CN118114031B - Radio waveform prediction method and system based on machine learning - Google Patents
Radio waveform prediction method and system based on machine learning Download PDFInfo
- Publication number
- CN118114031B CN118114031B CN202410519978.2A CN202410519978A CN118114031B CN 118114031 B CN118114031 B CN 118114031B CN 202410519978 A CN202410519978 A CN 202410519978A CN 118114031 B CN118114031 B CN 118114031B
- Authority
- CN
- China
- Prior art keywords
- model
- radio waveform
- signal
- lstm
- denoising
- 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
- 238000000034 method Methods 0.000 title claims abstract description 74
- 238000010801 machine learning Methods 0.000 title claims abstract description 21
- 238000011156 evaluation Methods 0.000 claims abstract description 33
- 230000000694 effects Effects 0.000 claims abstract description 18
- 238000010276 construction Methods 0.000 claims abstract description 12
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 11
- 230000008859 change Effects 0.000 claims abstract description 9
- 230000007246 mechanism Effects 0.000 claims abstract description 9
- 230000004913 activation Effects 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 21
- 238000012549 training Methods 0.000 claims description 18
- 230000008569 process Effects 0.000 claims description 16
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000005070 sampling Methods 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 9
- 230000009466 transformation Effects 0.000 claims description 8
- 239000000284 extract Substances 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 2
- 230000009471 action Effects 0.000 description 3
- 230000003044 adaptive effect Effects 0.000 description 3
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a radio waveform prediction method and a system based on machine learning. The invention belongs to the technical field of radio signal detection, in particular to a radio waveform prediction method and a radio waveform prediction system based on machine learning, wherein the scheme is based on wavelet decomposition to extract signal multi-scale characteristics; the noise in the signal is effectively restrained by adopting a self-adaptive threshold denoising method; a denoising evaluation mechanism is introduced to quantify denoising effect; by combining the reconstruction of the data set and the construction of an internal and external LSTM model, the internal rule and the characteristic of the data are more accurately captured; for searching initial parameters of the model, the model is quickly converged to a better solution by designing diversity factors and dynamically adjusting weights; adopting a self-adaptive searching strategy, and adjusting the searching direction and step length according to the diversity of the population and the change condition of the loss value; and finally, the accuracy of model prediction is greatly improved.
Description
Technical Field
The invention relates to the technical field of radio signal detection, in particular to a radio waveform prediction method and a radio waveform prediction system based on machine learning.
Background
The radio waveform prediction method is a method of predicting the waveform type of a radio signal by means of mathematical modeling, signal processing techniques, machine learning algorithms, and the like. However, the general radio waveform prediction method has the problems of excessive noise content of original signals, poor signal denoising effect and low accuracy of reconstructed signals; the general radio waveform prediction method has the problems of poor data set construction, poor model flexibility, poor expression capability and poor prediction effect caused by improper initial parameter setting of the model.
Disclosure of Invention
Aiming at the problems of excessive noise content of original signals, poor signal denoising effect and low accuracy of reconstructed signals in a general radio waveform prediction method, the invention provides a radio waveform prediction method and a system based on machine learning, and aims at solving the problems of excessive noise content of the original signals, poor signal denoising effect and low accuracy of the reconstructed signals in the general radio waveform prediction method; the adaptive threshold denoising method is adopted, so that noise in the signal is effectively suppressed, meanwhile, effective information of the signal is reserved, and the quality of the signal is improved; reconstructing the denoised wavelet coefficients into an original signal through inverse wavelet transformation, and reserving important characteristics of the original signal to enable the reconstructed signal to be smoother and more accurate; introducing a denoising evaluation mechanism, and quantifying denoising effect by evaluating the difference between the denoised signal and the original signal; aiming at the problems of poor data set construction, poor model flexibility, poor expression capability and poor prediction effect caused by improper initial parameter setting of a model in a general radio waveform prediction method, the scheme fully utilizes time sequence information of time sequence data by combining a reconstruction data set and construction of an internal and external LSTM model, and captures the internal rules and features of the data more accurately; the built model has stronger expression capability, the complexity and fitting capability of the model can be improved by increasing the number of network layers and the number of nodes, and the model is suitable for data sets with different complexity and scales; for searching initial parameters of the model, the parameter space can be effectively explored by designing diversity factors and dynamically adjusting weights, the model can be quickly converged to a better solution, and the training efficiency and performance of the model are improved; the self-adaptive searching strategy is adopted, parameters and weights can be dynamically adjusted in the searching process, the searching direction and the step length are adjusted according to the diversity of the population and the change condition of the loss value, and the searching flexibility and the searching robustness are improved; and finally, the accuracy of model prediction is greatly improved.
The technical scheme adopted by the invention is as follows: the invention provides a radio waveform prediction method based on machine learning, which comprises the following steps:
Step S1: collecting signals;
step S2: reconstructing signals;
Step S3: establishing a radio waveform prediction model;
step S4: radio waveform prediction.
Further, in step S1, the data acquisition is to acquire radio waveform history data; the radio waveform history data includes radio signal data, environment data, time data, and a radio waveform type; the radio waveform type is used as a tag.
Further, in step S2, the signal reconstruction specifically includes the following steps:
Step S21: decomposing the signal into sub-signals of 7 different frequency ranges based on a DB6 wavelet function, and extracting wavelet coefficients of each layer, including approximation coefficients and detail coefficients;
Step S22: adaptively acquiring a threshold for each layer, comprising:
Step S221: performing squaring operation on each wavelet coefficient, and then arranging the wavelet coefficients in a sequence from small to large to obtain a vector P= [ P 1,P2,...,PN ], wherein N represents the length of the wavelet coefficient; p 1、P2 and P N are the squares of the 1 st, 2 nd and nth wavelet coefficients, respectively;
Step S222: calculating a risk vector R based on the vector P, and finding the smallest R i in the risk vector as a risk value; the formula used is as follows:
;
Wherein R i is the ith risk value; p k is the square of the kth wavelet coefficient; sum (·) is the sum;
Step S223: the threshold is calculated using the formula:
;
wherein L is a threshold; median (·) is median taken; abs (·) is taken absolute; sqrt (·) is the square root; w (j-1, k) is the current wavelet coefficient;
Step S23: denoising the 7-layer signals obtained by decomposition according to the selected threshold value, wherein the formula is as follows:
;
Wherein X i is the i-th layer signal after denoising, and d i is the original i-th layer signal; li is the threshold of the i-th layer;
Step S24: signal reconstruction, comprising:
Step S241: for the highest level of detail coefficients and the lowest frequency approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
step S242: for each level of detail and approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
Step S243: repeating the steps S241 and S242 until all the reconstruction levels are completed;
step S25: denoising evaluation, namely presetting a denoising evaluation threshold value, and returning to the step S21 when the denoising evaluation value is lower than the denoising evaluation threshold value; when the denoising evaluation value is not lower than the denoising evaluation threshold value, denoising is completed; the formula used for denoising evaluation is as follows:
;
wherein QR is a denoising evaluation value; x (n) is the nth sample value of the original signal; x m (n) is the nth sample value of the denoised signal.
Further, in step S3, the building of the radio waveform prediction model is building of an internal-external LSTM; the method specifically comprises the following steps:
Step S31: reconstructing the data set, reconstructing the data set based on the time delay; randomly dividing a training set and a testing set, wherein the training set is used for training a model, and the testing set is used for verifying the performance of the model; the construction rule is to start from the first sampling point of the reconstructed signal, use the 1 st to 99 th sampling points as input samples, the 100 th sampling point as output sample, and so on; the input sample of the ith training set is the ith to the (i+98) th sampling points, the output sample is the (i+99) th sampling points, and input-output sample pairs are generated; the expression is as follows:
;
Wherein X is a reconstructed dataset; x 1 and x n are the 1 st and n th sample points, respectively; Is a time delay; m is the reconstruction dimension;
step S32: constructing an internal LSTM; the formula used is as follows:
;
;
;
;
;
;
;
;
In the method, in the process of the invention, Is the internal LSTM hidden state of the previous time step; And The internal LSTM inputs for the previous time step and the current time step, respectively; f t is the activation value of the forget gate; i t is the activation value of the input gate; is the activation value of the candidate memory cell; And The internal LSTM memory unit is respectively used for the current time step and the last time step; Is the activation value of the internal LSTM output gate; (. Cndot.) is a sigmoid activation function; tanh (·) is a hyperbolic tangent activation function; And Is the weight matrix of the internal LSTM forget gate; And Is a weight matrix of the internal LSTM input gate; And Is the weight matrix of the internal LSTM candidate memory cell; And Is a weight matrix of the output gates of the internal LSTM;、、 And Bias of internal LSTM forget gate, input gate, candidate memory cell and output gate, respectively;
step S33: an external LSTM was constructed using the following formula:
;
;
Wherein c t is the external LSTM current memory cell; o t is the activation value of the external LSTM output gate; w ox and W oh are weight matrices of external LSTM output gates; b o is the bias of the external LSTM output gate; x t is the external LSTM input for the current time step; h t-1 is the external LSTM hidden state of the previous time step;
step S34: the output layer is constructed using the following formula:
;
Wherein W yh represents a weight matrix of the output layer; y t is the output layer output;
Step S35: defining a loss function, updating network parameters based on an error back propagation algorithm, wherein the loss function adopts the following formula:
;
Wherein E is a loss value; t is the length of the time series; y t is the real tag at time step t; model predictive labels at time step t;
Step S36: optimizing model initial parameters, including:
step S361: constructing a parameter search space based on initial parameters of a model, randomly initializing a search population, calculating a loss value of a test set based on the position of a search individual, taking the reciprocal and then carrying out normalization processing to obtain a final individual fitness value;
step S362: the diversity factor is designed using the following formula:
;
Where d t is the diversity factor at the t-th iteration; n 2 is the search population number; d is the search dimension; is the search space longest diagonal length; is the position of the ith individual in the nth dimension at the t-th iteration; Is the population average position at the time of the t iteration of the d dimension;
Step S363: judging the diversity of the population, wherein the formula is as follows:
;
Wherein w, w min and w max are a movement weight, a minimum movement weight and a maximum movement weight, respectively; d low and d high are population diversity lower and upper limits, respectively;
step S364: updating the individual location; the formula used is as follows:
;
In the method, in the process of the invention, AndThe positions of the ith individual at the t-th iteration and the t-1 th iteration are respectively, and eta is the learning rate; best t-1 is the optimal individual position of the population at iteration t-1; is the optimal individual position of experience in the ith individual t-1 iteration; f i The fitness value of the individual and the average fitness value of the population are respectively;
step S365: searching and judging, wherein a loss threshold value is preset, when the loss value of a training set by a model trained based on the individual position is lower than the loss threshold value, searching is finished, and the individual position is a model initialization parameter, so that model establishment is completed; if the maximum iteration number is reached, the population position is reinitialized; otherwise, continuing the iterative search.
Further, in step S4, the radio waveform prediction is based on the established radio waveform prediction model, radio signal data, environment data and time data are collected in real time, and the type of radio waveform output by the model is used as a prediction result.
The invention provides a radio waveform prediction system based on machine learning, which comprises a signal acquisition model, a signal reconstruction model, a radio waveform prediction model establishment model and a radio waveform prediction model;
The signal acquisition model acquires radio waveform historical data and sends the data to the signal reconstruction model;
The signal reconstruction model extracts signal multi-scale features based on wavelet decomposition; adopting a self-adaptive threshold denoising method, and reconstructing the denoised wavelet coefficient into an original signal through inverse wavelet transformation; quantizing the denoising effect based on a denoising evaluation mechanism, and finally completing signal reconstruction; transmitting the data to a radio waveform prediction model building model;
the radio waveform prediction model building model is based on the reconstruction data set and builds an internal and external LSTM model; the weight is dynamically adjusted by designing diversity factors; the method comprises the steps of adopting a self-adaptive searching strategy, dynamically adjusting parameters and weights in the searching process, adjusting the searching direction and step length according to the diversity of population and the change condition of loss values, realizing the searching of initial parameters of a model, and finally establishing a radio waveform prediction model; and transmitting the data to a radio waveform prediction model;
the radio waveform prediction model is based on the established radio waveform prediction model, and radio waveform types output by the model are taken as prediction results through collecting radio signal data, environment data and time data in real time.
By adopting the scheme, the beneficial effects obtained by the invention are as follows:
(1) Aiming at the problems of excessive noise content of an original signal, poor signal denoising effect and low accuracy of a reconstructed signal in a general radio waveform prediction method, the method extracts multi-scale characteristics of the signal based on wavelet decomposition; the adaptive threshold denoising method is adopted, so that noise in the signal is effectively suppressed, meanwhile, effective information of the signal is reserved, and the quality of the signal is improved; reconstructing the denoised wavelet coefficients into an original signal through inverse wavelet transformation, and reserving important characteristics of the original signal to enable the reconstructed signal to be smoother and more accurate; and a denoising evaluation mechanism is introduced, and the denoising effect is quantized by evaluating the difference between the denoised signal and the original signal.
(2) Aiming at the problems of poor data set construction, poor model flexibility, poor expression capability and poor prediction effect caused by improper initial parameter setting of a model in a general radio waveform prediction method, the scheme fully utilizes time sequence information of time sequence data by combining a reconstruction data set and construction of an internal and external LSTM model, and captures the internal rules and features of the data more accurately; the built model has stronger expression capability, the complexity and fitting capability of the model can be improved by increasing the number of network layers and the number of nodes, and the model is suitable for data sets with different complexity and scales; for searching initial parameters of the model, the parameter space can be effectively explored by designing diversity factors and dynamically adjusting weights, the model can be quickly converged to a better solution, and the training efficiency and performance of the model are improved; the self-adaptive searching strategy is adopted, parameters and weights can be dynamically adjusted in the searching process, the searching direction and the step length are adjusted according to the diversity of the population and the change condition of the loss value, and the searching flexibility and the searching robustness are improved; and finally, the accuracy of model prediction is greatly improved.
Drawings
Fig. 1 is a flow chart of a radio waveform prediction method based on machine learning according to the present invention;
FIG. 2 is a schematic diagram of a machine learning based radio waveform prediction system according to the present invention;
fig. 3 is a flow chart of step S3.
The accompanying drawings are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all embodiments of the invention; all other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
In the description of the present invention, it should be understood that the terms "upper," "lower," "front," "rear," "left," "right," "top," "bottom," "inner," "outer," and the like indicate orientation or positional relationships based on those shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the system or element being referred to must have a particular orientation, be constructed and operate in a particular orientation, and thus should not be construed as limiting the present invention.
Referring to fig. 1, the method for predicting a radio waveform based on machine learning according to the present invention includes the following steps:
Step S1: signal acquisition, namely acquiring radio waveform historical data;
Step S2: reconstructing signals, and extracting multi-scale characteristics of the signals based on wavelet decomposition; adopting a self-adaptive threshold denoising method, and reconstructing the denoised wavelet coefficient into an original signal through inverse wavelet transformation; quantizing the denoising effect based on a denoising evaluation mechanism, and finally completing signal reconstruction;
Step S3: establishing a radio waveform prediction model, and establishing an internal and external LSTM model based on the reconstruction data set; the weight is dynamically adjusted by designing diversity factors; the method comprises the steps of adopting a self-adaptive searching strategy, dynamically adjusting parameters and weights in the searching process, adjusting the searching direction and step length according to the diversity of population and the change condition of loss values, realizing the searching of initial parameters of a model, and finally establishing a radio waveform prediction model;
step S4: radio waveform prediction.
Second embodiment referring to fig. 1, the embodiment is based on the above embodiment, and in step S1, the radio waveform history data includes radio signal data, environment data, time data, and a radio waveform type; the radio waveform type is used as a tag.
In the third embodiment, referring to fig. 1, the signal reconstruction specifically includes the following steps in step S2, where the embodiment is based on the above embodiment:
Step S21: decomposing the signal into sub-signals of 7 different frequency ranges based on a DB6 wavelet function, and extracting wavelet coefficients of each layer, including approximation coefficients and detail coefficients;
Step S22: adaptively acquiring a threshold for each layer, comprising:
Step S221: performing squaring operation on each wavelet coefficient, and then arranging the wavelet coefficients in a sequence from small to large to obtain a vector P= [ P 1,P2,...,PN ], wherein N represents the length of the wavelet coefficient; p 1、P2 and P N are the squares of the 1 st, 2 nd and nth wavelet coefficients, respectively;
Step S222: calculating a risk vector R based on the vector P, and finding the smallest R i in the risk vector as a risk value; the formula used is as follows:
;
Wherein R i is the ith risk value; p k is the square of the kth wavelet coefficient; sum (·) is the sum;
Step S223: the threshold is calculated using the formula:
;
wherein L is a threshold; median (·) is median taken; abs (·) is taken absolute; sqrt (·) is the square root; w (j-1, k) is the current wavelet coefficient;
Step S23: denoising the 7-layer signals obtained by decomposition according to the selected threshold value, wherein the formula is as follows:
;
Wherein X i is the i-th layer signal after denoising, and d i is the original i-th layer signal; li is the threshold of the i-th layer;
Step S24: signal reconstruction, comprising:
Step S241: for the highest level of detail coefficients and the lowest frequency approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
step S242: for each level of detail and approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
Step S243: repeating the steps S241 and S242 until all the reconstruction levels are completed;
step S25: denoising evaluation, namely presetting a denoising evaluation threshold value, and returning to the step S21 when the denoising evaluation value is lower than the denoising evaluation threshold value; when the denoising evaluation value is not lower than the denoising evaluation threshold value, denoising is completed; the formula used for denoising evaluation is as follows:
;
wherein QR is a denoising evaluation value; x (n) is the nth sample value of the original signal; x m (n) is the nth sample value of the denoised signal.
By executing the operation, aiming at the problems of excessive noise content of an original signal, poor signal denoising effect and low accuracy of a reconstructed signal in a general radio waveform prediction method, the method extracts multi-scale characteristics of the signal based on wavelet decomposition; the adaptive threshold denoising method is adopted, so that noise in the signal is effectively suppressed, meanwhile, effective information of the signal is reserved, and the quality of the signal is improved; reconstructing the denoised wavelet coefficients into an original signal through inverse wavelet transformation, and reserving important characteristics of the original signal to enable the reconstructed signal to be smoother and more accurate; and a denoising evaluation mechanism is introduced, and the denoising effect is quantized by evaluating the difference between the denoised signal and the original signal.
Fourth embodiment referring to fig. 1 and 3, the embodiment is based on the above embodiment, and in step S3, the radio waveform prediction model is built by building an internal-external LSTM; the method specifically comprises the following steps:
Step S31: reconstructing the data set, reconstructing the data set based on the time delay; randomly dividing a training set and a testing set, wherein the training set is used for training a model, and the testing set is used for verifying the performance of the model; the construction rule is to start from the first sampling point of the reconstructed signal, use the 1 st to 99 th sampling points as input samples, the 100 th sampling point as output sample, and so on; the input sample of the ith training set is the ith to the (i+98) th sampling points, the output sample is the (i+99) th sampling points, and input-output sample pairs are generated; the expression is as follows:
;
Wherein X is a reconstructed dataset; x 1 and x n are the 1 st and n th sample points, respectively; Is a time delay; m is the reconstruction dimension;
step S32: constructing an internal LSTM; the formula used is as follows:
;
;
;
;
;
;
;
;
In the method, in the process of the invention, Is the internal LSTM hidden state of the previous time step; And The internal LSTM inputs for the previous time step and the current time step, respectively; f t is the activation value of the forget gate; i t is the activation value of the input gate; is the activation value of the candidate memory cell; And The internal LSTM memory unit is respectively used for the current time step and the last time step; Is the activation value of the internal LSTM output gate; (. Cndot.) is a sigmoid activation function; tanh (·) is a hyperbolic tangent activation function; And Is the weight matrix of the internal LSTM forget gate; And Is a weight matrix of the internal LSTM input gate; And Is the weight matrix of the internal LSTM candidate memory cell; And Is a weight matrix of the output gates of the internal LSTM;、、 And Bias of internal LSTM forget gate, input gate, candidate memory cell and output gate, respectively;
step S33: an external LSTM was constructed using the following formula:
;
;
Wherein c t is the external LSTM current memory cell; o t is the activation value of the external LSTM output gate; w ox and W oh are weight matrices of external LSTM output gates; b o is the bias of the external LSTM output gate; x t is the external LSTM input for the current time step; h t-1 is the external LSTM hidden state of the previous time step;
step S34: the output layer is constructed using the following formula:
;
Wherein W yh represents a weight matrix of the output layer; y t is the output layer output;
Step S35: defining a loss function, updating network parameters based on an error back propagation algorithm, wherein the loss function adopts the following formula:
;
Wherein E is a loss value; t is the length of the time series; y t is the real tag at time step t; model predictive labels at time step t;
Step S36: optimizing model initial parameters, including:
step S361: constructing a parameter search space based on initial parameters of a model, randomly initializing a search population, calculating a loss value of a test set based on the position of a search individual, taking the reciprocal and then carrying out normalization processing to obtain a final individual fitness value;
step S362: the diversity factor is designed using the following formula:
;
Where d t is the diversity factor at the t-th iteration; n 2 is the search population number; d is the search dimension; is the search space longest diagonal length; is the position of the ith individual in the nth dimension at the t-th iteration; Is the population average position at the time of the t iteration of the d dimension;
Step S363: judging the diversity of the population, wherein the formula is as follows:
;
Wherein w, w min and w max are a movement weight, a minimum movement weight and a maximum movement weight, respectively; d low and d high are population diversity lower and upper limits, respectively;
step S364: updating the individual location; the formula used is as follows:
;
In the method, in the process of the invention, AndThe positions of the ith individual at the t-th iteration and the t-1 th iteration are respectively, and eta is the learning rate; best t-1 is the optimal individual position of the population at iteration t-1; is the optimal individual position of experience in the ith individual t-1 iteration; f i The fitness value of the individual and the average fitness value of the population are respectively;
step S365: searching and judging, wherein a loss threshold value is preset, when the loss value of a training set by a model trained based on the individual position is lower than the loss threshold value, searching is finished, and the individual position is a model initialization parameter, so that model establishment is completed; if the maximum iteration number is reached, the population position is reinitialized; otherwise, continuing the iterative search.
By executing the operations, aiming at the problems of poor data set construction, poor model flexibility, poor expression capability and poor prediction effect caused by improper initial parameter setting of a model in a general radio waveform prediction method, the method can more accurately capture the internal rules and features of the data by combining the reconstruction of the data set and the construction of an internal and external LSTM model and fully utilizing the time sequence information of time sequence data; the built model has stronger expression capability, the complexity and fitting capability of the model can be improved by increasing the number of network layers and the number of nodes, and the model is suitable for data sets with different complexity and scales; for searching initial parameters of the model, the parameter space can be effectively explored by designing diversity factors and dynamically adjusting weights, the model can be quickly converged to a better solution, and the training efficiency and performance of the model are improved; the self-adaptive searching strategy is adopted, parameters and weights can be dynamically adjusted in the searching process, the searching direction and the step length are adjusted according to the diversity of the population and the change condition of the loss value, and the searching flexibility and the searching robustness are improved; and finally, the accuracy of model prediction is greatly improved.
Fifth embodiment referring to fig. 1, the embodiment is based on the above embodiment, and in step S4, the radio waveform prediction is based on the established radio waveform prediction model, and radio signal data, environment data, and time data are collected in real time, and the type of radio waveform output by the model is used as a prediction result.
An embodiment six, referring to fig. 2, based on the above embodiment, the radio waveform prediction system based on machine learning provided by the present invention includes a signal acquisition model, a signal reconstruction model, a radio waveform prediction model establishment model, and a radio waveform prediction model;
The signal acquisition model acquires radio waveform historical data and sends the data to the signal reconstruction model;
The signal reconstruction model extracts signal multi-scale features based on wavelet decomposition; adopting a self-adaptive threshold denoising method, and reconstructing the denoised wavelet coefficient into an original signal through inverse wavelet transformation; quantizing the denoising effect based on a denoising evaluation mechanism, and finally completing signal reconstruction; transmitting the data to a radio waveform prediction model building model;
the radio waveform prediction model building model is based on the reconstruction data set and builds an internal and external LSTM model; the weight is dynamically adjusted by designing diversity factors; the method comprises the steps of adopting a self-adaptive searching strategy, dynamically adjusting parameters and weights in the searching process, adjusting the searching direction and step length according to the diversity of population and the change condition of loss values, realizing the searching of initial parameters of a model, and finally establishing a radio waveform prediction model; and transmitting the data to a radio waveform prediction model;
the radio waveform prediction model is based on the established radio waveform prediction model, and radio waveform types output by the model are taken as prediction results through collecting radio signal data, environment data and time data in real time.
It is noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made hereto without departing from the spirit and principles of the present invention.
The invention and its embodiments have been described above with no limitation, and the actual construction is not limited to the embodiments of the invention as shown in the drawings. In summary, if one of ordinary skill in the art is informed by this disclosure, a structural manner and an embodiment similar to the technical solution should not be creatively devised without departing from the gist of the present invention.
Claims (7)
1. A machine learning based radio waveform prediction method, characterized by: the method comprises the following steps:
Step S1: collecting signals;
Step S2: reconstructing signals, and extracting multi-scale characteristics of the signals based on wavelet decomposition; adopting a self-adaptive threshold denoising method, and reconstructing the denoised wavelet coefficient into an original signal through inverse wavelet transformation; quantizing the denoising effect based on a denoising evaluation mechanism, and finally completing signal reconstruction;
Step S3: establishing a radio waveform prediction model, and establishing an internal and external LSTM model based on the reconstruction data set; the weight is dynamically adjusted by designing diversity factors; the method comprises the steps of adopting a self-adaptive searching strategy, dynamically adjusting parameters and weights in the searching process, adjusting the searching direction and step length according to the diversity of population and the change condition of loss values, realizing the searching of initial parameters of a model, and finally establishing a radio waveform prediction model;
step S4: radio waveform prediction;
step S22, i.e. adaptively acquiring the threshold value of each layer, includes:
Step S221: performing squaring operation on each wavelet coefficient, and then arranging the wavelet coefficients in a sequence from small to large to obtain a vector P= [ P 1,P2,...,PN ], wherein N represents the length of the wavelet coefficient; p 1、P2 and P N are the squares of the 1 st, 2 nd and nth wavelet coefficients, respectively;
Step S222: calculating a risk vector R based on the vector P, and finding the smallest R i in the risk vector as a risk value; the formula used is as follows:
;
Wherein R i is the ith risk value; p k is the square of the kth wavelet coefficient; sum (·) is the sum;
Step S223: the threshold is calculated using the formula:
;
Wherein L is a threshold; median (·) is median taken; abs (·) is taken absolute; sqrt (·) is the square root; w (j-1, k) is the current wavelet coefficient.
2. The machine learning based radio waveform prediction method of claim 1, wherein: in step S3, said building a radio waveform prediction model is building an inside-outside LSTM; the method specifically comprises the following steps:
Step S31: reconstructing the data set, reconstructing the data set based on the time delay; randomly dividing a training set and a testing set, wherein the training set is used for training a model, and the testing set is used for verifying the performance of the model; the construction rule is to start from the first sampling point of the reconstructed signal, use the 1 st to 99 th sampling points as input samples, the 100 th sampling point as output sample, and so on; the input sample of the ith training set is the ith to the (i+98) th sampling points, the output sample is the (i+99) th sampling points, and input-output sample pairs are generated; the expression is as follows:
;
Wherein X is a reconstructed dataset; x 1 and x n are the 1 st and n th sample points, respectively; Is a time delay; m is the reconstruction dimension;
step S32: constructing an internal LSTM; the formula used is as follows:
;
;
;
;
;
;
;
;
In the method, in the process of the invention, Is the internal LSTM hidden state of the previous time step; And The internal LSTM inputs for the previous time step and the current time step, respectively; f t is the activation value of the forget gate; i t is the activation value of the input gate; is the activation value of the candidate memory cell; And The internal LSTM memory unit is respectively used for the current time step and the last time step; Is the activation value of the internal LSTM output gate; (. Cndot.) is a sigmoid activation function; tanh (·) is a hyperbolic tangent activation function; And Is the weight matrix of the internal LSTM forget gate; And Is a weight matrix of the internal LSTM input gate; And Is the weight matrix of the internal LSTM candidate memory cell; And Is a weight matrix of the output gates of the internal LSTM;、、 And Bias of internal LSTM forget gate, input gate, candidate memory cell and output gate, respectively;
step S33: an external LSTM was constructed using the following formula:
;
;
Wherein c t is the external LSTM current memory cell; o t is the activation value of the external LSTM output gate; w ox and W oh are weight matrices of external LSTM output gates; b o is the bias of the external LSTM output gate; x t is the external LSTM input for the current time step; h t-1 is the external LSTM hidden state of the previous time step;
step S34: the output layer is constructed using the following formula:
;
Wherein W yh represents a weight matrix of the output layer; y t is the output layer output;
Step S35: defining a loss function, updating network parameters based on an error back propagation algorithm, wherein the loss function adopts the following formula:
;
Wherein E is a loss value; t is the length of the time series; y t is the real tag at time step t; model predictive labels at time step t;
Step S36: optimizing initial parameters of the model.
3. The machine learning based radio waveform prediction method of claim 2, wherein: in step S36, the optimization model initial parameters include:
step S361: constructing a parameter search space based on initial parameters of a model, randomly initializing a search population, calculating a loss value of a test set based on the position of a search individual, taking the reciprocal and then carrying out normalization processing to obtain a final individual fitness value;
step S362: the diversity factor is designed using the following formula:
;
Where d t is the diversity factor at the t-th iteration; n 2 is the search population number; d is the search dimension; is the search space longest diagonal length; is the position of the ith individual in the nth dimension at the t-th iteration; Is the population average position at the time of the t iteration of the d dimension;
Step S363: judging the diversity of the population, wherein the formula is as follows:
;
Wherein w, w min and w max are a movement weight, a minimum movement weight and a maximum movement weight, respectively; d low and d high are population diversity lower and upper limits, respectively;
step S364: updating the individual location; the formula used is as follows:
;
In the method, in the process of the invention, AndThe positions of the ith individual at the t-th iteration and the t-1 th iteration are respectively, and eta is the learning rate; best t-1 is the optimal individual position of the population at iteration t-1; is the optimal individual position of experience in the ith individual t-1 iteration; f i The fitness value of the individual and the average fitness value of the population are respectively;
step S365: searching and judging, wherein a loss threshold value is preset, when the loss value of a training set by a model trained based on the individual position is lower than the loss threshold value, searching is finished, and the individual position is a model initialization parameter, so that model establishment is completed; if the maximum iteration number is reached, the population position is reinitialized; otherwise, continuing the iterative search.
4. The machine learning based radio waveform prediction method of claim 1, wherein: in step S2, the signal reconstruction specifically includes the following steps:
Step S21: decomposing the signal into sub-signals of 7 different frequency ranges based on a DB6 wavelet function, and extracting wavelet coefficients of each layer, including approximation coefficients and detail coefficients;
Step S22: adaptively acquiring a threshold value of each layer;
Step S23: denoising the 7-layer signals obtained by decomposition according to the selected threshold value, wherein the formula is as follows:
;
Wherein X i is the i-th layer signal after denoising, and d i is the original i-th layer signal; li is the threshold of the i-th layer;
Step S24: signal reconstruction, comprising:
Step S241: for the highest level of detail coefficients and the lowest frequency approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
step S242: for each level of detail and approximation coefficients, upsampling and convolving using an inverse high pass filter and an inverse low pass filter of the wavelet basis function to obtain a reconstructed signal;
Step S243: repeating the steps S241 and S242 until all the reconstruction levels are completed;
step S25: denoising evaluation, namely presetting a denoising evaluation threshold value, and returning to the step S21 when the denoising evaluation value is lower than the denoising evaluation threshold value; when the denoising evaluation value is not lower than the denoising evaluation threshold value, denoising is completed; the formula used for denoising evaluation is as follows:
;
wherein QR is a denoising evaluation value; x (n) is the nth sample value of the original signal; x m (n) is the nth sample value of the denoised signal.
5. The machine learning based radio waveform prediction method of claim 1, wherein: in step S4, the radio waveform prediction is based on the established radio waveform prediction model, and radio signal data, environment data and time data are collected in real time, and the radio waveform type output by the model is used as a prediction result.
6. The machine learning based radio waveform prediction method of claim 1, wherein: in step S1, the signal acquisition is to acquire radio waveform history data; the radio waveform history data includes radio signal data, environment data, time data, and a radio waveform type; the radio waveform type is used as a tag.
7. A machine learning based radio waveform prediction system for implementing the machine learning based radio waveform prediction method of any one of claims 1-6, characterized by: the method comprises a signal acquisition model, a signal reconstruction model, a radio waveform prediction model establishment model and a radio waveform prediction model;
The signal acquisition model acquires radio waveform historical data and sends the data to the signal reconstruction model;
The signal reconstruction model extracts signal multi-scale features based on wavelet decomposition; adopting a self-adaptive threshold denoising method, and reconstructing the denoised wavelet coefficient into an original signal through inverse wavelet transformation; quantizing the denoising effect based on a denoising evaluation mechanism, and finally completing signal reconstruction; transmitting the data to a radio waveform prediction model building model;
the radio waveform prediction model building model is based on the reconstruction data set and builds an internal and external LSTM model; the weight is dynamically adjusted by designing diversity factors; the method comprises the steps of adopting a self-adaptive searching strategy, dynamically adjusting parameters and weights in the searching process, adjusting the searching direction and step length according to the diversity of population and the change condition of loss values, realizing the searching of initial parameters of a model, and finally establishing a radio waveform prediction model; and transmitting the data to a radio waveform prediction model;
the radio waveform prediction model is based on the established radio waveform prediction model, and radio waveform types output by the model are taken as prediction results through collecting radio signal data, environment data and time data in real time.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410519978.2A CN118114031B (en) | 2024-04-28 | 2024-04-28 | Radio waveform prediction method and system based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202410519978.2A CN118114031B (en) | 2024-04-28 | 2024-04-28 | Radio waveform prediction method and system based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN118114031A CN118114031A (en) | 2024-05-31 |
CN118114031B true CN118114031B (en) | 2024-07-05 |
Family
ID=91219449
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202410519978.2A Active CN118114031B (en) | 2024-04-28 | 2024-04-28 | Radio waveform prediction method and system based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN118114031B (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111681667A (en) * | 2020-06-23 | 2020-09-18 | 青岛科技大学 | Underwater sound signal denoising method based on adaptive window filtering and wavelet threshold optimization |
CN112418014A (en) * | 2020-11-09 | 2021-02-26 | 南京信息工程大学滨江学院 | Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2017124044A1 (en) * | 2016-01-15 | 2017-07-20 | The Regents Of The University Of California | Machine-learning-based denoising of doppler ultrasound blood flow and intracranial pressure signal |
WO2019147953A1 (en) * | 2018-01-25 | 2019-08-01 | Ctrl-Labs Corporation | Methods and apparatus for mitigating neuromuscular signal artifacts |
US11687778B2 (en) * | 2020-01-06 | 2023-06-27 | The Research Foundation For The State University Of New York | Fakecatcher: detection of synthetic portrait videos using biological signals |
CN113160132A (en) * | 2021-03-10 | 2021-07-23 | 上海应用技术大学 | Detection processing method and system for weld defect image |
CN117369026B (en) * | 2023-12-06 | 2024-03-08 | 江苏省气象台 | Real-time high-precision cloud cluster residence time prediction method |
-
2024
- 2024-04-28 CN CN202410519978.2A patent/CN118114031B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111681667A (en) * | 2020-06-23 | 2020-09-18 | 青岛科技大学 | Underwater sound signal denoising method based on adaptive window filtering and wavelet threshold optimization |
CN112418014A (en) * | 2020-11-09 | 2021-02-26 | 南京信息工程大学滨江学院 | Modulation signal identification method based on wavelet transformation and convolution long-short term memory neural network |
Also Published As
Publication number | Publication date |
---|---|
CN118114031A (en) | 2024-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108900346B (en) | Wireless network flow prediction method based on LSTM network | |
CN108696331B (en) | Signal reconstruction method based on generation countermeasure network | |
CN111563706A (en) | Multivariable logistics freight volume prediction method based on LSTM network | |
CN111860982A (en) | Wind power plant short-term wind power prediction method based on VMD-FCM-GRU | |
CN112949828B (en) | Graph convolution neural network traffic prediction method and system based on graph learning | |
CN112560918B (en) | Dish identification method based on improved YOLO v3 | |
CN108629370B (en) | Classification recognition algorithm and device based on deep belief network | |
CN111832228B (en) | Vibration transmission system based on CNN-LSTM | |
CN104156943B (en) | Multi objective fuzzy cluster image change detection method based on non-dominant neighborhood immune algorithm | |
CN111931983A (en) | Precipitation prediction method and system | |
CN112686376A (en) | Node representation method based on timing diagram neural network and incremental learning method | |
CN115587666A (en) | Load prediction method and system based on seasonal trend decomposition and hybrid neural network | |
CN115510748A (en) | Landslide displacement prediction method based on variational modal decomposition and CNN-GRU | |
CN109492816B (en) | Coal and gas outburst dynamic prediction method based on hybrid intelligence | |
CN117196418B (en) | Reading teaching quality assessment method and system based on artificial intelligence | |
CN113033878B (en) | Landslide displacement prediction method based on multi-topology grading collaborative particle swarm LSTM | |
CN113627597A (en) | Countermeasure sample generation method and system based on general disturbance | |
CN118114031B (en) | Radio waveform prediction method and system based on machine learning | |
CN117370766A (en) | Satellite mission planning scheme evaluation method based on deep learning | |
CN117458480A (en) | Photovoltaic power generation power short-term prediction method and system based on improved LOF | |
CN116011338A (en) | Full waveform inversion method based on self-encoder and deep neural network | |
CN115169740A (en) | Sequence prediction method and system of pooled echo state network based on compressed sensing | |
CN114630207A (en) | Multi-sensing-node perception data collection method based on noise reduction self-encoder | |
CN115035304A (en) | Image description generation method and system based on course learning | |
Liu et al. | Non-local mean filtering algorithm based on deep learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |