on the mapped tiles and sending the result to a main process, wherein the main process merges the DEM.
The step 1-3 of the invention realizes the training of the CART model, and specifically comprises feature selection, sample data generation, model training and model localization storage; step 4-8, carrying out space division, data mapping and DEM generation and combination on point cloud data to be processed, wherein the step comprises the steps of firstly carrying out recursion division on the point cloud data by using a quadtree, carrying out dimension reduction processing on tiles in a two-dimensional space based on a Z-curve, then mapping the tiles to each process based on predicted tile calculation intensity and space coding, specifically realizing calculation of average calculation intensity which can be contained by each process, sequentially mapping the tiles to each process according to a coding sequence, and counting the sum of the calculation intensities of the tiles of each process until the sum is larger than the calculation intensity which can be contained by each process. The mapping strategy ensures that the sum of the computational intensities of the tiles to be processed by each process is substantially equal. By adopting CART model to predict the calculation intensity and combining the prediction calculation intensity, a multi-granularity Z-curve dividing method is adopted, so that better load balancing is realized.
When the method is specifically implemented, the technical scheme of the invention can adopt the computer software technology to realize the automatic operation flow. The relevant threshold values can be preset experimentally or empirically by a person skilled in the art.
It should be understood that parts of the specification not specifically set forth herein are all prior art.
It should be understood that the foregoing description of the preferred embodiments is not intended to limit the scope of the invention, but rather to limit the scope of the claims, and that those skilled in the art can make substitutions or modifications without departing from the scope of the invention as set forth in the appended claims.
Claims (7)
1. A parallelization point cloud generation DEM method based on machine learning prediction computation intensity is characterized in that: training a CART model, including feature selection, sample data generation, model training and storage; performing recursion division on point cloud data to be processed by using a quadtree, predicting calculation intensity of each tile by adopting a stored CART model, performing dimension reduction processing on tiles in a two-dimensional space based on a Z-curve, and mapping the tiles to each process for parallel interpolation DEM based on the predicted calculation intensity of the tiles and space coding;
the method is realized by adopting the following steps,
step 1, selecting sample characteristics, wherein the sample characteristics comprise the number of the tile point clouds, the number of the neighborhood point clouds, the number of DEM grids in the tile, the point cloud density and the point cloud distribution variance;
step 2, generating sample data, which comprises the steps of generating point cloud data with different areas and different numbers, interpolating the DEM according to the uniform resolution, counting the characteristics of each sample, and taking the time of interpolating the DEM by the point cloud as a sample label;
step 3, splitting sample data into a training set and a verification set, constructing a CART model based on the training set, performing post pruning treatment on the CART model based on the verification set, and locally storing the constructed CART model; the CART represents a classification regression tree;
step 4, dividing the point cloud into tiles based on the quadtree and a preset initial recursion parameter, reading a CART model stored locally, and predicting the calculation intensity of each tile;
step 5, carrying out space coding on the divided tiles based on Z-curves to realize conversion from two-dimensional space to one-dimensional space, and sequencing the tiles according to the Z-curve codes;
step 6, mapping the tiles to each process according to the coding sequence, wherein the mapping strategy is to sum the calculation intensities of all the tiles, divide the calculation intensities by the number of processes to obtain the average calculation intensity contained by each process, and finally map the tiles to each process in sequence according to the coding sequence;
step 7, calculating a parameter time difference ratio reflecting the load balancing condition, if the parameter is larger than a preset corresponding time difference ratio threshold, continuously dividing tiles with calculation intensity larger than a preset corresponding calculation intensity threshold, generating multi-granularity tiles, predicting calculation intensity, and returning to the step 5; otherwise, jumping to the step 8;
and 8, interpolating the DEM by each process based on the mapped tiles and sending the result to a main process, wherein the main process merges the DEM.
2. The machine learning predictive computation strength based parallelized point cloud DEM generation method of claim 1, wherein: the point cloud interpolation DEM is implemented in an IDW manner, which IDW represents inverse distance weighted interpolation.
3. The machine learning predictive computation strength based parallelized point cloud DEM generation method of claim 1, wherein: constructing a CART model based on a training set to generate a complete decision tree, wherein the implementation mode is that firstly, based on the input training set, a feature is selected to split data so that the error of the split data set is minimum; and then dividing the data into two subsets according to the characteristics, judging whether the difference value between the current node error and the node split subset error is smaller than a preset corresponding error change threshold value, if so, stopping dividing, otherwise, continuously judging whether the number of samples contained in the two subsets is smaller than a preset corresponding number threshold value, if so, stopping dividing, otherwise, continuing dividing.
4. The machine learning predictive computation strength based parallelized point cloud DEM generation method of claim 1, wherein: and carrying out post pruning processing on the CART model based on the verification set, wherein the implementation mode is that based on each subtree detected from bottom to top by the verification set, the verification errors of two leaf nodes under the subtree and the verification errors after combination are calculated, and if the verification errors after combination are smaller than the verification errors of the two leaf nodes, the two leaf nodes are combined to finish pruning processing.
5. The machine learning predictive computation strength based parallelized point cloud DEM generation method of claim 1, wherein: the time difference ratio is calculated by dividing the difference between the maximum time consumption and the minimum time consumption in the process by the maximum time consumption.
6. The machine learning predictive computation strength based parallelized point cloud DEM generation method of claim 1, wherein: in step 7, a backtracking parameter is set to control the granularity of tile division and additional division time, and prevent no solution.
7. The machine learning predictive computation intensity based parallelized point cloud DEM generation method of claim 1 or 2 or 3 or 4 or 5 or 6, characterized by: the implementation mode for predicting the calculation intensity of each tile is as follows:
firstly, counting the number of tile point clouds, the number of neighborhood point clouds and the number of DEM grids in the tiles, wherein the number of the tile point clouds, the number of the neighborhood point clouds and the number of the DEM grids in the tiles are divided, and the distribution variance of the point clouds is calculated; reading a CART model stored locally, and sequentially judging each characteristic of the tile to a leaf node according to the split characteristic selected in the CART model; finally, the value of the leaf node is the calculation intensity of the tile.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910151140.1A CN109934917B (en) | 2019-02-28 | 2019-02-28 | Parallelization point cloud generation DEM method based on machine learning prediction computation strength |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910151140.1A CN109934917B (en) | 2019-02-28 | 2019-02-28 | Parallelization point cloud generation DEM method based on machine learning prediction computation strength |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109934917A CN109934917A (en) | 2019-06-25 |
| CN109934917B true CN109934917B (en) | 2023-05-16 |
Family
ID=66986071
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910151140.1A Active CN109934917B (en) | 2019-02-28 | 2019-02-28 | Parallelization point cloud generation DEM method based on machine learning prediction computation strength |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109934917B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112181642B (en) * | 2020-09-16 | 2024-02-02 | 武汉大学 | Artificial intelligence optimization method for space calculation operation |
| CN113345092B (en) * | 2021-05-06 | 2022-07-05 | 武汉大学 | Automatic separation method for ground model and non-ground model of real-scene three-dimensional model |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268342A (en) * | 2013-05-21 | 2013-08-28 | 北京大学 | DEM dynamic visualization accelerating system and method based on CUDA |
| CN107548556A (en) * | 2015-04-21 | 2018-01-05 | Vid拓展公司 | Video coding based on artistic intent |
| CN109146195A (en) * | 2018-09-06 | 2019-01-04 | 北方***科技有限公司 | A kind of blast fragmentation size prediction technique based on cart tree regression algorithm |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170185950A1 (en) * | 2015-12-28 | 2017-06-29 | Draco Ltd. | System for monitoring carts and method |
-
2019
- 2019-02-28 CN CN201910151140.1A patent/CN109934917B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103268342A (en) * | 2013-05-21 | 2013-08-28 | 北京大学 | DEM dynamic visualization accelerating system and method based on CUDA |
| CN107548556A (en) * | 2015-04-21 | 2018-01-05 | Vid拓展公司 | Video coding based on artistic intent |
| CN109146195A (en) * | 2018-09-06 | 2019-01-04 | 北方***科技有限公司 | A kind of blast fragmentation size prediction technique based on cart tree regression algorithm |
Non-Patent Citations (1)
| Title |
|---|
| 决策树学习笔记（三）：CART算法，决策树总结;xiaoyu;《https://www.sohu.com/a/290362448_654419》;20190120;第3-5页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109934917A (en) | 2019-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107506865B (en) | Load prediction method and system based on LSSVM optimization | |
| CN112540849B (en) | Parameter configuration optimization method and system for distributed computing operation | |
| CN109934917B (en) | Parallelization point cloud generation DEM method based on machine learning prediction computation strength | |
| CN112187554A (en) | Operation and maintenance system fault positioning method and system based on Monte Carlo tree search | |
| CN111796917A (en) | Operator operation scheduling method and device | |
| CN116245019A (en) | Load prediction method, system, device and storage medium based on Bagging sampling and improved random forest algorithm | |
| CN106919504B (en) | Test data evolution generation method based on GA algorithm | |
| CN117092908B (en) | Intelligent control method and system for wastewater treatment equipment based on genetic algorithm | |
| CN117094535B (en) | Artificial intelligence-based energy supply management method and system | |
| CN116822742A (en) | Power load prediction method based on dynamic decomposition-reconstruction integrated processing | |
| KR102392131B1 (en) | Food-web network analysis-based ecosystem prediction evaluation system and operation method thereof | |
| CN106874215B (en) | Serialized storage optimization method based on Spark operator | |
| CN114298133A (en) | Short-term wind speed hybrid prediction method and device | |
| CN112181642B (en) | Artificial intelligence optimization method for space calculation operation | |
| CN113158134A (en) | Method and device for constructing non-invasive load identification model and storage medium | |
| CN111260077A (en) | Method and device for determining hyper-parameters of business processing model | |
| Chebykin et al. | Evolutionary neural cascade search across supernetworks | |
| Dhivya et al. | Weighted particle swarm optimization algorithm for randomized unit testing | |
| CN114399152B (en) | Method and device for optimizing comprehensive energy scheduling of industrial park | |
| CN117353302B (en) | New energy power generation power prediction method, device, equipment and medium | |
| CN117009092B (en) | Method and system for dynamically distributing compiling time resources based on multiple multi-arm slot machines | |
| CN117195001A (en) | HPC job power consumption prediction method and system based on power consumption curve and script information | |
| Wang et al. | Improved CBS algorithm based on the idea of pre-planning | |
| CN117172100A (en) | Modeling method for inversion model of self-memory differential equation of power system | |
| CN118017597A (en) | Photovoltaic power station output prediction method, device, equipment and medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |