CN110414718A - A kind of distribution network reliability index optimization method under deep learning - Google Patents
A kind of distribution network reliability index optimization method under deep learning Download PDFInfo
- Publication number
- CN110414718A CN110414718A CN201910599100.3A CN201910599100A CN110414718A CN 110414718 A CN110414718 A CN 110414718A CN 201910599100 A CN201910599100 A CN 201910599100A CN 110414718 A CN110414718 A CN 110414718A
- Authority
- CN
- China
- Prior art keywords
- distribution network
- deep learning
- reliability index
- sample
- network
- 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.)
- Pending
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
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The present invention relates to a kind of distribution network reliability index optimization methods under deep learning, comprising steps of 1) collecting sample data, construct training sample and test sample；2) deepness belief network structure is determined；3) training sample is inputted into deepness belief network, deepness belief network model is optimized；4) test sample is input to the deepness belief network model after optimization, obtains corresponding distribution network reliability index；5) critical value is determined, the distribution network reliability index that will acquire is compared with existing precise results, if relative error is greater than or equal to critical value, after then adjusting the network number of plies and node number, repeat step 3)~4), if relative error is less than critical value, the optimization that the distribution network reliability under deep learning obtains is completed.Compared with prior art, the present invention has many advantages, such as complete, scientific, reliably obtain the distribution network reliability index under the deep learning of optimization, and guarantees index accuracy, shortening calculating time.
Description
Technical field
The present invention relates to distribution network reliability analysis fields, more particularly, to the distribution network reliability under a kind of deep learning
Index optimization method.
Background technique
Power distribution network is a part indispensable in electric system, the conventional method packet of existing distribution network reliability analysis
Analytic method, simulation etc. are included, is all the calculating distribution network reliability index based on power distribution network component reliability parameter, it is difficult to
Suitable for all types of distribution net work structures, and it cannot be considered in terms of result precision and calculate the time.The advantage of intelligent algorithm exists
Objective law, fuzzy parameter, determination of blind number existing for being analyzed inside event by analogue simulator method etc., establishes input
Relationship between vector and output vector.In recent years, at home and abroad utilizing artificial intelligence under the continuously attempting to of experts and scholars
Method, which analyzes the reliability of distribution network system element and distribution network system, also achieves certain progress, mentions on this basis
The analysis and assessment algorithm such as Fuzzy Reliability and artificial neural network is gone out.Electric network composition is increasingly complicated, data volume is gradually increased, with
Artificial intelligence is also rapidly developing, becomes increasingly popular simultaneously for this, and wherein deep learning algorithm also reaches its maturity.However how to provide one
The electric network reliability index optimization method of kind more accurateization is still the major issue nowadays to be solved.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide under a kind of deep learning
Distribution network reliability index optimization method.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of distribution network reliability index optimization method under deep learning, including the following steps:
S1: collecting sample data construct training sample and test sample.
The data of training sample matrix include topology data, line parameter circuit value, the power distribution network component reliability of power distribution network
Parameter and load class parameter.The line parameter circuit value includes each line length, each circuit types, the distribution mesh element
Dependability parameter includes line failure rate, route average time for repair of breakdowns, transformer fault rate, transformer mean failure rate reparation
Time, circuit breaker failure rate, breaker average time for repair of breakdowns, bus-bar fault rate, bus average time for repair of breakdowns, segmentation
Switch fault rate, block switch fault correction time, the load class parameter include the total number of users of load point.
S2: deepness belief network structure, including the setting network number of plies, node number and original state parameter are determined.
Hidden layer is set four layers by the structure for determining deep learning network, wherein each node in hidden layer maximum value isM is sample distribution network structure structure number, and n is the number of single sample input data；By original state parameter
It is set as minimum, wherein the output feature vector of output layer is set as Q=3, and output feature vector includes that three classes power distribution network is reliable
(System average interruption frequency index, system System average interruption frequency refer to property index S AIFI
Mark), SAIDI (System average interruption duration index, system System average interruption duration),
ASAI (Average service availability index, averagely power supply Availability Index).Original state parameter includes
Cycle of training, learning rate, activation primitive, input layer number, output layer number of nodes, the bias vector of visible layer, hidden layer
Bias vector and weight.
S3: inputting deepness belief network for training sample, optimizes to deepness belief network model；
It is input in deepness belief network after obtaining incidence matrix to training sample, gradually trains in deepness belief network
The bias vector b of all RBM, weight matrix W, the bias vector a of visible layer and hidden layer, the update principle expression formula of parameter
Are as follows:
Wherein, ε is the learning rate of gradient descent method,<>reconFor the probability of visible layer or hidden layer point after once reconstructing
Cloth, v are that visible layer observes data, and h is implicit layer data；
The parameter that training obtains is input in BP network and carries out Reverse optimization training, makes weight matrix W and bias vector
A, b parameter updates again, obtains complete depth belief network model.
Formula is chosen in the bias vector a initialization of visible layer are as follows:
In formula, piThe ratio of total sample number shared by the sample for being 1 for ith feature value；
The bias vector b initialization of hidden layer takes 0；
Each numerical value is initialized as the random number of normal distribution N (0,0.1) in weight matrix W.
Preferably, in the training process, cycle of training be each layer RBM iteration 30 times, entire deep learning network iteration
5000 times, and ReLU function is chosen as activation primitive.
S4: test sample is input to the deepness belief network model after optimization, corresponding distribution network reliability is obtained and refers to
Mark；
S5: determining critical value, and the distribution network reliability index that step S4 is obtained is compared with existing precise results, if
Relative error is greater than or equal to critical value, then after adjusting the network number of plies and node number, step S3~S4 is repeated, if relative error
Less than critical value, then the optimization that the distribution network reliability under deep learning obtains is completed.Have precise results by commercial reliability
Software for calculation CYME, which is calculated, to be obtained.
Compared with prior art, the selection distribution network reliability influence factor of the method for the present invention maximum magnitude, by it is each because
The numerical value input deep learning network of element is trained, and the accuracy according to gained distribution network reliability index, to depth
It practises network structure to be adjusted, to obtain the deep learning network for being most suitable for distribution network reliability analysis；The method of the present invention energy
Distribution network reliability index under enough complete, science, the reliable deep learning for obtaining optimization, and this method can guarantee power distribution network
The accuracy of reliability index is applicable to a variety of distribution network structure structures, while the deep learning network after training can
Shorten the calculating time of mass data.
Detailed description of the invention
Fig. 1 is the flow diagram of the method for the present invention；
Fig. 2 is power distribution network network topological diagram；
Fig. 3 is that ten sample ASAI indexs are opposite compared to existing accurate result in test matrix in the embodiment of the present invention
Error line chart.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.Obviously, described embodiment is this
A part of the embodiment of invention, rather than whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art exist
Every other embodiment obtained under the premise of creative work is not made, all should belong to the scope of protection of the invention.
Embodiment
The present invention can restore after selecting breaker, bus, overhead transmission line, cable run, block switch etc. to repair
Reliability, network topology structure and network parameter of power distribution network component to normal operating conditions etc. are used as deep learning net
Network input.
Deep learning network includes deepness belief network, convolutional neural networks and Recognition with Recurrent Neural Network etc. at present.It is wherein deep
It is more flexible to spend belief network (Deep Belief Network, DBN) model, has merged unsupervised learning and supervised learning
Advantage, can from great amount of samples learning data set substantive characteristics, with other methods compatibility preferably, therefore the present invention
The depth conviction net formed is stacked using multiple limited Boltzmann machines (Restricted Boltzmann Machines, RBM)
Network model is for obtaining more optimal distribution network reliability index.Distribution network reliability based on deepness belief network refers to
Mark is obtained and is realized by TensorFlow deep learning frame.
The present invention relates to a kind of distribution network reliability index optimization methods under deep learning, as shown in Figure 1, specifically including
Following steps:
Step 1: collecting sample data or reference distribution network reliability sample database, construct training sample matrix and survey
Try sample matrix.
Sample is m * n matrix, and wherein m is sample distribution network structure structure number, and n is the number of state of electric distribution network parameter.
Wherein, have four major class distribution network datas as input: 1, the topological structure 2 of power distribution network, line parameter circuit value are (each line length, each
Circuit types) 3, power distribution network component reliability parameter (line failure rate, route average time for repair of breakdowns, transformer fault
Rate, transformer average time for repair of breakdowns, circuit breaker failure rate, breaker average time for repair of breakdowns, bus-bar fault rate, bus
Average time for repair of breakdowns, block switch failure rate, block switch fault correction time) 4, load class parameter (always use by load point
Amount).
By three classes distribution network reliability index S AIFI (System average interruption frequency
Index, system System average interruption frequency index), SAIDI (System average interruption duration index,
System System average interruption duration), ASAI (Average service availability index, averagely power supply availability
Index) output characteristic vector (vector element number Q=3) as deep learning network.
Step 2: by the topological structure of power distribution network as input by way of being converted into incidence matrix.
When, there are k node, incidence matrix can be then the rectangular of a k*k in a power distribution network network topology structure
Formula can determine the element value of each single item in matrix according to following formula, wherein if only one section of connection between node i and j,
And route does not include other nodes, element aijIt is indicated with 1, is otherwise just indicated with 0.
Power distribution network can be big according to power load distributing quantity, density variation, geographical conditions situation, position of source, voltage class
The structures such as situations such as small difference forms different structures, and foundation structure is for example tree-shaped, hand in hand, multi-joint network is segmented more.Such as
Shown in Fig. 2, the power distribution network network topology model of ten node trees is possessed for one, incidence matrix is a 10*10
Matrix.When the point in incidence matrix on diagonal line represents same node, element value should be set as 1, i.e., diagonal line element in matrix
Element is that 1, Fig. 2 interior joint 1 is only associated with node 2, so a21With a12Value is all set to 1, other onrelevant elements are all set to 0,
And so on, incidence matrix A corresponding to available Fig. 2 are as follows:
In power distribution network topological structure matrix, the size of numerical value represents the distance between associated nodes.Except opening up for power distribution network
It flutters outside structure, other three categories state of electric distribution network parameters are all in a manner of numerical value as the input of deep learning network.
Step 3: determining the structure of deepness belief network.
The network number of plies and node number are set, wherein each node layer number maximum value isIt is determined by following principle.
A. structure as compact as possible is taken under the premise of meeting required precision, that is, exhausts the hidden layer node that may lack
Number.Studies have shown that node in hidden layer is not only related with Shu Ru, the number of nodes of output layer, the complicated journey with problem to be solved
The factors such as the characteristic of degree, the pattern of transfer function and sample data are also related.
B. node in hidden layer is necessarily less than N-1 (wherein N is number of training), otherwise, the systematic error of network model
Unrelated with the characteristic of training sample and go to zero, that is, the network model established does not have generalization ability, without any practical value yet.
Therefore the number of nodes of input layer (variable number) is necessarily less than N-1.
C. number of training must be more than the connection flexible strategy of network model, and generally 2~10 times, otherwise, sample must divide
It is likely to obtain reliable neural network model at several parts and using the method for " trained in turn ".
Original state parameter is set as minimum, including cycle of training, learning rate, activation primitive, input layer number, output
Node layer number, biasing and weight, in which: input layer number is taken as 209, and output layer neuron number is taken as 3；Training week
Phase be each layer RBM iteration 30 times, entire deep learning network iteration 5000 times；RBM learning rate takes 0.0001, entire depth
It practises e-learning rate and takes 0.005, activation primitive chooses ReLU function.Relative to Sigmoid and Tanh activation primitive, ReLU is not only
Gradient decline, backpropagation can be made in deep learning network training process more efficient, gradient explosion and gradient is avoided to disappear
Problem, and calculating process can be simplified, without the influence of such as exponential function in other complicated activation primitives, while liveness
Dispersibility make neural network overall calculation complexity reduce, the time reduce.ReLU function is shown below:
F (x)=max (0, x)
The bias vector a initialization of visible layer takes:
Wherein, piIt is expressed as the ratio of total sample number shared by the sample that ith feature value is 1, hidden layer is biased towards
Amount b initialization takes 0；Each numerical value is initialized as the random number of normal distribution N (0,0.1) in weight matrix W.
Step 4: network model parameter optimization.
Training sample is inputted in deepness belief network, all RBM in deepness belief network, weight matrix are gradually trained
W and the update of bias vector a, b parameter are shown below:
Wherein, ε is the learning rate of gradient descent method,<>reconFor the probability of visible layer or hidden layer point after once reconstructing
Cloth, v are that visible layer observes data, and h is implicit layer data, is used for feature extraction.
The parameter that training obtains is input in BP network and carries out Reverse optimization training, makes weight matrix W and bias vector
A, b parameter updates again, obtains complete depth belief network model.
Step 5: input test matrix, obtains corresponding distribution network reliability index.
The reliability index of output is respectively SAIFI, SAIDI, ASAI, is presented with the matrix form of m × 3, by visual
Change process is compared result and existing precise results in the form of line chart, the error of observation.Have precise results by commercialization
Calculation of Reliability software CYME, which is calculated, to be obtained.
Step 6: determining critical value μ.
According to pertinent literature and experience, critical value μ takes 0.03, by gained output distribution net reliability index relative error with
μ is compared: being adjusted the network number of plies and node number if Error Absolute Value is greater than or equal to μ, is repeated Step 4: five；If small
In μ, then the distribution network reliability optimization under deep learning is completed.
Since test matrix sample size is more, ten samples are taken at random in test matrix, its ASAI index is compared
In existing accurate result, by visualization process, its relative error is shown in the form of line chart, as shown in figure 3, maximum value is
0.05%, it is much smaller than allowable error range 3%, meets result requirement.
The present embodiment is illustrated the method for the present invention by taking a power distribution network sample as an example.
Choose four major class totally ten four kinds of input datas, respectively each line length, each circuit grade, line fault
Rate, route average time for repair of breakdowns, transformer fault rate, transformer average time for repair of breakdowns, circuit breaker failure rate, open circuit
Device average time for repair of breakdowns, bus-bar fault rate, bus average time for repair of breakdowns, block switch failure rate, block switch event
Hinder repair time, the total number of users of load point, power distribution network topological structure.Wherein, a certain typical 10 joint net frame structure of selection
It is line failure rate in case, route average time for repair of breakdowns, transformer fault rate, transformer average time for repair of breakdowns, disconnected
Road device failure rate, breaker average time for repair of breakdowns, bus-bar fault rate, bus average time for repair of breakdowns, block switch event
Barrier rate, block switch fault correction time, the data of the total number of users of load point be respectively 1.97,9,0,200,0.01,4,
0.13,10.48,0.06,2,1232}。
For the distribution network structure structure of 10 nodes, the topology knot of the power distribution network indicated with 10 × 10 matrix B is obtained
Structure is shown below:
The sample shares 209 input feature vectors, and first layer visible layer has 209 neurons corresponding, therefore input layer is refreshing
209 are taken as through first number；Output layer neuron number is taken as 3；Every layer RBM iteration 30 times in training process, entire DBN iteration
5000 times；RBM learning rate takes 0.0001, and entire DBN learning rate takes 0.005；Activation primitive chooses ReLU function；Visible layer it is inclined
Vector a initialization is set to take:
Pass through 4 layers of hidden layer h1, h2, h3, h4, (being corresponding in turn to 600,1000,1500,2000 neurons) reaches output
3 neurons of layer, the numerical value of three classes distribution network reliability output-index SAIFI, SAIDI, ASAI are 0.953,8.249,
0.99906.The distribution network reliability optimization being finally completed under deep learning.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
The staff for being familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right
It is required that protection scope subject to.
Claims (10)
1. a kind of distribution network reliability index optimization method under deep learning, which is characterized in that this method includes the following steps:
1) collecting sample data construct training sample and test sample；
2) deepness belief network structure, including the setting network number of plies, node number and original state parameter are determined；
3) training sample is inputted into deepness belief network, deepness belief network model is optimized；
4) test sample is input to the deepness belief network model after optimization, obtains corresponding distribution network reliability index；
5) it determines critical value, the distribution network reliability index that step 4) obtains is compared with existing precise results, if relatively
Error is greater than or equal to critical value, then after adjusting the network number of plies and node number, repeats step 3)~4), if relative error is less than
Critical value then completes the optimization that the distribution network reliability under deep learning obtains.
2. the distribution network reliability index optimization method under a kind of deep learning according to claim 1, which is characterized in that
The data of training sample matrix include the topology data of power distribution network, line parameter circuit value, power distribution network component reliability parameter and negative
Lotus class parameter.
3. the distribution network reliability index optimization method under a kind of deep learning according to claim 2, which is characterized in that
The line parameter circuit value includes each line length, each circuit types, and the power distribution network component reliability parameter includes line
Road failure rate, route average time for repair of breakdowns, transformer fault rate, transformer average time for repair of breakdowns, circuit breaker failure
Rate, breaker average time for repair of breakdowns, bus-bar fault rate, bus average time for repair of breakdowns, block switch failure rate, segmentation
Switch fault repair time, the load class parameter include the total number of users of load point.
4. the distribution network reliability index optimization method under a kind of deep learning according to claim 3, which is characterized in that
The particular content of step 2) are as follows:
Hidden layer is set four layers by the structure for determining deep learning network, wherein each node in hidden layer maximum value is
M is sample distribution network structure structure number, and n is the number of single sample input data；Original state parameter is set as minimum,
Wherein the output feature vector of output layer is set as Q=3, output feature vector include three classes distribution network reliability index S AIFI,
SAIDI、ASAI。
5. the distribution network reliability index optimization method under a kind of deep learning according to claim 4, which is characterized in that
Original state parameter include cycle of training, learning rate, activation primitive, input layer number, output layer number of nodes, visible layer it is inclined
Set the bias vector and weight of vector, hidden layer.
6. the distribution network reliability index optimization method under a kind of deep learning according to claim 5, which is characterized in that
The particular content of step 3) are as follows:
It is input in deepness belief network, gradually trains all in deepness belief network after obtaining incidence matrix to training sample
The bias vector b of RBM, weight matrix W, the bias vector a of visible layer and hidden layer, the update principle expression formula of parameter are as follows:
Wherein, ε is the learning rate of gradient descent method,<>reconFor visible layer or the probability distribution of hidden layer after once reconstructing, v
Data are observed for visible layer, h is implicit layer data；
The parameter that training obtains is input in BP network and carries out Reverse optimization training, joins weight matrix W and bias vector a, b
Number updates again, obtains complete depth belief network model.
7. the distribution network reliability index optimization method under a kind of deep learning according to claim 6, which is characterized in that
Formula is chosen in the bias vector a initialization of visible layer are as follows:
In formula, piThe ratio of total sample number shared by the sample for being 1 for ith feature value；
The bias vector b initialization of hidden layer takes 0；
Each numerical value is initialized as the random number of normal distribution N (0,0.1) in weight matrix W.
8. the distribution network reliability index optimization method under a kind of deep learning according to claim 6, which is characterized in that
Step 3) in the training process, cycle of training be each layer RBM iteration 30 times, entire deep learning network iteration 5000 times.
9. the distribution network reliability index optimization method under a kind of deep learning according to claim 6, which is characterized in that
ReLU function is chosen as activation primitive.
10. the distribution network reliability index optimization method under a kind of deep learning according to claim 1, feature exist
In in step 5), existing precise results are calculated by commercial reliability software for calculation CYME to be obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910599100.3A CN110414718A (en) | 2019-07-04 | 2019-07-04 | A kind of distribution network reliability index optimization method under deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910599100.3A CN110414718A (en) | 2019-07-04 | 2019-07-04 | A kind of distribution network reliability index optimization method under deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110414718A true CN110414718A (en) | 2019-11-05 |
Family
ID=68360345
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910599100.3A Pending CN110414718A (en) | 2019-07-04 | 2019-07-04 | A kind of distribution network reliability index optimization method under deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110414718A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111083151A (en) * | 2019-12-23 | 2020-04-28 | 深圳供电局有限公司 | Attack identification method based on deep belief network and wind power management system |
| CN111177101A (en) * | 2019-12-18 | 2020-05-19 | 广西电网有限责任公司电力科学研究院 | Power distribution network multidimensional visualization platform based on big data architecture |
| CN111881628A (en) * | 2020-08-06 | 2020-11-03 | 云南电网有限责任公司电力科学研究院 | Method and system for optimizing electrical parameters of square wave voltage source and storage medium |
| CN112053032A (en) * | 2020-08-05 | 2020-12-08 | 上海工程技术大学 | Micro-grid reliability evaluation method based on load change |
| CN112327101A (en) * | 2020-10-30 | 2021-02-05 | 国网上海市电力公司 | Power distribution network reliability detection method and system based on long-time and short-time memory neural network |
| CN112651183A (en) * | 2021-01-19 | 2021-04-13 | 广西大学 | Reliability evaluation method for quantum distributed countermeasure unified deep hash network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794534A (en) * | 2015-04-16 | 2015-07-22 | 国网山东省电力公司临沂供电公司 | Power grid security situation predicting method based on improved deep learning model |
| CN107506590A (en) * | 2017-08-26 | 2017-12-22 | 郑州大学 | A kind of angiocardiopathy forecast model based on improvement depth belief network |
| WO2018076475A1 (en) * | 2016-10-26 | 2018-05-03 | 广东产品质量监督检验研究院 | Photovoltaic assembly accelerated degradation model established based on deep approach of learning, and method for predicting photovoltaic assembly lifetime |
| CN108694479A (en) * | 2018-07-11 | 2018-10-23 | 南京邮电大学 | Consider the distribution network reliability prediction technique that weather influences time between overhaul |
| CN109214416A (en) * | 2018-07-23 | 2019-01-15 | 华南理工大学 | A kind of multidimensional information fusion Diagnosis Method of Transformer Faults based on deep learning |
| WO2019015461A1 (en) * | 2017-07-18 | 2019-01-24 | 中国银联股份有限公司 | Risk identification method and system based on transfer deep learning |
-
2019
- 2019-07-04 CN CN201910599100.3A patent/CN110414718A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104794534A (en) * | 2015-04-16 | 2015-07-22 | 国网山东省电力公司临沂供电公司 | Power grid security situation predicting method based on improved deep learning model |
| WO2018076475A1 (en) * | 2016-10-26 | 2018-05-03 | 广东产品质量监督检验研究院 | Photovoltaic assembly accelerated degradation model established based on deep approach of learning, and method for predicting photovoltaic assembly lifetime |
| WO2019015461A1 (en) * | 2017-07-18 | 2019-01-24 | 中国银联股份有限公司 | Risk identification method and system based on transfer deep learning |
| CN107506590A (en) * | 2017-08-26 | 2017-12-22 | 郑州大学 | A kind of angiocardiopathy forecast model based on improvement depth belief network |
| CN108694479A (en) * | 2018-07-11 | 2018-10-23 | 南京邮电大学 | Consider the distribution network reliability prediction technique that weather influences time between overhaul |
| CN109214416A (en) * | 2018-07-23 | 2019-01-15 | 华南理工大学 | A kind of multidimensional information fusion Diagnosis Method of Transformer Faults based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| KINGZONE_2008: "神经网络中隐层数和隐层节点数问题的讨论", Retrieved from the Internet <URL:https://blog.csdn.net/kingzone_2008/article/details/81291507> * |
| 孔祥玉: "基于深度信念网络的短期负荷预测方法", 《电力系统自动化》 * |
| 孔祥玉: "基于深度信念网络的短期负荷预测方法", 《电力系统自动化》, 10 March 2018 (2018-03-10), pages 133 - 139 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111177101A (en) * | 2019-12-18 | 2020-05-19 | 广西电网有限责任公司电力科学研究院 | Power distribution network multidimensional visualization platform based on big data architecture |
| CN111083151A (en) * | 2019-12-23 | 2020-04-28 | 深圳供电局有限公司 | Attack identification method based on deep belief network and wind power management system |
| CN111083151B (en) * | 2019-12-23 | 2021-05-25 | 深圳供电局有限公司 | Attack identification method based on deep belief network and wind power management system |
| CN112053032A (en) * | 2020-08-05 | 2020-12-08 | 上海工程技术大学 | Micro-grid reliability evaluation method based on load change |
| CN111881628A (en) * | 2020-08-06 | 2020-11-03 | 云南电网有限责任公司电力科学研究院 | Method and system for optimizing electrical parameters of square wave voltage source and storage medium |
| CN111881628B (en) * | 2020-08-06 | 2023-01-20 | 云南电网有限责任公司电力科学研究院 | Method and system for optimizing electrical parameters of square wave voltage source and storage medium |
| CN112327101A (en) * | 2020-10-30 | 2021-02-05 | 国网上海市电力公司 | Power distribution network reliability detection method and system based on long-time and short-time memory neural network |
| CN112651183A (en) * | 2021-01-19 | 2021-04-13 | 广西大学 | Reliability evaluation method for quantum distributed countermeasure unified deep hash network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110414718A (en) | A kind of distribution network reliability index optimization method under deep learning | |
| Song et al. | New chaotic PSO-based neural network predictive control for nonlinear process | |
| CN104636801A (en) | Transmission line audible noise prediction method based on BP neural network optimization | |
| CN107506590A (en) | A kind of angiocardiopathy forecast model based on improvement depth belief network | |
| CN112382082A (en) | Method and system for predicting traffic running state in congested area | |
| CN110223234A (en) | Depth residual error network image super resolution ratio reconstruction method based on cascade shrinkage expansion | |
| CN109490814A (en) | Metering automation terminal fault diagnostic method based on deep learning and Support Vector data description | |
| CN113780002A (en) | Knowledge reasoning method and device based on graph representation learning and deep reinforcement learning | |
| CN109255469A (en) | Merge the Forecasting Flood method of stack self-encoding encoder and support vector regression | |
| CN114912357A (en) | Multi-task reinforcement learning user operation method and system based on user model learning | |
| Fan et al. | Neighborhood centroid opposite-based learning Harris Hawks optimization for training neural networks | |
| CN116667369B (en) | Distributed photovoltaic voltage control method based on graph convolution neural network | |
| CN112287120A (en) | Knowledge inference method of dual-system knowledge graph based on reinforcement learning | |
| CN116432359A (en) | Variable topology network tide calculation method based on meta transfer learning | |
| Kottas et al. | A new method for weight updating in Fuzzy cognitive Maps using system Feedback | |
| Papageorgiou et al. | A weight adaptation method for fuzzy cognitive maps to a process control problem | |
| CN112861332B (en) | Cluster dynamics prediction method based on graph network | |
| Bideh et al. | Automatic fault diagnosis of computer networks based on a combination BP neural network and fuzzy logic | |
| CN108898227A (en) | Learning rate calculation method and device, disaggregated model calculation method and device | |
| CN114254828A (en) | Power load prediction method based on hybrid convolution feature extractor and GRU | |
| CN114662989A (en) | Transient stability self-adaptive evaluation method and device for power system | |
| CN114372418A (en) | Wind power space-time situation description model establishing method | |
| Chun-Mei | Using fuzzy cognitive map for system control | |
| Abraham et al. | An intelligent forex monitoring system | |
| JP4267726B2 (en) | Device for determining relationship between operation signal and operation amount in control device, control device, data generation device, input / output characteristic determination device, and correlation evaluation device |
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 |