CN106570561B - A kind of insoluble sediment density forecasting system of insulator surface and method - Google Patents
A kind of insoluble sediment density forecasting system of insulator surface and method Download PDFInfo
- Publication number
- CN106570561B CN106570561B CN201610983705.9A CN201610983705A CN106570561B CN 106570561 B CN106570561 B CN 106570561B CN 201610983705 A CN201610983705 A CN 201610983705A CN 106570561 B CN106570561 B CN 106570561B
- Authority
- CN
- China
- Prior art keywords
- neural network
- nsdd
- data
- insulator
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- 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
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Insulators (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of insoluble sediment density forecasting system of insulator surface and method, system includes raw data acquisition unit, tandem type grey neural network predicting unit, parallel connection type grey neural network predicting unit, embedded type grey neural network predicting unit, NSDD predicted value output unit and NSDD prewarning unit;By the insoluble sediment density data of insulator surface and local meteorological data investment tandem type grey neural network predicting unit, parallel connection type grey neural network predicting unit and embedded type grey neural network predicting unit, insulator NSDD numerical value is predicted respectively by this single predicting unit;Then judged using prediction accuracy of the test samples to these three predicting units, the output of the predicting unit high using prediction accuracy is as insulator NSDD predicted value;By NSDD prewarning unit according to whether there are two and the above predicting unit export predicted value reach default grading forewarning system threshold value to issue early warning.
Description
Technical field
The invention belongs to electric system external insulation technical fields, insoluble heavy more particularly, to a kind of insulator surface
Product object density prediction system and method.
Background technique
Insulator under normal working voltage is held under the bad weathers such as wet weather, dense fog due to the accumulation of surface filth object
Pollution flashover accident easily occurs, constitutes a serious threat to the safe and stable operation of electric system.To insulator on transmission line of electricity
Pollution degree, which carries out prediction, to be highly desirable, so as to the generation of timely Prevent from Dirt Flash accident.Usually using insoluble sediment density
(Non Soluble Deposit Density, NSDD) referred to as ash is close assesses pollution severity of insulators degree.
Gray model is because sample data needed for it is modeled is few, simple regardless of the regularity of distribution and variation tendency, modeling
Convenient advantage is widely used in terms of pollution severity of insulators degree prediction with operation, but since gray system lacks
Weary self study, self-organizing and adaptive ability, it is weaker to the processing capacity of information, it is unable to complete independently prediction task.
Summary of the invention
Aiming at the above defects or improvement requirements of the prior art, the present invention provides a kind of insoluble depositions of insulator surface
Object density prediction system and method can be widely applied to appoint its object is to combine gray model and neural network, provide one kind
The insoluble sediment density prediction technique and pre-warning function of meaning model insulator.
To achieve the above object, according to one aspect of the present invention, a kind of insoluble deposit of insulator surface is provided
Density prediction system, including raw data acquisition unit, tandem type grey neural network predicting unit, parallel connection type gray neural net
Network predicting unit, embedded type grey neural network predicting unit and NSDD predicted value output unit;
Wherein, raw data acquisition unit is used to obtain insulator NSDD data and the meteorological data on power transmission line;Series connection
Type grey neural network predicting unit, parallel connection type grey neural network predicting unit and embedded type grey neural network predicting unit
It is respectively used to predict the insulator NSDD on power transmission line according to insulator NSDD data on power transmission line and meteorological data;It is right
It answers, tandem type grey neural network predicting unit exports the first predicted value, the output of parallel connection type grey neural network predicting unit
Second predicted value, embedded type grey neural network predicting unit export third predicted value;
NSDD predicted value output unit be used for by above three predicted value and sample data progress and it is right, according to comparison result
From above three predicting unit choose the highest predicting unit of prediction accuracy, using the predicting unit export predicted value as
Insulator NSDD predicted value.
Preferably, the insoluble sediment density forecasting system of above-mentioned insulator surface further includes NSDD prewarning unit;
NSDD prewarning unit is used for according to above-mentioned first predicted value, the second predicted value, third predicted value and preset early warning
Threshold value generates pre-warning signal;Specifically, when two or more pre- in the first predicted value, the second predicted value, third predicted value
Measured value reaches threshold value of warning, generates pre-warning signal.
Preferably, the insoluble sediment density forecasting system of above-mentioned insulator surface, tandem type grey neural network are pre-
Surveying unit includes the first GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) model and neural network arranged side by side;
Wherein, GM (1,1) model is one kind of gray model, is one only comprising univariate differential equation of first order;Mind
It include input layer, hidden layer and output layer through network, transmission function uses Sigmoid function;
First GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) model input interface and initial data
Acquisition unit is connected;The input terminal of neural network and the first GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) mould
The output interface of type is connected;
First GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) model are respectively according to the insulation on power transmission line
Sub- NSDD data carry out insulator NSDD prediction, obtain three groups of ashing prediction results;It is pre- according to this three groups of ashing by neural network
It surveys result and carries out insulator NSDD prediction, obtain the first predicted value.
Preferably, the insoluble sediment density forecasting system of above-mentioned insulator surface, parallel connection type grey neural network are pre-
Surveying unit includes GM (1,1) model, neural network and combined prediction network;Wherein, GM (1,1) model is in parallel with neural network;
GM (1,1) model is connected with raw data acquisition unit with the input terminal of neural network;Combined prediction network
Input terminal is connected with the output end of the output end of GM (1,1) model, neural network;
GM (1,1) model and neural network respectively according on power transmission line insulator NSDD data and meteorological data carry out it is exhausted
Edge NSDD prediction;It is pre- that combined prediction network is weighted to obtain second to the prediction result of GM (1,1) model and neural network
Measured value.
Preferably, the insoluble sediment density forecasting system of above-mentioned insulator surface, embedded type grey neural network are pre-
Surveying unit includes the podzolic horizon being sequentially connected in series, neural network and albefaction layer;
Wherein, podzolic horizon be used for on original power transmission line insulator NSDD data and meteorological data carry out cumulative transformation
And smoothing processing;Neural network is used to carry out insulator NSDD prediction according to the smoothed out data that add up, and albefaction layer is used for mind
Output data through network carries out regressive and converts reduction treatment, obtains third predicted value.
It is another aspect of this invention to provide that being based on the insoluble sediment density forecasting system of above-mentioned insulator surface, provide
A kind of insulator surface insoluble sediment density prediction technique, includes the following steps:
(1) three sequence length different GM (1,1) models are established according to collected original insulator NSDD data;
With the prediction NSDD value of these three GM (1,1) models for input quantity, to measure NSDD as output quantity, investment neural network is instructed
Practice the best initial weights and threshold value for obtaining neural network;Insulator NSDD prediction is carried out with trained neural network, obtains first
Predicted value;
Wherein, measurement NSDD refers to collected NSDD data;
(2) insulator NSDD prediction is carried out by gray model and neural network model respectively, obtains two initial predicteds
Data;The weight coefficient of the two initial predicted data is determined according to test samples;According to weight coefficient to two initial predicteds
Data are weighted processing, obtain the second predicted value;
In this step, for the timing node k of quasi- prediction NSDD value, with (k-10) a measurement of 1~(k-10) node
NSDD data are as training sample, using 10 measurement NSDD data values of (k-10)~k as test samples;
(3) neural network is trained, and ashing processing is carried out to original insulator NSDD data;After ashing is handled
The neural network that has gone into training of data, insulator NSDD prediction is carried out by trained neural network;To neural network
Output data carries out whitening processing, obtains third predicted value;Wherein, ashing processing includes cumulative variation and smoothing processing;Albefaction
Processing refers to that regressive converts;
(4) above-mentioned first predicted value, the second predicted value and third predicted value and test samples are carried out and right, according to comparison
As a result the highest predicting unit of prediction accuracy is chosen from above three predicting unit, the predicted value exported with the predicting unit
As final insulator NSDD predicted value.
Preferably, the insoluble sediment density prediction technique of above-mentioned insulator surface, step (3) include following sub-step
It is rapid:
(3.1) before using meteorological data, the original insulator NSDD data before timing node m and timing node m
The ashing data of 10 timing nodes neural network is trained, obtain best initial weights and threshold value;According to best initial weights and
Threshold value constructs to obtain trained neural network;
Wherein, m is the timing node of training prediction insulator NSDD data value, the input data that training neural network uses
Including meteorological data, the original insulator NSDD data of 1~timing node of timing node (m-1), 10 times of (m-10)~m
Treated ashing data that the original insulator NSDD data of node are ashed;Meteorological data includes wind speed, precipitation, relatively wet
Degree;The output data that training neural network uses refers to the original insulator NSDD number of (m-9)~(m+1) totally 10 timing nodes
According to it is ashed treated ashing data;
(3.2) ashing processing is carried out to the original insulator NSDD data of m node;Data after ashing are gone into training
Good neural network obtains initial prediction;Whitening processing is carried out to initial prediction, obtains the second predicted value.
Preferably, the insoluble sediment density prediction technique of above-mentioned insulator surface further includes step (5):
(5) the first predicted value, the second predicted value, third predicted value are compared with preset threshold value of warning respectively, when
First predicted value, the second predicted value, the two or more in third predicted value reach threshold value of warning, generate pre-warning signal.
In general, through the invention it is contemplated above technical scheme is compared with the prior art, can obtain down and show
Beneficial effect:
(1) the insoluble sediment density forecasting system of insulator surface provided by the invention and method, by gray model with
Neural network model organically blends, the unique method not only modeled with gray system with Small Sample Database, but also has neural network
Model has the advantages that adaptive ability to non-linear, non-precision rule;It can model from different angles, different by combination
The different information of system is obtained, achieve the purpose that improve precision of prediction and increase the reliability of stability and result, makes combination die
Type has stronger robustness for the variation of data structure, effectively compensates for the defect of Individual forecast method inaccuracy;
(2) the insoluble sediment density forecasting system of insulator surface provided by the invention and method, the series connection used
Neural network is connect by type grey neural network predicting unit in series in systems with gray model, with gray model
The input as neural network is exported, can be used for the prediction of complication system robust parsing, when greatly reducing the training of neural network
It is long, and solve the problems, such as that single neural network prediction easily falls into local minimum;
(3) the insoluble sediment density forecasting system of insulator surface provided by the invention and method pass through weighting
Gray model and neural network are combined by mode, are constructed parallel connection type grey neural network predicting unit, are overcome single mould
The defect of type information easy to be lost, reduces randomness, has the effect of improving precision of prediction;
(4) the insoluble sediment density forecasting system of insulator surface provided by the invention and method, embedded type grey
Neural network prediction unit weakens the randomness of initial data by setting podzolic horizon, is easily the non-linear excitation letter of neural network
It is several to be approached, e-learning duration is greatly shortened, accelerates convergence process while improving precision of prediction;
(5) the insoluble sediment density forecasting system of insulator surface provided by the invention and method, it is auxiliary with gray system
Constructing neural network is helped, since the message structure of gray system includes certainty information and unascertained information, uses gray system
In certainty information carry out auxiliary construction neural network, the structure of neural network is instructed by certainty information, improves nerve
The learning algorithm of network;
(6) the insoluble sediment density forecasting system of insulator surface provided by the invention and method, neural network can
Effectively enhancing gray system;It, can only since empty set (not including the time zone of information) occurs in information time zone in gray system
Approximate, not exclusively determining grey differential equation is established, and is difficult to directly use Grey Differential Equation in practical applications, is needed
Grey differential equation is parsed;Constructing neural network of the present invention parses the grey parameter of grey differential equation, from grey colour system
It extracts sample in known data of uniting to be trained neural network, the ash that parsing is therefrom extracted when neural network convergence is micro-
Divide equation parameter, obtain the determining differential equation, realizes accurate prediction.
Detailed description of the invention
Fig. 1 is the insoluble sediment density forecasting system schematic diagram of insulator surface that embodiment provides;
Fig. 2 is the prediction flow diagram of GM in embodiment (1,1) model;
Fig. 3 is the Artificial Neural Network Structures schematic diagram in embodiment;
Fig. 4 is the neural network prediction flow diagram in embodiment;
Fig. 5 is the structural schematic diagram of the tandem type Grey Neural Network Model in embodiment;
Fig. 6 is the structural schematic diagram of the parallel connection type Grey Neural Network Model in embodiment;
Fig. 7 is the structural schematic diagram of the embedded type Grey Neural Network Model in embodiment.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.As long as in addition, technical characteristic involved in the various embodiments of the present invention described below
Not constituting a conflict with each other can be combined with each other.
The insoluble sediment density forecasting system of insulator surface provided in an embodiment of the present invention, as shown in Figure 1, including original
Beginning data acquisition unit, tandem type grey neural network predicting unit, parallel connection type grey neural network predicting unit, embedded type ash
Color neural network prediction unit, NSDD predicted value output unit and NSDD prewarning unit;
Wherein, raw data acquisition unit is for obtaining insulator NSDD data and meteorological data on power transmission line;Tandem type
Grey neural network predicting unit, parallel connection type grey neural network predicting unit and embedded type grey neural network predicting unit point
Insulator NSDD is not predicted according to insulator NSDD data on power transmission line and meteorological data, it is corresponding, it is pre- to obtain first
Measured value, the second predicted value, third predicted value;In the present embodiment, using the close on-line monitoring equipment of optical sensor power transmission and transformation ash or its
Its on-Line Monitor Device acquires insulator NSDD data on power transmission line;Meteorological data is collected around insulator to be predicted
Data;
NSDD predicted value output unit carries out above three predicted value and sample data and right, predicts from above three single
The highest predicting unit of prediction accuracy is chosen in member, is predicted using the predicted value that the predicting unit exports as insulator NSDD
Value;
NSDD prewarning unit is according to the first predicted value, the second predicted value, third predicted value and preset grading forewarning system threshold value
Generate pre-warning signal;Specifically, when two or more predicted values in the first predicted value, the second predicted value, third predicted value
Reach grading forewarning system threshold value, generates pre-warning signal.
The pre- flow gauge of GM employed in the present embodiment (1,1) model is as schematically shown in Figure 2, as follows according to scheming:
Original data sequence X is inputted first(0)=[x(0)(1),x(0)(2),…,x(0)(n)];
Then judge above-mentioned original data sequence X(0)Whether modeling conditions are met;If it is not, then carrying out the smooth place of data
Reason, if so, doing accumulation process to initial data obtains cumulative sequence X(1);x(1)=[x(1)(1),x(1)(2),…,x(1)(n)];
Wherein
Then it constructs about X(1)Linear first-order differential equationJoined through least square method solution
Number a and b;
Wherein,
Thus to obtain the predicted value of (1) X,
It restores to obtain the predicted value of (0) X through regressive,
The structure of neural network model employed in the present embodiment is as schematically shown in Figure 3, including input layer, hidden layer,
Output layer;WijFor the weight of input layer to hidden layer, θjFor the threshold value of hidden layer neuron, VijFor the company of hidden layer to output layer
Weight is connect,For the threshold value of output layer;The connection weight W of neural networkij、VijAnd threshold θj、It is obtained by training;
The input of each neuron of hidden layerWherein XiFor the input of each input node of neural network
Amount;
The transmission function of neural network uses Sigmoid function f (x)=1/ (1+e-x);
The output of hidden layer
The input of output layer neuron
The output of output layer neuron, the i.e. predicted value of neural network
The process predicted by above-mentioned neural network is as shown in figure 4, include the following steps:
The basic structure of neural network is constructed first: the input layer, hidden layer and output layer number of nodes of neural network are set,
Input the initial weight and threshold value of hidden layer, the initial weight and threshold value of input and output layer;
Then a sample is chosen from sample database to train neural network;In trained process, according to hidden layer
Error and output and output layer error and output, constantly to adjust the weight and threshold value and output layer of hidden layer
Weight and threshold value;
Terminate mind in the case where meeting the requirement of study number and error condition by the neural network that whole sample trainings are crossed
Training through network.
In embodiment, the structure of used tandem type grey neural network predicting unit is as shown in Figure 5;Including 3 GM
(1,1) model, GM1、GM2And GM3And 3 × 7 × 1 type neural network;
The neural network includes 3 input layers, 7 hidden layer nodes, 1 output node layer;3 GM (1,1) moulds
Type GM1, GM2, GM3For the input quantity of neural network;The transmission function of output layer to hidden layer, hidden layer to output layer is
Sigmoid type function, the installation warrants Kolmogorov theorem of hidden layer;The output of neural network is the first predicted value.
The step of carrying out NSDD prediction using above-mentioned tandem type grey neural network predicting unit is specific as follows:
(1.1) 3 GM (1,1) models, 3 GM (1,1) models are established using the NSDD data that data acquisition module obtains
Sequence length be respectively 10,8 and 6;That is when (k-20)~(k-10), (k-18)~(k-10) and (k-16)~(k-10) a
The data of intermediate node;Wherein, k refers to the timing node for needing to carry out insulator NSDD prediction;
(1.2) it is predicted respectively with this 3 GM (1,1) models, obtain 3 groups of NSDD prediction data, every group includes (k-
10) 10 NSDD prediction data values of~k timing node;
(1.3) the shared NSDD predicted value of above-mentioned 3 GM (1,1) model is this 16 NSDD data of (k-16)~k, with this
Input quantity of 16 NSDD prediction data values as neural network, using the original insulator NSDD data that actual acquisition arrives as mind
Output quantity through network is trained neural network, obtain neural network best initial weights and threshold value;In embodiment, this step
Suddenly the neural network structure used is 3 × 7 × 1 type;
(1.4) it is carried out using insulator NSDD value of the trained neural network to the future time instance after k timing node
Prediction obtains the first predicted value.
In embodiment, the structure of used parallel connection type grey neural network predicting unit as shown in fig. 6, include GM (1,
1) model, neural network and combined prediction network;GM (1,1) model is in parallel with neural network, and the output of the two is pre- as combination
The input of survey grid network, the output of combined prediction network are the second predicted value;It is predicted using the parallel connection type grey neural network single
It is specific as follows that member carries out the step of NSDD prediction:
(2.1) it is predicted, is obtained initial pre- respectively with neural network model using Grey models GM (1,1) model
Survey NSDD value y1And y2;
(2.2) test value Y is utilized(0)(t) initial predicted NSDD value y is individually subtracted1And y2, obtain prediction error e1And e2;
The weight coefficient ω of GM (1,1) model is obtained according to prediction error calculation1With the weight coefficient of neural network model
ω2;ω1+ω2=1;
(2.3) according to yc=ω1y1+ω2y2Obtain the second predicted value.
Wherein, weight coefficient ω1With weight coefficient ω2It is obtained according to following methods:
According to checking sequence Y(0)(t) e is obtained1、e2And ecFor
The second predicted value of insulator NSDD ycVariance be
To Var (ec) minimizing acquisition
Due to ω2=1- ω1Therefore, weight coefficient ω1With weight coefficient ω2It is respectively as follows:
Wherein, Var (e1)=σ11, Var (e2)=σ22, cov (e1,e2)=σ12。
The structure of embedded type grey neural network predicting unit is as shown in fig. 7, be sequentially connected in series employed in the present embodiment
Podzolic horizon, neural network and albefaction layer;Podzolic horizon does cumulative transformation and smoothing processing to initial data, and albefaction layer is to nerve net
The output data of network carries out regressive and converts reduction treatment;Wherein, the structure of neural network is 4 × 9 × 1 type;Neural network it is defeated
Entering layer is meteorological data, NSDD data;Wherein NSDD data are the tired of (k-10) node~kth node totally 10 timing nodes
Addend evidence;Its hidden layer is arranged according to Kolmogorov theorem;Output layer be prediction NSDD data, value be (k-9)~
The cumulative data of (k+1) totally 10 timing nodes;
The step of carrying out NSDD prediction using the embedded type grey neural network predicting unit is specific as follows:
(3.1) to original data sequence X(0)=(x(0)(1),x(0)(2),…,x(0)(n)) it carries out cumulative transformation and obtains 1-
AGO sequence X(1)=(x(1)(1),x(1)(2),…,x(1)(n));
Wherein, x(0)..., (k) >=0, k=1,2 n;
(3.2) 3 smoothing processings are carried out to above-mentioned 1-AGO sequence;
For k=2,3 ..., the node of n-1:
Smoothing processing formula is
To two endpoints of k=1 and k=n:
Smoothing processing formula is
(3.3) it is predicted by neural network according to the result of ashing processing;
(3.4) to the output sequence X of neural network(3)=(x(3)(1),x(3)(2),…,x(3)(n)) regressive is carried out to become
It changes,
Obtain third predicted value X(4)=(x(4)(1),x(4)(2),…,x(4)(n)), wherein x(3)..., (k) >=0, k=1,2
n;
x(3)(k)=x(2)(k)-x(2)(k-1)。
In the present embodiment, early warning and alert can be carried out according to the predicted value of above-mentioned each unit using NSDD prewarning unit;Early warning
A, B, C, D totally 4 warning grades are arranged in unit.
Wherein, when there are two the NSDD predicted values of predicting unit output to reach possible hair in three predicting units of this system
Insulator NSDD numerical value ρ when raw pollution flashoverF95% when, i.e. 95% ρF, system A grades of early warning of sending;When three prediction moulds of this system
NSDD predicted value in type there are two prediction model reaches insulator NSDD numerical value ρ when pollution flashover may occurF90% when, i.e.,
90% ρF, system B grades of early warning of sending;When the NSDD predicted value in three prediction models of this system there are two prediction model reaches
Insulator NSDD numerical value ρ when pollution flashover may occurF85% when, i.e. 85% ρF, system C grades of early warning of sending;When three of this system
NSDD predicted value in prediction model there are two prediction model reaches insulator NSDD numerical value ρ when pollution flashover may occurF80%
When, i.e. 80% ρF, system D grades of early warning of sending.
Insulator NSDD value when by NSDD prewarning unit by insulator NSDD predicted value and generation pollution flashover compares next life
At warning information for operations staff processing, can play the role of it is timely and effective prevent transmission line of electricity occur pollution flashover accident.
As it will be easily appreciated by one skilled in the art that the foregoing is merely illustrative of the preferred embodiments of the present invention, not to
The limitation present invention, any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should all include
Within protection scope of the present invention.
Claims (8)
1. a kind of insoluble sediment density forecasting system of insulator surface, which is characterized in that including raw data acquisition unit,
Tandem type grey neural network predicting unit, parallel connection type grey neural network predicting unit, the prediction of embedded type grey neural network
Unit and NSDD predicted value output unit;
The raw data acquisition unit is used to acquire original insulator NSDD data and the meteorological data on power transmission line;
The tandem type grey neural network predicting unit is used for according to collected original insulator NSDD data, based on series connection
Type Grey Neural Network Model obtains the first predicted value and specifically establishes three according to collected original insulator NSDD data
A sequence length different GM (1,1) model;Using the prediction NSDD value of these three GM (1,1) models output as the defeated of neural network
Enter amount, neural network is trained using collected original insulator NSDD data as the output quantity of neural network;With training
Good neural network carries out insulator NSDD prediction, obtains the first predicted value;
The parallel connection type grey neural network predicting unit is used for according to collected original insulator NSDD data, based on parallel connection
Type Grey Neural Network Model obtains the second predicted value, and specifically, the input of gray model and neural network is collected
Original insulator NSDD data, gray model export an initial prediction, and neural network exports another initial prediction, root
The weight coefficient that two initial predicted data are determined according to test samples, according to the weight coefficient to two initial predicted data into
Row weighting processing, obtains the second predicted value;
The embedded type grey neural network predicting unit is used for according to collected original insulator NSDD data, based on insertion
Type Grey Neural Network Model obtains third predicted value and is specifically ashed to collected original insulator NSDD data
Processing, is trained neural network, and using ashing treated data and meteorological data as input data, use is trained
Neural network carries out insulator NSDD prediction;Whitening processing is carried out to the output data of neural network, obtains third predicted value;Institute
Stating ashing processing includes cumulative variation and smoothing processing;The whitening processing refers to that regressive converts;
The NSDD predicted value output unit is used for first predicted value, the second predicted value and third predicted value and examines sample
Originally it is compared, chooses prediction accuracy highest one from three predicted values according to comparison result and predicted as insulator NSDD
Value.
2. the insoluble sediment density forecasting system of insulator surface as described in claim 1, which is characterized in that further include
NSDD prewarning unit;
The NSDD prewarning unit is used to generate pre-warning signal according to three predicted values and preset threshold value of warning;Specifically
Ground then generates pre-warning signal, the threshold value of warning when two or three predicted values of three predicted values reach threshold value of warning
It is set according to insulator NSDD numerical value when pollution flashover occurs.
3. the insoluble sediment density forecasting system of insulator surface as claimed in claim 1 or 2, which is characterized in that described
Tandem type grey neural network predicting unit includes: the first GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) mould
Type is in parallel, then neural network of connecting;
First GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) model input interface with it is described original
Data acquisition unit is connected;The input terminal of the neural network and the first GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM
(1,1) output interface of model is connected;
First GM (1,1) model, the 2nd GM (1,1) model, the 3rd GM (1,1) model are respectively according to the insulation on power transmission line
Sub- NSDD data carry out insulator NSDD prediction, obtain three groups of ashing prediction results;By neural network according to three groups of ashing
Prediction result carries out insulator NSDD prediction, obtains the first predicted value.
4. the insoluble sediment density forecasting system of insulator surface as claimed in claim 1 or 2, which is characterized in that described
Parallel connection type grey neural network predicting unit includes: that GM (1,1) model and neural network are in parallel, then tandem compound predicts network;
GM (1,1) model is in parallel with neural network;GM (1,1) model and the input terminal of neural network are and original number
It is connected according to acquisition unit;The input terminal of the combined prediction network and the output end of GM (1,1) model, the output end of neural network
It is connected;
GM (1,1) model and neural network for respectively according on power transmission line insulator NSDD data and meteorological data into
Row insulator NSDD prediction;The combined prediction network is for adding the prediction result of GM (1,1) model and neural network
Power obtains the second predicted value.
5. the insoluble sediment density forecasting system of insulator surface as claimed in claim 1 or 2, which is characterized in that described
Embedded type grey neural network predicting unit includes the podzolic horizon being sequentially connected in series, neural network and albefaction layer;
The podzolic horizon be used for on power transmission line insulator NSDD data and meteorological data carry out cumulative transformation and smoothing processing;
The neural network is used to carry out insulator NSDD prediction according to the smoothed out data that add up, and the albefaction layer is used for nerve net
The output data of network carries out regressive and converts reduction treatment, obtains third predicted value.
6. a kind of based on the exhausted of the insoluble sediment density forecasting system of the described in any item insulator surfaces of Claims 1 to 5
The insoluble sediment density prediction technique in edge sublist face, which comprises the steps of:
(1) the original insulator NSDD data and meteorological data on power transmission line are acquired;
(2) according to collected original insulator NSDD data, the first prediction is obtained based on tandem type Grey Neural Network Model
Value, specifically, establishes three sequence length different GM (1,1) models according to collected original insulator NSDD data;With
The prediction NSDD value of these three GM (1,1) models output is the input quantity of neural network, with collected original insulator NSDD
Data are that the output quantity of neural network is trained neural network;It is pre- that insulator NSDD is carried out with trained neural network
It surveys, obtains the first predicted value;
(3) according to collected original insulator NSDD data, the second prediction is obtained based on parallel connection type Grey Neural Network Model
Value, specifically, the input of gray model and neural network are collected original insulator NSDD data, gray model output
One initial prediction, neural network export another initial prediction, determine two initial predicted data according to test samples
Weight coefficient, processing is weighted to two initial predicted data according to the weight coefficient, obtains the second predicted value;
(4) it according to collected original insulator NSDD data and meteorological data, is obtained based on embedded type Grey Neural Network Model
Third predicted value is taken, specifically, ashing processing is carried out to collected original insulator NSDD data, neural network is instructed
Practice, be used as input data using ashing treated data and meteorological data, using trained neural network progress insulator
NSDD prediction;Whitening processing is carried out to the output data of neural network, obtains third predicted value;The ashing processing includes cumulative
Variation and smoothing processing;The whitening processing refers to that regressive converts;
(5) first predicted value, the second predicted value and third predicted value and test samples are carried out and right, according to comparison result
Prediction accuracy highest one is chosen from three predicted values is used as insulator NSDD predicted value.
7. the insoluble sediment density prediction technique of insulator surface as claimed in claim 6, which is characterized in that the step
(4) include following sub-step:
(4.1) using the original insulator NSDD data before meteorological data, timing node m and 10 before timing node m
The ashing data of a timing node are trained neural network, obtain best initial weights and threshold value;According to best initial weights and threshold value
Building obtains trained neural network;
Wherein, m is the timing node of training prediction insulator NSDD data value, and the input data that training neural network uses includes
Meteorological data, the original insulator NSDD data of 1~timing node of timing node (m-1), 10 timing nodes of (m-10)~m
Original insulator NSDD data it is ashed treated ashing data;Meteorological data includes wind speed, precipitation, relative humidity;
The output data that training neural network uses refers to the original insulator NSDD data of (m-9)~(m+1) totally 10 timing nodes
It is ashed treated ashing data;
(4.2) ashing processing is carried out to the original insulator NSDD data of m node;Data after ashing have been gone into training
Neural network obtains initial prediction;Whitening processing is carried out to initial prediction, obtains third predicted value.
8. the insoluble sediment density prediction technique of insulator surface as claimed in claim 6 is it is characterized in that, further include step
Suddenly (6):
First predicted value, the second predicted value, third predicted value are compared with preset threshold value of warning respectively, when first
Predicted value, the second predicted value, two or three in third predicted value reach threshold value of warning, generate pre-warning signal, the early warning
Threshold value is set according to insulator NSDD numerical value when pollution flashover occurs.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610983705.9A CN106570561B (en) | 2016-11-08 | 2016-11-08 | A kind of insoluble sediment density forecasting system of insulator surface and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610983705.9A CN106570561B (en) | 2016-11-08 | 2016-11-08 | A kind of insoluble sediment density forecasting system of insulator surface and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106570561A CN106570561A (en) | 2017-04-19 |
CN106570561B true CN106570561B (en) | 2019-04-12 |
Family
ID=58540708
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610983705.9A Active CN106570561B (en) | 2016-11-08 | 2016-11-08 | A kind of insoluble sediment density forecasting system of insulator surface and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106570561B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR20220007166A (en) * | 2019-05-16 | 2022-01-18 | 다이킨 고교 가부시키가이샤 | Learning model creation method, program, storage medium, learning completion model |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102590677A (en) * | 2012-02-28 | 2012-07-18 | 浙江省电力试验研究院 | Analyzing and processing method for test data of manual pollution flashover of insulator |
CN104502410A (en) * | 2013-07-21 | 2015-04-08 | 国家电网公司 | Prediction method for insulator equivalent salt deposit density and non-soluble deposit density by least squares support vector machine and genetic algorithm |
CN104992055A (en) * | 2015-06-16 | 2015-10-21 | 哈尔滨工业大学 | Pollution-flashover-caused trip probability calculation method of overhead lines in sand and dust environment |
-
2016
- 2016-11-08 CN CN201610983705.9A patent/CN106570561B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102590677A (en) * | 2012-02-28 | 2012-07-18 | 浙江省电力试验研究院 | Analyzing and processing method for test data of manual pollution flashover of insulator |
CN104502410A (en) * | 2013-07-21 | 2015-04-08 | 国家电网公司 | Prediction method for insulator equivalent salt deposit density and non-soluble deposit density by least squares support vector machine and genetic algorithm |
CN104992055A (en) * | 2015-06-16 | 2015-10-21 | 哈尔滨工业大学 | Pollution-flashover-caused trip probability calculation method of overhead lines in sand and dust environment |
Non-Patent Citations (1)
Title |
---|
用神经网络预测基于气象因素的绝缘子等值附盐密度;张寒,文习山,丁辉;《高压电器》;20031231;第39卷(第6期);全文 |
Also Published As
Publication number | Publication date |
---|---|
CN106570561A (en) | 2017-04-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107169628B (en) | Power distribution network reliability assessment method based on big data mutual information attribute reduction | |
CN106017551A (en) | Intelligent transmission line integrated monitoring analysis and early warning method | |
Wang et al. | Time‐varying failure rate simulation model of transmission lines and its application in power system risk assessment considering seasonal alternating meteorological disasters | |
CN104318347A (en) | Power transmission line icing state assessment method based on information fusion of multiple sensors | |
CN113327022B (en) | Lightning protection safety risk management system and method | |
CN108182514A (en) | A kind of power grid icing waves Risk Forecast Method, system and storage medium | |
CN106599417A (en) | Method for identifying urban power grid feeder load based on artificial neural network | |
CN106485333A (en) | A kind of transmission line of electricity running status appraisal procedure | |
CN105719094A (en) | State evaluation method of power transmission equipment | |
CN114896872A (en) | Comprehensive evaluation method for icing state of high-voltage transmission line | |
CN113988655A (en) | Power transmission line running state evaluation method considering multiple meteorological factors | |
CN106815635A (en) | A kind of forecasting system and method for insulator surface equivalent salt deposit density | |
CN104502410A (en) | Prediction method for insulator equivalent salt deposit density and non-soluble deposit density by least squares support vector machine and genetic algorithm | |
CN106570582A (en) | Method and system for building transmission line dancing tripping risk prediction network model | |
CN115759445A (en) | Machine learning and cloud model-based classified flood random forecasting method | |
CN114357670A (en) | Power distribution network power consumption data abnormity early warning method based on BLS and self-encoder | |
CN102175202A (en) | Ice cover thickness predicting method based on fuzzy logic | |
CN114912355A (en) | Method and device for predicting short-term icing of power transmission line and storage medium | |
CN104732107B (en) | Using medium parameter as the residual life of sleeve of transformer Forecasting Methodology for assessing parameter | |
CN111222687A (en) | Power load probability prediction method and device based on heterogeneous neural network | |
CN106570561B (en) | A kind of insoluble sediment density forecasting system of insulator surface and method | |
CN107688862A (en) | Insulator equivalent salt density accumulation rate Forecasting Methodology based on BA GRNN | |
CN106650060A (en) | Prediction method of internal resistance attenuation coefficient for photovoltaic cells | |
CN115577854A (en) | Quantile regression wind speed interval prediction method based on EEMD-RBF combination | |
CN115730526A (en) | Intelligent monitoring and predicting method and system for airport electric load |
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 |