CN108573059B - Time sequence classification method and device based on feature sampling - Google Patents
Time sequence classification method and device based on feature sampling Download PDFInfo
- Publication number
- CN108573059B CN108573059B CN201810384213.7A CN201810384213A CN108573059B CN 108573059 B CN108573059 B CN 108573059B CN 201810384213 A CN201810384213 A CN 201810384213A CN 108573059 B CN108573059 B CN 108573059B
- Authority
- CN
- China
- Prior art keywords
- sampling
- data
- time series
- data set
- training
- 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
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to the technical field of data processing, and provides a time series classification method and a device based on feature sampling, wherein the method comprises the following steps: converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method, and converting a time sequence data set for testing into a testing data set with characteristics of equal length; performing model training by using the training data set with the equal-length features by adopting an ensemble learning classification method; and carrying out time series classification on the test data sets with the characteristics of equal length by using the trained model. According to the invention, time sequence data sets with different lengths are converted into data sets with characteristics of equal length through a characteristic sampling method, and then the data sets are classified by adopting an ensemble learning classification method, so that the accuracy of time sequence classification is improved, and the large-scale time sequence data can be accurately classified.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a time series classification method and device based on feature sampling.
Background
Time series classification has a wide range of applications in many fields, such as Hidden Markov Models (HMMs) and Dynamic Time Warping (DTW) in speech processing and speech recognition. In the database direction, we refer to a database consisting of values that change over time, as a time series database, and the work of data mining in such a time series database is referred to as time series mining. The time series classification problem is important for time series mining, and compared with common and conventional classification data, the time series has the characteristics of unfixed length, strong front-back dependency of the sequence data and more noise data.
The time series classification problem cannot directly apply a common classification learning algorithm to perform classification learning because of the special attributes of the data series. Conventional classification learning algorithms, such as Support Vector Machines (SVMs), Logistic Regression (LR), etc., do not work well in dealing with time series problems. At present, the algorithm with a better effect is a Dynamic Time Warping (DTW) algorithm, the idea of the DTW algorithm is to find the optimal matching of two sequences by using a dynamic programming algorithm so as to obtain the category of the sequence to be predicted, the DTW algorithm is proposed by Berndt and Clifford, and a large number of experiments and actual effects prove that the DTW has good performance and accuracy on most time sequence classification problems.
Although many algorithms for time series classification exist at present, the algorithms cannot meet the requirements of the time series classification problem in terms of generalization and accuracy, which is mainly reflected in the following aspects:
1. many time series data conversion and classification algorithms proposed at present are effective for small-scale time series data, and are not applicable for large-scale time series data due to limitations of conditions such as memory and processing time.
2. Time series data have complex attributes such as local similarity and dependency relationship, and the current time series processing algorithm cannot process the problem of local similarity.
3. At present, most time classification algorithms are based on a single linear model or a tree model, and the single model is weak in performance and low in accuracy. The model adopted by the method is a random forest model, a plurality of decision trees are contained in the model, and results are voted by the decision trees together in training and prediction, so that the defect of weak showing capability of a single model is overcome.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a time series classification method and apparatus based on feature sampling, aiming at one or more of the above defects in time series classification in the prior art.
In order to solve the above technical problem, the present invention provides a time series classification method based on feature sampling, including:
converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method, and converting a time sequence data set for testing into a testing data set with characteristics of equal length;
performing model training by using the training data set with the equal-length features by adopting an ensemble learning classification method;
and carrying out time series classification on the test data sets with the characteristics of equal length by using the trained model.
Optionally, the feature sampling method is a segmented feature sampling method, and includes:
setting the segment length l1Number of segments m1And the spacing between segments g 1;
for each time series data in the time series data set, sampling is carried out from the starting position of the time series data, and continuous l is selected1Taking the time series data as the first section of characteristic data, and then setting the position of the beginning of each section of sampling as the position of the beginning of the previous section of sampling plus the interval g between the sections1Selecting l from each segment1Sampling the sequence data, and converting each time sequence data into m1And (4) characteristic data.
Optionally, the ensemble learning classification method is a bagging-based random forest classifier.
Optionally, the feature sampling method is a random feature sampling method, and includes:
setting the number l of random feature samples2Number of samplings m2And an initial sampling position b1；
For each time series data in the time series data set, from an initial sampling location b of the time series data1Starting with the selection of successive ones of l2The time sequence data is used as a first piece of feature data, subsequent sampling positions are randomly determined after the first piece of feature data, and then continuous l are selected2The time-series data is used as the second feature data,so as to m2After the bar characteristic data is selected, ensuring that the selected characteristic data has a front-back sequence in the sampling process, repeating the selection operation from the time sequence starting position when the selected characteristic dimension exceeds the longest length of the time sequence, and converting each bar of time sequence data into m after sampling2Bar feature data.
Optionally, the feature sampling method is an equal time interval sampling method, and includes:
setting the time interval g2Initial sampling position b1Number of sampling features m3；
For each time series data in the time series data set, from an initial sampling location b of the time series data1At the beginning, according to the time interval g2Sampling is carried out in sequence, and the characteristic number of the sampling is m3When the sampling interval exceeds the maximum length of the time sequence, sampling is continued from the starting position of the time sequence, and after the sampling at equal time intervals, each piece of time sequence data is converted into a piece of characteristic data.
The invention also provides a time sequence classification device based on the characteristic sampling, which at least comprises: the device comprises a data sampling unit, a model training unit and a model prediction unit;
the data sampling unit is used for converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method and converting a time sequence data set for testing into a testing data set with characteristics of equal length;
the model training unit is used for performing model training by using the training data set with the equal-length characteristics by adopting an ensemble learning classification method;
and the model prediction unit is used for carrying out time series classification on the test data sets with the characteristics of equal length by using the trained model.
Optionally, the feature sampling method adopted by the data sampling unit is a segmented feature sampling method, including:
setting the segment length l1Number of segments m1And is segmentedInterval g1；
For each time series data in the time series data set, sampling is carried out from the starting position of the time series data, and continuous l is selected1Taking the time series data as the first section of characteristic data, and then setting the position of the beginning of each section of sampling as the position of the beginning of the previous section of sampling plus the interval g between the sections1Selecting l from each segment1Sampling the sequence data, and converting each time sequence data into m1And (4) characteristic data.
Optionally, the ensemble learning classification method for the model training unit sampling is a bagging-based random forest classifier.
Optionally, the feature sampling method adopted by the data sampling unit is a segmented feature sampling method, which includes:
setting the number l of random feature samples2Number of samplings m2And an initial sampling position b1；
For each time series data in the time series data set, from an initial sampling location b of the time series data1Starting with the selection of successive ones of l2The time sequence data is used as a first piece of feature data, subsequent sampling positions are randomly determined after the first piece of feature data, and then continuous l are selected2Using the time series data as the second feature data, and doing so until m2After the bar characteristic data is selected, ensuring that the selected characteristic data has a front-back sequence in the sampling process, repeating the selection operation from the time sequence starting position when the selected characteristic dimension exceeds the longest length of the time sequence, and converting each bar of time sequence data into m after sampling2Bar feature data.
Optionally, the feature sampling method adopted by the data sampling unit is a segmented feature sampling method, which includes:
setting the time interval g2Initial sampling position b1Number of sampling features m3；
For each time series data in the set of time series data, an initial sample from the time series dataPosition b1Firstly, sampling is carried out in sequence according to a time interval g, and the characteristic number of the sampling is m3When the sampling interval exceeds the maximum length of the time sequence, sampling is continued from the starting position of the time sequence, and after the sampling at equal time intervals, each piece of time sequence data is converted into a piece of characteristic data.
Setting the time interval g2Sampling position b2Number of sampling features m3；
For each piece of time-series data in the time-series data set, sampling is sequentially performed at a time interval g from a sampling position b of the time-series data, and m is sampled3Secondly, when the sampling interval exceeds the maximum length of the time sequence, sampling is continued from the starting position of the time sequence, and after the sampling is processed by the equal time interval sampling method, each piece of time sequence data is converted into m3Bar feature data.
The time sequence classification method and device based on the characteristic sampling provided by the embodiment of the invention at least have the following beneficial effects:
1. according to the invention, time sequence data sets with different lengths are converted into data sets with characteristics of equal length through a characteristic sampling method, and then classification is carried out by integrating a learning classification method, so that the accuracy of time sequence classification is improved, and the large-scale time sequence data can be accurately classified.
2. Furthermore, the invention can select the combination of the segmented feature sampling method and the random forest classifier, and the random forest classifier has the advantages of strong generalization, high classification accuracy, strong fitting capability, distributed realization and the like.
3. The invention also provides other two applicable characteristic sampling methods, namely a random characteristic sampling method and an equal time interval sampling method, whether the time sequence can be converted into the characteristic data with the same dimensionality or not.
Drawings
FIG. 1 is a flowchart of a method for feature sample based time series classification according to an embodiment of the present invention;
FIG. 2a is a schematic diagram of a segmented feature sampling method;
FIG. 2b is a schematic diagram of a random feature sampling method;
FIG. 2c is a schematic diagram of an equal time interval sampling method;
FIG. 3 is a flowchart of a feature sample-based time series classification method according to a sixth embodiment of the present invention;
fig. 4 is a schematic diagram of a time-series classification apparatus based on feature sampling according to a seventh embodiment of the present invention.
In the figure: 401: a data sampling unit; 402: a model training unit; 403: and a model prediction unit.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.
Example one
As shown in fig. 1, the method for classifying a time series based on feature sampling according to the embodiment of the present invention may include the following steps:
step S101: converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method, and converting a time sequence data set for testing into a testing data set with characteristics of equal length;
step S102: performing model training by using the training data set with the equal-length features by adopting an ensemble learning classification method;
step S103: and carrying out time series classification on the test data sets with the characteristics of equal length by using the trained model.
The invention mainly aims at the characteristic selection problem of the time sequence problem. Given a time series of sample sets, each sample contains a time series XiThe purpose of time series classification is to predict the label for a given new time series. Because the time series have the characteristic of unequal length and the correlation exists before and after the time series data, for example, the time series belonging to the same category may have very high similarity of only a part of fragments, the invention mainly utilizes the characteristic of local similarity in the time series to carry out random sampling on the time series.
The time series classification method based on the characteristic sampling provided by the embodiment of the invention can be used for sampling data aiming at different data sets and then classifying the data by applying a machine learning algorithm, thereby improving the accuracy of time series classification.
It should be understood that the present invention is not limited to the execution sequence of the steps in fig. 1, for example, the step of "converting the test time-series data set into the test data set with equal-length features" in step S101 in the above method may also be performed after the model training in step S102 is completed.
Example two
On the basis of the time series classification method based on feature sampling provided in the first embodiment, the feature sampling method in step S101 is a segmented feature sampling method, which can be specifically implemented as follows:
first, a segment length l is set1Number of segments m1And the interval g between segments1；
Then, for each time series data in the time series data set, as shown in FIG. 2a, sampling is performed from the start position of the time series data, and a continuous l is selected1The time series data is used as the first section of characteristic data, and the position of the beginning of each section of sampling is the position of the beginning of the previous section of sampling plus the segmentationInterval g between1Selecting l from each segment1Sampling the sequence data, and converting each time sequence data into m1Individual characteristic data (i.e., sub-time series T1, T2, T3 … …). Preferably,/1＝5，g1＝2，m1＝3。
EXAMPLE III
On the basis of the time series classification method based on the feature sampling provided in the second embodiment, the ensemble learning classification method adopted in step S102 is a bagging-based random forest classifier.
After the sampling method is determined, the selection and design of the classifier are also important research contents of the invention. The current mainstream classification methods include basic classification methods such as decision trees and logistic regression, and also include integrated learning methods such as XGboost and gradient lifting trees. The algorithms such as decision trees, logistic regression and the like are easy to write, but the effect is inferior to that of the ensemble learning classification method in accuracy, so that the effect of the random forest classifier is optimal by adopting the ensemble learning classification method and combining experimental results in the selection of the classifier. The main reason is that the random forest classifier has the advantages of strong generalization, high classification accuracy, strong fitting capability, distributed realization and the like, and meanwhile, when the random forest algorithm is used for model training, random column sampling can be carried out on the trained feature data, so that the random column sampling is in accordance with the concept of segmented sampling.
Example four
On the basis of the time series classification method based on feature sampling provided in the first embodiment, the feature sampling method adopted in step S101 is a random feature sampling method, which can be specifically implemented as follows:
setting the number l of random feature samples2Number of samplings m2And an initial sampling position b1；
For each time series data in the time series data set, as shown in FIG. 2b, the initial sampling position b of the time series data is selected1Starting with the selection of successive ones of l2The time-series data is used as a first feature data, and the subsequent time-series data is randomly determined after the first feature dataSampling position b2、b3Etc. and then selecting successive ones of l2The time series data is used as a second characteristic data … … and a third characteristic data … …, and the time series data is m2After the bar characteristic data is selected, strictly ensuring that the selected characteristic data has a front-back sequence in the sampling process, repeating the selection operation from the time sequence starting position when the selected characteristic dimension exceeds the longest length of the time sequence, and converting each time sequence data into m after sampling2The bar feature data (i.e., the sub-time series T1, T2, T3 … …). Preferably,/2＝5，g1＝2。
EXAMPLE five
On the basis of the time series classification method based on feature sampling provided in the first embodiment, the feature sampling method adopted in step S101 is an equal time interval sampling method, which can be specifically implemented as follows:
setting the time interval g2Initial sampling position b1Number of sampling features m3；
For each time series data in the time series data set, as shown in FIG. 2c, the initial sampling position b of the time series data is selected1At the beginning, sampling is carried out in sequence according to a time interval g, and the number of the samples is m3When the sampling position exceeds the maximum length of the time series, the sampling is continued from the time series starting position, and after the processing of the equal time interval sampling method, each piece of time series data is converted into a piece of characteristic data (namely, a sub-time series T1). Preferably, g2＝2，m3＝5。
For the classification of time series data, the common classification method is not suitable because the time series data may have different feature dimensions, and the common machine learning classification method requires the same feature dimension when performing model training and classification. Therefore, how to convert the time series into feature data having the same dimension becomes the research focus of the present invention. Aiming at the characteristic that the time sequences are not equal in length, the invention provides a sampling method for converting the time sequence data with different lengths into training data with equal lengths. The time series data are classified by combining a sampling method and a component classifier, and a segmented sampling method, a random feature sampling method and an equal time interval sampling method are provided aiming at the sampling method. The three sampling methods are mainly provided aiming at the characteristics that time sequences are not equal in length and have strong time relevance, and are converted into feature data with equal length after sampling, so that classifiers such as a gradient decision tree, a random forest and the like are used for classification.
After the three different sampling methods are carried out, the three to five different sampling methods can convert the unequal-time sequence data sets into data sets with equal-length characteristics, so that model training and model prediction are carried out, and the classification accuracy is obtained. For the three sampling methods, the invention carries out experiments in sequence and compares the advantages and disadvantages of the three sampling methods. From the accuracy of classification, the segmented feature sampling method of the third embodiment is superior to the sampling methods of the fourth and fifth embodiments. The reason is that in the sampling method of the third embodiment, the characteristics of strong correlation in time series and the similarity of the time series partial fragments are sufficiently combined. Although the sampling method in the fourth embodiment considers the characteristic of the similarity of the segments of the time sequence, the random sampling causes the possibility of confusion of time correlation between dimensions, and in the sampling method in the fifth embodiment, the equal time interval sampling method considers the time correlation of the time sequence, but the interval sampling may cause the loss of important segments, so that the theoretical analysis and the experimental effect are integrated, and the segmented feature sampling method in the third embodiment is preferably adopted.
EXAMPLE six
With reference to the first to the fifth embodiments, the present invention provides a time series classification method based on feature sampling, as shown in fig. 3, which may include the following steps:
step 301: carrying out normalization processing on a time sequence data set for training;
step S302: converting the time sequence data set for training after normalization processing into a training data set with characteristics of equal length by a characteristic sampling method; the characteristic sampling method in this step may adopt the aforementioned segmented characteristic sampling method, random characteristic sampling method and equal time interval sampling method. More preferably, the feature sampling method adopts a segmented feature sampling method;
step S303: determining parameters in the feature sampling method by using a cross validation method to obtain the optimal parameters of the feature sampling method;
in the foregoing embodiments of the present invention, three characteristic sampling methods are proposed, and the three characteristic sampling methods include an interval (e.g., the interval g between the foregoing segments)1And a time interval g2) Initial sampling position b1Number of times (e.g. number of segments m as described above)1And the number of times of sampling m2) And the parameters can be determined by adopting the following cross validation method. Firstly, dividing a training set into two parts, wherein one part is used for training, the other part is used for predicting, firstly, defining a value range for parameters needing to determine values, and for an interval g between segments1And a time interval g2Setting the value range to be 1-10, and aiming at the initial sampling position b1The value can be determined by adopting a random generation method. Number of segments m1And the number of times of sampling m2It can be set to the number of classes in the training set plus one, since the final result is voted from the m training samples. Segment length l1Number of random feature samples l2And number of sampling features m3The length of the time sequence is related, the value range is set to be 0.85-1 of the length of the time sequence, and each iteration is increased by 0.01. And traversing in the value range in sequence, and selecting a parameter set with the highest accuracy in the test set as the final value of the parameter.
Step S304: performing model training by using the training data set with the equal-length characteristics obtained in the step S302 by using a random forest classifier to obtain a trained model;
step 305: normalizing the time series data set for testing;
step S306: converting the time sequence data set for training after normalization processing into a training data set with characteristics of equal length by a characteristic sampling method; the type of the feature sampling method is the same as that in the step S302, and the optimal parameters determined in the step S303 are adopted.
Step S307: and S304, carrying out time sequence classification on the test data set with the characteristics of equal length by using the model trained in the step S304, voting according to a classification result, and obtaining a prediction result, namely obtaining a classification label corresponding to the time sequence in the test time sequence data set.
EXAMPLE seven
As shown in fig. 4, the apparatus for classifying a time series based on feature samples according to an embodiment of the present invention may include: a data sampling unit 401, a model training unit 402, and a model prediction unit 803;
the data sampling unit 401 is configured to convert the time series data set for training into a training data set with characteristics of equal length by using a characteristic sampling method, and convert the time series data set for testing into a testing data set with characteristics of equal length. The operation performed by the data sampling unit 401 is the same as that in step S101, and is not described herein again;
a model training unit 402, configured to perform model training by using the training data set with equal length features by using an ensemble learning classification method; the operation performed by the model training unit 402 is the same as the operation performed in step S102, and is not described herein again;
a model prediction unit 403, configured to perform time series classification on the test data sets with equal-length features by using the trained model; the operation performed by the model prediction unit 403 is the same as that performed in step S103, and will not be described herein again.
It should be noted that the time series classification device based on feature sampling provided in the embodiment of the present invention may be implemented by software, or implemented by hardware, or implemented by a combination of hardware and software. Taking a software implementation as an example, as shown in fig. 4, as a logical apparatus, the apparatus is formed by reading a corresponding computer program instruction in a non-volatile memory into a memory by a CPU of a device in which the apparatus is located and running the computer program instruction.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (2)
1. A time series classification method based on feature sampling is characterized by comprising the following steps:
converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method, and converting a time sequence data set for testing into a testing data set with characteristics of equal length;
performing model training by using the training data set with the equal-length features by adopting an ensemble learning classification method;
using the trained model to perform time series classification on the test data sets with the characteristics of equal length;
the feature sampling method is a segmented feature sampling method, and comprises the following steps:
setting the segment length l1Number of segments m1And the interval g between segments1；
For each time series data in the time series data set, sampling is carried out from the starting position of the time series data, and continuous l is selected1Taking the time series data as the first section of characteristic data, and then setting the position of the beginning of each section of sampling as the position of the beginning of the previous section of sampling plus the interval g between the sections1Selecting l from each segment1Sampling the sequence data, and converting each time sequence data into m1A piece of feature data;
the ensemble learning classification method is a bagging-based random forest classifier.
2. A time series classification apparatus based on feature sampling, comprising at least: the device comprises a data sampling unit, a model training unit and a model prediction unit;
the data sampling unit is used for converting a time sequence data set for training into a training data set with characteristics of equal length by a characteristic sampling method and converting a time sequence data set for testing into a testing data set with characteristics of equal length;
the model training unit is used for performing model training by using the training data set with the equal-length characteristics by adopting an ensemble learning classification method;
the model prediction unit is used for carrying out time series classification on the test data sets with the characteristics of equal length by using the trained model;
the characteristic sampling method adopted by the data sampling unit is a segmented characteristic sampling method, and comprises the following steps:
setting the segment length l1Number of segments m1And the interval g between segments1；
For each time series data in the time series data set, sampling is carried out from the starting position of the time series data, and continuous l is selected1Taking the time series data as the first section of characteristic data, and then setting the position of the beginning of each section of sampling as the position of the beginning of the previous section of sampling plus the interval g between the sections1Selecting l from each segment1Sampling the sequence data, and converting each time sequence data into m1A piece of feature data;
the ensemble learning classification method for the model training unit to sample is a bagging-based random forest classifier.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810384213.7A CN108573059B (en) | 2018-04-26 | 2018-04-26 | Time sequence classification method and device based on feature sampling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810384213.7A CN108573059B (en) | 2018-04-26 | 2018-04-26 | Time sequence classification method and device based on feature sampling |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108573059A CN108573059A (en) | 2018-09-25 |
| CN108573059B true CN108573059B (en) | 2021-02-19 |
Family
ID=63575385
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810384213.7A Active CN108573059B (en) | 2018-04-26 | 2018-04-26 | Time sequence classification method and device based on feature sampling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108573059B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109710065B (en) * | 2018-12-18 | 2021-12-28 | 苏州大学 | Method for recognizing walking regulation intention based on brain hemoglobin information |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880621A (en) * | 2011-07-14 | 2013-01-16 | 富士通株式会社 | Method and device for extracting similar sub time sequences |
| CN103279643A (en) * | 2013-04-26 | 2013-09-04 | 华北电力大学（保定） | Method for calculating time sequence similarity |
| CN103577562A (en) * | 2013-10-24 | 2014-02-12 | 河海大学 | Multi-measurement time series similarity analysis method |
| CN104102680A (en) * | 2013-04-15 | 2014-10-15 | 肖瑞 | Coding indexing mode for time sequences |
| CN104809226A (en) * | 2015-05-07 | 2015-07-29 | 武汉大学 | Method for early classifying imbalance multi-variable time sequence data |
| CN106960099A (en) * | 2017-03-28 | 2017-07-18 | 清华大学 | A kind of manipulator grasp stability recognition methods based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9396253B2 (en) * | 2013-09-27 | 2016-07-19 | International Business Machines Corporation | Activity based analytics |
-
2018
- 2018-04-26 CN CN201810384213.7A patent/CN108573059B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102880621A (en) * | 2011-07-14 | 2013-01-16 | 富士通株式会社 | Method and device for extracting similar sub time sequences |
| CN104102680A (en) * | 2013-04-15 | 2014-10-15 | 肖瑞 | Coding indexing mode for time sequences |
| CN103279643A (en) * | 2013-04-26 | 2013-09-04 | 华北电力大学（保定） | Method for calculating time sequence similarity |
| CN103577562A (en) * | 2013-10-24 | 2014-02-12 | 河海大学 | Multi-measurement time series similarity analysis method |
| CN104809226A (en) * | 2015-05-07 | 2015-07-29 | 武汉大学 | Method for early classifying imbalance multi-variable time sequence data |
| CN106960099A (en) * | 2017-03-28 | 2017-07-18 | 清华大学 | A kind of manipulator grasp stability recognition methods based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| Multi-Scale Convolutional Neural Networks for Time Series Classification;ZhiCheng Cui;《https://arxiv.org/pdf/1603.06995》;20160511;正文2.2与3.2节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108573059A (en) | 2018-09-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109308912B (en) | Music style recognition method, device, computer equipment and storage medium | |
| CN110019421B (en) | Time sequence data classification method based on data characteristic fragments | |
| CN1726532A (en) | Sensor based speech recognizer selection, adaptation and combination | |
| Srikanthan et al. | Implementing the dynamic time warping algorithm in multithreaded environments for real time and unsupervised pattern discovery | |
| US20210074270A1 (en) | Keyword spotting apparatus, method, and computer-readable recording medium thereof | |
| Cholewa et al. | Estimation of the number of states for gesture recognition with Hidden Markov Models based on the number of critical points in time sequence | |
| US20210004727A1 (en) | Hyper-parameter tuning method for machine learning algorithms using pattern recognition and reduced search space approach | |
| Du et al. | A quantitative analysis framework for recurrent neural network | |
| CN111309577B (en) | Spark-oriented batch application execution time prediction model construction method | |
| CN108573059B (en) | Time sequence classification method and device based on feature sampling | |
| JP2012181579A (en) | Pattern classification learning device | |
| Fotso et al. | Grasp heuristic for time series compression with piecewise aggregate approximation | |
| Prudêncio et al. | Combining meta-learning and active selection of datasetoids for algorithm selection | |
| CN111897734A (en) | Fuzzy test case selection method and device based on online incremental learning | |
| Zhang et al. | Time series classification by shapelet dictionary learning with SVM‐based ensemble classifier | |
| Schiff et al. | Characterizing the latent space of molecular deep generative models with persistent homology metrics | |
| Brown et al. | Automatic construction of accurate bioacoustics workflows under time constraints using a surrogate model | |
| CN115221045A (en) | Multi-target software defect prediction method based on multi-task and multi-view learning | |
| Wang et al. | Shapelet classification algorithm based on efficient subsequence matching | |
| JP2018132678A (en) | Turn-taking timing identification apparatus, turn-taking timing identification method, program and recording medium | |
| Gaiceanu et al. | On CNN applied to speech-to-text–comparative analysis of different gradient based optimizers | |
| Jindal et al. | Ensemble Based-Cross Project Defect Prediction | |
| CN117690455B (en) | Sliding window-based partial synthesis fake voice detection method and system | |
| Luzyanin et al. | Problems of tiny changes analysis in complex time series using dynamic time warping algorithm | |
| Fotso et al. | Parameter free piecewise dynamic time warping for time series classification |
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 |