CN104463323A - Data prediction method and apparatus - Google Patents
Data prediction method and apparatus Download PDFInfo
- Publication number
- CN104463323A CN104463323A CN201310424568.1A CN201310424568A CN104463323A CN 104463323 A CN104463323 A CN 104463323A CN 201310424568 A CN201310424568 A CN 201310424568A CN 104463323 A CN104463323 A CN 104463323A
- Authority
- CN
- China
- Prior art keywords
- input
- data
- dynamic pond
- matrix
- pond
- 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
Landscapes
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention relates to the field of computer networks, and provides a data prediction method and apparatus, for improving the speed and precision of data prediction. The method comprises: introducing adjacent feedback into a dynamic pool nerve cell in an echo state network, constructing an annular adjacent feedback dynamic pool, training the echo state network, and afterwards, performing the data prediction by use of the echo state network. The data prediction method and apparatus provided by the embodiments of the invention are applied to the data prediction by use of the echo state network.
Description
Technical field
The present invention relates to computer network field, particularly relate to a kind of data predication method and device.
Background technology
Network traffics, especially video flow, have very strong self-similarity and sudden.Predicting network flow plays vital effect in Strengthens network management and control, raising network resource utilization, strick precaution Large-scale automatic attack and network failure etc., by the prediction to network traffics, the variation tendency of future network flow can be obtained, realize congestion control to network, and provide effective foundation for the reasonable distribution of Internet resources.Meanwhile, potential Cyberthreat, fault etc. can also be detected, realize intrusion detection and the fault management of network.
In the prior art, traditional recurrent neural network is a kind of desirable Forecast of Nonlinear Time Series instrument, adopts error gradient descent method to train this network, has now been widely used in the prediction of Internet video flow.
State in realization in the process of video traffic prediction, inventor finds that in prior art, at least there are the following problems:
The amendment of error gradient descent method to the connection weight of recurrent neural network progressively realizes, and may occur that gradient information is degenerated, thus cannot ensure convergence, thus cannot ensure precision of prediction;
The right value update process computation amount of training algorithm is very large.Under normal circumstances, because needs circulation upgrades some weights, make the training need time of network too much, cause the speed of prediction excessively slow, cause gradient descent algorithm generally can only be applied to the training of the smaller recurrent neural network of scale.
Summary of the invention
Embodiments of the invention provide a kind of video traffic prediction method and apparatus, can improve predetermined speed and the precision of prediction of data.
For achieving the above object, embodiments of the invention adopt following technical scheme:
First aspect, provides a kind of data predication method, and described method comprises:
Obtain input data;
Described input data are input in echo state network, wherein, dynamic pond neuron in described echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, described dynamic pond nervus opticus unit is to described dynamic pond peripheral sensory neuron feedback data, and described dynamic pond peripheral sensory neuron is adjacent with described dynamic pond nervus opticus unit;
Predicted data is obtained by described echo state network.
Second aspect, provide a kind of data prediction device, described device comprises:
Input data capture unit, for obtaining input data;
Predicting unit, for after described input data capture unit obtains described input data, described input data are input in echo state network, wherein, dynamic pond neuron in described echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, nervus opticus unit in described dynamic pond is to described dynamic pond peripheral sensory neuron feedback data, and described dynamic pond peripheral sensory neuron is adjacent with described dynamic pond nervus opticus unit;
Predicted data is obtained by described echo state network.
The embodiment of the present invention provides a kind of data predication method and device, first training data is utilized to train echo state network, dynamic pond neuron in described echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, described dynamic pond nervus opticus unit is to described dynamic pond peripheral sensory neuron feedback data, described dynamic pond peripheral sensory neuron is adjacent with described dynamic pond nervus opticus unit, then input data are input in the echo state network after training, thus acquisition predicted data, data prediction speed and precision of prediction can be improved.
Accompanying drawing explanation
In order to be illustrated more clearly in the embodiment of the present invention or technical scheme of the prior art, be briefly described to the accompanying drawing used required in embodiment or description of the prior art below, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.
A kind of data predication method schematic flow sheet one that Fig. 1 provides for the embodiment of the present invention;
A kind of data predication method schematic flow sheet two that Fig. 2 provides for the embodiment of the present invention;
Fig. 3 is the structural representation of a kind of echo state network provided by the invention;
Fig. 4 is the structural representation one of a kind of data prediction device provided by the invention;
Fig. 5 is the structural representation two of a kind of data prediction device provided by the invention;
Fig. 6 is the structural representation three of a kind of data prediction device provided by the invention;
Fig. 7 is the structural representation four of a kind of data prediction device provided by the invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is only the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.
The embodiment of the present invention provides a kind of data predication method, and as shown in Figure 1, described method comprises:
S101, acquisition input data.
S102, input data are input in echo state network, wherein, dynamic pond neuron in echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, dynamic pond nervus opticus unit is to dynamic pond peripheral sensory neuron feedback data, and dynamic pond peripheral sensory neuron is adjacent with dynamic pond nervus opticus unit.
S103, obtain predicted data by echo state network.
The embodiment of the present invention provides a kind of data predication method, first training data is utilized to train echo state network, dynamic pond neuron in echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, dynamic pond nervus opticus unit is to dynamic pond peripheral sensory neuron feedback data, dynamic pond peripheral sensory neuron is adjacent with dynamic pond nervus opticus unit, then input data are input in the echo state network after training, thus acquisition predicted data, data prediction speed and precision of prediction can be improved.
The embodiment of the present invention provides a kind of data predication method, and as shown in Figure 2, method comprises:
S201, set up novel echo state network forecast model.
Concrete, realize by following methods:
This echo state network is divided into three layers, and as shown in Figure 3, i.e. input layer, dynamically pond and output layer, input layer is by K input block u (t)=(u
1(t), u
2(t) ...., u
k(t))
tcomposition, dynamic pond is by N number of dynamic pond neuron x (t)=(x
1(t), x
2(t) ..., x
n(t))
tcomposition, output layer is by L output unit y (t)=(y
1(t), y
2(t) ..., y
l(t))
tcomposition;
At input layer and the weight matrix W that dynamically connects between pond
in, connect between dynamic pond intrinsic nerve unit weight matrix
w, connect between dynamic pond and output layer weight matrix W
out;
This model has a dynamic pool structure determined, neuron is wherein connected in an annular manner, and each neuron is connected by the adjacent mode fed back with the neuron before it.The internal weights matrix W in the dynamic pond of loop feedback built has following symmetrical feature:
For arbitrary i=1,2 ..., N, sub-diagonal line W
i, i+1=r, W
i+1, i=r, upper right corner element W
1, N=r, lower left corner element W
n, 1=r, wherein r ∈ (0,1).
Exemplary, as the scale N=5 in dynamic pond, its internal weights matrix W can be expressed as:
In t, each neuron in dynamic pond or unit have a foment, and dynamically the foment of pool unit is upgraded by following state updating equation:
(1-1) x(t+1)=f
in(W
inu(t+1)+Wx(t))
Here, W
inrepresent that N × K inputs weight matrix and N × N dynamic pond internal weights matrix respectively with W, f
inrepresent the neuronic excitation function in dynamic pond (being generally two tans or other S type functions, also may be identity function);
According to above-mentioned dynamic pond foment x (t), the output of this model can be calculated by following equation:
(1-2) y(n+1)=W
outx(n+1)
Here, W
outrepresent that L × N exports weight matrix.
By increasing adjacent feedback between adjacent neurons in the topology of most simple annular dynamic pond, add the degree of rarefication (degree of rarefication is the important indicator affecting echo state network None-linear approximation ability) in dynamic pond to a certain extent, therefore this model has better estimated performance.
In addition, input weight matrix W is set
inin all nonzero element absolute values identical, all equal the non-zero weights r ∈ (0,1) in the weight matrix of dynamic pond, the mode that its weights symbol is mapped by random device or Logistic produces; Significantly, we only need adjustment free parameter r, greatly simplifie traditional echo state network.
S202, train this novel echo state network model.
Concrete, by realizing with under type:
Determine the input and output of echo state network, namely according to predicted data feature and the demand of prediction, build the length of sample and the step number of prediction, thus obtain the training sample of inputoutput pair;
Initialization is carried out to this echo state network, namely the parameter in dynamic pond is set, comprises: the dynamically scale in pond, weight matrix, degree of rarefication, the dimensional variation factor etc.
Wherein produce the input weight matrix and dynamic pond internal weights matrix determined, and no longer change at whole echo state network training process; The method that input weights symbol is mapped by random method or Logistic produces; Internal weights Spectral radius radius is set between 0 to 1 to ensure the echo status attribute of echo state network.
Output weight matrix is trained, namely inputs training sample, and recording the network state of this echo state network sometime, till knowing that the training time terminates or training objective reaches.
Exemplary, this echo state network is utilized to predict video flow data, after this echo state network forecast model has been set up, determine the input and output of ESN, namely according to the feature of video flow data and the demand of prediction, build the length L of sample and the step number h of prediction, thus obtain the training sample of inputoutput pair, can be expressed as:
(1-3) {u
train(k),y
train(k+h),k=1,2,...,T}；
Initialization is carried out to this echo state network, namely the parameter in dynamic pond is set, comprises: the dynamically scale N in pond, weight matrix, degree of rarefication SD, the dimensional variation factor etc.Wherein produce the input weight matrix W determined
inwith dynamic pond internal weights matrix W, and no longer change at whole echo state network training process; The method that input weights symbol is mapped by random method or Logistic produces; The spectral radius arranging W meets 0< λ <1 to ensure the echo status attribute of echo state network;
To output weight matrix W
outtrain, i.e. the training sample data u of input video flow
traint (), at t sometime
0collect the network state x (t) (because the network state before t0 is relevant to initial value) of this echo state network, until moment a terminates, according to formula (1-2) and (1-3), then the dynamic pond state of this echo state network is:
(1-4) X={x
train(t
0),x
train(t
0+1),...,x
train(a)}
Actual output is:
(1-5)
And the output expected is:
(1-6) Y={y
train(t
0+h),y
train(t
0+h+1),...,y
train(a+h)}
Therefore, the target of training can be expressed as:
(1-7)
Finally, by the method for ridge regression, can calculate and export weight matrix W
out, complete the training process of this model, that is:
(1-8) W
out=(X
TX+λ
2I)
-1X
TY
Wherein, I represents identity matrix, and λ is penalty factor.
S203, utilize through training after this echo state network data are predicted.
Concrete, realize by following methods:
The input data of needs prediction are input in this echo state network after training, can obtain inputting predicted data corresponding to data by formula (1-1) and formula (1-2).
Exemplary, utilize this echo state network to predict above-mentioned video flow data, input video flow data u (t), wherein t=a+1, a+2 ...., according to formula (1-1) and (1-2), the predicted value of video flow can be obtained.
The embodiment of the present invention provides a kind of data predication method, first training data is utilized to train echo state network, dynamic pond neuron in echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, dynamic pond nervus opticus unit is to dynamic pond peripheral sensory neuron feedback data, dynamic pond peripheral sensory neuron is adjacent with dynamic pond nervus opticus unit, then input data are input in the echo state network after training, thus acquisition predicted data, data prediction speed and precision of prediction can be improved.
The embodiment of the present invention provides a kind of data prediction device 01, and as shown in Figure 4, device 01 comprises:
Input data capture unit 011, for obtaining input data;
Predicting unit 012, for after input data capture unit obtains input data, input data are input in echo state network, wherein, dynamic pond neuron in echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, nervus opticus unit in dynamic pond is to dynamic pond peripheral sensory neuron feedback data, and dynamic pond peripheral sensory neuron is adjacent with dynamic pond nervus opticus unit;
Predicted data is obtained by echo state network.
Preferably, echo state network comprises input layer, dynamically pond and output layer, and input layer comprises input block, and dynamic pond comprises dynamic pond neuron, and output layer comprises output unit.
Further, as shown in Figure 5, device 01 also comprises:
Weight matrix sets up unit 013, for setting up the first connection weight value matrix between input layer and dynamic pond, setting up the second connection weight value matrix, set up the 3rd connection weight value matrix between dynamic pond and output layer between the neuron of dynamic pond;
The sub-diagonal line W of the first connection weight value matrix
i, i+1=r, W
i+1, i=r, upper right corner element W
1, N=r, lower left corner element W
n, 1=r, wherein, W represents the first connection weight value matrix, r ∈ (0,1), and i represents the natural number being greater than zero, and N represents the dimension of the first connection weight value matrix.
Preferably, input layer comprises at least one input block, and output layer comprises at least one output unit;
Input layer and output layer support the I/O mode of single-input single-output, single input and multi-output, multiple input single output and multiple-input and multiple-output.
Further, as shown in Figure 6, before input data are input to echo state network, device 01 also comprises:
Matrix setting unit 014, for arranging the element value of the first weight matrix and the second weight matrix, wherein, the absolute value of the nonzero element in the first weight matrix and the second weight matrix is equal;
The symbol of the element value of the first weight matrix is obtained by random device or chaotic maps logistic method;
Training data acquiring unit 015, for obtaining training data;
Training unit 016, for after training data acquiring unit gets training data, utilizes training data to train echo state network.
Further, training data acquiring unit 015 specifically for:
According to the input feature of data and the demand of prediction, build the length of training data and the step number of prediction, obtain training data.
Further, as shown in Figure 7, before utilizing video flow training data training echo state network, device also comprises:
Neural network setting unit 017, for arranging the degree of rarefication in the scale in dynamic pond, the spectral radius of the second connection weight value matrix, the flexible yardstick of input and dynamic pond.
Further, training unit 016 specifically for:
Input training data, echo state network is according to following formula Regeneration dynamics pond neuron state:
x(n+1)＝f(W
inu(n+1)+Wx(n))，
Wherein, x (n) represents dynamic pond neuron, W
inrepresent the first connection weight value matrix, W represents the second connection weight value matrix, and u (n) represents training data, and f represents the neuronic excitation function in dynamic pond, and excitation function comprises two tan, other S type function and identity functions;
According to the method for ridge regression, according to following formula optimization the 3rd connection weight value matrix:
W
out=(X
TX+λ
2I)
-1X
Ty，
Wherein, W
outrepresent the 3rd connection weight value matrix, I represents identity matrix, and λ represents penalty factor, and y represents desired output, and X represents dynamic pond neuron state.
Further, predicting unit 012 specifically for:
Predicted data is obtained by following formula:
Wherein, x (n) is prediction input data,
for prediction exports data, W
outrepresent the 3rd connection weight value matrix.
Further, predicting unit 012 specifically for:
Echo state network is utilized to carry out multi-step prediction to input data.
The embodiment of the present invention provides a kind of data prediction device, first training data is utilized to train echo state network, dynamic pond neuron in echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, dynamic pond nervus opticus unit is to dynamic pond peripheral sensory neuron feedback data, dynamic pond peripheral sensory neuron is adjacent with dynamic pond nervus opticus unit, then input data are input in the echo state network after training, thus acquisition predicted data, data prediction speed and precision of prediction can be improved.
One of ordinary skill in the art will appreciate that: all or part of step realizing said method embodiment can have been come by the hardware that programmed instruction is relevant, aforesaid program can be stored in a computer read/write memory medium, this program, when performing, performs the step comprising said method embodiment; And aforesaid storage medium comprises: ROM, RAM, magnetic disc or CD etc. various can be program code stored medium.
The above; be only the specific embodiment of the present invention, but protection scope of the present invention is not limited thereto, is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; change can be expected easily or replace, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with the protection domain of described claim.
Claims (10)
1. a data predication method, is characterized in that, described method comprises:
Obtain input data;
Described input data are input in echo state network, wherein, dynamic pond neuron in described echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, described dynamic pond nervus opticus unit is to described dynamic pond peripheral sensory neuron feedback data, and described dynamic pond peripheral sensory neuron is adjacent with described dynamic pond nervus opticus unit;
Predicted data is obtained by described echo state network.
2. method according to claim 1, it is characterized in that, described echo state network comprises input layer, described dynamic pond and output layer, and described input layer comprises input block, and described dynamic pond comprises dynamic pond neuron, and described output layer comprises output unit;
Between described input layer and described dynamic pond, set up the first connection weight value matrix, between the neuron of described dynamic pond, set up the second connection weight value matrix, between described dynamic pond and described output layer, set up the 3rd connection weight value matrix;
The sub-diagonal line W of described first connection weight value matrix
i, i+1=r, W
i+1, i=r, upper right corner element W
1, N=r, lower left corner element W
n, 1=r, wherein, W represents described first connection weight value matrix, r ∈ (0,1), and i represents the natural number being greater than zero, and N represents the dimension of described first connection weight value matrix;
Described input layer comprises at least one input block, and described output layer comprises at least one output unit;
Described input layer and described output layer support the I/O mode of single-input single-output, single input and multi-output, multiple input single output and multiple-input and multiple-output.
3. method according to claim 1, it is characterized in that, before described input data are input to echo state network, described method also comprises:
Arrange the element value of described first weight matrix and described second weight matrix, wherein, the absolute value of the nonzero element in described first weight matrix and described second weight matrix is equal;
The symbol of the element value of described first weight matrix is obtained by random device or chaotic maps logistic method;
Obtain training data, described acquisition training data, comprises according to the described feature of input data and the demand of prediction, builds length and the prediction step number of described training data, obtains described training data;
Utilize described training data to train described echo state network, describedly utilize described training data to train described echo state network, comprise the described training data of input, described echo state network upgrades described dynamic pond neuron state according to following formula:
x(n+1)=f(W
inu(n+1)+Wx(n))，
Wherein, x (n) represents described dynamic pond neuron, W
inrepresent described first connection weight value matrix, W represents described second connection weight value matrix, u (n) represents described training data, and f represents the neuronic excitation function in described dynamic pond, and described excitation function comprises two tan, other S type function and identity functions;
According to the method for ridge regression, according to the 3rd connection weight value matrix described in following formula optimization:
W
out=(X
TX+λ
2I)
-1X
Ty，
Wherein, W
outrepresent described 3rd connection weight value matrix, I represents identity matrix, and λ represents penalty factor, and y represents desired output, and X represents described dynamic pond neuron state.
4. method according to claim 3, is characterized in that, utilizing before described training data trains described echo state network, described method also comprises:
The degree of rarefication in the scale in described dynamic pond, the spectral radius of described second connection weight value matrix, the flexible yardstick of input and described dynamic pond is set.
5. method according to claim 1, is characterized in that, describedly obtains predicted data by described echo state network and comprises:
Described predicted data is obtained by following formula:
Wherein, x (n) represents described input data,
represent described predicted data, W
outrepresent described 3rd connection weight value matrix;
Described echo state network is utilized to carry out multi-step prediction to described input data.
6. a video traffic prediction device, is characterized in that, described device comprises:
Input data capture unit, for obtaining input data;
Predicting unit, for after described input data capture unit obtains described input data, described input data are input in echo state network, wherein, dynamic pond neuron in described echo state network is connected formation ring topology successively, and dynamically pond peripheral sensory neuron transmits data to dynamic pond nervus opticus unit, nervus opticus unit in described dynamic pond is to described dynamic pond peripheral sensory neuron feedback data, and described dynamic pond peripheral sensory neuron is adjacent with described dynamic pond nervus opticus unit;
Predicted data is obtained by described echo state network.
7. device according to claim 6, it is characterized in that, described echo state network comprises input layer, described dynamic pond and output layer, and described input layer comprises input block, and described dynamic pond comprises dynamic pond neuron, and described output layer comprises output unit;
Described device also comprises weight matrix and sets up unit, for setting up the first connection weight value matrix between described input layer and described dynamic pond, between the neuron of described dynamic pond, set up the second connection weight value matrix, between described dynamic pond and described output layer, set up the 3rd connection weight value matrix;
The sub-diagonal line W of described first connection weight value matrix
i, i+1=r, W
i+1, i=r, upper right corner element W
1, N=r, lower left corner element W
n, 1=r, wherein, W represents described first connection weight value matrix, r ∈ (0,1), and i represents the natural number being greater than zero, and N represents the dimension of described first connection weight value matrix;
Described input layer comprises at least one input block, and described output layer comprises at least one output unit;
Described input layer and described output layer support the I/O mode of single-input single-output, single input and multi-output, multiple input single output and multiple-input and multiple-output.
8. device according to claim 6, it is characterized in that, before described input data are input to echo state network, described device also comprises:
Matrix setting unit, for arranging the element value of described first weight matrix and described second weight matrix, wherein, the absolute value of the nonzero element in described first weight matrix and described second weight matrix is equal;
The symbol of the element value of described first weight matrix is obtained by random device or logistic method;
Training data acquiring unit, for obtaining training data, comprising according to the described feature of input data and the demand of prediction, building the length of described training data and the step number of prediction, obtaining described training data;
Training unit, for after described training data acquiring unit gets described training data, utilize described training data to train described echo state network, comprise the described training data of input, described echo state network upgrades described dynamic pond neuron state according to following formula:
x(n+1)=f(W
inu(n+1)+Wx(n))，
Wherein, x (n) represents described dynamic pond neuron, W
inrepresent described first connection weight value matrix, W represents described second connection weight value matrix, u (n) represents described training data, and f represents the neuronic excitation function in described dynamic pond, and described excitation function comprises two tan, other S type function and identity functions;
According to the method for ridge regression, according to the 3rd connection weight value matrix described in following formula optimization:
W
out=(X
TX+λ
2I)
-1X
Ty，
Wherein, W
outrepresent described 3rd connection weight value matrix, I represents identity matrix, and λ represents penalty factor, and y represents desired output, and X represents described dynamic pond neuron state.
9. device according to claim 8, is characterized in that, utilizing before described video flow training data trains described echo state network, described device also comprises:
Neural network setting unit, for arranging the degree of rarefication in the scale in described dynamic pond, the spectral radius of described second connection weight value matrix, the flexible yardstick of input and described dynamic pond.
10. device according to claim 6, is characterized in that, described predicting unit specifically for:
Described predicted data is obtained by following formula:
Wherein, x (n) is prediction input data,
for prediction exports data, W
outrepresent described 3rd connection weight value matrix;
Described echo state network is utilized to carry out multi-step prediction to described input data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310424568.1A CN104463323A (en) | 2013-09-17 | 2013-09-17 | Data prediction method and apparatus |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310424568.1A CN104463323A (en) | 2013-09-17 | 2013-09-17 | Data prediction method and apparatus |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104463323A true CN104463323A (en) | 2015-03-25 |
Family
ID=52909328
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310424568.1A Pending CN104463323A (en) | 2013-09-17 | 2013-09-17 | Data prediction method and apparatus |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104463323A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107300388A (en) * | 2017-06-04 | 2017-10-27 | 东南大学 | Tourism route planing method of riding based on Q learning algorithms and echo state network |
| WO2018090580A1 (en) * | 2016-11-17 | 2018-05-24 | 北京智芯微电子科技有限公司 | Method and apparatus for sensing optical access network service stream and computer storage medium |
| CN108540331A (en) * | 2018-04-26 | 2018-09-14 | 南京邮电大学 | Based on the network flow prediction method for improving ESN |
| CN109978238A (en) * | 2019-03-05 | 2019-07-05 | 电子科技大学 | Destination prediction technique based on depth echo state network |
| CN110855331A (en) * | 2019-08-30 | 2020-02-28 | 北京邮电大学 | Machine learning-based proactive antenna inclination angle adjusting mechanism |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101541016A (en) * | 2009-05-06 | 2009-09-23 | 华为技术有限公司 | Method for predicting data and equipment |
| CN101729315A (en) * | 2009-12-24 | 2010-06-09 | 北京邮电大学 | Network flow-predicting method and device based on wavelet package decomposition and fuzzy neural network |
| CN102075352A (en) * | 2010-12-17 | 2011-05-25 | 北京邮电大学 | Method and device for predicting network user behavior |
-
2013
- 2013-09-17 CN CN201310424568.1A patent/CN104463323A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101541016A (en) * | 2009-05-06 | 2009-09-23 | 华为技术有限公司 | Method for predicting data and equipment |
| CN101729315A (en) * | 2009-12-24 | 2010-06-09 | 北京邮电大学 | Network flow-predicting method and device based on wavelet package decomposition and fuzzy neural network |
| CN102075352A (en) * | 2010-12-17 | 2011-05-25 | 北京邮电大学 | Method and device for predicting network user behavior |
Non-Patent Citations (3)
| Title |
|---|
| XIAO-CHUAN SUN ET AL: "Modeling deterministic echo state network with loop reservoir", 《JOURNAL OF ZHEJIANG UNIVERSITY-SCIENCE C (COMPUTERS & ELECTRONICS)》 * |
| 付先平等: "《多媒体技术及应用 第2版》", 30 September 2012, 北京：清华大学出版社 * |
| 韩敏: "《基于储备池的非线性系统预测理论与分析方法》", 31 December 2011, 北京：中国水利水电出版社 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018090580A1 (en) * | 2016-11-17 | 2018-05-24 | 北京智芯微电子科技有限公司 | Method and apparatus for sensing optical access network service stream and computer storage medium |
| CN107300388A (en) * | 2017-06-04 | 2017-10-27 | 东南大学 | Tourism route planing method of riding based on Q learning algorithms and echo state network |
| CN107300388B (en) * | 2017-06-04 | 2020-05-08 | 东南大学 | Riding tour route planning method based on Q-learning algorithm and echo state network |
| CN108540331A (en) * | 2018-04-26 | 2018-09-14 | 南京邮电大学 | Based on the network flow prediction method for improving ESN |
| CN108540331B (en) * | 2018-04-26 | 2021-07-13 | 南京邮电大学 | Network flow prediction method based on improved ESN |
| CN109978238A (en) * | 2019-03-05 | 2019-07-05 | 电子科技大学 | Destination prediction technique based on depth echo state network |
| CN109978238B (en) * | 2019-03-05 | 2022-05-03 | 电子科技大学 | Destination prediction method based on deep echo state network |
| CN110855331A (en) * | 2019-08-30 | 2020-02-28 | 北京邮电大学 | Machine learning-based proactive antenna inclination angle adjusting mechanism |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104463323A (en) | Data prediction method and apparatus | |
| CN102682345B (en) | Based on the traffic flow forecasting method of two variable optimal learning rate Fast Learning neural network | |
| CN112419726B (en) | Urban traffic signal control system based on traffic flow prediction | |
| CN104217258B (en) | A kind of electric load sigma-t Forecasting Methodology | |
| TW201531966A (en) | Configuring sparse neuronal networks | |
| Tayyab et al. | Discharge forecasting by applying artificial neural networks at the Jinsha River Basin, China | |
| CN103530317B (en) | A kind of copy management method of energy consumption self adaptation in cloud storage system | |
| CN106803135A (en) | The Forecasting Methodology and device of a kind of photovoltaic power generation system output power | |
| CN108280998A (en) | Short-time Traffic Flow Forecasting Methods based on historical data dynamic select | |
| Jin et al. | Forewarning of sustainable utilization of regional water resources: a model based on BP neural network and set pair analysis | |
| CN104504442A (en) | Neural network optimization method | |
| Majumder et al. | Application of novel intuitionistic fuzzy BWAHP process for analysing the efficiency of water treatment plant | |
| CN105487376A (en) | Optimal control method based on data-driven single network structure | |
| Dai et al. | Neural network based online traffic signal controller design with reinforcement training | |
| CN104035037A (en) | On-line estimating method for SOH of new energy automobile power battery | |
| CN107145935A (en) | A kind of Smith Prediction Control methods based on modified neutral net | |
| JP2020501252A (en) | Signal processing method and signal processing device based on spike neural network | |
| Yu et al. | Clustered complex echo state networks for traffic forecasting with prior knowledge | |
| Yao et al. | A stability criterion for discrete-time fractional-order echo state network and its application | |
| CN116758767B (en) | Traffic signal lamp control method based on multi-strategy reinforcement learning | |
| CN104915714A (en) | Predication method and device based on echo state network (ESN) | |
| CN104951871A (en) | Intelligent evaluation method and system for power simulation training | |
| Yao et al. | Ensembles of echo state networks for time series prediction | |
| CN102930341B (en) | Optimal training method of collaborative filtering recommendation model | |
| CN109102698B (en) | Method for predicting short-term traffic flow in road network based on integrated LSSVR model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150325 |