CN103209417B - Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net - Google Patents
Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net Download PDFInfo
- Publication number
- CN103209417B CN103209417B CN201310069826.9A CN201310069826A CN103209417B CN 103209417 B CN103209417 B CN 103209417B CN 201310069826 A CN201310069826 A CN 201310069826A CN 103209417 B CN103209417 B CN 103209417B
- Authority
- CN
- China
- Prior art keywords
- neural net
- frequency range
- sequential
- presets
- occupancy state
- 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
Abstract
The invention discloses a kind of Forecasting Methodology and device of the spectrum occupancy state based on neural net.The Forecasting Methodology of the described spectrum occupancy state based on neural net comprises the following steps: step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter; Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer number being less than i; Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range, there is computation complexity low, realize simple advantage.
Description
Technical field
The present invention relates to communication technical field, particularly relate to Forecasting Methodology and the device of spectrum occupancy state.
Background technology
In dynamic frequency spectrum deployment process, Secondary Users need temporarily to be undertaken communicating to reach the object improving the availability of frequency spectrum by the idle frequency spectrum of main users.Be can dope spectrum occupancy state accurately based on the prerequisite communicated under above-mentioned dynamic frequency spectrum deployment, and find out the communication of idle frequency spectrum for Secondary Users.Existing spectrum occupancy state Forecasting Methodology such as the Forecasting Methodology such as Kalman filter model, hidden Markov model is all predict based on the current and/or spectrum occupancy state of history sequential and the distributed intelligence of frequency spectrum.And Secondary Users hold and obtain frequency spectrum to take distributed intelligence difficulty comparatively large in concrete implementation process, and take distribution cause that amount of calculation is large, computation complexity is high and the problem such as not accurate enough that predicts the outcome owing to relating to frequency spectrum.
Summary of the invention
(1) goal of the invention
The invention provides a kind of only based on frequency range seizure condition required in history sequential, the Forecasting Methodology of amount of calculation is little, computation complexity is low, accuracy rate the is high spectrum occupancy state based on neural net and device.
(2) technical scheme
For reaching above-mentioned purpose, the Forecasting Methodology that the present invention is based on the spectrum occupancy state of neural net comprises the following steps successively:
Step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., L, b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
Preferably, before described step S1, also step S0 is comprised;
Described step S0: gather { b
i, b
i-1..., b
i-mand h
i+1.
Preferably, between described step S0 and described step S1, also comprise the step that the data gathered in described step S0 are normalized.
Preferably, described initial parameter and described Prediction Parameters all at least comprise the weight coefficient between neural net adjacent two layers;
In described step S2 according to comparative result with back-propagation algorithm correction initial parameter.
Preferably, in described step S3, passing threshold criterion judges described H
i+2the corresponding i+2 moment presets the seizure condition of frequency range.
For achieving the above object, the prediction unit that the present invention is based on the spectrum occupancy state of neural net comprises:
Neural net builds module, in order to build neural net and to adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Neural metwork training module, in order to input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Spectrum occupancy state prediction module, in order to while neuralward network input set { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
Further, the prediction unit of the described spectrum occupancy state based on neural net also comprises collection { b
i, b
i-1..., b
i-mand h
i+1data acquisition module.
Further, described data acquisition module also comprises the normalization submodule that the data in order to be gathered are normalized.
Further, described initial parameter and described Prediction Parameters all at least comprise the weight coefficient between neural net adjacent two layers;
Described neural metwork training module comprise in order to according to comparative result with the parameters revision submodule of back-propagation algorithm correction initial parameter.
Further, spectrum occupancy state prediction module comprises and judges described H in order to passing threshold criterion
i+2the corresponding i+2 moment presets the decision sub-module of the seizure condition of frequency range.
(3) the present invention is based on the Forecasting Methodology of the spectrum occupancy state of neural net and the beneficial effect of device
First: the Forecasting Methodology and the device that the present invention is based on the spectrum occupancy state of neural network, the mapping relations between frequency band signals intensity and described prediction sequential spectrum occupancy state are preset in m+1 sequential before adopting the prediction in the neural network radio communication after Pruning Algorithm is determined and passed through training adjustment, thus Secondary Users are when predicting spectrum occupancy state, what do not need to obtain frequency range as conventional method takies distributed intelligence, therefore there is not the problem that Secondary Users obtain frequency range distributed intelligence difficulty.
Second: due to frequency range take distribution usually meet as during the formulae discovery such as Gaussian Profile, Poisson distribution inevitably needs carry out as the complex calculation such as differential, integration.The present invention is based on Forecasting Methodology and the device of the spectrum occupancy state of neural network, in forecasting process, no longer relate to frequency range take distributed intelligence, and neural net is the formation of simulation human brain and the input-output mappings model that the nonlinear network be made up of simple computation unit forms, so calculate the weight computing between normally comparatively simple functional operation and network adjacent two layers, thus have calculate simple, amount of calculation is little and calculate advantage accurately.
3rd: the Forecasting Methodology and the device that the present invention is based on the spectrum occupancy state of neural network, because the learning ability of neural net is strong, there is very strong generalization ability, so different frequency range can be applied to, prediction that the frequency spectrum of different sequential takies, and Prediction Parameters adjusts in real time thus the result obtained is more accurate, thus relative to traditional Forecasting Methodology and device, there is stronger applicability, the information of forecasting provided is more practical.
Accompanying drawing explanation
Fig. 1 is the flow chart based on the Forecasting Methodology of the spectrum occupancy state of neural net described in the embodiment of the present invention three;
Fig. 2 is the flow chart of the prediction unit of the spectrum occupancy state based on neural net described in second embodiment of the invention.
Embodiment
Below in conjunction with Figure of description and embodiment, the Forecasting Methodology to the spectrum occupancy state that the present invention is based on neural net is described further.
Embodiment one:
The signal strength signal intensity of a certain frequency range can be used for weighing the seizure condition of this frequency range, therefore signal strength signal intensity can be used as the occupied information that obtain this frequency that go the related operations such as decay after of input by neural net; Again by observing, the variation tendency of the seizure condition of several continuous sequential abstract can dope the seizure condition information of this frequency range of next sequential of several continuous sequential described by corresponding calculating and conversion.
The present embodiment, based on the Forecasting Methodology of the spectrum occupancy state of neural net, is characterized in that, the Forecasting Methodology of the described spectrum occupancy state based on neural net comprises the following steps:
Step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter; Neural net comprises input layer, intermediate layer and output layer successively; Every one deck is all made up of some neurons, does not mutually have data interaction between the neuron of same layer, carries out data transmission between neuron between layers, for showing the mapping relations between constrained input; Adopt Pruning Algorithm effectively can reject unnecessary neuron and/or neuronic connection in each layer of neural net, thus between the neural net constrained input reaching structure simple, the mapping result of mapping relations is accurately and realize simple and reliable effect;
Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i; The value of i determines the initial value of set, and the value of m determines the length of set; And the training of neural network is carried out with the seizure condition of default frequency range in the signal strength signal intensity of a continuous print m+1 sequential next sequential that is input and m+1 sequential for desired value in the middle of this step, make the predicted value of the output of neural net and desired value meet certain error; If namely the initial parameter that both errors are less than default error then neural net is final Prediction Parameters; If be greater than default error then to need to adjust corresponding parameter, then need again to input { b
i, b
i-1..., b
i-mcalculate result, and by operation result and h
i+1again compare, again according to comparative result corrected parameter, finally make operation result be less than default error, thus by repeatedly revising initial parameter, be final Prediction Parameters with the parameter making final predicted value and desired value be less than the default error moment; Determine the final argument of neural net by this method, realize simple and for the final object accurately that predicts the outcome.
Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
In concrete communication process, wireless frequency spectrum divides with frequency range as carrier wave, and some frequency range is used for uplink communication, some is for downlink communication, also some is for specific area communication etc., therefore in different radio communications, the spectral range of each communications band is different.Therefore the present embodiment needs to arrange default frequency range for different communication according to the demand of telex network kind, because neural net has good generalization ability well can realize the correction of the Prediction Parameters for the different sequential of different frequency range, thus the prediction of Spectrum Occupancy Information for different frequency range and different sequential.
It is comprehensively above-mentioned,
First the present embodiment is based on the Forecasting Methodology of the spectrum occupancy state of neural network, simulate default frequency range in history sequential by neural net characterize the signal strength signal intensity of seizure condition and predict that sequential presets the mapping relations between frequency range seizure condition, therefore do not relate to the distributed intelligence of default frequency range, therefore there is not the problem that Secondary Users obtain frequency range distributed intelligence difficulty;
Secondly, the distributed intelligence of general frequency range meets and to grade the high computing of complexity as needs inevasible when Gaussian Profile, Poisson distribution etc. calculate carry out differential, integration and convolution, and method described in the present embodiment does not relate to the application of frequency range distributed intelligence, therefore the computing such as differential, integration of high complexity can not be related to, and neural net itself have calculate simple, calculate rapid etc. characteristic, so it is low that this method has computation complexity relative to conventional method, and the advantage that relative computation complexity is little.
Again, method described in the present embodiment be in real time for different frequency range different sequential adjustment Prediction Parameters, and the factor such as the adjustment of Prediction Parameters and current network environment is closely related, so Prediction Parameters and Forecasting Methodology are precisely, thus it is also more accurate to predict the outcome;
In addition, method described in the present embodiment, neural net is adopted to predict, because neural net itself has very strong generalization ability, can fast and effectively for two kinds of different frequency ranges by the training in step S2, the prediction of different networks environment intermediate frequency spectrum seizure condition, thus relative to traditional method practicality and applicability all stronger.
Embodiment two:
The present embodiment comprises the following steps based on the Forecasting Methodology of the spectrum occupancy state of neural net:
Step S0: gather { b
i, b
i-1..., b
i-mand h
i+1.The value of i and m determines set
length.Described in concrete implementation process, the value of m is generally the positive integer of 4 ~ 6.
Step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
Due to the needs of step S2, described { b
i+1, b
i..., b
i-m+1and h
i+1can be inputted by peripheral hardware, also can gather voluntarily, and select the method gathered voluntarily in the present embodiment, concrete acquisition method can make actual measurement or empirical data.
Embodiment three:
As shown in Figure 1, the present embodiment is on the basis of a upper embodiment, this enforcement, based on the Forecasting Methodology of the spectrum occupancy state of neural net, also comprises the step to the data line normalized gathered in described step S0 between described step S0 and described step S1.Make the data unit of input unified by normalized process and be positioned at the process range of neural net, thus improving the computational speed of neural net, shortening computing time.
In concrete implementation process, described initial parameter and described Prediction Parameters include the weight coefficient parameter between neuron number, each neuronic excitation function, learning rate, maximum iteration time, default error and each neuronal layers.
As the further optimization of the present embodiment, in described step S3 according to comparative result with back-propagation algorithm correction initial parameter.Adopt back-propagation algorithm correction initial parameter, realize simple and fast.
Embodiment four:
The present embodiment comprises the following steps based on the Forecasting Methodology of the spectrum occupancy state of neural net:
Step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, passing threshold criterion judges that described prediction characterizes variable H
i+2the seizure condition of corresponding default frequency range i+2 sequential.
Other decision methods such as function judgement can be adopted in concrete implementation process, and threshold determination method implements simple advantage, thus make the Forecasting Methodology of the spectrum occupancy state based on neural net described in the present embodiment, have that predicted value is accurate, amount of calculation is little, computation complexity low while be provided with and judge simple and clear and realize easy advantage.
Below in conjunction with Figure of description and embodiment, the prediction unit to the spectrum occupancy state that the present invention is based on neural net is described further.
First embodiment:
The present embodiment comprises based on the prediction unit of the spectrum occupancy state of neural net:
The prediction unit of the described spectrum occupancy state based on neural net comprises:
Neural net builds module, in order to build neural net and to adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Neural metwork training module, in order to input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Spectrum occupancy state prediction module, in order to neuralward network input set { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
In the present embodiment for the device of spectrum occupancy state prediction, build the prediction of module, neural metwork training module and spectrum occupancy state prediction module realization to frequency spectrum by neural net, compared with conventional apparatus, there is computation complexity low, amount of calculation is little, predicts advantage accurately.
Second embodiment:
As shown in Figure 2, the present embodiment comprises based on the prediction unit of the spectrum occupancy state of neural net:
Data acquisition module, in order to gather { b
i, b
i-1..., b
i-mand h
i+1;
Neural net builds module, in order to build neural net and to adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Neural metwork training module, in order to input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Spectrum occupancy state prediction module, in order to while neuralward network input set { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range.
The present embodiment has set up data acquisition module based on the prediction unit of the spectrum occupancy state of neural net on the basis of a upper embodiment, for gathering information needed, thus just can realize the collection of data without the need to the help of peripheral hardware.
3rd embodiment:
The present embodiment is on the basis of a upper embodiment, and the data acquisition module that further tool is described, described data acquisition module also comprises a normalization submodule be normalized in order to the data gathered.Gathered data can be normalized by arranging of normalization submodule, thus make the forecasting efficiency of device and prediction effect better.
4th embodiment:
The present embodiment comprises based on the prediction unit of the spectrum occupancy state of neural net:
Data acquisition module, in order to gather { b
i, b
i-1..., b
i-mand h
i+1;
Neural net builds module, in order to build neural net and to adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Neural metwork training module, in order to input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i; Described initial parameter and described Prediction Parameters all at least comprise the weight coefficient between neural net adjacent two layers; Described neural metwork training module comprise in order to according to comparative result with the parameters revision submodule of back-propagation algorithm correction initial parameter.
Spectrum occupancy state prediction module, in order to while neuralward network input set { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range; Spectrum occupancy state prediction module comprises and judges described H in order to passing threshold criterion
i+2the corresponding i+2 moment presets the decision sub-module of the seizure condition of frequency range.
Comprehensively above-mentioned, the Forecasting Methodology of the spectrum occupancy state based on neural net of the present invention and device, when carrying out the prediction of Spectrum Occupancy Information, without the need to obtaining current spectrum distribution state information, just directly can dope the duty state of required frequency range, and adopt neural net to carry out predicting to have amount of calculation little with the mapping relations between the strength signal of frequency spectrum corresponding to frequency range and seizure condition, the low and advantage accurately that predicts the outcome of computation complexity.
Above execution mode is only for illustration of the present invention; and be not limitation of the present invention; the those of ordinary skill of relevant technical field; without departing from the spirit and scope of the present invention; can also make a variety of changes and modification; therefore all equivalent technical schemes also belong to category of the present invention, and scope of patent protection of the present invention should be defined by the claims.
Claims (10)
1. based on a Forecasting Methodology for the spectrum occupancy state of neural net, it is characterized in that, the Forecasting Methodology of the described spectrum occupancy state based on neural net comprises the following steps successively:
Step S1: build neural net and adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Step S2: input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, i is positive integer, and m is the positive integer being less than i;
Step S3: the set of neuralward network input simultaneously { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range;
Wherein, describedly obtain parameter preset according to comparative result correction initial parameter and comprise: if the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be less than default error, then namely described initial parameter is final Prediction Parameters;
If the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be greater than default error, then again input set { b
i, b
i-1..., b
i-min each element, and by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1again compare, again obtain Prediction Parameters according to comparative result correction initial parameter, finally make the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be less than default error.
2. the Forecasting Methodology of the spectrum occupancy state based on neural net according to claim 1, is characterized in that, before described step S1, also comprise step S0;
Described step S0: gather { b
i, b
i-1..., b
i-mand h
i+1.
3. the Forecasting Methodology of the spectrum occupancy state based on neural net according to claim 2, is characterized in that, also comprises the step be normalized the data gathered in described step S0 between described step S0 and described step S1.
4. the Forecasting Methodology of the spectrum occupancy state based on neural net according to claim 1,2 or 3, is characterized in that, described initial parameter and described Prediction Parameters all at least comprise the weight coefficient between neural net adjacent two layers;
In described step S2 according to comparative result with back-propagation algorithm correction initial parameter.
5. the Forecasting Methodology of the spectrum occupancy state based on neural net according to claim 1,2 or 3, is characterized in that, in described step S3, passing threshold criterion judges described H
i+2the corresponding i+2 moment presets the seizure condition of frequency range.
6. based on a prediction unit for the spectrum occupancy state of neural net, it is characterized in that, the prediction unit of the described spectrum occupancy state based on neural net comprises:
Neural net builds module, in order to build neural net and to adopt each layer of Pruning Algorithm determination neural net and each neuronic initial parameter;
Neural metwork training module, in order to input set { b in neuralward network
i, b
i-1..., b
i-min each element, by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1compare; Prediction Parameters is obtained according to comparative result correction initial parameter; b
ibe the signal strength signal intensity that the i-th sequential presets frequency range, the positive integer that i is, m is the positive integer being less than i;
Spectrum occupancy state prediction module, in order to while neuralward network input set { b
i+1, b
i..., b
i-m+1in each element, neural net exports the prediction that sign i-th+2 sequential presets the seizure condition of frequency range and characterizes variable H
i+2, characterize variable H according to prediction
i+2judge that i+2 sequential presets the seizure condition of frequency range;
Wherein, described in described neural metwork training module obtains parameter preset according to comparative result correction initial parameter and comprises: if the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be less than default error, then namely described initial parameter is final Prediction Parameters;
If the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be greater than default error, then again input set { b
i, b
i-1..., b
i-min each element, and by the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1again compare, again obtain Prediction Parameters according to comparative result correction initial parameter, finally make the Output rusults of neural net with to characterize the sign variable h that the i-th+1 sequential presets frequency range seizure condition
i+1error be less than default error.
7. the prediction unit of the spectrum occupancy state based on neural net according to claim 6, is characterized in that, the prediction unit of the described spectrum occupancy state based on neural net also comprises collection { b
i, b
i-1..., b
i-mand h
i+1data acquisition module.
8. the prediction unit of the spectrum occupancy state based on neural net according to claim 7, is characterized in that, described data acquisition module also comprises the normalization submodule that the data in order to be gathered are normalized.
9., according to the prediction unit of the arbitrary described spectrum occupancy state based on neural net of claim 6-7, it is characterized in that, described initial parameter and described Prediction Parameters all at least comprise the weight coefficient between neural net adjacent two layers;
Described neural metwork training module comprise in order to according to comparative result with the parameters revision submodule of back-propagation algorithm correction initial parameter.
10. according to the prediction unit of the arbitrary described spectrum occupancy state based on neural net of claim 6-7, it is characterized in that, spectrum occupancy state prediction module comprises and judges described H in order to passing threshold criterion
i+2the corresponding i+2 moment presets the decision sub-module of the seizure condition of frequency range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310069826.9A CN103209417B (en) | 2013-03-05 | 2013-03-05 | Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310069826.9A CN103209417B (en) | 2013-03-05 | 2013-03-05 | Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103209417A CN103209417A (en) | 2013-07-17 |
| CN103209417B true CN103209417B (en) | 2016-01-20 |
Family
ID=48756472
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310069826.9A Active CN103209417B (en) | 2013-03-05 | 2013-03-05 | Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103209417B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105915299B (en) * | 2016-03-16 | 2018-08-14 | 重庆邮电大学 | Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band |
| US11887001B2 (en) | 2016-09-26 | 2024-01-30 | Intel Corporation | Method and apparatus for reducing the parameter density of a deep neural network (DNN) |
| EP3340129B1 (en) * | 2016-12-21 | 2019-01-30 | Axis AB | Artificial neural network class-based pruning |
| CN115552413A (en) * | 2020-06-11 | 2022-12-30 | 阿里巴巴集团控股有限公司 | Pruning hardware unit for training neural networks |
| CN112261719B (en) * | 2020-07-24 | 2022-02-11 | 大连理智科技有限公司 | Area positioning method combining SLAM technology with deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005076010A2 (en) * | 2004-02-06 | 2005-08-18 | Council Of Scientific And Industrial Research | Computational method for identifying adhesin and adhesin-like proteins of therapeutic potential |
| CN102056183A (en) * | 2010-12-10 | 2011-05-11 | 北京交通大学 | Network flow prediction method and device based on cognitive network |
| CN102090097A (en) * | 2008-02-14 | 2011-06-08 | 香港科技集团有限公司 | Robust cooperative spectrum sensing for cognitive radios |
-
2013
- 2013-03-05 CN CN201310069826.9A patent/CN103209417B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2005076010A2 (en) * | 2004-02-06 | 2005-08-18 | Council Of Scientific And Industrial Research | Computational method for identifying adhesin and adhesin-like proteins of therapeutic potential |
| CN102090097A (en) * | 2008-02-14 | 2011-06-08 | 香港科技集团有限公司 | Robust cooperative spectrum sensing for cognitive radios |
| CN102056183A (en) * | 2010-12-10 | 2011-05-11 | 北京交通大学 | Network flow prediction method and device based on cognitive network |
Non-Patent Citations (1)
| Title |
|---|
| 认知车载网络中基于QoS 保障的频谱共享新策略;张磊等;《通 信 学 报》;20111130;第32卷(第11期);第159-167页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103209417A (en) | 2013-07-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103209417B (en) | Based on Forecasting Methodology and the device of the spectrum occupancy state of neural net | |
| CN109299430A (en) | The short-term wind speed forecasting method with extreme learning machine is decomposed based on two stages | |
| CN108776683A (en) | A kind of electric power operation/maintenance data cleaning method based on isolated forest algorithm and neural network | |
| CN108171259A (en) | A kind of wisdom tip water quality monitoring system and method based on NB-IoT | |
| CN102682345B (en) | Based on the traffic flow forecasting method of two variable optimal learning rate Fast Learning neural network | |
| CN109302262A (en) | A kind of communication anti-interference method determining Gradient Reinforcement Learning based on depth | |
| CN106254330A (en) | A kind of software defined network intrusion detection method based on BP neutral net | |
| CN110119766B (en) | Green pepper greenhouse temperature intelligent early warning device of multi-combination intelligent model | |
| Dong et al. | A learner based on neural network for cognitive radio | |
| CN108732926A (en) | Networked system method for estimating state based on insufficient information | |
| CN107734512A (en) | A kind of network selecting method based on the analysis of gray scale relevance presenting levelses | |
| CN110163254B (en) | Cucumber greenhouse output intelligent prediction device based on recurrent neural network | |
| CN110119086B (en) | Tomato greenhouse environmental parameter intelligent monitoring device based on ANFIS neural network | |
| CN102592171A (en) | Method and device for predicting cognitive network performance based on BP (Back Propagation) neural network | |
| CN104469833A (en) | Heterogeneous network operation and maintenance management method based on user perception | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN101706888B (en) | Method for predicting travel time | |
| CN108595803A (en) | Shale gas well liquid loading pressure prediction method based on recurrent neural network | |
| CN110212551B (en) | Micro-grid reactive power automatic control method based on convolutional neural network | |
| CN112153617B (en) | Terminal equipment transmission power control method based on integrated neural network | |
| CN107656152A (en) | One kind is based on GA SVM BP Diagnosis Method of Transformer Faults | |
| CN108053052A (en) | A kind of oil truck oil and gas leakage speed intelligent monitor system | |
| CN107528649A (en) | A kind of Forecasting Methodology of the cognitive radio networks frequency spectrum based on Bagging | |
| CN103945398A (en) | Network coverage and capacity optimizing system and optimizing method based on fuzzy neural network | |
| CN101771477A (en) | Mobile phone radio frequency emission power correcting system and mobile phone radio frequency emission power correcting method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |