WO2021258636A1 - 基于深度分层模糊算法的环保设备识别方法与系统 - Google Patents

基于深度分层模糊算法的环保设备识别方法与系统 Download PDF

Info

Publication number
WO2021258636A1
WO2021258636A1 PCT/CN2020/132231 CN2020132231W WO2021258636A1 WO 2021258636 A1 WO2021258636 A1 WO 2021258636A1 CN 2020132231 W CN2020132231 W CN 2020132231W WO 2021258636 A1 WO2021258636 A1 WO 2021258636A1
Authority
WO
WIPO (PCT)
Prior art keywords
environmental protection
protection equipment
fuzzy
harmonic signal
data
Prior art date
Application number
PCT/CN2020/132231
Other languages
English (en)
French (fr)
Inventor
李成栋
邓晓平
张桂青
阎俏
Original Assignee
山东建筑大学
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 山东建筑大学 filed Critical 山东建筑大学
Priority to US17/782,792 priority Critical patent/US20230014095A1/en
Publication of WO2021258636A1 publication Critical patent/WO2021258636A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06QINFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10Services
    • G06Q50/26Government or public services
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00Computing arrangements based on specific mathematical models
    • G06N7/02Computing arrangements based on specific mathematical models using fuzzy logic
    • G06N7/023Learning or tuning the parameters of a fuzzy system
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/047Probabilistic or stochastic networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/24Classification techniques
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00Computing arrangements using knowledge-based models
    • G06N5/01Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16YINFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y20/00Information sensed or collected by the things
    • G16Y20/20Information sensed or collected by the things relating to the thing itself
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16YINFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00IoT characterised by the purpose of the information processing
    • G16Y40/10Detection; Monitoring
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16YINFORMATION AND COMMUNICATION TECHNOLOGY SPECIALLY ADAPTED FOR THE INTERNET OF THINGS [IoT]
    • G16Y40/00IoT characterised by the purpose of the information processing
    • G16Y40/20Analytics; Diagnosis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/12Classification; Matching

Definitions

  • the invention relates to the identification of environmental protection equipment, is a method for identifying whether it is a corresponding environmental protection equipment by analyzing the electricity consumption data of the equipment, and belongs to the field of data mining analysis.
  • the present invention proposes a method and system that can identify environmental protection equipment in real time.
  • an embodiment of the present invention provides a method for identifying environmental protection equipment based on a deep layered fuzzy algorithm, which includes the following steps:
  • the steps of the step (1) are as follows:
  • a training sample database D is constructed.
  • the steps of the step (2) are as follows:
  • Step 1 Extract feature vector
  • Step 2 Build a deep layered fuzzy system
  • the feature vector in the training sample set D 1 As the input vector of the system, namely The category label y m is used as the output vector of each fuzzy subsystem.
  • Construct the input-output data pair of the i-th fuzzy subsystem of the first layer According to the data pair, determine the value range of the fuzzy set [min x 0 ,max x 0 ]. Within this range, the input vector can be further divided into q fuzzy sets A 1 , A 2 ,..., A q .
  • the i-th fuzzy subsystem of the first layer can be expressed as: It can be further obtained by using the existing standard formula And simplify the expression to:
  • the output of the first layer As the input vector of the fuzzy subsystem of the second layer, the output vector is still y m , and the fuzzy subsystem of the second layer is designed according to the same design method as that of the first layer. By analogy, the design of the last layer of fuzzy subsystem is completed, and the entire deep layered fuzzy system is built.
  • the data in the training sample database D is divided into two parts: training set D 1 and test set D 2 , and both training set D 1 and test set D 2 perform feature vector extraction Step: By inputting the harmonic signal data in the test set D 2 into the recognition model, and comparing the recognition result with the label, it is tested whether the accuracy of the recognition model can meet the requirements. If the accuracy cannot meet the requirements, more sample data needs to be used to retrain the recognition model until the accuracy meets the requirements.
  • the harmonic signal data collected from the tested equipment is input into the established recognition model.
  • This model first extracts the feature vector of the harmonic signal data. Then input the extracted feature vector into the deep layered fuzzy system to obtain the type label, determine whether it is the corresponding environmental protection device, and use the display device to output the recognition result obtained by the analysis.
  • an embodiment of the present invention also provides an environmental protection equipment identification system based on a deep layered fuzzy algorithm, which is used to implement the steps of the above-mentioned environmental protection equipment identification method based on a deep layered fuzzy algorithm during execution, including:
  • Signal recognition module which is used to execute the method of step (3).
  • the invention uses the harmonic detector installed on the environmental protection equipment to collect the harmonic signal of the equipment in real time, and then through the decomposition and identification of the harmonic signal, it is judged whether the detected equipment is the corresponding environmental protection equipment type, and it can also reflect indirectly Whether the environmental protection equipment is activated.
  • This invention avoids cumbersome inspections by law enforcement personnel, and avoids the problem of enterprises sneaking off environmental protection equipment and secretly replacing the tested equipment. It improves the efficiency of daily inspections of environmental protection equipment, and at the same time strengthens environmental protection inspections, which is helpful for environmental protection policies. implement.
  • the equipment can be detected in real time to determine whether it is the corresponding environmental protection equipment, avoiding the problem of sneaking the detection equipment to non-environmental protection equipment and the mismatch of environmental protection equipment types.
  • This method also reflects in real time whether the environmental protection equipment is activated, so as to avoid the problem of unlawfully stopping the environmental protection equipment.
  • the deep hierarchical fuzzy system designed based on the least square method has better accuracy and calculation speed, and solves the problem of dimensional catastrophe rule explosion.
  • the cloud-side collaborative architecture is adopted, which facilitates data collection and centralized processing and analysis, and saves resources more than conventional methods.
  • Figure 1 is a schematic diagram of the architecture of the cloud-side collaboration system of the present invention.
  • Figure 2 is a structural diagram of the deep layered fuzzy system of the present invention.
  • Figure 3 is a flow chart of the identification of the present invention.
  • the present invention proposes an environmental protection equipment identification method and system based on a deep layered fuzzy algorithm, which combines a cloud-side collaboration architecture and a depth based on the least squares method.
  • the method of hierarchical fuzzy system is a deep layered fuzzy algorithm.
  • the present invention is composed of three modules, which are a data acquisition module, a recognition model construction module, and a signal recognition module.
  • the data acquisition module is responsible for collecting the harmonic signals of environmental protection equipment through the harmonic detector, uploading it to the "cloud platform", and constructing a training sample database based on the actual surveyed environmental protection equipment type information;
  • the recognition model building module is configured to pass local
  • the mean decomposition method extracts the feature vector of the data in the training sample database, and uses the training sample database to train a deep hierarchical fuzzy system based on the least square method to construct a recognition model;
  • the signal recognition module is configured to use the recognition model Evaluate the input harmonic signal data to determine whether the tested equipment is the corresponding environmental protection equipment.
  • This module is responsible for using the harmonic detector to collect the harmonic signal data of the equipment, and collect the corresponding environmental protection equipment type information through field investigation, and upload all the data information to the cloud platform through the communication network.
  • the collected data information is used to construct a training sample database.
  • a training sample database D is constructed using the collected harmonic signal data x m (t) and the category label y m corresponding to each harmonic signal data. This database is used to build and train recognition models.
  • This module is responsible for extracting the feature vector of the data in the training sample database through the local mean decomposition method, and using the deep hierarchical fuzzy system constructed by the least square method to classify the harmonic signal, thereby constructing the recognition model.
  • This model is used to identify and analyze the harmonic signal data of the equipment that needs to be tested.
  • the data in the training sample database D is divided into two parts: 80% of the data is used as the training set D 1 , and the remaining part of the data is used as the test set D 2 .
  • PF component Perform local mean decomposition for each harmonic signal x m (t) in the training sample database to find the PF component (after the local mean decomposition method, an envelope signal and a pure frequency modulation signal are generated, and the product of the two is the final result PF (Product function) component.
  • the PF component obtained in the first run can be recorded as the PF 1 component.
  • Both the training set D 1 and the test set D 2 are subjected to the above-mentioned data processing process.
  • the fuzzy subsystem is constructed based on the least square method, and finally the entire deep layered fuzzy system is built.
  • the structure of the deep layered fuzzy system is shown in Figure 2.
  • the feature vector in the training sample set D 1 As the input vector of the system, namely The category label y m is used as the output vector of each fuzzy subsystem.
  • Construct the input-output data pair of the i-th fuzzy subsystem of the first layer It is obtained by a moving window of length w (convolution operator), which starts from the first data of the input vector and moves one step at a time until all data is covered. 0 represents the input of the first layer.
  • the input vector can be further divided into q fuzzy sets A 1 , A 2 ,..., A q .
  • the fuzzy sets A 1 , A 2 ,..., A q can be obtained by using the existing calculation formulas of the triangular fuzzy sets (other fuzzy sets can also be used).
  • the i-th fuzzy subsystem of the first layer can be expressed as: It can be further obtained by using the existing standard formula And simplify the expression to:
  • S(c) represents a defined function of parameter c; y m represents the correct output result. Seek its optimal solution problem.
  • the output of the first layer As the input vector of the fuzzy subsystem of the second layer, the output vector is still y m , and the fuzzy subsystem of the second layer is designed according to the same design method as that of the first layer. By analogy, the design of the last layer of fuzzy subsystem is completed, and the entire deep layered fuzzy system is built.
  • Feature vector extraction based on local mean decomposition method and deep layered fuzzy system classification constitute the recognition model.
  • This module is responsible for identifying and analyzing the collected harmonic signal data by using the constructed recognition model, to determine the model of the tested equipment, and then to determine whether it is an environmental protection equipment and whether it is the corresponding environmental protection equipment type.

Abstract

一种基于深度分层模糊算法的环保设备识别方法与系统,所述方法包括以下步骤:(1)通过谐波检测仪采集环保设备的谐波信号数据,并实地采集对应的环保设备类型信息,用于构建训练样本数据库;(2)通过局部均值分解方法提取训练样本数据库中的数据的特征向量,并利用训练样本数据库训练基于最小二乘法构建的深度分层模糊系统,以此构建出识别模型;(3)利用识别模型评估输入的谐波信号数据,以判断被检测设备是否为对应的环保设备。上述方法可以实时检测设备,判断其是否为对应的环保设备,避免了偷换检测仪器至非环保设备上以及环保设备类型不匹配的问题。

Description

基于深度分层模糊算法的环保设备识别方法与系统 技术领域
本发明涉及环保设备的识别,是一种通过分析设备用电数据来识别是否为对应环保设备的方法,属于数据挖掘分析领域。
背景技术
这里的陈述仅提供与本发明相关的背景技术,而不必然地构成现有技术。
近几年,企业环保问题受到格外关注,国家也不断修订《环保法》以加强环保监督。在企业方面,配备必要的环保治污设备是企业环保达标的基础。环保部门执法人员也会对企业环保设备进行相关的日常检查。
目前,对企业环保设备进行检查时,执法人员需要亲临现场,由于需要检查的企业和项目众多,导致部分地区的执法人员在检查企业环保设备时敷衍了事。由于开启环保设备会提高企业的生产成本,所以很多企业为应付检查,仅在检查开始前才开启环保设备;还有部分企业虽然在检查时设备运转良好,各项检查也均达到标准要求,却在执法人员离开后关停环保设备;利用谐波检测仪检查的方法则存在仪器被偷换安装到其他非环保设备上的问题。通过目前部分存在的现象可以发现,环保部门对企业环保设备的日常检查方式存在着缺陷。
发明内容
针对现有技术存在的不足,为了能够准确、快速的解决目前环保设备日常检查中存在的问题,本发明提出一种可以实时对环保设备进行识别的方法与系统。
为了实现上述目的,本发明是通过如下的技术方案来实现:
第一方面,本发明的实施例提供了一种基于深度分层模糊算法的环保设备识别方法,包括以下步骤:
(1)通过谐波检测仪采集环保设备的谐波信号数据,并实地采集对应的环保设备类型信息,用于构建训练样本数据库;
(2)通过局部均值分解方法提取训练样本数据库中的数据的特征向量,并利用训练样本数据库训练基于最小二乘法构建的深度分层模糊系统,以此构建出识别模型;
(3)利用识别模型评估输入的谐波信号数据,以判断被检测设备是否为对 应的环保设备。
作为进一步的技术方案,所述步骤(1)步骤如下:
采集若干个信号周期的谐波信号数据x m(t),然后将此数据上传至云平台;
并将收集所有谐波检测仪(设备节点m(m=1,2,...,n))对应的环保设备的类型信息,将设备的类型作为类别标签y m。其中,y m∈{1,2,...,k,k+1}(k≤n),标签1,2,...,k代表k种不同类型的环保设备,标签k+1代表非环保设备。对应关系为:
Figure PCTCN2020132231-appb-000001
利用采集到的谐波信号数据x m(t)和每个谐波信号数据对应的类别标签y m,构建出训练样本数据库D。
作为进一步的技术方案,所述步骤(2)步骤如下:
步骤1:提取特征向量
对训练样本数据库中的每个谐波信号x m(t)进行局部均值分解求出PF分量,取PF 1,PF 2,PF 3分量,并求出谐波信号x m(t)的PF r(r=1,2,3)分量的瞬时幅值a r(t)和瞬时频率f r(t),进一步利用平均值法求得各自的平均值
Figure PCTCN2020132231-appb-000002
Figure PCTCN2020132231-appb-000003
利用谐波信号x m(t)的PF r分量的
Figure PCTCN2020132231-appb-000004
Figure PCTCN2020132231-appb-000005
构建特征向量PF m,即
Figure PCTCN2020132231-appb-000006
步骤2:搭建深度分层模糊系统
先对系统的总体参数进行设置,手动确定层数L、移动步长s和卷积窗的长度w。
将训练样本集D 1中的特征向量
Figure PCTCN2020132231-appb-000007
作为系统的输入向量,即
Figure PCTCN2020132231-appb-000008
类别标签y m作为每一个模糊子系统的输出向量。
构建出第一层第i个模糊子系统的输入-输出数据对:
Figure PCTCN2020132231-appb-000009
根据该数据对,确定模糊集合的值域 [min x 0,max x 0]。在此值域内,输入向量就可以进一步划分为q个模糊集合A 1,A 2,...,A q
第一层第i个模糊子系统可以表示为:
Figure PCTCN2020132231-appb-000010
利用已有标准公式进一步可得到
Figure PCTCN2020132231-appb-000011
的表达式并进行简化得:
Figure PCTCN2020132231-appb-000012
对于上式中的参数
Figure PCTCN2020132231-appb-000013
采用最小二乘法进行设计,可以将其转化为:
Figure PCTCN2020132231-appb-000014
求其最优解问题。
解得参数矩阵c,第一层第i个模糊子系统设计完成,按照以上方法,完成第一层模糊子系统的搭建。
将第一层的输出
Figure PCTCN2020132231-appb-000015
作为第二层模糊子系统的输入向量,输出向量仍为y m,按照与第一层相同的设计方法来设计第二层的模糊子系统。以此类推,完成最后一层模糊子系统的设计,整个深度分层模糊系统搭建完成。
作为进一步的技术方案,所述步骤(2)中将训练样本数据库D中的数据分为两部分:训练集D 1和测试集D 2,训练集D 1和测试集D 2均执行提取特征向量步骤;通过将测试集D 2中的谐波信号数据输入到识别模型中,通过将识别结果与标签进行比较,测试出识别模型的精确度是否能够满足需求。如果精确度不能满足要求,则需要利用更多的样本数据重新对识别模型进行训练,直至精确度达到需求。
作为进一步的技术方案,所述步骤(3)中将从被检测设备中采集到的谐波信号数据输入到已构建出的识别模型中,此模型首先对谐波信号数据进行特征向量的提取,然后将提取出的特征向量输入到深度分层模糊系统中,得出类型标签,判断出是否为对应的环保设备,利用显示设备输出分析得到的识别结果。
第二方面,本发明实施例还提供了一种基于深度分层模糊算法的环保设备识别系统,用于在执行时实现上述的基于深度分层模糊算法的环保设备识别方法的 步骤,包括:
数据采集模块,该模块用于执行步骤(1)的方法;
识别模型构建模块,该模块用于执行步骤(2)的方法;
信号识别模块,该模块用于执行步骤(3)的方法。
本发明利用安装在环保设备上的谐波检测仪实时采集设备的谐波信号,然后通过对谐波信号的分解与识别判断出所检测的设备是否为对应的环保设备类型,同时也能间接的反映环保设备是否启动。此发明避免执法人员进行繁琐的检查,并且可避免企业偷停环保设备及偷换被检测设备的问题,提高了环保设备日常检查效率,同时也增强了环保检查力度,有助于对环保政策的落实。
上述本发明的实施例的有益效果如下:
(1)可以实时检测设备,判断其是否为对应的环保设备,避免了偷换检测仪器至非环保设备上以及环保设备类型不匹配的问题。
(2)此方法同时也实时的反映出环保设备是否启动,避免发生偷停环保设备的问题。
(3)大大降低了执法人员日常检查工作的繁琐性,提高了日常检查效率,增强了环保检查力度。
(4)基于最小二乘法设计的深度分层模糊系统具有更好的精度、计算速度,并且解决了维数灾难规则爆炸问题。
(5)采用云边协同架构,方便了数据的采集和集中处理分析,并且较常规方式更加节约资源。
附图说明
构成本发明的一部分的说明书附图用来提供对本发明的进一步理解,本发明的示意性实施例及其说明用于解释本发明,并不构成对本发明的不当限定。
图1是本发明云边协同系统架构示意图;
图2是本发明深度分层模糊系统结构图;
图3是本发明识别流程图。
具体实施方式
应该指出,以下详细说明都是例示性的,旨在对本发明提供进一步的说明。 除非另有指明,本发明使用的所有技术和科学术语具有与本发明所属技术领域的普通技术人员通常理解的相同含义。
需要注意的是,这里所使用的术语仅是为了描述具体实施方式,而非意图限制根据本发明的示例性实施方式。如在这里所使用的,除非本发明另外明确指出,否则单数形式也意图包括复数形式,此外,还应当理解的是,当在本说明书中使用术语“包含”和/或“包括”时,其指明存在特征、步骤、操作、器件、组件和/或它们的组合;
为了能够准确、快速的解决目前环保设备日常检查中存在的问题,本发明提出一种基于深度分层模糊算法的环保设备识别方法与系统,结合了云边协同的架构以及基于最小二乘法的深度分层模糊系统的方法。通过在企业的环保设备上安装谐波检测仪,采集环保设备的谐波信号数据,并通过实地调查采集每个谐波信号数据对应的环保设备的类型信息,所有数据信息通过通信网络传输到“云平台”;在“云平台”中,利用采集到的数据信息构建训练样本数据库,结合局部均值分解方法和深度分层模糊系统构建出识别模型;将采集到的谐波信号数据输入至识别模型中进行评估,进而可以判断被检测设备是否为对应的环保设备。
本发明由三大模块组成,分别为数据采集模块、识别模型构建模块和信号识别模块。数据采集模块负责通过谐波检测仪采集环保设备的谐波信号,上传至“云平台”,并结合实际调查的环保设备类型信息构建出训练样本数据库;识别模型构建模块被配置为用于通过局部均值分解方法提取训练样本数据库中的数据的特征向量,并利用训练样本数据库训练基于最小二乘法构建的深度分层模糊系统,以此构建出识别模型;信号识别模块被配置为用于利用识别模型评估输入的谐波信号数据,以判断被检测设备是否为对应的环保设备。
以下是对三个模块的详细介绍:
模块一:数据采集模块
本模块负责利用谐波检测仪采集设备的谐波信号数据,并通过实地调查采集对应的环保设备类型信息,将所有数据信息通过通信网络上传至云平台。采集到的数据信息被用来构建训练样本数据库。
采用云边协同的系统架构(示意图如图1所示),利用安装在环保设备上的谐波检测仪(设备节点m(m=1,2,...,n))采集10个信号周期的谐波信号数据 x m(t),式中t指采集数据时连续的时间值,然后利用通信网络将此数据上传至云平台。
通过现场实际调查,收集所有设备节点对应的环保设备的类型信息,将设备的类型作为类别标签y m。其中,y m∈{1,2,...,k,k+1}(k≤n),标签1,2,...,k代表k种不同类型的环保设备,标签k+1代表非环保设备。对应关系为:
Figure PCTCN2020132231-appb-000016
在云平台中,利用采集到的谐波信号数据x m(t)和每个谐波信号数据对应的类别标签y m,构建出训练样本数据库D。此数据库被用来构建、训练识别模型。
模块二:识别模型构建模块
本模块负责通过局部均值分解方法提取训练样本数据库中的数据的特征向量,并利用通过最小二乘法构建出的深度分层模糊系统进行谐波信号分类,以此构建出识别模型。此模型被用于对需要进行检测的设备的谐波信号数据进行识别分析。
1、提取特征向量
将训练样本数据库D中的数据分为两部分:80%的数据作为训练集D 1,剩余部分的数据作为测试集D 2
对训练样本数据库中的每个谐波信号x m(t)进行局部均值分解求出PF分量(经过局部均值分解的方法产生一个包络信号与一个纯调频信号,两者的乘积得到最终结果PF(Product function)分量。第一次运行得到的PF分量可记为PF 1分量,将原始信号减去此分量重新进行一次局部均值分解,便可得到PF 2分量,以此类推),取PF 1,PF 2,PF 3分量,并求出谐波信号x m(t)的PF r(r=1,2,3)分量的瞬时幅值a r(t)和瞬时频率f r(t),进一步利用平均值法求得各自的平均值
Figure PCTCN2020132231-appb-000017
Figure PCTCN2020132231-appb-000018
利用谐波信号x m(t)的PF r分量的
Figure PCTCN2020132231-appb-000019
Figure PCTCN2020132231-appb-000020
构建特征向量PF m,即
Figure PCTCN2020132231-appb-000021
训练集D 1和测试集D 2均做上述数据处理过程。
2、搭建深度分层模糊系统
基于最小二乘法构建模糊子系统,最终搭建出整个深度分层模糊系统。深度分层模糊系统结构如图2所示。
首先,先对系统的总体参数进行设置,手动确定层数L(设为3)、移动步长s(设为2)和卷积窗的长度w(设为2)。
将训练样本集D 1中的特征向量
Figure PCTCN2020132231-appb-000022
作为系统的输入向量,即
Figure PCTCN2020132231-appb-000023
类别标签y m作为每一个模糊子系统的输出向量。
构建出第一层第i个模糊子系统的输入-输出数据对:
Figure PCTCN2020132231-appb-000024
其通过长度为w(卷积算子)的移动窗口得到,该窗口从输入向量的第一个数据开始,一次移动一个步长,直至覆盖所有数据。0代表第一层的输入。
根据该数据对,确定模糊集合的值域[min x 0,max x 0]。在此值域内,输入向量就可以进一步划分为q个模糊集合A 1,A 2,...,A q。模糊集合A 1,A 2,...,A q可通过采用的三角模糊集合(也可采用其他模糊集合)已有计算公式得到。
第一层第i个模糊子系统可以表示为:
Figure PCTCN2020132231-appb-000025
利用已有标准公式进一步可得到
Figure PCTCN2020132231-appb-000026
的表达式并进行简化得:
Figure PCTCN2020132231-appb-000027
式中,
Figure PCTCN2020132231-appb-000028
指的是第1层的第i个模糊子系统,可见图2。
Figure PCTCN2020132231-appb-000029
指的是第1层第i个子系统实际运行得到的输出结果。c为参数矩阵。
对于上式中的参数
Figure PCTCN2020132231-appb-000030
采用最小二乘法进行设计,可以将其转化为:
Figure PCTCN2020132231-appb-000031
式中,S(c)代表定义的一个关于参数c的函数;y m代表正确的输出结果。求其最优解问题。
解得参数矩阵c,第一层第i个模糊子系统设计完成,按照以上方法,完成 第一层模糊子系统的搭建。
将第一层的输出
Figure PCTCN2020132231-appb-000032
作为第二层模糊子系统的输入向量,输出向量仍为y m,按照与第一层相同的设计方法来设计第二层的模糊子系统。以此类推,完成最后一层模糊子系统的设计,整个深度分层模糊系统搭建完成。
基于局部均值分解方法的特征向量提取和深度分层模糊系统分类两大部分构成识别模型。
通过将测试集D 2中的谐波信号数据输入到识别模型中,通过将识别结果与标签进行比较,测试出识别模型的精确度是否能够满足需求。如果精确度不能满足要求,则需要利用更多的样本数据重新对识别模型进行训练,直至精确度达到需求。
模块三:信号识别模块
本模块负责利用已构建出的识别模型对所采集到的谐波信号数据进行识别分析,判断出被检测设备的型号,进而判断出是否为环保设备以及是否为应当对应的环保设备类型。
将从被检测设备中采集到的谐波信号数据输入到已构建出的识别模型中,此模型首先对谐波信号数据进行特征向量的提取,然后将提取出的特征向量输入到深度分层模糊系统中,得出类型标签,判断出是否为对应的环保设备,利用显示设备输出分析得到的识别结果。
本发明的整体步骤如图3所示。
以上所述仅为本发明的优选实施例而已,并不用于限制本发明,对于本领域的技术人员来说,本发明可以有各种更改和变化。凡在本发明的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。

Claims (6)

  1. 一种基于深度分层模糊算法的环保设备识别方法,其特征在于,包括以下步骤:
    (1)通过谐波检测仪采集环保设备的谐波信号数据,并实地采集对应的环保设备类型信息,用于构建训练样本数据库;
    (2)通过局部均值分解方法提取训练样本数据库中的数据的特征向量,并利用训练样本数据库训练基于最小二乘法构建的深度分层模糊系统,以此构建出识别模型;
    (3)利用识别模型评估输入的谐波信号数据,以判断被检测设备是否为对应的环保设备。
  2. 根据权利要求1所述的基于深度分层模糊算法的环保设备识别方法,其特征在于,步骤(1)步骤如下:
    采集若干个信号周期的谐波信号数据x m(t),然后将此数据上传至云平台;
    并将收集所有谐波检测仪(设备节点m(m=1,2,...,n))对应的环保设备的类型信息,将设备的类型作为类别标签y m;其中,y m∈{1,2,...,k,k+1}(k≤n),标签1,2,...,k代表k种不同类型的环保设备,标签k+1代表非环保设备;对应关系为:
    Figure PCTCN2020132231-appb-100001
    利用采集到的谐波信号数据x m(t)和每个谐波信号数据对应的类别标签y m,构建出训练样本数据库D。
  3. 根据权利要求1所述的基于深度分层模糊算法的环保设备识别方法,其特征在于,步骤(2)步骤如下:
    步骤1:提取特征向量
    对训练样本数据库中的每个谐波信号x m(t)进行局部均值分解求出PF分量,取PF 1,PF 2,PF 3分量,并求出谐波信号x m(t)的PF r(r=1,2,3)分量的瞬时幅值a r(t)和瞬时频率f r(t),进一步利用平均值法求得各自的平均值
    Figure PCTCN2020132231-appb-100002
    Figure PCTCN2020132231-appb-100003
    利用谐波信号x m(t)的PF r分量的
    Figure PCTCN2020132231-appb-100004
    Figure PCTCN2020132231-appb-100005
    构建特征向量PF m,即
    Figure PCTCN2020132231-appb-100006
    步骤2:搭建深度分层模糊系统
    先对系统的总体参数进行设置,手动确定层数L、移动步长s和卷积窗的长度w;
    将训练样本集D 1中的特征向量
    Figure PCTCN2020132231-appb-100007
    作为系统的输入向量,即
    Figure PCTCN2020132231-appb-100008
    类别标签y m作为每一个模糊子系统的输出向量;
    构建出第一层第i个模糊子系统的输入-输出数据对:
    Figure PCTCN2020132231-appb-100009
    根据该数据对,确定模糊集合的值域[min x 0,max x 0];在此值域内,输入向量就可以进一步划分为q个模糊集合A 1,A 2,...,A q
    第一层第i个模糊子系统可以表示为:
    Figure PCTCN2020132231-appb-100010
    利用已有标准公式进一步可得到
    Figure PCTCN2020132231-appb-100011
    的表达式并进行简化得:
    Figure PCTCN2020132231-appb-100012
    对于上式中的参数
    Figure PCTCN2020132231-appb-100013
    采用最小二乘法进行设计,将其转化为:
    Figure PCTCN2020132231-appb-100014
    求其最优解问题;
    解得参数矩阵c,第一层第i个模糊子系统设计完成,按照以上方法,完成第一层模糊子系统的搭建;
    将第一层的输出
    Figure PCTCN2020132231-appb-100015
    作为第二层模糊子系统的输入向量,输出向量仍为y m,按照与第一层相同的设计方法来设计第二层的模糊子系统;以此类推,完成最后一层模糊子系统的设计,整个深度分层模糊系统搭建完成。
  4. 根据权利要求3所述的基于深度分层模糊算法的环保设备识别方法,其特征在于,步骤(2)中将训练样本数据库D中的数据分为两部分:训练集D 1和测 试集D 2,训练集D 1和测试集D 2均执行提取特征向量步骤;通过将测试集D 2中的谐波信号数据输入到识别模型中,通过将识别结果与标签进行比较,测试出识别模型的精确度是否能够满足需求;如果精确度不能满足要求,则需要利用更多的样本数据重新对识别模型进行训练,直至精确度达到需求。
  5. 根据权利要求1所述的基于深度分层模糊算法的环保设备识别方法,其特征在于,步骤(3)中将从被检测设备中采集到的谐波信号数据输入到已构建出的识别模型中,此模型首先对谐波信号数据进行特征向量的提取,然后将提取出的特征向量输入到深度分层模糊系统中,得出类型标签,判断出是否为对应的环保设备,利用显示设备输出分析得到的识别结果。
  6. 一种基于深度分层模糊算法的环保设备识别系统,其特征在于,用于在执行时实现权利要求1-5任一项所述的基于深度分层模糊算法的环保设备识别方法的步骤,包括:
    数据采集模块,该模块用于执行步骤(1)的方法;
    识别模型构建模块,该模块用于执行步骤(2)的方法;
    信号识别模块,该模块用于执行步骤(3)的方法。
PCT/CN2020/132231 2020-06-24 2020-11-27 基于深度分层模糊算法的环保设备识别方法与系统 WO2021258636A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/782,792 US20230014095A1 (en) 2020-06-24 2020-11-27 Method and system for recognizing environmental protection equipment based on deep hierarchical fuzzy algorithm

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010588932.8 2020-06-24
CN202010588932.8A CN111724290B (zh) 2020-06-24 2020-06-24 基于深度分层模糊算法的环保设备识别方法与系统

Publications (1)

Publication Number Publication Date
WO2021258636A1 true WO2021258636A1 (zh) 2021-12-30

Family

ID=72568769

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/132231 WO2021258636A1 (zh) 2020-06-24 2020-11-27 基于深度分层模糊算法的环保设备识别方法与系统

Country Status (3)

Country Link
US (1) US20230014095A1 (zh)
CN (1) CN111724290B (zh)
WO (1) WO2021258636A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111724290B (zh) * 2020-06-24 2023-09-26 山东建筑大学 基于深度分层模糊算法的环保设备识别方法与系统
CN116509337A (zh) * 2023-06-27 2023-08-01 安徽星辰智跃科技有限责任公司 基于局部分解的睡眠周期性检测及调节方法、系统和装置
CN117035562B (zh) * 2023-10-10 2024-01-30 云境商务智能研究院南京有限公司 基于电力大数据的环保智慧监测方法及数据分析设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101118420A (zh) * 2007-09-18 2008-02-06 郑州大学 基于分层模糊系统的石油钻井工程事故预警系统
US20140351188A1 (en) * 2013-05-22 2014-11-27 Exablade Corporation Prefetch system and method
CN109034054A (zh) * 2018-07-24 2018-12-18 华北电力大学 基于lstm的谐波多标签分类方法
CN110633870A (zh) * 2019-09-24 2019-12-31 国家电网有限公司 一种谐波预警方法、谐波预警装置及终端设备
CN111724290A (zh) * 2020-06-24 2020-09-29 山东建筑大学 基于深度分层模糊算法的环保设备识别方法与系统

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101118420A (zh) * 2007-09-18 2008-02-06 郑州大学 基于分层模糊系统的石油钻井工程事故预警系统
US20140351188A1 (en) * 2013-05-22 2014-11-27 Exablade Corporation Prefetch system and method
CN109034054A (zh) * 2018-07-24 2018-12-18 华北电力大学 基于lstm的谐波多标签分类方法
CN110633870A (zh) * 2019-09-24 2019-12-31 国家电网有限公司 一种谐波预警方法、谐波预警装置及终端设备
CN111724290A (zh) * 2020-06-24 2020-09-29 山东建筑大学 基于深度分层模糊算法的环保设备识别方法与系统

Also Published As

Publication number Publication date
CN111724290A (zh) 2020-09-29
CN111724290B (zh) 2023-09-26
US20230014095A1 (en) 2023-01-19

Similar Documents

Publication Publication Date Title
WO2021258636A1 (zh) 基于深度分层模糊算法的环保设备识别方法与系统
WO2022077605A1 (zh) 一种风力机叶片图像损伤检测和定位方法
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
CN108809948A (zh) 一种基于深度学习的异常网络连接检测方法
CN109543542A (zh) 一种特定场所人员着装是否规范的判定方法
CN112134871A (zh) 一种能源互联网信息支撑网络的异常流量检测装置及方法
CN111551888A (zh) 一种改进型AdaBoost算法的电能表计量数据故障分析方法
CN115187527A (zh) 一种多源混合型特高频局部放电图谱的分离识别方法
He et al. Intelligent Fault Analysis With AIOps Technology
CN110837532A (zh) 一种基于大数据平台对充电桩窃电行为的检测方法
Li et al. A learning-based comprehensive evaluation model for traffic data quality in intelligent transportation systems
CN114700587B (zh) 一种基于模糊推理和边缘计算的漏焊缺陷实时检测方法及系统
CN113962308A (zh) 一种航空设备故障预测方法
CN109613109A (zh) 一种管道漏磁检测数据自动分析系统
CN111190072A (zh) 集抄系统诊断模型建立方法、故障诊断方法及装置
CN112653675A (zh) 一种基于深度学习的智能入侵检测方法及其装置
Luo et al. Design of Computer Recognition System Based on Graphic Image
CN110765900A (zh) 一种基于dssd的自动检测违章建筑方法及系统
CN115880472A (zh) 一种电力红外图像数据智能诊断分析系统
CN109902209A (zh) 一种基于空间智能的特种承压设备用户三维可视化方法
CN105573984A (zh) 社会经济指标的识别方法及装置
CN114235653A (zh) 基于端云协同的大气颗粒污染物时空预测云平台
Liu et al. Substation intelligent monitoring system based on pattern recognition
CN203930500U (zh) 一种红外热成像检测数据分析诊断平台
Pu et al. Anomaly Detection of Substation Engineering Video Object Based on Cloud Platform

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: 20941816

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: 20941816

Country of ref document: EP

Kind code of ref document: A1