CN109934917A - Predict that the parallelization point cloud for calculating intensity generates DEM method based on machine learning - Google Patents
Predict that the parallelization point cloud for calculating intensity generates DEM method based on machine learning Download PDFInfo
- Publication number
- CN109934917A CN109934917A CN201910151140.1A CN201910151140A CN109934917A CN 109934917 A CN109934917 A CN 109934917A CN 201910151140 A CN201910151140 A CN 201910151140A CN 109934917 A CN109934917 A CN 109934917A
- Authority
- CN
- China
- Prior art keywords
- tile
- point cloud
- intensity
- dem
- machine learning
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Investigation Of Foundation Soil And Reinforcement Of Foundation Soil By Compacting Or Drainage (AREA)
- Geophysics And Detection Of Objects (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention provides a kind of parallelization point cloud generation DEM method for predicting to calculate intensity based on machine learning, it is characterised in that: carries out the training of CART model, including the generation of feature selecting, sample data, model training and storage；Recurrence division is carried out to point cloud data to be processed using quaternary tree, intensity is calculated using each tile of the CART model prediction of storage, the tile of two-dimensional space is subjected to dimension-reduction treatment based on Z- curve, the tile for being then based on prediction calculates intensity and space encoding and tile is mapped to each task parallelism interpolation DEM.Technical solution of the present invention has high efficiency and feasibility.The present invention is compared with the point cloud of conventional serial generates DEM, predict that the parallelization point cloud for calculating intensity generates DEM method using based on machine learning, performance boost is able to achieve the breakthrough of magnitude, it is effectively saved and executes the time, particularly suitable for the data processing of magnanimity dense point cloud, geoscience applications real time implementation is supported.
Description
Technical field
It is a kind of rapidly and efficiently based on point cloud generation DEM the invention belongs to network geographic information system applied technical field
The method of (Digital Elevation Model, digital elevation model), specifically one kind can be real in a manner of concurrently
The now method that point cloud generates DEM can be achieved the breakthrough of magnitude, propose method compared to traditional serial processing mode on the time
Thought can be used for other geographical spatial data process fields.
Background technique
In recent years with the fast development of data acquisition technology, the acquisition of dense point cloud is more and more common, however how
It efficiently and rapidly handles original point cloud data and therefrom extracts useful information and for example quickly generate DEM using a cloud, at present still
It is so a challenge.Extensive use has been obtained in GIS (GIS-Geographic Information System) field in high-performance calculation, as an allusion quotation
Data-intensive and computation-intensive the application of type, point cloud generation DEM can be obtained considerable performance by parallel computation and mentioned
It rises.However point cloud, as a kind of spatial data, distribution usually has special heterogeneity, and puts the meter of cloud interpolation DEM algorithm
It calculates intensity and does not also depend solely on cloud quantity, meter of the parallelization process according to conventional data division mode and inaccuracy
Calculate strength assessment method, it will cause serious load imbalance phenomenon, reduce parallelization efficiency.Therefore, how accurate evaluation
Algorithm calculates intensity, realizes that fast and efficiently generating DEM based on point cloud becomes current urgent need.
Summary of the invention
It predicts to calculate the parallelization point Yun Sheng of intensity based on machine learning to solve the above problems, the present invention provides a kind of
It at DEM method, realizes that accurate evaluation algorithm calculates intensity, and DEM is fast and efficiently generated based on point cloud.
The technical solution adopted by the present invention includes a kind of parallelization point Yun Shengcheng for predicting to calculate intensity based on machine learning
DEM method carries out the training of CART model, including the generation of feature selecting, sample data, model training and storage；Use four
Fork tree carries out recurrence division to point cloud data to be processed, calculates intensity using each tile of the CART model prediction of storage, is based on
The tile of two-dimensional space is carried out dimension-reduction treatment by Z- curve, and the tile for being then based on prediction calculates intensity and space encoding for tile
It is mapped to each task parallelism interpolation DEM.
Moreover, realized using following steps,
Step 1, select sample characteristics, including by DEM number of grid in tile point cloud quantity, neighborhood point cloud quantity, tile,
Point cloud density and point cloud distribution variance are as sample characteristics；
Step 2, sample data generates, including generating the point cloud data of different area, different number and pressing unified resolution
Rate interpolation DEM, counts each sample characteristics, using the time of cloud interpolation DEM as sample label；
Step 3, sample data is split as training set and verifying collects, CART model is constructed based on training set, based on verifying
Beta pruning is handled after collection carries out CART model, the CART model of localization storage building；The CART presentation class regression tree；
Step 4, a cloud is divided by tile based on quaternary tree and preset initial Recursive parameter, reads localization storage
CART model, predict that each tile calculates intensity；
Step 5, Z- curve is based on to the tile of division to be spatially encoded, realize that two dimension arrives the conversion of the one-dimensional space, and
Tile is sorted according to Z- curve encoding；
Step 6, tile is mapped to each process according to coded sequence, mapping policy is to calculate intensity to all tiles first
Summation, then obtain the average computation intensity that each process is accommodated divided by process number, finally according to coded sequence by tile according to
It is secondary to be mapped to each process；
Step 7, the parameter time difference ratio of reflection non-load balanced case is calculated, if relational threshold when the parameter is greater than preset corresponding
Value is then greater than the preset corresponding tile for calculating intensity threshold to calculating intensity and continues to divide, generates more granularity tiles and predict
Calculate intensity, return step 5；Otherwise, step 8 is jumped to；
Step 8, tile interpolation DEM of each process based on mapping and host process is sent the result to, host process merges DEM.
Moreover, realizing that cloud interpolation a DEM, the IDW indicate inverse distance-weighting interpolation using IDW mode.
Moreover, constructing CART model based on training set, complete decision tree is generated, implementation is as follows,
It is primarily based on the training set of input, a feature progress data splitting is chosen and makes the data set error after division most
It is small；Then data are divided by two subsets according to this feature, judge present node error and subset error after node split
Whether difference is less than preset corresponding error change threshold value, stops dividing if being less than, otherwise continues to judge that two subsets are wrapped
Whether the sample size contained is respectively less than preset respective numbers threshold value, stops dividing if being less than, otherwise continues to divide.
Moreover, beta pruning is handled after being carried out based on verifying collection to CART model, implementation is, bottom-up based on verifying collection
Detection each subtree, the validation error and the validation error after merging for calculating under the subtree two leaf nodes such as merge
Validation error afterwards then merges two leaf nodes less than the validation error of two leaf nodes, completes beta pruning processing.
Moreover, the calculation of time difference ratio is difference maximum time-consuming in process and that minimum is time-consuming divided by maximum time-consuming.
Moreover, the granularity of division and additional division time that one is recalled state modulator tile is arranged, together in step 7
When prevent no solution.
Moreover, it is as follows to predict that each tile calculates intensity implementation:
DEM number of grid in the tile point cloud quantity, neighborhood point cloud quantity, tile of each tile after dividing is counted first,
Point cloud density and point cloud distribution variance；Then the CART model for reading localization storage, according to the division chosen in CART model
Feature successively judges each feature of tile until leaf node；Finally, the value of leaf node is the calculating intensity of tile.
The present invention with conventional serial point cloud generate DEM compared with, using based on machine learning predict calculate intensity and
Rowization point cloud generates DEM method, and performance boost is able to achieve the breakthrough of magnitude, is effectively saved the execution time of algorithm, especially suitable
Together in the data processing of magnanimity dense point cloud, geoscience applications real time implementation is supported, there is important economic value.
Specific embodiment
For the ease of those of ordinary skill in the art understand and implement the present invention, below with reference to embodiment to the present invention make into
The detailed description of one step, it should be understood that implementation example described herein is merely to illustrate and explain the present invention, and is not used to limit
The fixed present invention.
It predicts that the parallelization point cloud for calculating intensity generates DEM method based on machine learning the present invention provides a kind of, realizes
Accurate evaluation algorithm calculates intensity, and fast and efficiently generates DEM based on point cloud.
The technical solution adopted by the present invention is that a kind of predict that the parallelization point cloud for calculating intensity generates DEM based on machine learning
Method carries out the training of CART model, including the generation of feature selecting, sample data, model training and storage；Use quaternary tree
Recurrence division is carried out to point cloud data to be processed, intensity is calculated using each tile of the CART model prediction of storage, it is bent based on Z-
The tile of two-dimensional space is carried out dimension-reduction treatment by line, and the tile for being then based on prediction calculates intensity and space encoding maps tile
To each process.
Moreover, realized using following steps,
Step 1, IDW (Inverse Distance Weighted, in inverse distance-weighting is selected in selection algorithm feature, experiment
Insert) realize point cloud interpolation DEM, will affect algorithm and calculate the factor of intensity includes tile point cloud quantity, neighborhood point cloud quantity, tile
Middle DEM number of grid, point Yun Midu put cloud distribution variance as sample characteristics；
The factor that IDW algorithm calculates intensity that influences has been comprehensively considered in this step, and not only using cloud quantity as commenting
The index of intensity is calculated in estimation.Wherein contiguous range is defined as the big rectangle that tile spatial dimension extends out distance to a declared goal, the distance one
As for the maximum search radius in IDW algorithm.
IDW algorithmic procedure is to initialize initial search radius, search radius growth step pitch, maximum search radius first, so
Afterwards to each DEM grid, judge whether the point cloud quantity in initial search radius is greater than interpolation and needs quantity to be used, if being less than
Then gradually increase by increasing step pitch, until put in radius cloud quantity reach interpolation need quantity to be used and based on IDW interpolation it is public
Formula generates DEM, if search radius is greater than maximum search radius time point cloud and quantity required for interpolation is still not achieved, without interior
It inserts.When it is implemented, the prior art can be used in IDW interpolation formula, it will not go into details by the present invention.
Step 2, sample data generates, and generates different area, the point cloud data of different number and by unified resolution ratio
DEM is inserted, sample data tile is obtained, counts various features (the tile point cloud quantity, neighborhood point cloud quantity, tile of each sample
Middle DEM number of grid, point cloud density and point cloud distribution variance), using the time of cloud interpolation DEM as sample label；
Step 3, sample data is split as training set and verifying collects, CART is constructed based on training set
(Classification and Regression Tree, post-class processing) model carries out CART model based on verifying collection
Beta pruning is handled afterwards, the CART model of localization storage building；
The case where CART model of training has been comprehensively considered in this step to generate over-fitting to training set, by sample data
It splits for independent training set and verifying collection.Training set is used for model training, and verifying collection then carries out hedge clipper to the model of generation
Branch.
Training set generates complete decision tree for model training, and complete decision tree is the CART model under special parameter,
Two parameters are defined when generating CART model is respectively used to control error change and the minimum sample number of cutting feature, error change
Meaning if to be current node error be less than corresponding error change threshold value to the difference of subset error after node split, no longer into
Line splitting；If the meaning of the minimum sample number of cutting feature be division after node in include sample number be less than cutting feature it is minimum
Sample number is then no longer divided.Complete decision tree is exactly that error change threshold value is 0, the feelings that the minimum sample number of cutting feature is 1
It generates under condition, i.e. fully nonlinear water wave as much as possible, prepares for rear beta pruning.
Training generates the specific implementation of complete decision tree are as follows:
It is primarily based on the training set of input, a feature progress data splitting is chosen and makes the data set error after division most
Small, using the variance of predicted value in data set, when division first traverses in sample in each feature data set error in embodiment
Then each value carries out bipartition point to sample according to the value, statistics divide after two samples error and, finally choosing makes
Errors due and that the smallest characteristic value divide sample；
Then data are divided by two subsets according to this feature, subset is missed after judging present node error and node split
Whether the difference of difference is less than preset corresponding error change threshold value, stops dividing if being less than, otherwise continues to judge two subsets
Whether the sample size for being included is respectively less than preset respective numbers threshold value, stops dividing if being less than, otherwise continues to divide.It is real
Subset error in example is applied using the variance of predicted value in subset, subset error is two subsets after dividing after node split
Error and.
Verifying collection carries out hedge clipper branch, specific implementation to the model of generation are as follows:
Collect bottom-up each subtree of detection based on verifying, calculate the validation error of two leaf nodes under the subtree with
And the validation error after merging, the validation error such as the validation error after merging less than two leaf nodes then merge two leaf segments
Point completes beta pruning processing.Validation error is the squared difference of predicted value and actual value.
Step 4, a cloud is divided by tile based on quaternary tree and preset initial Recursive parameter, reads localization storage
CART model, predict that each tile calculates intensity；
When it is implemented, initial Recursive parameter can be preset.Initial Recursive parameter value principle is after dividing recurrence
The area of tile fall in sample space as far as possible, realize more accurately prediction.
Present invention employs CART model prediction computation intensity, realize the accurate evaluation that parallel task calculates intensity.In advance
It is as follows to survey implementation:
First statistics divide after each tile tile point cloud quantity, neighborhood point cloud quantity, DEM number of grid in tile,
Point cloud density and point cloud distribution variance；Then the CART model for reading localization storage, according to the division chosen in CART model
Feature successively judges each feature of tile until leaf node；Finally, the value of leaf node is the calculating intensity of tile.
Step 5, Z- curve is based on to the tile of division to be spatially encoded, realize that two dimension arrives the conversion of the one-dimensional space, and
Tile is sorted according to Z- curve encoding；
Being spatially encoded based on Z- curve is the prior art, conversion and guarantee for realizing two dimension to the one-dimensional space
Certain spacial proximity, it will not go into details by the present invention.
Step 6, tile is mapped to each process according to coded sequence, mapping policy is to calculate intensity to all tiles first
Summation, then obtain the calculating intensity that each process is accommodated divided by number of processes, finally according to coded sequence by tile successively
It is mapped to each process；
In this step, the tile based on prediction calculates intensity and space encoding and tile is mapped to each process, implements
To predict that each tile calculates intensity according to step 4 first, calculate all tiles calculating intensity and, averagely arrive each process, obtain
The calculating intensity that each process is accommodated；Then tile is sequentially mapped to each process by coded sequence, counts each process tile
Intensity summation is calculated, the calculating intensity that can be accommodated until being greater than process.Mapping policy ensure that be dealt with watt of each process
The calculating intensity summation of piece is of substantially equal.
Step 7, the parameter time difference ratio of reflection non-load balanced case is calculated, if relational threshold when the parameter is greater than preset corresponding
Value is then greater than the preset corresponding tile for calculating intensity threshold to calculating intensity and continues to divide, generates more granularity tiles and predict
Calculate intensity, return step 5；Otherwise, step 8 is jumped to.
Further, when this step can add the granularity of division and additional division of a backtracking state modulator tile
Between, while the case where prevent without solving；
This step is consistent with division mode in step 4, continues to continue to divide by the way of quaternary tree to tile.It needs to infuse
Meaning is to use time difference ratio (threshold value is generally 0.1) to determine whether continuing to divide herein, it may appear that condition is unsatisfactory for, and is drawn always
Divide.So needing to add the granularity of division and additional division time that one is recalled state modulator tile, come while anti-
Only without solving the case where.Step number is recalled depending on institute's patient additional division time in practical application, while needing to examine
Considering the tile after dividing should fall in sample space as much as possible.
Non-load balanced case is assessed than parameter using the time difference in this step, the parameter is for time-consuming maximum in process and most
The difference of small time-consuming is divided by maximum time-consuming, it is considered that more excellent less than 0.1.When the parameter is unsatisfactory for specified criteria just in the step
Recall and continue to divide tile, it is therefore desirable to add the division time that a backtracking step number is come outside quota, and prevent the time difference
The case where specified criteria is unable to satisfy than always generation.
Step 8, tile interpolation DEM of each process based on mapping and host process is sent the result to, host process merges DEM.
Step 1-3 of the present invention realizes the training of CART model, specifically includes feature selecting, sample data generates, model
The localization storage of trained and model；Point cloud data to be processed is carried out space division, data mapping and DEM by step 4-8
Generation with merge, including quaternary tree has been used to carry out recurrence division to point cloud data first, based on Z- curve will two dimension it is empty
Between tile carry out dimension-reduction treatment, the tile for being then based on prediction calculates intensity and space encoding and tile is mapped to each process,
It is implemented as calculating the average computation intensity that each process should accommodate, tile is sequentially mapped to each process by coded sequence,
It counts each process tile and calculates intensity summation, the calculating intensity that can be accommodated until being greater than each process.Mapping policy guarantee respectively into
The calculating intensity summation of journey tile to be dealt with is of substantially equal.By using CART model prediction computation intensity, in conjunction with prediction
Calculating intensity realize preferable load balancing using more granularity Z- curve division methods.
When it is implemented, computer software technology, which can be used, in technical solution of the present invention realizes automatic running process.This field
Technical staff can preset relevant threshold value according to experiment or experience.
It should be understood that the part that this specification does not elaborate belongs to the prior art.
It should be understood that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered to this
The limitation of invention patent protection range, those skilled in the art under the inspiration of the present invention, are not departing from power of the present invention
Benefit requires to make replacement or deformation under protected ambit, fall within the scope of protection of the present invention, this hair
It is bright range is claimed to be determined by the appended claims.
Claims (8)
1. a kind of predict that the parallelization point cloud for calculating intensity generates DEM method based on machine learning, it is characterised in that: carry out CART
The training of model, including the generation of feature selecting, sample data, model training and storage；Using quaternary tree to point to be processed
Cloud data carry out recurrence division, calculate intensity using each tile of the CART model prediction of storage, are based on Z- curve for two-dimensional space
Tile carry out dimension-reduction treatment, the tile for being then based on prediction calculates intensity and space encoding and tile is mapped to each task parallelism
Interpolation DEM.
2. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 1
It is: is realized using following steps,
Step 1, sample characteristics are selected, including by DEM number of grid in tile point cloud quantity, neighborhood point cloud quantity, tile, point cloud
Density and point cloud distribution variance are as sample characteristics；
Step 2, sample data generates, including generating the point cloud data of different area, different number and by unified resolution ratio
DEM is inserted, each sample characteristics are counted, using the time of cloud interpolation DEM as sample label；
Step 3, sample data is split as training set and verifying collects, CART model is constructed based on training set, based on verifying collection pair
Beta pruning is handled after CART model carries out, the CART model of localization storage building；The CART presentation class regression tree；
Step 4, a cloud is divided by tile based on quaternary tree and preset initial Recursive parameter, reads localization storage
CART model predicts that each tile calculates intensity；
Step 5, to the tile of division be based on Z- curve be spatially encoded, realize two dimension arrive the one-dimensional space conversion, and will watt
Piece sorts according to Z- curve encoding；
Step 6, tile is mapped to each process according to coded sequence, mapping policy is to calculate intensity to all tiles first to ask
With, then obtain the average computation intensity that each process is accommodated divided by process number, finally according to coded sequence by tile successively
It is mapped to each process；
Step 7, the parameter time difference ratio of reflection non-load balanced case is calculated, if the parameter is greater than the preset corresponding time difference than threshold value,
Then it is greater than the preset corresponding tile for calculating intensity threshold to calculating intensity to continue to divide, generates more granularity tiles and predict to calculate
Intensity, return step 5；Otherwise, step 8 is jumped to；
Step 8, tile interpolation DEM of each process based on mapping and host process is sent the result to, host process merges DEM.
3. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 2
It is: realizes that cloud interpolation a DEM, the IDW indicate inverse distance-weighting interpolation using IDW mode.
4. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 2
It is: CART model is constructed based on training set, generates complete decision tree, implementation is as follows,
It is primarily based on the training set of input, a feature progress data splitting is chosen and makes the data set error after division minimum；
Then data are divided by two subsets according to this feature, judge the difference of subset error after present node error and node split
Whether it is less than preset corresponding error change threshold value, stops dividing if being less than, otherwise continue to judge that two subsets are included
Whether sample size is respectively less than preset respective numbers threshold value, stops dividing if being less than, otherwise continues to divide.
5. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 2
Be: beta pruning is handled after being carried out based on verifying collection to CART model, and implementation is, every based on the bottom-up detection of verifying collection
One subtree, the validation error and the validation error after merging for calculating under the subtree two leaf nodes, such as the verifying after merging
Error then merges two leaf nodes less than the validation error of two leaf nodes, completes beta pruning processing.
6. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 2
Be: the calculation of time difference ratio is difference maximum time-consuming in process and that minimum is time-consuming divided by maximum time-consuming.
7. predicting that the parallelization point cloud for calculating intensity generates DEM method, feature based on machine learning according to claim 2
It is: in step 7, the granularity of division and additional division time that one is recalled state modulator tile is set, while preventing nothing
Solution.
8. the according to claim 1 or 2 or 3 or 4 or 5 or 6 or 7 parallelization point clouds for predicting to calculate intensity based on machine learning
Generate DEM method, it is characterised in that: it is as follows to predict that each tile calculates intensity implementation:
DEM number of grid in the tile point cloud quantity, neighborhood point cloud quantity, tile of each tile after dividing is counted first, puts cloud
Density and point cloud distribution variance；Then the CART model for reading localization storage, according to the disruptive features chosen in CART model,
Successively judge each feature of tile until leaf node；Finally, the value of leaf node is the calculating intensity of 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 true CN109934917A (en) | 2019-06-25 |
| CN109934917B 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) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112181642A (en) * | 2020-09-16 | 2021-01-05 | 武汉大学 | Artificial intelligence optimization method for space calculation operation |
| CN113345092A (en) * | 2021-05-06 | 2021-09-03 | 武汉大学 | Automatic separation method for ground model and non-ground model of real-scene three-dimensional model |
Citations (4)
| 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 |
| US20170185950A1 (en) * | 2015-12-28 | 2017-06-29 | Draco Ltd. | System for monitoring carts and method |
| 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 | 北方爆破科技有限公司 | A kind of blast fragmentation size prediction technique based on cart tree regression algorithm |
-
2019
- 2019-02-28 CN CN201910151140.1A patent/CN109934917B/en active Active
Patent Citations (4)
| 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 |
| US20170185950A1 (en) * | 2015-12-28 | 2017-06-29 | Draco Ltd. | System for monitoring carts and method |
| CN109146195A (en) * | 2018-09-06 | 2019-01-04 | 北方爆破科技有限公司 | A kind of blast fragmentation size prediction technique based on cart tree regression algorithm |
Non-Patent Citations (1)
| Title |
|---|
| XIAOYU: "决策树学习笔记（三）：CART算法，决策树总结", 《HTTPS://WWW.SOHU.COM/A/290362448_654419》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112181642A (en) * | 2020-09-16 | 2021-01-05 | 武汉大学 | Artificial intelligence optimization method for space calculation operation |
| CN112181642B (en) * | 2020-09-16 | 2024-02-02 | 武汉大学 | Artificial intelligence optimization method for space calculation operation |
| CN113345092A (en) * | 2021-05-06 | 2021-09-03 | 武汉大学 | Automatic separation method for ground model and non-ground model of real-scene three-dimensional model |
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109934917B (en) | 2023-05-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Applications of multi-objective dimension-based firefly algorithm to optimize the power losses, emission, and cost in power systems | |
| CN111103565B (en) | Data transformation method and system based on intelligent electric energy meter metering error analysis | |
| CN109145957A (en) | The identification and processing method and processing device of power distribution network abnormal index based on big data | |
| CN107124394A (en) | A kind of powerline network security postures Forecasting Methodology and system | |
| CN108233359B (en) | Partition optimization method and system for extra-high voltage alternating current-direct current feed-in receiving end power grid | |
| CN109102146B (en) | Electric power system risk assessment acceleration method based on multi-parameter linear programming | |
| CN104035816A (en) | Cloud computing task scheduling method based on improved NSGA-II | |
| CN111612528A (en) | Method, device and equipment for determining user classification model and storage medium | |
| CN105205052B (en) | A kind of data digging method and device | |
| CN102984137A (en) | Multi-target server scheduling method based on multi-target genetic algorithm | |
| CN108133260A (en) | The workflow schedule method of multi-objective particle swarm optimization based on real-time status monitoring | |
| CN109934917A (en) | Predict that the parallelization point cloud for calculating intensity generates DEM method based on machine learning | |
| Moraes et al. | A novel multi-objective evolutionary algorithm based on subpopulations for the bi-objective traveling salesman problem | |
| CN109298930A (en) | A kind of cloud workflow schedule method and device based on multiple-objection optimization | |
| Mariani et al. | A hybrid shuffled complex evolution approach with pattern search for unconstrained optimization | |
| Ploussard et al. | A search space reduction method for transmission expansion planning using an iterative refinement of the DC load flow model | |
| CN106469114B (en) | A kind of Parallel Computing Performance detection system and its method towards communication test | |
| Stiskalek et al. | The scatter in the galaxy–halo connection: a machine learning analysis | |
| CN112087316A (en) | Network anomaly root cause positioning method based on anomaly data analysis | |
| CN106993296A (en) | The performance estimating method and device of terminal | |
| CN114117917B (en) | Multi-objective optimization ship magnetic dipole array modeling method | |
| Gerster et al. | Comparison of random sampling and heuristic optimization-based methods for determining the flexibility potential at vertical system interconnections | |
| CN109871998B (en) | Power distribution network line loss rate prediction method and device based on expert sample library | |
| CN114118608A (en) | Power grid business matching optimization method and device, terminal and storage medium | |
| CN113381869B (en) | Fault root cause analysis method and device, and computer readable storage 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 |