CN113988205B - Method and system for judging electric precipitation working condition - Google Patents

Method and system for judging electric precipitation working condition Download PDF

Info

Publication number
CN113988205B
CN113988205B CN202111315328.9A CN202111315328A CN113988205B CN 113988205 B CN113988205 B CN 113988205B CN 202111315328 A CN202111315328 A CN 202111315328A CN 113988205 B CN113988205 B CN 113988205B
Authority
CN
China
Prior art keywords
training
neural network
output layer
node
normalized
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
Application number
CN202111315328.9A
Other languages
Chinese (zh)
Other versions
CN113988205A (en
Inventor
钱云亮
黄成鑫
谢小杰
李仁贵
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Fujian Longking Co Ltd.
Original Assignee
Fujian Longking Co Ltd.
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 Fujian Longking Co Ltd. filed Critical Fujian Longking Co Ltd.
Priority to CN202111315328.9A priority Critical patent/CN113988205B/en
Publication of CN113988205A publication Critical patent/CN113988205A/en
Application granted granted Critical
Publication of CN113988205B publication Critical patent/CN113988205B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B21/00—Alarms responsive to a single specified undesired or abnormal condition and not otherwise provided for
    • G08B21/18—Status alarms
    • G08B21/24—Reminder alarms, e.g. anti-loss alarms
    • G—PHYSICS
    • G08—SIGNALLING
    • G08B—SIGNALLING OR CALLING SYSTEMS; ORDER TELEGRAPHS; ALARM SYSTEMS
    • G08B29/00—Checking or monitoring of signalling or alarm systems; Prevention or correction of operating errors, e.g. preventing unauthorised operation
    • G08B29/18—Prevention or correction of operating errors
    • G08B29/185—Signal analysis techniques for reducing or preventing false alarms or for enhancing the reliability of the system
    • G08B29/186—Fuzzy logic; neural networks
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A50/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE in human health protection, e.g. against extreme weather
    • Y02A50/20—Air quality improvement or preservation, e.g. vehicle emission control or emission reduction by using catalytic converters
    • Y02A50/2351—Atmospheric particulate matter [PM], e.g. carbon smoke microparticles, smog, aerosol particles, dust

Abstract

The invention discloses a method and a system for judging the working condition of electric precipitation, which are used for obtaining the current dust removal operation parameter under the current electric precipitation working condition, processing the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter, extracting an n-dimensional characteristic vector from the normalized operation parameter, inputting the n-dimensional characteristic vector to an SOM neural network model, mapping the n-dimensional characteristic vector to an output layer node to obtain a unique hot code value serving as an early warning value of the current working condition, searching an early warning type corresponding to the unique hot code value from a target vector matrix, and realizing the judgment of the electric precipitation working condition. According to the invention, the n-dimensional characteristic vector corresponding to the current dust removal operation parameter is input to the SOM neural network model, so that the n-dimensional characteristic vector is mapped to the output layer node to obtain the early warning value of the current working condition, and the output layer node basically keeps the probability statistical characteristic of the n-dimensional characteristic vector, therefore, the cluster analysis of the change trend of the whole dust removal operation parameter can be realized, and the accuracy of judging the electric dust removal working condition is improved.

Description

Method and system for judging electric precipitation working condition
Technical Field
The invention relates to the technical field of flue gas treatment, in particular to a method and a system for judging the working condition of electric precipitation.
Background
The electric dust removal system is a mechatronic system and consists of an electric system and a mechanical system, wherein the electric system comprises a high-voltage power supply system (such as an electrostatic dust collector), a low-voltage heating system and a low-voltage rapping system, and the mechanical system comprises a cathode wire, an anode plate, a pressure-bearing insulator, an ash bucket and the like. The main dust removal principle of the electric dust removal system is as follows: the high-voltage electrostatic field generated by the electrostatic dust collector acts on dust particles in the flue gas flow to charge the dust particles, and the charged dust particles move and gather towards a dust collecting plate of the electrostatic dust collector under the action of the high-voltage electrostatic field. The low-voltage rapping equipment acts according to a certain time sequence to make charged dust particles fall off and enter an ash hopper below the low-voltage rapping equipment, the pneumatic ash conveying equipment operates according to a preset working mode to convey accumulated ash in the ash hopper to an ash warehouse, and the operation is repeated so as to achieve the aim of removing dust.
For guaranteeing electric precipitation system's steady operation, need detect the electric precipitation operating mode, generally carry out the analysis to electric precipitation operating mode parameter and confirm whole electric precipitation system's running state, electric precipitation operating mode parameter includes: secondary voltage, sparking rate, ammonia escape problems, sampling loop and power supply output fluctuation size, etc. And alarming and reminding when the electric precipitation working condition parameters reach or exceed a preset threshold, wherein the preset threshold of each electric precipitation working condition parameter is generally obtained according to historical experience summary.
Therefore, the prior art adopts quantization logic when detecting the working condition of electric precipitation, and has a clear boundary. However, the quantization logic cannot reflect the relationship between the operation parameters and the operation state of the electric dust removal system, and false alarm and missed alarm are easy to occur, especially when the electric dust removal system is in a critical operation state, the situation of early warning false alarm or early warning missed alarm is easy to occur.
Disclosure of Invention
In view of the above, the invention discloses a method and a system for judging an electric precipitation working condition, which are used for mapping an n-dimensional feature vector corresponding to a current precipitation running parameter to an output layer node of an SOM neural network model to obtain a current working condition early warning value by inputting the n-dimensional feature vector into the SOM neural network model, wherein the output layer node basically keeps the probability statistical characteristic of the n-dimensional feature vector, so that the SOM neural network model can realize cluster analysis on the whole precipitation running parameter variation trend of the electric precipitation working condition, thereby effectively improving the accuracy of judging the electric precipitation working condition and effectively avoiding the occurrence of early warning false alarm or early warning missing alarm.
A method for judging the working condition of electric precipitation comprises the following steps:
acquiring current dust removal operation parameters under the current electric dust removal working condition;
performing normalization pretreatment on the current dedusting operation parameters by adopting a Z-score method to obtain normalized dedusting operation parameters;
extracting n-dimensional feature vectors from the normalized dust removal operation parameters, wherein each dimension in the n-dimensional feature vectors is an operation parameter, and n is a positive integer;
inputting the n-dimensional characteristic vector in the normalized dust removal operation parameter into a pre-constructed SOM neural network model, and mapping the n-dimensional characteristic vector to an output layer node to obtain a single-hot coded value serving as a current working condition early warning value;
and finding the early warning type corresponding to the one-hot coded value from a predetermined target vector matrix.
Optionally, the construction process of the SOM neural network model includes:
acquiring historical electric precipitation operation parameters, and determining a training data set from the historical electric precipitation operation parameters;
carrying out normalization processing on the training data set by adopting a Z-score method to obtain a normalized training set;
constructing an SOM neural network initial model, wherein the SOM neural network initial model comprises an input layer and an output layer, and the output layer is of a two-dimensional square network structure;
acquiring a training sample from the normalized training set, and extracting an n-dimensional feature vector from the training sample;
inputting the n-dimensional feature vectors in the training samples into the SOM neural network initial model, calculating the similarity between the feature vectors of all dimensions and the weight vectors of the output layer nodes aiming at each output layer node, and determining the output layer node with the highest similarity as a winner node;
determining a preset neighborhood radius of the winner node as a winner node neighborhood, and determining output layer nodes contained in the winner node neighborhood;
updating the weight vector corresponding to each output layer node contained in the win node neighborhood;
judging whether a preset iteration end condition is met or not;
if so, determining that the training of the SOM neural network initial model is completed, and obtaining a final SOM neural network model.
Optionally, the normalizing the training data set by using the Z-score method to obtain a normalized training set specifically includes:
the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For the normalized training set, X is the historical electric precipitation operating parameter, mu, in the training data set x Is the average value, sigma, of the historical electric precipitation operation parameters in the training data set x And the standard deviation of the historical electric precipitation operation parameters in the training data set is obtained.
Optionally, obtaining a training sample from the normalized training set, and determining an n-dimensional feature vector of the training sample, specifically including:
initializing the network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
and acquiring one training sample from the initialization training set by adopting a batch method, and determining the n-dimensional feature vector of the training sample.
Optionally, the preset iteration end condition includes: the iteration times reach preset times, or the learning rate of the initial model of the SOM neural network is smaller than a preset threshold value.
Optionally, the expression of the node number K in the SOM neural network initial model is as follows:
Figure GDA0003695736820000031
in the formula, N is the number of training samples in the normalized training set.
Optionally, the calculating the similarity between the feature vector of each dimension and the output layer node weight vector specifically includes:
and calculating the similarity of the feature vector of each dimension and the output layer node weight vector by adopting an Euclidean distance calculation method.
Optionally, the updating the weight vector corresponding to each output layer node included in the winner-node neighborhood specifically includes:
W ij (t+1)＝ W ij (t)+η(t,N)[x i p -W ij (t)]；
in the formula (I), the compound is shown in the specification, W ij (t +1) is the i-dimensional value of the weight of the jth node at the time t +1 in the neighbor of the winner node, W ij (t) is the i-dimensional value, x, of the weight of the jth node at time t in the neighborhood of the winner node i p For normalizing the p-th training sample in the training set, i.e. the current training sample, i ═ 1,2,3, … N is the dimension of the feature vector, j ∈ N j* (t), η (t, N) is a function of the training time t and the topological distance N between the jth neuron in the neighborhood of the winning node and the winning neuron j ═ η (t) e -N 。
A judging system for electric dust removal working conditions comprises:
the parameter acquisition unit is used for acquiring current dust removal operation parameters under the current electric dust removal working condition;
the normalization unit is used for carrying out normalization pretreatment on the current dedusting operation parameter by adopting a Z-score method to obtain a normalized dedusting operation parameter;
the vector extraction unit is used for extracting n-dimensional feature vectors from the normalized dust removal operation parameters, each dimension in the n-dimensional feature vectors is an operation parameter, and n is a positive integer;
the early warning value determining unit is used for inputting the n-dimensional characteristic vector in the normalized dust removal operation parameter into a pre-constructed SOM neural network model and mapping the n-dimensional characteristic vector to an output layer node to obtain a single hot coded value serving as a current working condition early warning value;
and the searching unit is used for searching the early warning type corresponding to the one-hot coded value from a predetermined target vector matrix.
Optionally, the method further includes:
the model construction unit is used for constructing the SOM neural network model;
the model building unit includes:
the parameter acquisition subunit is used for acquiring historical electric precipitation operation parameters and determining a training data set from the historical electric precipitation operation parameters;
the normalization subunit is used for performing normalization processing on the training data set by adopting a Z-score method to obtain a normalized training set;
the SOM neural network initial model comprises an initial model building subunit, a first model building subunit and a second model building subunit, wherein the initial model building subunit is used for building an SOM neural network initial model, the SOM neural network initial model comprises an input layer and an output layer, and the output layer is of a two-dimensional square network structure;
the sample acquisition subunit is used for acquiring a training sample from the normalized training set and extracting an n-dimensional feature vector from the training sample;
a winner node determining subunit, configured to input the n-dimensional feature vectors in the training samples to the SOM neural network initial model, calculate, for each output layer node, a similarity between the feature vector of each dimension and the output layer node weight vector, and determine an output layer node with a highest similarity as a winner node;
a neighborhood determining subunit, configured to determine a preset neighborhood radius of the winner node as a winner node neighborhood, and determine output layer nodes included in the winner node neighborhood;
the weight vector updating subunit is used for updating the weight vectors corresponding to the output layer nodes contained in the win node neighborhood;
a judging subunit, configured to judge whether a preset iteration end condition is satisfied;
and the model determining subunit is used for determining that the training of the initial model of the SOM neural network is finished under the condition that the judging subunit judges that the initial model of the SOM neural network is finished, so as to obtain a final SOM neural network model.
Optionally, the normalization subunit is specifically configured to:
the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For the normalized training set, X is the historical electric precipitation operating parameter, mu, in the training data set x Is the average value, sigma, of the historical electric precipitation operation parameters in the training data set x And the standard deviation of the historical electric precipitation operation parameters in the training data set is obtained.
Optionally, the sample acquiring subunit is specifically configured to:
initializing the network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
and acquiring one training sample from the initialization training set by adopting a batch method, and determining the n-dimensional feature vector of the training sample.
According to the technical scheme, the invention discloses a method and a system for judging the electric precipitation working condition, which are used for obtaining the current dust removal operation parameter under the current electric precipitation working condition, carrying out normalization pretreatment on the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter, extracting an n-dimensional characteristic vector from the normalized dust removal operation parameter, inputting the n-dimensional characteristic vector into a pre-constructed SOM neural network model, mapping the n-dimensional characteristic vector to an output layer node to obtain a single-hot-code value serving as an early-warning value of the current working condition, and searching the early-warning type corresponding to the single-hot-code value from a predetermined target vector matrix so as to realize the judgment on the electric precipitation working condition. The SOM belongs to a neural network for unsupervised learning, and the invention can realize that the n-dimensional characteristic vector corresponding to the current dust removal operation parameter is mapped to the output layer node of the SOM neural network model to obtain the early warning value of the current working condition by inputting the n-dimensional characteristic vector into the SOM neural network model, and the output layer node basically keeps the probability statistical characteristic of the n-dimensional characteristic vector, so the SOM neural network model can realize the cluster analysis of the whole dust removal operation parameter variation trend of the electric dust removal working condition, thereby effectively improving the accuracy of judging the electric dust removal working condition and effectively avoiding the occurrence of early warning false alarm or early warning missing alarm.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the disclosed drawings without creative efforts.
FIG. 1 is a flow chart of a method for determining the working conditions of electric precipitation, which is disclosed by the embodiment of the invention;
FIG. 2 is a flow chart of a method for constructing an SOM neural network model according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a system for determining the electric precipitation conditions, according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a model building unit according to an embodiment of the present invention.
Detailed Description
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 only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The embodiment of the invention discloses a method and a system for judging the working condition of electric precipitation, which are used for obtaining the current dust removal operation parameter under the current electric precipitation working condition, carrying out normalization pretreatment on the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter, extracting an n-dimensional characteristic vector from the normalized dust removal operation parameter, inputting the n-dimensional characteristic vector into a pre-constructed SOM neural network model, mapping the n-dimensional characteristic vector to an output layer node to obtain an independent thermal coding value serving as the early warning value of the current working condition, and searching the early warning type corresponding to the independent thermal coding value from a predetermined target vector matrix so as to realize the judgment on the electric precipitation working condition. The SOM belongs to a neural network for unsupervised learning, and the invention can realize that the n-dimensional characteristic vector corresponding to the current dust removal operation parameter is mapped to the output layer node of the SOM neural network model to obtain the early warning value of the current working condition by inputting the n-dimensional characteristic vector into the SOM neural network model, and the output layer node basically keeps the probability statistical characteristic of the n-dimensional characteristic vector, so the SOM neural network model can realize the cluster analysis of the whole dust removal operation parameter variation trend of the electric dust removal working condition, thereby effectively improving the accuracy of judging the electric dust removal working condition and effectively avoiding the occurrence of early warning false alarm or early warning missing alarm.
Referring to fig. 1, a flow chart of a method for determining an electric precipitation condition disclosed in the embodiment of the present invention includes:
s101, obtaining current dust removal operation parameters under the current electric dust removal working condition;
wherein, current electric precipitation operation parameter includes: current primary voltage, current primary current, current secondary voltage, current secondary current, current secondary voltage peak value, current sparking rate and the like.
S102, carrying out normalization pretreatment on the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter;
s103, extracting n-dimensional feature vectors from the normalized dust removal operation parameters;
wherein each dimension of the n-dimensional feature vector is an operating parameter, such as a secondary voltage U 2 And n is a positive integer.
Step S104, inputting the n-dimensional feature vector into a pre-constructed SOM neural network model, and mapping the n-dimensional feature vector to an output layer node to obtain a single hot code value serving as a current working condition early warning value;
the One-Hot coded value in this embodiment is also the One-Hot coded value.
The SOM (Self-Organizing mapping) algorithm is an unsupervised learning algorithm for clustering and high-dimensional visualization, and is an artificial neural network developed by simulating the characteristics of human brain on signal processing.
And S105, finding the early warning type corresponding to the one-hot coded value from a predetermined target vector matrix.
And the early warning type represents the current electric precipitation working condition.
The early warning types in this embodiment include: low secondary voltage, high cremation rate, high ammonia escape problem, and large fluctuation of sampling loop and power supply output.
It should be noted that, for each early warning type, a One-Hot encoded value is provided in the target vector matrix, for example, the One-Hot encoded value is [00100], and the corresponding early warning type is to investigate the problem of ammonia escape, where the target vector matrix in this embodiment is specifically shown in table 1.
TABLE 1
Early warning number Type of early warning One-Hot encoded value
1 Low secondary voltage [00001]
2 The sparking rate is too high [00010]
3 Troubleshooting ammonia escape problems [00100]
4 Checking sampling loop [01000]
5 Excessive fluctuation of power output [10000]
In summary, the invention discloses a method for judging the electric precipitation working condition, which comprises the steps of obtaining the current dust removal operation parameter under the current electric precipitation working condition, carrying out normalization pretreatment on the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter, extracting an n-dimensional characteristic vector from the normalized dust removal operation parameter, inputting the n-dimensional characteristic vector into a pre-constructed SOM neural network model, mapping the n-dimensional characteristic vector to an output layer node to obtain a unique hot code value serving as an early warning value of the current working condition, and searching the early warning type corresponding to the unique hot code value from a predetermined target vector matrix, thereby realizing the judgment on the electric precipitation working condition. The SOM belongs to a neural network for unsupervised learning, and the invention can realize that the n-dimensional characteristic vector is mapped to the output layer node of the SOM neural network model to obtain the early warning value of the current working condition by inputting the n-dimensional characteristic vector corresponding to the current dust removal operating parameter into the SOM neural network model.
Referring to fig. 2, a flow chart of a method for constructing an SOM neural network model disclosed in the embodiment of the present invention includes:
step S201, obtaining historical electric precipitation operation parameters, and determining a training data set from the historical electric precipitation operation parameters;
wherein, historical electric precipitation operation parameters can include: primary voltage U1, primary current I1, secondary voltage U2, secondary current I2, secondary voltage peak U2peak, sparking Rate Spark Rate, and the like.
In practical application, the historical electric precipitation operation parameters can be divided into a training data set and a testing data set according to a preset proportion, wherein the value of the preset proportion is determined according to actual needs, such as 7:3, and the invention is not limited herein.
S202, carrying out normalization processing on the training data set by adopting a Z-score method to obtain a normalized training set;
specifically, the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For normalized training set, X is the historical electrostatic precipitation operating parameter, μ, in the training data set x Is the average, sigma, of historical electrical precipitation operating parameters in the training dataset x The standard deviation of the historical electric precipitation operation parameters in the training data set is shown.
S203, constructing an SOM neural network initial model;
the network structure of the initial model of the SOM neural network has 2 layers, which are respectively: the device comprises an input layer and an output layer (also called competition layer), wherein the output layer is of a two-dimensional square network structure, the total number of nodes is K, and the number of the nodes K is obtained according to the following formula:
Figure GDA0003695736820000091
in the formula, K is the number of nodes in the initial model of the SOM neural network, and N is the number of training samples in the normalized training set.
Supposing that the number of training samples is 500, the node number is obtained according to a formulaIs composed of
Figure GDA0003695736820000092
Two-dimensional square has side length of
Figure GDA0003695736820000093
Rounding up yields an output layer with 11 × 11 neuron numbers.
Step S204, obtaining a training sample from the normalized training set, and extracting an n-dimensional feature vector from the training sample;
specifically, (1) initializing a network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
wherein the network weight matrix comprises a plurality of output layer node weight vectors.
Principal Component Analysis (PCA), one of the most commonly used dimension reduction methods, converts a set of variable data that may have correlation into a set of linearly uncorrelated variables by orthogonal transformation, and the converted variables are called Principal components.
(2) And acquiring a training sample from the initialization training set by adopting a batch method, and determining the n-dimensional characteristic vector of the training sample.
Step S205, inputting the n-dimensional feature vectors in the training samples into an SOM neural network initial model, calculating the similarity between the feature vectors of all dimensions and the weight vectors of the output layer nodes aiming at each output layer node, and determining the output layer node with the highest similarity as a winner node;
in practical application, the characteristic vector X can be calculated by adopting an Euclidean distance calculation method i And an output layer node weight vector W j The similarity is as follows:
Figure GDA0003695736820000094
x k representing a feature vector X i K-th dimension value of, w k Representing output layer node weight vector W j Is characterized by a k-th dimension value, k being from 1 to nVector X i Of (c) is calculated.
Step S206, determining the preset neighborhood radius of the winner node as a winner node neighborhood, and determining the output layer nodes contained in the winner node neighborhood;
in the invention, a high-speed function is selected as a neighborhood function, the update amplitude of a neighborhood node is calculated through the neighborhood function, the update amplitude is also a preset neighborhood radius, and the update amplitude is generally in inverse proportion to the distance between the neighborhood node and a winner node.
Step S207, updating the weight vector corresponding to each output layer node contained in the win node neighborhood;
wherein, the updated weight vector of each output layer node is obtained according to the following formula:
W ij (t+1)＝ W ij (t)+η(t,N)[x i p -W ij (t)]；
in the formula (I), the compound is shown in the specification, W ij (t +1) is the i-dimensional value of the weight of the jth node at the time t +1 in the neighbor of the winner node, W ij (t) is the i-dimensional value, x, of the weight of the jth node at time t in the neighborhood of the winner node i p For normalizing the p-th training sample in the training set, i.e. the current training sample, i ═ 1,2,3, … N is the dimension of the feature vector, j ∈ N j* (t), η (t, N) is a function of the training time t and the topological distance N between the jth neuron in the neighborhood of the winning node and the winning neuron j ═ η (t) e -N 。
And step S208, judging whether a preset iteration end condition is met, if so, executing step S209, otherwise, returning to execute step S202.
Wherein the iteration end condition comprises: the iteration times reach the preset times, or the learning rate of the initial model of the SOM neural network is smaller than the preset threshold, and the values of the preset times and the preset threshold are determined according to the actual needs, which is not limited herein.
And S209, determining that the training of the SOM neural network initial model is finished to obtain a final SOM neural network model.
In order to verify that the disclosed judging method for electric precipitation improves the accuracy and effectiveness of early warning diagnosis, the invention also verifies the SOM neural network model, and the method specifically comprises the following steps:
(1) establishing a target vector matrix by adopting an One-Hot method according to the diagnosis logic sample;
the target vector matrix can be seen in table 1.
(2) Extracting preset data containing a working condition early warning value from training data, encoding the preset data by adopting a preset encoding rule, inputting the encoded preset data to the SOM neural network model, mapping the encoded preset data to an output layer, and labeling nodes of the output layer.
(3) And comparing and verifying the effective prediction times of the SOM neural network model and the diagnosis logic sample after the output layer nodes are labeled by adopting a test data set.
See Table 2 for details
Type of early warning SOM neural network model Diagnostic logic sample
Low secondary voltage 19 13
The sparking rate is too high 11 8
Troubleshooting ammonia escape problems 5 3
Checking sampling loop 8 5
Excessive fluctuation of power output 6 2
Total up to 49 31
As can be seen from table 2, the number of effective early-warning times of the SOM neural network model is 49, and the number of effective prediction times of the diagnostic logic sample is 31. Obviously, the SOM neural network model improves the early warning diagnosis accuracy and effectiveness of the electric precipitation working condition, and brings convenience to the electric precipitation working condition analysis and the subsequent equipment maintenance.
Corresponding to the embodiment of the method, the invention also discloses a system for judging the working condition of the electric precipitation.
Referring to fig. 3, a schematic structural diagram of a system for determining electric precipitation conditions disclosed in the embodiment of the present invention includes:
a parameter obtaining unit 301, configured to obtain a current dust removal operating parameter under a current electric dust removal working condition;
wherein, current electric precipitation operation parameter includes: current primary voltage, current primary current, current secondary voltage, current secondary current, current secondary voltage peak value, current sparking rate and the like.
A normalization unit 302, configured to perform normalization preprocessing on the current dust removal operation parameter by using a Z-score method to obtain a normalized dust removal operation parameter;
a vector extraction unit 303, configured to extract an n-dimensional feature vector from the normalized dust removal operating parameters, where each dimension in the n-dimensional feature vector is an operating parameter, and n is a positive integer;
the early warning value determining unit 304 is configured to input the n-dimensional feature vector in the normalized dust removal operation parameter to a pre-constructed SOM neural network model, and map the n-dimensional feature vector to an output layer node to obtain a one-hot coded value serving as a current working condition early warning value;
the One-Hot coded value in this embodiment is also the One-Hot coded value.
The SOM (Self-Organizing mapping) algorithm is an unsupervised learning algorithm for clustering and high-dimensional visualization, and is an artificial neural network developed by simulating the characteristics of human brain on signal processing.
The searching unit 305 is configured to search, from a predetermined target vector matrix, an early warning type corresponding to the one-hot coded value.
And the early warning type represents the current electric precipitation working condition.
The early warning types in this embodiment include: low secondary voltage, high cremation rate, high ammonia escape problem, and large fluctuation of sampling loop and power supply output.
It should be noted that, for each early warning type, a One-Hot encoded value is provided in the target vector matrix, for example, the One-Hot encoded value is [00100], and the corresponding early warning type is to investigate the problem of ammonia escape, where the target vector matrix in this embodiment is specifically shown in table 1.
In summary, the invention discloses a system for judging the electric precipitation working condition, which obtains the current dust removal operation parameter under the current electric precipitation working condition, performs normalization pretreatment on the current dust removal operation parameter by adopting a Z-score method to obtain a normalized dust removal operation parameter, extracts an n-dimensional characteristic vector from the normalized dust removal operation parameter, inputs the n-dimensional characteristic vector into a pre-constructed SOM neural network model, maps the n-dimensional characteristic vector to an output layer node to obtain a unique hot code value serving as an early warning value of the current working condition, and searches an early warning type corresponding to the unique hot code value from a predetermined target vector matrix, thereby realizing the judgment on the electric precipitation working condition. The SOM belongs to a neural network for unsupervised learning, and the invention can realize that the n-dimensional characteristic vector is mapped to the output layer node of the SOM neural network model to obtain the early warning value of the current working condition by inputting the n-dimensional characteristic vector corresponding to the current dust removal operating parameter into the SOM neural network model.
To further optimize the above embodiment, the determination system may further include:
and the model building unit is used for building the SOM neural network model.
Referring to fig. 4, a schematic structural diagram of a model building unit disclosed in the embodiment of the present invention is shown, where the model building unit includes:
a parameter obtaining subunit 401, configured to obtain historical electric precipitation operation parameters, and determine a training data set from the historical electric precipitation operation parameters;
a normalization subunit 402, configured to perform normalization processing on the training data set by using a Z-score method to obtain a normalized training set;
an initial model constructing subunit 403, configured to construct an SOM neural network initial model, where the SOM neural network initial model includes an input layer and an output layer, and the output layer is a two-dimensional square network structure;
a sample obtaining subunit 404, configured to obtain a training sample from the normalized training set, and extract an n-dimensional feature vector from the training sample;
a winner node determining subunit 405, configured to input the n-dimensional feature vectors in the training samples into the SOM neural network initial model, calculate, for each output layer node, a similarity between the feature vector of each dimension and the output layer node weight vector, and determine an output layer node with the highest similarity as a winner node;
a neighborhood determining subunit 406, configured to determine a preset neighborhood radius of the winner node as a winner node neighborhood, and determine an output layer node included in the winner node neighborhood;
a weight vector updating subunit 407, configured to update weight vectors corresponding to output layer nodes included in the winner-node neighborhood;
a judging subunit 408, configured to judge whether a preset iteration end condition is met;
and a model determining subunit 409, configured to determine that training of the initial SOM neural network model is completed to obtain a final SOM neural network model when the determining subunit 408 determines that the initial SOM neural network model is the final SOM neural network model.
Wherein, the normalizing subunit 402 may specifically be configured to:
the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For the normalized training set, X is the historical electric precipitation operating parameter, mu, in the training data set x Is the average value, sigma, of the historical electric precipitation operation parameters in the training data set x And the standard deviation of the historical electric precipitation operation parameters in the training data set is obtained.
The sample acquiring subunit 404 may specifically be configured to:
initializing the network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
and acquiring one training sample from the initialization training set by adopting a batch method, and determining the n-dimensional feature vector of the training sample.
It should be noted that, for the specific working principle of each component in the system embodiment, please refer to the corresponding part of the method embodiment, which is not described herein again.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (12)

1. A method for judging the working condition of electric precipitation is characterized by comprising the following steps:
acquiring current dust removal operation parameters under the current electric dust removal working condition, wherein the current dust removal operation parameters comprise: a present primary voltage, a present primary current, a present secondary voltage, a present secondary current, a present secondary voltage peak value, and a present sparking rate;
performing normalization pretreatment on the current dedusting operation parameters by adopting a Z-score method to obtain normalized dedusting operation parameters;
extracting n-dimensional feature vectors from the normalized dust removal operation parameters, wherein each dimension in the n-dimensional feature vectors is an operation parameter, and n is a positive integer;
inputting the n-dimensional characteristic vector in the normalized dust removal operation parameter into a pre-constructed SOM neural network model, and mapping the n-dimensional characteristic vector to an output layer node to obtain a single-hot coded value serving as a current working condition early warning value;
and finding the early warning type corresponding to the One-Hot code value from a predetermined target vector matrix, wherein the target vector matrix has One-Hot code value corresponding to each early warning type.
2. The method of claim 1, wherein the construction process of the SOM neural network model comprises:
acquiring historical electric precipitation operation parameters, and determining a training data set from the historical electric precipitation operation parameters;
carrying out normalization processing on the training data set by adopting a Z-score method to obtain a normalized training set;
constructing an SOM neural network initial model, wherein the SOM neural network initial model comprises an input layer and an output layer, and the output layer is of a two-dimensional square network structure;
acquiring a training sample from the normalized training set, and extracting an n-dimensional feature vector from the training sample;
inputting the n-dimensional feature vectors in the training samples into the SOM neural network initial model, calculating the similarity between the feature vectors of all dimensions and the weight vectors of the output layer nodes aiming at each output layer node, and determining the output layer node with the highest similarity as a winner node;
determining a preset neighborhood radius of the winner node as a winner node neighborhood, and determining output layer nodes contained in the winner node neighborhood;
updating the weight vector corresponding to each output layer node contained in the win node neighborhood;
judging whether a preset iteration end condition is met or not;
if so, determining that the training of the SOM neural network initial model is completed, and obtaining a final SOM neural network model.
3. The method according to claim 2, wherein the normalizing the training data set by the Z-score method to obtain a normalized training set includes:
the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For the normalized training set, X is the historical electric precipitation operating parameter, mu, in the training data set x Is the average value, sigma, of the historical electric precipitation operation parameters in the training data set x And the standard deviation of the historical electric precipitation operation parameters in the training data set is obtained.
4. The method according to claim 2, wherein obtaining a training sample from the normalized training set and determining an n-dimensional feature vector of the training sample specifically includes:
initializing the network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
and acquiring one training sample from the initialization training set by adopting a batch method, and determining the n-dimensional feature vector of the training sample.
5. The method of claim 2, wherein the predetermined iteration end condition comprises: the iteration times reach preset times, or the learning rate of the initial model of the SOM neural network is smaller than a preset threshold value.
6. The determination method according to claim 2, wherein the expression of the number of nodes K in the SOM neural network initial model is as follows:
Figure FDA0003712521490000021
in the formula, N is the number of training samples in the normalized training set.
7. The method according to claim 2, wherein the calculating the similarity between the feature vector of each dimension and the output layer node weight vector specifically includes:
and calculating the similarity of the feature vector of each dimension and the output layer node weight vector by adopting an Euclidean distance calculation method.
8. The method according to claim 2, wherein the updating the weight vector corresponding to each output layer node included in the winning node neighborhood includes:
W ij (t+1)＝ W ij (t)+η(t,N)[x i p - W ij (t)]；
in the formula (I), the compound is shown in the specification, W ij (t +1) is the i-dimensional value of the weight of the jth node at the time t +1 in the neighborhood of the winner node, W ij (t) is the i-dimensional value, x, of the weight of the jth node at time t in the neighborhood of the winner node i p For normalizing the p-th training sample in the training set, i.e. the current training sample, i ═ 1,2,3, … N is the dimension of the feature vector, j ∈ N j* (t), η (t, N) is a function of the training time t and the topological distance N between the jth neuron in the neighborhood of the winning node and the winning neuron j ═ η (t) e -N 。
9. The utility model provides a judgement system of electric precipitation operating mode which characterized in that includes:
the parameter acquisition unit is used for acquiring current dust removal operation parameters under the current electric dust removal working condition, wherein the current dust removal operation parameters comprise: a present primary voltage, a present primary current, a present secondary voltage, a present secondary current, a present secondary voltage peak, and a present sparking rate;
the normalization unit is used for carrying out normalization pretreatment on the current dedusting operation parameter by adopting a Z-score method to obtain a normalized dedusting operation parameter;
the vector extraction unit is used for extracting n-dimensional feature vectors from the normalized dust removal operation parameters, each dimension in the n-dimensional feature vectors is an operation parameter, and n is a positive integer;
the early warning value determining unit is used for inputting the n-dimensional characteristic vector in the normalized dust removal operation parameter into a pre-constructed SOM neural network model and mapping the n-dimensional characteristic vector to an output layer node to obtain a single-hot coded value serving as a current working condition early warning value;
and the searching unit is used for searching the early warning type corresponding to the One-Hot code value from a predetermined target vector matrix, wherein the target vector matrix has One-Hot code value corresponding to each early warning type.
10. The determination system according to claim 9, further comprising:
the model construction unit is used for constructing the SOM neural network model;
the model building unit includes:
the parameter acquisition subunit is used for acquiring historical electric precipitation operation parameters and determining a training data set from the historical electric precipitation operation parameters;
the normalization subunit is used for performing normalization processing on the training data set by adopting a Z-score method to obtain a normalized training set;
the SOM neural network initial model comprises an initial model building subunit, a first model building subunit and a second model building subunit, wherein the initial model building subunit is used for building an SOM neural network initial model, the SOM neural network initial model comprises an input layer and an output layer, and the output layer is of a two-dimensional square network structure;
the sample acquisition subunit is used for acquiring a training sample from the normalized training set and extracting an n-dimensional feature vector from the training sample;
a winner node determining subunit, configured to input the n-dimensional feature vectors in the training samples to the SOM neural network initial model, calculate, for each output layer node, a similarity between the feature vector of each dimension and the output layer node weight vector, and determine an output layer node with a highest similarity as a winner node;
a neighborhood determining subunit, configured to determine a preset neighborhood radius of the winner node as a winner node neighborhood, and determine an output layer node included in the winner node neighborhood;
the weight vector updating subunit is used for updating the weight vectors corresponding to the output layer nodes contained in the win node neighborhood;
the judging subunit is used for judging whether a preset iteration ending condition is met or not;
and the model determining subunit is used for determining that the training of the initial model of the SOM neural network is finished under the condition that the judging subunit judges that the initial model of the SOM neural network is finished, so as to obtain a final SOM neural network model.
11. The decision system according to claim 10, wherein the normalizing subunit is specifically configured to:
the expression of the normalized training set obtained by the Z-score method is as follows:
X * ＝(X-μ x )/σ x ；
in the formula, X * For the normalized training set, X is the historical electric precipitation operating parameter, mu, in the training data set x Is the average value, sigma, of the historical electric precipitation operation parameters in the training data set x And the standard deviation of the historical electric precipitation operation parameters in the training data set is obtained.
12. The determination system according to claim 10, wherein the sample acquisition subunit is specifically configured to:
initializing the network weight matrix in the normalized training set by adopting a principal component analysis method to obtain an initialized training set;
and acquiring one training sample from the initialization training set by adopting a batch method, and determining the n-dimensional feature vector of the training sample.
CN202111315328.9A 2021-11-08 2021-11-08 Method and system for judging electric precipitation working condition Active CN113988205B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111315328.9A CN113988205B (en) 2021-11-08 2021-11-08 Method and system for judging electric precipitation working condition

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111315328.9A CN113988205B (en) 2021-11-08 2021-11-08 Method and system for judging electric precipitation working condition

Publications (2)

Publication Number Publication Date
CN113988205A CN113988205A (en) 2022-01-28
CN113988205B true CN113988205B (en) 2022-09-20

Family

ID=79747185

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111315328.9A Active CN113988205B (en) 2021-11-08 2021-11-08 Method and system for judging electric precipitation working condition

Country Status (1)

Country Link
CN (1) CN113988205B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116967640B (en) * 2023-09-22 2024-01-05 杭州众能光电科技有限公司 Perovskite battery layer following dust removal control device and method

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE0850040A1 (en) * 2008-10-16 2010-04-17 Uddeholm Tooling Ab Steel material and process for making them
CN103914735A (en) * 2014-04-17 2014-07-09 北京泰乐德信息技术有限公司 Failure recognition method and system based on neural network self-learning
CN106443310A (en) * 2016-11-22 2017-02-22 国网四川省电力公司广安供电公司 Transformer fault detection method based on SOM (Self Organizing Map) neural network
CN107798235A (en) * 2017-10-30 2018-03-13 清华大学 Unsupervised abnormal access detection method and device based on one hot encoding mechanisms
CN108648827A (en) * 2018-05-11 2018-10-12 北京邮电大学 Cardiovascular and cerebrovascular disease Risk Forecast Method and device
CN108694502A (en) * 2018-05-10 2018-10-23 清华大学 A kind of robot building unit self-adapting dispatching method based on XGBoost algorithms
CN111624887A (en) * 2020-06-08 2020-09-04 福建龙净环保股份有限公司 Electric dust removal control method and related device
WO2021073152A1 (en) * 2019-10-14 2021-04-22 平安科技（深圳）有限公司 Data label generation method and apparatus based on neural network, and terminal and medium
WO2021203854A1 (en) * 2020-04-09 2021-10-14 深圳壹账通智能科技有限公司 User classification method and apparatus, computer device and storage medium

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2003228706A (en) * 2002-02-05 2003-08-15 Fuji Xerox Co Ltd Data classifying device

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE0850040A1 (en) * 2008-10-16 2010-04-17 Uddeholm Tooling Ab Steel material and process for making them
CN103914735A (en) * 2014-04-17 2014-07-09 北京泰乐德信息技术有限公司 Failure recognition method and system based on neural network self-learning
CN106443310A (en) * 2016-11-22 2017-02-22 国网四川省电力公司广安供电公司 Transformer fault detection method based on SOM (Self Organizing Map) neural network
CN107798235A (en) * 2017-10-30 2018-03-13 清华大学 Unsupervised abnormal access detection method and device based on one hot encoding mechanisms
CN108694502A (en) * 2018-05-10 2018-10-23 清华大学 A kind of robot building unit self-adapting dispatching method based on XGBoost algorithms
CN108648827A (en) * 2018-05-11 2018-10-12 北京邮电大学 Cardiovascular and cerebrovascular disease Risk Forecast Method and device
WO2021073152A1 (en) * 2019-10-14 2021-04-22 平安科技（深圳）有限公司 Data label generation method and apparatus based on neural network, and terminal and medium
WO2021203854A1 (en) * 2020-04-09 2021-10-14 深圳壹账通智能科技有限公司 User classification method and apparatus, computer device and storage medium
CN111624887A (en) * 2020-06-08 2020-09-04 福建龙净环保股份有限公司 Electric dust removal control method and related device

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
《Life_Cycle_Identification_and_Analysis_of_Microblog_hot_Topics》;Yue He 等;《2018 10th International Conference on Intelligent Human-Machine Systems and Cybernetics (IHMSC)》;20180826;第2卷;全文 *
《基于复杂结构深度神经网络的入侵检测技术研究》;崔建京;《中国优秀硕士学位论文全文数据库》;20201215;全文 *
一种结构自适应的径向基函数神经网络;许新征;《计算机工程与应用》;20070511(第14期);全文 *

Also Published As

Publication number Publication date
CN113988205A (en) 2022-01-28

Similar Documents

Publication Publication Date Title
CN112699913A (en) Transformer area household variable relation abnormity diagnosis method and device
CN108073158A (en) Based on PCA and KNN density algorithm Wind turbines Method for Bearing Fault Diagnosis
CN112383052B (en) Power grid fault repairing method and device based on power internet of things
CN110879917A (en) Electric power system transient stability self-adaptive evaluation method based on transfer learning
CN112084237A (en) Power system abnormity prediction method based on machine learning and big data analysis
CN113378990B (en) Flow data anomaly detection method based on deep learning
CN113191253A (en) Non-invasive load identification method based on feature fusion under edge machine learning
CN113988205B (en) Method and system for judging electric precipitation working condition
CN111126820A (en) Electricity stealing prevention method and system
CN115618261A (en) Photovoltaic access power distribution network electric leakage identification method based on NCA and SSA-KELM
CN114266301A (en) Intelligent power equipment fault prediction method based on graph convolution neural network
CN111275069B (en) Non-invasive load monitoring method
CN113010504A (en) Electric power data anomaly detection method and system based on LSTM and improved K-means algorithm
CN113657030A (en) Method for predicting remaining service life of lithium battery based on Gaussian process regression
CN113344406A (en) Power quality monitoring reliability assessment method for intelligent fusion terminal in distribution network area
CN111026741A (en) Data cleaning method and device based on time series similarity
CN113884807B (en) Power distribution network fault prediction method based on random forest and multi-layer architecture clustering
CN110348489A (en) A kind of partial discharge of transformer mode identification method based on autoencoder network
CN115965135A (en) New energy prediction error modeling method and system based on naive Bayes classification
CN115859594A (en) Health evaluation algorithm of power transformation equipment based on hierarchical analysis
CN112883639A (en) GIS equipment service life prediction device and method based on machine learning
CN114066239A (en) User power consumption abnormity detection method and device
CN112949497A (en) GIS partial discharge mode identification method based on improved generalized regression neural network
Wang et al. Data-Driven Outage Restoration Time Prediction via Transfer Learning with Cluster Ensembles
Du et al. Shape matching and recognition base on genetic algorithm and application to plant species identification

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