WO2020147286A1 - 一种边缘端的嵌入式时间序列决策树分类方法及系统 - Google Patents

一种边缘端的嵌入式时间序列决策树分类方法及系统 Download PDF

Info

Publication number
WO2020147286A1
WO2020147286A1 PCT/CN2019/096748 CN2019096748W WO2020147286A1 WO 2020147286 A1 WO2020147286 A1 WO 2020147286A1 CN 2019096748 W CN2019096748 W CN 2019096748W WO 2020147286 A1 WO2020147286 A1 WO 2020147286A1
Authority
WO
WIPO (PCT)
Prior art keywords
edge
data
decision tree
classification
vfdt
Prior art date
Application number
PCT/CN2019/096748
Other languages
English (en)
French (fr)
Inventor
李锐
王相成
宗云兵
于治楼
段强
Original Assignee
山东浪潮人工智能研究院有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 山东浪潮人工智能研究院有限公司 filed Critical 山东浪潮人工智能研究院有限公司
Publication of WO2020147286A1 publication Critical patent/WO2020147286A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition

Definitions

  • the invention relates to a real-time data classification modeling at the edge, in particular to an embedded time sequence decision tree classification method and system at the edge.
  • Categorizing data is a common requirement. Unlike clustering algorithms, classification algorithms require training data for modeling. After the modeling is completed, the model can be used for continuous forecasting.
  • the most classic classification algorithm is the decision tree.
  • the decision tree uses the information gain method in informatics to judge the importance of sample variables, and then distinguishes the samples according to the importance ranking.
  • the disadvantage of this method is that it is difficult to perform one-time modeling processing on massive data because it requires large memory and is time-consuming.
  • the patent document with the patent number CN104318270A discloses a land cover classification method based on MODIS time series data.
  • the method is specifically carried out according to the following steps: 1. Establish the original curve; 2. Filter the original curve to fit the initial curve ; 3. Establish a two-dimensional array of cloudless images of the initial curve pixels; 4. Set a threshold T, where Yi ⁇ yi; 5. Processed original curve; 6. Get the reconstructed NDVI annual change curve; 7. Extracting vegetation growing season parameters to form characteristic images; 8. Determine the final voting classification results and other steps; the present invention is applied to the field of land cover classification based on MODIS time series data.
  • this technical solution has to solve the problems of the traditional method, the negative deviation of the vegetation index, and the reduced accuracy of the SG reconstruction result. It cannot realize the one-time modeling processing of massive data, and at the same time ensure the efficiency of the memory requirements and the short time.
  • the technical task of the present invention is to provide an edge-end embedded time series decision tree classification method and system to solve the problem of how to implement one-time modeling processing of massive data while ensuring the efficiency and short time consumption of memory requirements.
  • an edge-end embedded time series decision tree classification method which applies the VFDT algorithm to the edge-end calculation, realizes the edge-end calculation of large data volume and can achieve the correct Real-time processing of demanding requirements; the specific steps are as follows:
  • VFDT algorithm very fast decision tree or time series decision tree
  • the edge end transmits the classification results to the cloud; the edge end can only transmit the analysis results obtained to the cloud, or the original data can also be transmitted to the cloud together; if only the results are transmitted, this avoids transmitting a large amount of data at the edge The time cost of coming.
  • the edge terminal collects data through a sensor, and the sensor is installed at a location where data collection is required to complete data collection, and the sensor sends the collected data to the edge terminal in real time.
  • a temperature sensor is installed in a place that can directly detect temperature. .
  • the VFDT algorithm analyzes and processes the streaming data in real time, makes classification judgments according to the characteristics of the incoming data, and judges whether the decision tree needs to be updated.
  • the steps for updating the decision tree are as follows:
  • HT is a decision tree with a single leaf node l_1;
  • step (8) If it is not the same type, proceed to step (8) in the next step;
  • step (11) If yes, perform step (11) in the next step;
  • the VFDT algorithm is programmed as a VFDT algorithm software system using C/C++ language, and the VFDT algorithm software system is deployed as an intelligent system for data processing to the edge device.
  • the edge end transmits the classification result to the cloud in real time or stores it for a specified time, and the specified time is a time period set at the edge end according to user requirements.
  • An edge-end embedded time-series decision tree classification system includes edge-end devices, sensors, and cloud devices.
  • the sensors and edge-end devices are connected wirelessly and exchange data.
  • the cloud devices and edge-end devices are connected wirelessly or wiredly. transfer data;
  • Edge devices are used to collect data, store data, process data and return classification results, and at the same time determine whether the decision tree needs to be updated
  • the cloud device is used to receive the classification result.
  • an edge data collection module an intelligent data processing module based on the VFDT algorithm, a data storage module, and a classification result transmission module are deployed in the edge data device;
  • the edge data collection module is used to collect data collected by sensors
  • the intelligent data processing module based on the VFDT algorithm is used to classify and process the data collected by the sensor using the VFDT algorithm;
  • the data storage module is used to store the data collected by the sensor
  • the classification result transmission module is used to transmit the classification result of the intelligent data processing module based on the VFDT algorithm to the cloud.
  • VFDT algorithm Very Fast Decision Tree, very fast decision tree
  • VFDT is a modeling algorithm for streaming data, which is an extension of the decision tree in real-time data
  • VFDT is a decision-making based on Hoeffding's inequality
  • the tree method uses the statistical inequality Hoeffding to determine whether a node should be used as the basis for classification judgment.
  • the present invention applies the VFDT algorithm to the edge calculation to achieve the calculation of the large amount of data at the edge and meet the high demand for real-time processing. ;
  • the data generated at the edge of the network is gradually increasing. If we can process and analyze the data at the edge of the network, then this computing model will be more efficient.
  • Such computing requirements cannot be met by cloud computing, because many scenarios require data to be processed quickly at the edge; the demand for edge computing mainly comes from the promotion of cloud services, the promotion of the Internet of Things and the demand for terminal use; Efficient and fast can meet the needs of many real-time scenarios.
  • the present invention processes massive real-time streaming data at the edge computing end.
  • the method adopted is a decision tree based on time series algorithm, namely VFDT algorithm, which does not require large memory to store data and models. Timely processing of real-time data is very suitable for edge computing;
  • the current main methods or algorithms are based on traditional batch modeling methods, that is, one-time modeling and multiple use; the present invention can classify data in real time, for example, distinguish two types Data; can also update the model based on real-time data; traditional methods, such as decision trees, need to model the entire amount of data; and the present invention can model a part of the data, and continuously update the model based on new data. Responding to changes in data is of great value to the layout of the Internet of Things, and is an important tool for edge computing in the era of the Internet of Things.
  • Figure 1 is a flow chart of the edge-end embedded time series decision tree classification method
  • Figure 2 is a flowchart of the decision tree update process
  • Fig. 3 is a structural block diagram of the embedded time series decision tree classification at the edge.
  • the edge-end embedded time series decision tree classification method of the present invention is to apply the VFDT algorithm to the edge-end calculations to achieve the calculation of the large amount of data at the edge-end and achieve high requirements for real-time processing
  • the specific steps are as follows:
  • the edge collects data and stores the data in the storage device at the edge; the edge collects data through sensors, and the sensor is installed at the location where data collection is required to complete the data collection.
  • the sensor sends the collected data to the edge in real time, such as temperature
  • the sensor is installed where it can directly detect the temperature.
  • VFDT algorithm very fast decision tree or time series decision tree
  • the VFDT algorithm uses C/C++ language to program the VFDT algorithm software system, and the VFDT algorithm software system as the data processing
  • the intelligent system is deployed to edge devices.
  • the VFDT algorithm analyzes and processes the streaming data in real time, classifies and judges according to the characteristics of the incoming data, and judges whether the decision tree should be updated, as shown in Figure 2, the steps are as follows:
  • HT is a decision tree with a single leaf node l_1;
  • step (8) If it is not the same type, proceed to step (8) in the next step;
  • step (11) If yes, perform step (11) in the next step;
  • the edge end transmits the classification results to the cloud in real time or after storing for a specified time.
  • the specified time is a time period set at the edge end according to user needs; the edge end can only transmit the obtained analysis results to the cloud or the original data It is also transmitted to the cloud together; if only the results are transmitted, this avoids the time overhead of transmitting large amounts of data at the edge.
  • the edge-end embedded time series decision tree classification system of the present invention includes edge-end devices, sensors, and cloud devices.
  • the sensors and edge-end devices are wirelessly connected and exchange data.
  • the cloud device and the edge End devices connect and transmit data through wireless or wired connections; edge devices are used to collect data, store data, process data, and return classification results, while determining whether the decision tree needs to be updated; sensors are used to collect data; cloud devices are used to receive classification results .
  • edge data collection module intelligent data processing module based on VFDT algorithm, data storage module and classification result transmission module are deployed in edge data equipment; edge data collection module is used to collect data collected by sensors; intelligent data processing based on VFDT algorithm The module is used to use the VFDT algorithm to classify the data collected by the sensor; the data storage module is used to store the data collected by the sensor; the classification result transmission module is used to transmit the classification results of the intelligent data processing module based on the VFDT algorithm to the cloud.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本发明公开了一种边缘端的嵌入式时间序列决策树分类方法及系统,属于边缘端的实时数据分类建模,本发明要解决的技术问题为如何实现对海量数据进行一次性建模处理,同时确保内存需求量效且耗时短,采用的技术方案为:①一种边缘端的嵌入式时间序列决策树分类方法,步骤如下:S1、边缘端收集数据并将数据存储到边缘端的存储设备中;S2、利用VFDT算法对边缘端收集数据进行智能分类处理;S3、边缘端获取分类结果;S4、边缘端将分类结果传输到云端。②一种边缘端的嵌入式时间序列决策树分类系统,该系统包括边缘端设备、传感器和云端设备,传感器和边缘端设备通过无线连接并互传数据,云端设备与边缘端设备通过无线或有线连接并传输数据。

Description

一种边缘端的嵌入式时间序列决策树分类方法及系统 技术领域
本发明涉及一种边缘端的实时数据分类建模,具体地说是一种边缘端的嵌入式时间序列决策树分类方法及系统。
背景技术
对数据进行分类是常见的一个需求。不同于聚类算法,分类算法需要训练数据进行建模。建模完成之后,就可以使用这个模型进行不断的预测使用。
最经典的分类算法是决策树,决策树利用信息学中信息增益的方法判断样本变量的重要性,然后根据重要性排序进行样本的区分。这样的方法的缺点在于很难对海量数据进行一次性的建模处理,因为需要大的内存并且耗时。
近些年,随着大数据的出现,对流式数据的处理和研究成为热门领域。如何实现对海量数据进行一次性建模处理,同时确保内存需求量效且耗时短是目前急需解决的技术问题。
专利号为CN104318270A的专利文献公开了一种基于MODIS时间序列数据的土地覆盖分类方法,该方法具体是按照以下步骤进行的:1、建立原始曲线;2、对原始曲线进行滤波拟合成初始曲线;3、建立初始曲线像元的无云影像二维数组;4、设置为阈值T,其中,Yi≠yi;5、处理过的原始曲线;6、得到重建后的NDVI年变化曲线;7、提取植被生长季参数组成特征影像;8、决定最终投票分类结果等步骤进行的;本发明应用于基于MODIS时间序列数据的土地覆盖分类领域。但是该技术方案要解决传统方法用时长、植被指数的负偏差以及SG重建结果准确性降低的问题,不能实现对海量数据进行一次性建模处理,同时确保内存需求量效且耗时短。
发明内容
本发明的技术任务是提供一种边缘端的嵌入式时间序列决策树分类方法及系统,来解决如何实现对海量数据进行一次性建模处理,同时确保内存需求量效且耗时短的问题。
本发明的技术任务是按以下方式实现的,一种边缘端的嵌入式时间序列决策树分类方法,该方法是将VFDT算法应用到边缘端的计算中,实现边缘端大数据量的计算且能够达到对实时处理要求高的需求;具体步骤如下:
S1、边缘端收集数据并将数据存储到边缘端的存储设备中;
S2、利用VFDT算法(非常快速决策树或者叫时间序列决策树)对边缘端收集数据进行智能分类处理;
S3、边缘端获取分类结果;
S4、边缘端将分类结果传输到云端;边缘端可以只将得到的分析结果传输到云端,也可以将原始数据也一起传输到云端;如果只传输结果,这样避免了在边缘端传输大量数据带来的时间开销。
作为优选,所述步骤S1中边缘端通过传感器采集数据,传感器安装到需要进行数据采集的位置完成数据采集,传感器实时将采集的数据发送到边缘端,例如温度传感器安装在能够直接检测温度的地方。
作为优选,所述步骤S2中VFDT算法实时的对流式数据进行分析处理,根据流入数据的特性进行分类判断,并判断决策树是否要进行更新。
更优地,所述决策树进行更新的步骤如下:
(1)、HT为有单个叶子结点l_1的决策树;
(2)、赋值:
Figure PCTCN2019096748-appb-000001
(3)、按预测S中最频繁一类把
Figure PCTCN2019096748-appb-000002
值赋给
Figure PCTCN2019096748-appb-000003
(4)、对于每一类y k,每个x ij值,赋值n ijk(l 1)=0;
(5)、对于每个例子(x,y k),使用HT分类(x,y)成为叶子节点;对于每个x ij,增加n ijk(l);
(6)、标记l;
(7)、判断l中的例子是否为同一类:
①、若不是同一类,则下一步执行步骤(8);
(8)、对每个属性
Figure PCTCN2019096748-appb-000004
计算
Figure PCTCN2019096748-appb-000005
使用n ijk(l)计数;
(9)、按最高的
Figure PCTCN2019096748-appb-000006
为X a赋值,据第二高的
Figure PCTCN2019096748-appb-000007
赋值X b,计算∈;
(10)、判断是否是
Figure PCTCN2019096748-appb-000008
Figure PCTCN2019096748-appb-000009
①、若是,则下一步执行步骤(11);
(11)、用在X a分离的全局节点代替l;
(12)、对于分离的每一分支添加l m,X m=X-{X a},据l m的最频繁一类将
Figure PCTCN2019096748-appb-000010
值赋给
Figure PCTCN2019096748-appb-000011
对于每一属性
Figure PCTCN2019096748-appb-000012
中每一类y k和x ij赋值n ijk(l m)=0;
(13)、输出决策树HT,完成决策树的更新。
更优地,所述步骤S2中VFDT算法利用C/C++语言编程为VFDT算法软件系统,将VFDT算法软件系统作为数据处理的智能系统部署到边缘端设备中。
更优地,步骤S4中边缘端将分类结果实时或者存储指定时间后传输到云端,指定时间是根据用户需求在边缘端设定的时间段。
一种边缘端的嵌入式时间序列决策树分类系统,该系统包括边缘端设备、传感器和云端设备,传感器和边缘端设备通过无线连接并互传数据,云端设备与边缘端设备通过无线或有线连接并传输数据;
边缘端设备用于收集数据、存储数据、处理数据并返回分类结果,同时判断决策树是否需要更新;
传感器用于采集数据;
云端设备用于接收分类结果。
作为优选,所述边缘数据设备内部署有边缘数据收集模块、基于VFDT算法的智能数据处理模块、数据存储模块以及分类结果传输模块;
其中,边缘数据收集模块用于收集传感器采集的数据;
基于VFDT算法的智能数据处理模块用于利用VFDT算法对传感器采集的数据进行分类处理;
数据存储模块用于存储传感器采集的数据;
分类结果传输模块用于将基于VFDT算法的智能数据处理模块的分类结果传输到云端。
本发明的边缘端的嵌入式时间序列决策树分类方法及系统具有以下优点
(一)、VFDT算法(Very Fast Decision Tree,非常快速决策树)是流式数据的一种建模算法,是对决策树在实时数据上的一种拓展,VFDT是一种基于Hoeffding不等式建立决策树的方法,利用统计不等式Hoeffding来判断一个节点是否应该作为分类判断的依据,本发明将VFDT算法应用到边缘端的计算中,实现边缘端大数据量的计算且能够达到对实时处理要求高的需求;
(二)、网络边缘产生的数据正在逐步增加,如果我们能够在网络的边缘结点去处理、分析数据,那么这种计算模型会更高效。这样的计算要求是云计算 所不能满足的,因为很多场景需要数据在边缘端得到快速的处理;边缘计算的需求主要来源于云服务的推动、物联网的推动和终端使用的需求;边缘计算的高效、快速可以满足很多实时场景的需求,本发明在边缘计算端处理海量实时流式数据,所采用方法为基于时间序列算法的决策树,即VFDT算法,可以不需要大内存保存数据和模型,对实时数据进行及时处理,非常适合边缘计算;
(三)、随着数据量的增大,对数据的实时处理提出了强烈的需求,甚至需要数据在边缘端得到智能的处理,并把结果发送到云端;边缘端的数据处理方法目前非常缺少,针对流式海量数据的处理方法更是稀缺,目前的主要方法或者算法是基于传统的批量建模方法,即一次建模多次使用;本发明可以实时的对数据进行分类,例如,区分两类数据;还可以根据实时数据进行模型的更新;传统方法,例如决策树,需要对全量数据进行建模;而本发明可以针对一部分数据进行建模,并不断的根据新的数据进行模型更新,能够应对数据的变化,对物联网布局很有价值,是物联网时代边缘计算的重要工具。
附图说明
下面结合附图对本发明进一步说明。
附图1为边缘端的嵌入式时间序列决策树分类方法流程框图;
附图2为决策树更新的流程框图;
附图3为边缘端的嵌入式时间序列决策树分类的结构框图。
具体实施方式
参照说明书附图和具体实施例对本发明的一种边缘端的嵌入式时间序列决策树分类方法及系统作以下详细地说明。
实施例1:
如附图1所示,本发明的边缘端的嵌入式时间序列决策树分类方法,该方法是将VFDT算法应用到边缘端的计算中,实现边缘端大数据量的计算且能够达到对实时处理要求高的需求;具体步骤如下:
S1、边缘端收集数据并将数据存储到边缘端的存储设备中;边缘端通过传感器采集数据,传感器安装到需要进行数据采集的位置完成数据采集,传感器实时将采集的数据发送到边缘端,例如温度传感器安装在能够直接检测温度的地方。
S2、利用VFDT算法(非常快速决策树或者叫时间序列决策树)对边缘端收集数据进行智能分类处理;VFDT算法利用C/C++语言编程为VFDT算法软件系统,将VFDT算法软件系统作为数据处理的智能系统部署到边缘端设备中。VFDT算法实时的对流式数据进行分析处理,根据流入数据的特性进行分类判断,并判 断决策树是否要进行更新,如附图2所示,步骤如下:
(1)、HT为有单个叶子结点l_1的决策树;
(2)、赋值:
Figure PCTCN2019096748-appb-000013
(3)、按预测S中最频繁一类把
Figure PCTCN2019096748-appb-000014
值赋给
Figure PCTCN2019096748-appb-000015
(4)、对于每一类y k,每个x ij值,赋值n ijk(l 1)=0;
(5)、对于每个例子(x,y k),使用HT分类(x,y)成为叶子节点;对于每个x ij,增加n ijk(l);
(6)、标记l;
(7)、判断l中的例子是否为同一类:
①、若不是同一类,则下一步执行步骤(8);
(8)、对每个属性
Figure PCTCN2019096748-appb-000016
计算
Figure PCTCN2019096748-appb-000017
使用n ijk(l)计数;
(9)、按最高的
Figure PCTCN2019096748-appb-000018
为X a赋值,据第二高的
Figure PCTCN2019096748-appb-000019
赋值X b,计算∈;
(10)、判断是否是
Figure PCTCN2019096748-appb-000020
Figure PCTCN2019096748-appb-000021
①、若是,则下一步执行步骤(11);
(11)、用在X a分离的全局节点代替l;
(12)、对于分离的每一分支添加l m,X m=X-{X a},据l m的最频繁一类将
Figure PCTCN2019096748-appb-000022
值赋给
Figure PCTCN2019096748-appb-000023
对于每一属性
Figure PCTCN2019096748-appb-000024
中每一类y k和x ij赋值n ijk(l m)=0;
(13)、输出决策树HT,完成决策树的更新。
S3、边缘端获取分类结果;
S4、边缘端将分类结果实时或者存储指定时间后传输到云端,指定时间是根据用户需求在边缘端设定的时间段;边缘端可以只将得到的分析结果传输到云端,也可以将原始数据也一起传输到云端;如果只传输结果,这样避免了在边缘端传输大量数据带来的时间开销。
实施例2:
如附图3所示,本发明的边缘端的嵌入式时间序列决策树分类系统,该系统包括边缘端设备、传感器和云端设备,传感器和边缘端设备通过无线连接并互传数据,云端设备与边缘端设备通过无线或有线连接并传输数据;边缘端设备用于收集数据、存储数据、处理数据并返回分类结果,同时判断决策树是否需要更新;传感器用于采集数据;云端设备用于接收分类结果。
其中,边缘数据设备内部署有边缘数据收集模块、基于VFDT算法的智能数据处理模块、数据存储模块以及分类结果传输模块;边缘数据收集模块用于收集传感器采集的数据;基于VFDT算法的智能数据处理模块用于利用VFDT算法对传感器采集的数据进行分类处理;数据存储模块用于存储传感器采集的数据;分类结果传输模块用于将基于VFDT算法的智能数据处理模块的分类结果传输到云端。
最后应说明的是:以上各实施例仅用以说明本发明的技术方案,而非对其限制;尽管参照前述各实施例对本发明进行了详细的说明,本领域的普通技术人员应当理解:其依然可以对前述各实施例所记载的技术方案进行修改,或者对其中部分或者全部技术特征进行等同替换;而这些修改或者替换,并不使相应技术方案的本质脱离本发明各实施例技术方案的范围。

Claims (8)

  1. 一种边缘端的嵌入式时间序列决策树分类方法,其特征在于,该方法是将VFDT算法应用到边缘端的计算中,实现边缘端大数据量的计算且能够达到对实时处理要求高的需求;具体步骤如下:
    S1、边缘端收集数据并将数据存储到边缘端的存储设备中;
    S2、利用VFDT算法对边缘端收集数据进行智能分类处理;
    S3、边缘端获取分类结果;
    S4、边缘端将分类结果传输到云端。
  2. 根据权利要求1所述的边缘端的嵌入式时间序列决策树分类方法,其特征在于,所述步骤S1中边缘端通过传感器采集数据,传感器安装到需要进行数据采集的位置完成数据采集,传感器实时将采集的数据发送到边缘端。
  3. 根据权利要求1或2所述的边缘端的嵌入式时间序列决策树分类方法,其特征在于,所述步骤S2中VFDT算法实时的对流式数据进行分析处理,根据流入数据的特性进行分类判断,并判断决策树是否要进行更新。
  4. 根据权利要求3所述的边缘端的嵌入式时间序列决策树分类方法,其特征在于,所述决策树进行更新的步骤如下:
    (1)、HT为有单个叶子结点l_1的决策树;
    (2)、赋值:
    Figure PCTCN2019096748-appb-100001
    (3)、按预测S中最频繁一类把
    Figure PCTCN2019096748-appb-100002
    值赋给
    Figure PCTCN2019096748-appb-100003
    (4)、对于每一类y k,每个x ij值,赋值n ijk(l 1)=0;
    (5)、对于每个例子(x,y k),使用HT分类(x,y)成为叶子节点;对于每个x ij,增加n ijk(l);
    (6)、标记l;
    (7)、判断l中的例子是否为同一类:
    ①、若不是同一类,则下一步执行步骤(8);
    (8)、对每个属性
    Figure PCTCN2019096748-appb-100004
    计算
    Figure PCTCN2019096748-appb-100005
    使用n ijk(l)计数;
    (9)、按最高的
    Figure PCTCN2019096748-appb-100006
    为X a赋值,据第二高的
    Figure PCTCN2019096748-appb-100007
    赋值X b,计算∈;
    (10)、判断是否是
    Figure PCTCN2019096748-appb-100008
    Figure PCTCN2019096748-appb-100009
    ①、若是,则下一步执行步骤(11);
    (11)、用在X a分离的全局节点代替l;
    (12)、对于分离的每一分支添加l m,X m=X-{X a},据l m的最频繁一类将
    Figure PCTCN2019096748-appb-100010
    值赋给
    Figure PCTCN2019096748-appb-100011
    对于每一属性
    Figure PCTCN2019096748-appb-100012
    中每一类y k和x ij赋值n ijk(l m)=0;
    (13)、输出决策树HT,完成决策树的更新。
  5. 根据权利要求4所述的边缘端的嵌入式时间序列决策树分类方法,其特征在于,所述步骤S2中VFDT算法利用C/C++语言编程为VFDT算法软件系统,将VFDT算法软件系统作为数据处理的智能系统部署到边缘端设备中。
  6. 根据权利要求5所述的一种边缘端的嵌入式时间序列决策树分类方法及系统,其特征在于,步骤S4中边缘端将分类结果实时或者存储指定时间后传输到云端,指定时间是根据用户需求在边缘端设定的时间段。
  7. 一种边缘端的嵌入式时间序列决策树分类系统,其特征在于,该系统包括边缘端设备、传感器和云端设备,传感器和边缘端设备通过无线连接并互传数据,云端设备与边缘端设备通过无线或有线连接并传输数据;
    边缘端设备用于收集数据、存储数据、处理数据并返回分类结果,同时判断决策树是否需要更新;
    传感器用于采集数据;
    云端设备用于接收分类结果。
  8. 根据权利要求7所述的边缘端的嵌入式时间序列决策树分类系统,其特征在于,所述边缘数据设备内部署有边缘数据收集模块、基于VFDT算法的智能数据处理模块、数据存储模块以及分类结果传输模块;
    其中,边缘数据收集模块用于收集传感器采集的数据;
    基于VFDT算法的智能数据处理模块用于利用VFDT算法对传感器采集的数据进行分类处理;
    数据存储模块用于存储传感器采集的数据;
    分类结果传输模块用于将基于VFDT算法的智能数据处理模块的分类结果传输到云端。
PCT/CN2019/096748 2019-01-17 2019-07-19 一种边缘端的嵌入式时间序列决策树分类方法及系统 WO2020147286A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910044039.6 2019-01-17
CN201910044039.6A CN109784408A (zh) 2019-01-17 2019-01-17 一种边缘端的嵌入式时间序列决策树分类方法及系统

Publications (1)

Publication Number Publication Date
WO2020147286A1 true WO2020147286A1 (zh) 2020-07-23

Family

ID=66501531

Family Applications (2)

Application Number Title Priority Date Filing Date
PCT/CN2019/096748 WO2020147286A1 (zh) 2019-01-17 2019-07-19 一种边缘端的嵌入式时间序列决策树分类方法及系统
PCT/CN2019/112075 WO2020147353A1 (zh) 2019-01-17 2019-10-21 一种边缘端的嵌入式时间序列决策树分类方法及系统

Family Applications After (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/112075 WO2020147353A1 (zh) 2019-01-17 2019-10-21 一种边缘端的嵌入式时间序列决策树分类方法及系统

Country Status (2)

Country Link
CN (1) CN109784408A (zh)
WO (2) WO2020147286A1 (zh)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109784408A (zh) * 2019-01-17 2019-05-21 济南浪潮高新科技投资发展有限公司 一种边缘端的嵌入式时间序列决策树分类方法及系统
CN113128694B (zh) * 2019-12-31 2024-07-19 北京超星未来科技有限公司 机器学习中的数据获取、数据处理的方法、装置及系统
CN112988327A (zh) * 2021-03-04 2021-06-18 杭州谐云科技有限公司 一种基于云边协同的容器安全管理方法和系统
CN114827143B (zh) * 2022-04-11 2023-12-12 广东省离散智造科技创新有限公司 一种离散数据采集方法、系统及云端服务器、边缘计算端
CN115116224B (zh) * 2022-06-24 2023-08-18 山东旗帜信息有限公司 一种边缘端数据采集传输系统及方法
CN115827621B (zh) * 2023-02-17 2023-05-16 河北雄安睿天科技有限公司 基于云计算和数据分析的水务数据管理系统

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170236074A1 (en) * 2016-02-10 2017-08-17 Sas Institute Inc. Kernel parameter selection in support vector data description for outlier identification
CN107766889A (zh) * 2017-10-26 2018-03-06 济南浪潮高新科技投资发展有限公司 一种云端边缘计算融合的深度学习计算系统和方法
CN107797867A (zh) * 2017-10-31 2018-03-13 济南浪潮高新科技投资发展有限公司 一种增强边缘侧智能计算能力的方法及装置
CN108932535A (zh) * 2018-07-13 2018-12-04 南方电网科学研究院有限责任公司 一种基于机器学习的边缘计算克隆节点识别方法
CN109784408A (zh) * 2019-01-17 2019-05-21 济南浪潮高新科技投资发展有限公司 一种边缘端的嵌入式时间序列决策树分类方法及系统

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7630950B2 (en) * 2006-08-18 2009-12-08 International Business Machines Corporation System and method for learning models from scarce and skewed training data
CN102737124B (zh) * 2012-06-15 2017-02-15 复旦大学 在时间序列数据库中查找给定时间序列的近似序列的方法
CN104159071A (zh) * 2014-07-11 2014-11-19 深圳瞭望通达科技有限公司 一种基于云服务的智能目标识别装置、系统及方法
CN106290709A (zh) * 2015-06-02 2017-01-04 齐鲁工业大学 一种变电站气体无线监测报警控制方法
CN105677728A (zh) * 2015-12-28 2016-06-15 广东正美家具科技有限公司 物体图像识别分类管理方法
CN105824913A (zh) * 2016-03-15 2016-08-03 南京邮电大学 基于Storm的决策树分类CVFDT并行化算法
CN107784251A (zh) * 2016-08-25 2018-03-09 大连楼兰科技股份有限公司 基于图像识别技术对驾驶行为评价的方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170236074A1 (en) * 2016-02-10 2017-08-17 Sas Institute Inc. Kernel parameter selection in support vector data description for outlier identification
CN107766889A (zh) * 2017-10-26 2018-03-06 济南浪潮高新科技投资发展有限公司 一种云端边缘计算融合的深度学习计算系统和方法
CN107797867A (zh) * 2017-10-31 2018-03-13 济南浪潮高新科技投资发展有限公司 一种增强边缘侧智能计算能力的方法及装置
CN108932535A (zh) * 2018-07-13 2018-12-04 南方电网科学研究院有限责任公司 一种基于机器学习的边缘计算克隆节点识别方法
CN109784408A (zh) * 2019-01-17 2019-05-21 济南浪潮高新科技投资发展有限公司 一种边缘端的嵌入式时间序列决策树分类方法及系统

Also Published As

Publication number Publication date
WO2020147353A1 (zh) 2020-07-23
CN109784408A (zh) 2019-05-21

Similar Documents

Publication Publication Date Title
WO2020147286A1 (zh) 一种边缘端的嵌入式时间序列决策树分类方法及系统
Xu et al. Wheat ear counting using K-means clustering segmentation and convolutional neural network
WO2022062419A1 (zh) 基于非督导金字塔相似性学习的目标重识别方法及系统
CN113408087A (zh) 一种基于云边系统和视频智能分析的变电站巡视方法
Cao et al. CNN-based intelligent safety surveillance in green IoT applications
CN112367273B (zh) 基于知识蒸馏的深度神经网络模型的流量分类方法及装置
CN111127423B (zh) 一种基于cnn-bp神经网络算法水稻病虫害识别方法
CN114462623B (zh) 基于边缘计算的数据分析方法、系统及平台
CN111985325A (zh) 特高压环境评价中的航拍小目标快速识别方法
CN103473128A (zh) 一种用于mashup应用推荐的协同过滤方法
CN110070071A (zh) 一种基于图像识别的生姜茎叶病虫害识别方法及系统
CN113688787A (zh) 花生叶片病害识别方法
CN117136765A (zh) 基于智能农业的大棚控制系统及方法
CN117670527B (zh) 基于地块数据的农户贷款授信额度确定方法及系统
CN117854156A (zh) 一种特征提取模型的训练方法和相关装置
CN112001290B (zh) 一种基于yolo算法的稻飞虱迁飞路径预测方法
CN117557994A (zh) 一种基于改进型YOLOv7网络的茶叶虫害识别方法
CN112348360A (zh) 一种基于大数据技术的中药生产工艺参数分析系统
CN113014645A (zh) 基于物联网的草场环境监测系统
CN115482419B (zh) 一种海洋渔业产品的数据采集分析方法及系统
CN115328870B (zh) 一种面向云制造的数据共享方法及系统
CN109583478A (zh) 一种智能蜂群聚类方法及车辆目标检测方法
CN114219051B (zh) 图像分类方法、分类模型的训练方法、装置及电子设备
CN113554010B (zh) 一种电网线路故障识别模型训练方法
CN113191569A (zh) 一种基于大数据的企业管理方法及系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19910788

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19910788

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19910788

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 04.02.2022)

122 Ep: pct application non-entry in european phase

Ref document number: 19910788

Country of ref document: EP

Kind code of ref document: A1