CN116016068B - Data-driven-based mutual-frequency intelligent intervention signal representation method and system - Google Patents
Data-driven-based mutual-frequency intelligent intervention signal representation method and system Download PDFInfo
- Publication number
- CN116016068B CN116016068B CN202211137836.7A CN202211137836A CN116016068B CN 116016068 B CN116016068 B CN 116016068B CN 202211137836 A CN202211137836 A CN 202211137836A CN 116016068 B CN116016068 B CN 116016068B
- Authority
- CN
- China
- Prior art keywords
- frequency
- output
- low
- component
- lstm network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
- Transmitters (AREA)
Abstract
The invention discloses a method and a system for representing mutual-frequency intelligent intervention signals based on data driving, which are characterized in that the intervention operation is carried out on the input of one or more low-frequency units through the output of one or more high-frequency units, the intervention operation is carried out on the input of one or more high-frequency units through the output of one or more low-frequency units, so that the interaction processing of a low-frequency component and a high-frequency component by a high-frequency intervention model, namely the intervention of the high-frequency component and the intervention of the high-frequency component by the low-frequency component, the mutual intervention and the mutual driving of the high-frequency component and the low-frequency component are realized, the obtained high-frequency output component fuses the implicit information of the low-frequency component, and the low-frequency output component fuses the implicit information of the high-frequency component, so that the signal information can be represented more accurately, and the signal characteristics obtained based on the high-frequency output component and the low-frequency output component can effectively represent signals.
Description
Technical Field
The invention relates to the technical field of communication, in particular to a method and a system for representing a mutual frequency intelligent intervention signal based on data driving.
Background
In a non-cooperative communication system, only the modulation scheme of the acquired signal can correctly demodulate the signal. The related key technologies such as signal identification and target identification are necessary components of a non-cooperative communication system, are effective means for intercepting enemy signals, are important steps of spectrum detection, and have important theoretical significance and application value in the military and civil fields such as unmanned aerial vehicle cluster combat, electromagnetic spectrum sensing, radio communication and the like. With the rapid development of communication technology, electromagnetic signals are characterized by more quantity, large density, complex form and the like, so that the communication environment is more complex, effective signal representation is difficult to obtain, the performance improvement of a related intelligent system is limited, and the following three aspects are mainly embodied: firstly, the increasing coverage frequency of the radiation source signals leads to the increasing number and variety of unknown signals; secondly, along with the improvement of the communication technology level, a large number of devices with complex systems start to appear, and the generated electromagnetic signals are complex in form and variable in frequency; thirdly, the working frequency range of the communication equipment is continuously widened, and the working system is increasingly complex, so that different radiation sources are overlapped in frequency range and time domain. Conventional signal representation methods generally follow a pattern recognition framework, and mainly comprise a plurality of independent processing modules such as data preprocessing, feature extraction, feature selection, classifier design and the like, and specifically:
The method comprises the steps of preprocessing signals. For example: data filtering and noise reduction, multipath signal discrimination, carrier frequency estimation, normalization, data alignment and the like;
and extracting features and selecting features. For example: signal transient characteristics, steady-state characteristics, transformation domain characteristics, characteristic optimization, characteristic library establishment and update and the like;
the classifier design method. For example: a variety of classifier designs suitable for engineering applications, etc.;
however, when the characteristics of the signals are extracted by using the traditional method, and then the signals are identified by using classical identification algorithms such as SVM, ELM and the like, the corresponding model performance difference is obvious. The main reason is that the signal conceals the implicit information along with the time change process, and the traditional method cannot fully represent the implicit characteristic, so that the signal characteristic is not fully mined, and the traditional identification method gradually loses effectiveness.
In recent years, the computing power of computers has been greatly enhanced, and deep learning has been rapidly developed. The deep learning automatically learns and adjusts the weight and bias in the neural network according to the back propagation algorithm, and is essentially different from the traditional model-based method in terms of solving the problem, so that the input sample can be adaptively learned, and the characteristic extraction process from the bottom layer to the upper layer, and specifically to the abstract layer can be realized. In addition, the deep learning technologies such as a transducer have strong generalization capability, and the trained deep network model can adapt to complex communication environments and is a powerful tool for effectively representing signals.
However, the deep learning method does not have the capability of displaying the hidden knowledge in the mining from the signal, and the signal characteristics obtained by only implicit mining are difficult to effectively characterize the signal.
Disclosure of Invention
The invention aims to provide a data-driven mutual frequency intelligent intervention signal representation method and system, which are used for solving the problems in the prior art.
In a first aspect, an embodiment of the present invention provides a method for representing a mutual-frequency intelligent intervention signal based on data driving, where the method includes:
obtaining the time-frequency characteristics of the variable frequency signals;
extracting a low-frequency component and a high-frequency component in the time-frequency characteristic;
performing interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component;
the high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
Optionally, the high frequency network comprises 2 high frequency units, and the low frequency network comprises 2 low frequency units;
the 2 low frequency units are a first LSTM network and a third LSTM network respectively; the 2 high frequency units are a second LSTM network and a fourth LSTM network respectively;
the input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the second LSTM network and/or the output of the fourth LSTM network on the output of the first LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network.
Optionally, the first fused component is obtained based on an output intervention operation of the second LSTM network on the output of the first LSTM network, including:
performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component;
and carrying out convolution operation on the first connection component to obtain a first fusion component.
Optionally, the input of the third LSTM network includes a first fused component after the intervention operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network, including:
Performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component;
performing connection operation on the output of the first LSTM network and the output of the fourth LSTM network to obtain a second connection component;
performing connection operation on the first connection component and the second connection component to obtain a third connection component;
performing convolution operation on the third connection component to obtain a first fusion component; the dimensions of the first fused component are the same as the dimensions of the low frequency component.
Optionally, the high frequency network comprises 3 high frequency units, and the low frequency network comprises 3 low frequency units;
the 3 low frequency units are a first LSTM network, a third LSTM network and a fifth LSTM network respectively; the 3 high frequency units are a second LSTM network, a fourth LSTM network and a sixth LSTM network, respectively;
the input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after the intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network;
The input of the fifth LSTM network comprises a third fusion component obtained after the intervention operation of the output of the third LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network;
the input of the sixth LSTM network comprises a fourth fusion component obtained by performing intervention operation on the output of the fourth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network.
Optionally, after the interaction processing is performed on the low-frequency component and the high-frequency component through the pre-trained high-low frequency intervention model to obtain the high-frequency output component and the low-frequency output component, the data-driven mutual frequency intelligent intervention signal representation method further includes:
performing splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal;
performing convolution operation on the spliced signals to obtain recovered time-frequency domain signals;
the dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency characteristic.
In a second aspect, an embodiment of the present invention further provides a data-driven mutual-frequency intelligent intervention signal representation system, where the system includes:
the acquisition module is used for acquiring the time-frequency characteristics of the variable frequency signals;
The extraction module is used for extracting a low-frequency component and a high-frequency component in the time-frequency characteristic;
the interaction module is used for carrying out interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component;
the high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
Optionally, the high frequency network comprises 2 high frequency units, and the low frequency network comprises 2 low frequency units;
the 2 low frequency units are a first LSTM network and a third LSTM network respectively; the 2 high frequency units are a second LSTM network and a fourth LSTM network respectively;
The input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the second LSTM network and/or the output of the fourth LSTM network on the output of the first LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network.
Optionally, the system further comprises:
the splicing module is used for carrying out splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal;
the recovery module is used for carrying out convolution operation on the spliced signals to obtain recovered time-frequency domain signals;
the dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency characteristic.
In a third aspect, an embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor implements the steps of any one of the methods described above when the processor executes the program.
Compared with the prior art, the embodiment of the invention achieves the following beneficial effects:
The embodiment of the invention provides a data-driven mutual-frequency intelligent intervention signal representation method and system, which are used for extracting signal characteristics, extracting low-frequency components and high-frequency components in the time-frequency characteristics by obtaining the time-frequency characteristics of a variable-frequency signal, and carrying out interaction processing on the low-frequency components and the high-frequency components by a pre-trained high-low-frequency intervention model to obtain a high-frequency output component and a low-frequency output component. The high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
By adopting the scheme, the input of one or more low-frequency units is interfered by the output of one or more high-frequency units, the input of one or more high-frequency units is interfered by the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency interference model, namely the interference of the low-frequency components by the high-frequency components and the interference of the high-frequency components by the low-frequency components, the mutual interference and the mutual driving between the high-frequency components and the low-frequency components are realized, the obtained high-frequency output components are fused with the implicit information of the low-frequency components, the low-frequency output components are fused with the implicit information of the high-frequency components, the high-frequency components and the low-frequency components can more accurately represent the signal information, and the signal characteristics obtained based on the high-frequency output components and the low-frequency output components can effectively represent the signal.
Drawings
Fig. 1 is a flowchart of a method for representing a mutual-frequency intelligent intervention signal based on data driving according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a frequency separation process according to an embodiment of the present invention.
Fig. 3 is a basic unit structure diagram of an LSTM network provided in an embodiment of the present invention.
Fig. 4 is a schematic diagram of a high-frequency network intervening low-frequency network according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of a low-frequency network intervening high-frequency network according to an embodiment of the present invention.
Fig. 6 is a schematic diagram of another high frequency network intervention low frequency network provided by an embodiment of the present invention.
Fig. 7 is a schematic diagram of another low frequency network intervention high frequency network provided by an embodiment of the present invention.
Fig. 8 is a schematic block diagram of an electronic device according to an embodiment of the present invention.
The marks in the figure: 500-buses; 501-a receiver; 502-a processor; 503-a transmitter; 504-memory; 505-bus interface.
Detailed Description
Examples
In the prior art, the signal characteristic of the stealth is extracted mainly on the basis of a convolutional neural network, however, the signal characteristic is difficult to effectively characterize by only implicitly mining the signal characteristic by using the convolutional neural network or a long-short-term memory network. The invention considers that the evolution process of the signal can be represented by the frequency, namely the effective dimension, and the low-frequency component and the high-frequency component of the signal are explicitly extracted from the frequency domain, different branches are respectively constructed to represent the low-frequency component and the high-frequency component, and the high-frequency and low-frequency explicit interaction is realized in the model construction process, so that the performance of the original data-driven deep learning method is improved by utilizing the training process of the high-frequency or low-frequency knowledge driven deep learning model.
The present invention will be described in detail with reference to the accompanying drawings.
Example 1
The embodiment of the invention provides a data-driven mutual-frequency intelligent intervention signal representation method, which is used for extracting signal characteristics, wherein the extracted signal characteristics can accurately represent the characteristics of signals, and the data-driven mutual-frequency intelligent intervention signal representation method can also be called as a data-driven mutual-frequency intelligent intervention signal extraction method. As shown in fig. 1, the method includes:
s101: and obtaining the time-frequency characteristic of the frequency-converted signal.
In the embodiment of the present invention, in the illustration of the variable frequency signal, the abscissa axis represents time and the ordinate axis represents time domain amplitude. In the illustration of the time-frequency characteristic, the abscissa indicates time and the ordinate indicates frequency. As shown in fig. 2.
The specific mode for obtaining the time-frequency characteristics of the variable frequency signal is as follows:
the variable frequency signal is processed by carrier frequency estimation, down-conversion, sampling and the like to obtain data s (m), a sliding window W (m) is constructed, and the finite point number in the window is subjected to Fourier transformation to obtain the variable frequency signal, namely
Wherein the variable frequency signal is a non-stationary signal, S (f, k) represents a time-frequency characteristic; m represents the data length of the variable frequency signal, M is a positive integer greater than 2, M represents the sequence number of the data, M is a non-negative integer less than M-1, f represents the frequency, and k is the time (or window sliding step). j is a mathematical representation of the constituent imaginary numbers, called j operator, typically j=sqrt (-1), which represents rotating a complex number 90 degrees counter-clockwise.
S102: the low frequency component and the high frequency component in the time-frequency characteristic are extracted.
In the embodiment of the invention, the low-frequency component represents a component with the frequency lower than or equal to a set threshold in the time-frequency characteristic, and the value of the set threshold can be 10Hz. The high frequency component represents a component having a frequency higher than a set threshold in the time-frequency characteristic. As shown in fig. 2.
S103: and carrying out interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component.
In the embodiment of the invention, the input high-low frequency intervention model can be a high-frequency component sequence containing a plurality of high-frequency components and a low-frequency component sequence containing a plurality of low-frequency components. The plurality of high-frequency components are arranged in time sequence to form a high-frequency component sequence, and the plurality of low-frequency components are arranged in time sequence to form a low-frequency component sequence.
The high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
By adopting the scheme, the input of one or more low-frequency units is interfered by the output of one or more high-frequency units, the input of one or more high-frequency units is interfered by the output of one or more low-frequency units, so that the interaction processing of the low-frequency component and the high-frequency component by the high-frequency interference model, namely the interference of the low-frequency component by the high-frequency component and the interference of the high-frequency component by the low-frequency component, is realized, the obtained high-frequency component fuses the implicit information of the low-frequency component, the low-frequency component fuses the implicit information of the high-frequency component, the high-frequency component and the low-frequency component can more accurately represent the signal information, and the signal characteristics obtained based on the high-frequency component and the low-frequency component can effectively represent the signal.
In the embodiment of the invention, the high-frequency component can be used for intervention on the low-frequency component, or the low-frequency component can be used for intervention on the high-frequency component.
As an optional implementation manner, after step S103, the method for representing a mutual frequency intelligent intervention signal based on data driving further includes:
and performing splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal.
The splicing operation may employ a splicing function concat (high frequency output component, low frequency output component).
And carrying out convolution operation on the spliced signals to obtain recovered time-frequency domain signals.
It can be expressed that the recovered time-frequency domain signal=conv (spliced signal), i.e., conv (high frequency output component, low frequency output component)), wherein the dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency feature.
By adopting the scheme, the obtained recovered time-frequency domain signal can accurately represent the signal characteristics, and the representation capability and accuracy of the recovered frequency domain signal on the signal are improved.
As an alternative embodiment, the high frequency network comprises 2 high frequency units and the low frequency network comprises 2 low frequency units;
the 2 low frequency units are a first LSTM network and a third LSTM network, respectively. The 2 high frequency units are a second LSTM network and a fourth LSTM network, respectively.
Wherein the input of the first LSTM network is a low frequency component and the input of the second LSTM network is a high frequency component.
The input of the third LSTM network comprises a first fusion component obtained by performing intervention operation on the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network. The input of the fourth LSTM network comprises a second fusion component obtained by performing intervention operation on the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network.
In the embodiment of the invention, the first LSTM network, the second LSTM network, the third LSTM network and the fourth LSTM network all adopt Long Short-Term Memory (LSTM) structures. I.e. each low frequency unit and each high frequency unit is an LSTM network.
The LSTM network structure comprises 3 gating mechanisms, namely input gates i t Output door o t And forget door f t . The input gate is used for controlling the input at the current moment and how much information needs to be stored in the state output at the previous moment, the output gate is used for determining how much information needs to be used as the state output at the current moment in the internal state at the current moment, and the forgetting gate is used for controlling how much information needs to be forgotten in the internal state at the previous moment. The specific network structure is shown in fig. 3.
The calculation formulas of the 3 gates of the LSTM network structure are shown as formulas (1), (2) and (3) respectively:
i t ＝σ(W ih h t-1 +W ix x t +b i ) (1)
o t ＝σ(W oh h t-1 +W ox x t +b o ) (2)
f t ＝σ(W fh h t-1 +W fx x t +b f ) (3)
wherein sigma (·) represents a Logistic function, t represents the current time, h t-1 Is the state output of the previous moment, x t Is the input of the current moment, W represents h t-1 And x t The corresponding weight matrix is specifically: w (W) ih Indicating h in the input gate t-1 Weight matrix, W of (2) ix Representing the inputX in entrance t Weight matrix of b) i Representing the bias vector of the input gate. W (W) oh Indicating h in the output gate t-1 Weight matrix, W of (2) ox Representing x in the output gate t Weight matrix of b) o Representing the deviation vector of the output gate. W (W) fh Indicating h in forgetting door t-1 Weight matrix, W of (2) fx Indicating x in forgetting door t Weight matrix of b) f A bias vector representing a forgetting gate.
In LSTM network, candidate state c_in is introduced t That is, the input is converted, and the calculation formula is shown as formula (4):
c_in t ＝tanh(W ch h t-1 +W cx x t +b c ) (4)
in period c_in t Representing candidate states, W, of the LSTM network at time t ch H in table candidate state t-1 Weight matrix, W of (2) cx Representing x in candidate state t Weight matrix of b) c Representing the bias vector in the candidate state.
The candidate state is obtained by a nonlinear activation function.
Internal state c at the present moment t As shown in formula (5):
c t ＝f t ⊙c t-1 +i t ⊙c_in t (5)
wherein c t Indicates the internal state at the current time, +. t-1 Is the internal state at the last moment. c t The control unit transmits information to the external output of the unit while the internal information of the control unit is cyclically transmitted, and the information contains the information accumulated in the previous state.
Output state h of LSTM network t The method comprises the following steps:
h t ＝o t ⊙tanh(c t ) (6)
the output values of the LSTM network 3 gates are all between 0 and 1, and the information transmission is controlled according to a certain proportion. For forgetting door f t The closer the value is to 0, the more the state information is for the previous time The more the amount of forgetting of the rest, but input the gate i t Controlled c_in t The current time state is still affected.
In summary, the operation mode of one LSTM network basic unit is: first, by inputting x t And outputting the state h at the previous time t-1 To calculate 3 gates; then, the state c_in is calculated by the values of the 2 gates t And updates the internal state c at the current time t The method comprises the steps of carrying out a first treatment on the surface of the Finally, calculating the output state h at the current moment through an output gate and a nonlinear function t 。
Since the memory unit in LSTM is to selectively forget the information at the previous time, the time span for storing the information is longer than the short-term memory for rewriting the state at each time, and shorter than the long-term memory for updating the parameters from the whole training data set, so called a long-term and short-term memory network. The 3 gating establishes a self-loop for the internal state of the LSTM cells relative to the recursive computation of the RNN on system state establishment. The gating mechanism of the LSTM can store information expressed for a long time, and establish long-distance time sequence dependency relationship, so that time sequence feature learning is realized.
After the above description of the memory cells in the LSTM, it can be known that the inputs of the LSTM structure cell include the state output h at the previous time (last time) t-1 And x t The third LSTM network is an LSTM structure, and the operation mode of the third LSTM network is already described above. In the embodiment of the present invention, the input of the third LSTM network includes a first fusion component obtained after the intervention operation on the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network, and then the input and output of the third LSTM network are described in detail below with reference to fig. 4.
As an optional implementation manner, the interaction processing is performed on the low-frequency component and the high-frequency component through a pre-trained high-frequency and low-frequency intervention model to obtain a high-frequency output component and a low-frequency output component, which includes:
and performing intervention operation on the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network to obtain a first fusion component. As shown in fig. 4.
The first fused component is taken as an input to a third LSTM network.
And performing intervention operation on the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network to obtain a second fusion component. As shown in fig. 5.
The second fused component is used as an input to a fourth LSTM network.
Optionally, the output of the third LSTM network is taken as the last output low frequency output component. High frequency output component with output of fourth LSTM network as final output
As another alternative embodiment, the output of the third LSTM network is interfered based on the output of the second LSTM network and/or the output of the fourth LSTM network to obtain a final output low-frequency output component. And performing intervention operation on the output of the fourth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network to obtain a finally output high-frequency component.
In combination with the description of the LSTM network structure, in the embodiment of the present application, for the third LSTM network, the input includes the low frequency component and the first fusion component at the current time, that is, the first fusion component is used as h in the formulas (1) - (6) t-1 With the low frequency component at the current time as x t H is output in formula (6) t Namely the low frequency output component extracted by the embodiment of the application.
Similarly, for the fourth LSTM network, the input includes the high frequency component at the current time and the second fused component, i.e., the second fused component is used as h in the above formulas (1) - (6) t-1 With the high frequency component at the current time as x t H is output in formula (6) t Namely, the high-frequency output component extracted by the embodiment of the application.
As an alternative embodiment, performing an intervention operation on the output of the first LSTM network based on the output of the second LSTM network to obtain a first fusion component includes:
Performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component; performing convolution operation on the first connection component to obtain a first fusion component, and referring to formula (7):
l1＝conv(concat(H1,L1)) (7)
where L1 represents a first fused component, H1 represents an output of a second LSTM network, L1 represents an output of the first LSTM network, concat () represents a connection operation for the output of the first LSTM network and the output of the second LSTM network, conv () represents a convolution operation for the main purpose of reducing the first connected component so that the dimension of the first connected component is the same as the output of the first LSTM network.
The specific operation of obtaining the first fused component by performing the intervention operation on the output of the first LSTM network based on the output of the fourth LSTM network is the same as the first fused component obtained by performing the intervention operation on the output of the first LSTM network based on the output of the second LSTM network, and only the output of the fourth LSTM network needs to be represented by H1.
Performing an intervening operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network to obtain a first fused component, including:
performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component;
Performing connection operation on the output of the first LSTM network and the output of the fourth LSTM network to obtain a second connection component;
performing connection operation on the first connection component and the second connection component to obtain a third connection component;
performing convolution operation on the third connection component to obtain a first fusion component; the dimensions of the first fused component are the same as the dimensions of the low frequency component.
Specifically, the operation mode of performing the intervention operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network to obtain the first fusion component is shown in formula (8):
l1＝conv(concat(concat(H1,L1),concat(H2,L1))) (8)
where H2 represents the output of the fourth LSTM network.
As an alternative embodiment, the operation of performing the intervention operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network to obtain the first fusion component is as shown in formula (9):
l1＝conv(concat(conv(concat(H1,L1)),conv(concat(H2,L1)))) (9)
that is, performing an intervention operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network to obtain a first fused component includes: performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component; performing convolution operation on the first connection component to obtain a first dimension reduction component, wherein the dimension of the first dimension reduction component is the same as the dimension of the output of the first LSTM network; performing connection operation on the output of the first LSTM network and the output of the fourth LSTM network to obtain a second connection component; performing convolution operation on the second connection component to obtain a second dimension reduction component, wherein the dimension of the second dimension reduction component is the same as the dimension of the output of the first LSTM network; performing connection operation on the first dimension reduction component and the second dimension reduction component to obtain a third connection component; performing convolution operation on the third connection component to obtain a first fusion component; the dimension of the first fused component is the same as the dimension of the output of the first LSTM network.
In the embodiment of the invention, for the operation of obtaining the first fusion component, under the condition of the same dimension, splicing is performed first and then convolution is performed; under the condition of different dimensions, convolution is performed first, then splicing is performed, and then convolution is performed.
In the embodiment of the present invention, the specific implementation manner of performing the intervention operation on the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network to obtain the second fusion component is similar to the specific implementation manner of performing the intervention operation on the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network to obtain the first fusion component, and the specific reference is made to the above-described manner and will not be repeated herein.
By the scheme, the interference of the low-frequency component on the high-frequency component and the interference of the high-frequency component on the low-frequency component can be realized, the extracted high-frequency output component and low-frequency output component can accurately represent the characteristics of the signal, and the effectiveness of signal extraction is improved.
As a further alternative embodiment, the high frequency network comprises 3 high frequency units and the low frequency network comprises 3 low frequency units;
the 3 low frequency units are a first LSTM network, a third LSTM network and a fifth LSTM network respectively; the 3 high frequency units are a second LSTM network, a fourth LSTM network and a sixth LSTM network, respectively;
The input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component.
The input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network; the input of the fifth LSTM network comprises a third fusion component obtained after intervention operation is performed on the output of the third LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after the intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network;
the input of the sixth LSTM network comprises a fourth fusion component obtained by performing intervention operation on the output of the fourth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network.
I.e. the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network in the low frequency network is interfered by the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network in the high frequency network, as shown in fig. 6. And performing intervention operation on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network in the high-frequency network through the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network in the low-frequency network. As shown in fig. 7.
In the embodiment of the present invention, the interaction processing is performed on the low frequency component and the high frequency component through the pre-trained high and low frequency intervention model, so as to obtain a high frequency output component and a low frequency output component, which may further include:
and performing intervention operation on the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network to obtain a first fusion component.
According to the formulas (1) - (6), the first fusion component is used as h in the input of the third LSTM network t-1 . The specific manner of the intervention operation may be referred to in the following manner, in addition to any one of the manners shown in the above formulas (7) to (9):
the first fusion component may be obtained in a manner as shown in any one of formulas (10), (11):
l1＝conv(concat(concat(H1,L1),concat(H2,L1，concat(H3,L1))) (10)
l1＝conv(concat(conv(concat(H1,L1)),conv(concat(H2,L1))，conv(concat(H3,L1))))(11)
where H3 represents the output of the sixth LSTM network.
And performing an intervention operation on the output of the third LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network to obtain a third fusion component.
According to the formulas (1) - (6), the third fusion component is used as h in the input of the fifth LSTM network t-1 。
And performing an intervention operation on the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network to obtain a second fusion component.
Taking the second fusion component as h in the input of the fourth LSTM network according to the formulas (1) - (6) t-1 。
And performing an intervention operation on the output of the fourth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network to obtain a fourth fusion component.
Taking the fourth fusion component as h in the input of the sixth LSTM network according to the formulas (1) - (6) t-1 。
Alternatively, the output of the fifth LSTM network is taken as the last output low-frequency output component, and the output of the sixth LSTM network is taken as the last output high-frequency output component.
In the embodiment of the present invention, the obtaining manners of the second fusion component, the third fusion component and the fourth fusion component are similar to those of the first fusion component, and are not described herein.
As another alternative implementation manner, the output of the fifth LSTM network is obtained after the intervention operation is performed on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network, and the finally output high-frequency output component is obtained after the intervention operation is performed on the output of the sixth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network.
In summary, in the method for representing a mutual-frequency intelligent intervention signal based on data driving provided by the embodiment of the invention, the high-frequency intervention model includes a high-frequency network and a low-frequency network, the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component;
the high-frequency network comprises N high-frequency LSTM networks, and the low-frequency network comprises N low-frequency LSTM networks; n high-frequency LSTM networks are connected in series end to end, and N low-frequency LSTM networks are connected in series end to end; n is a positive integer greater than or equal to 2;
in the low frequency network, the input of the 1 st low frequency LSTM network is the low frequency component at the 1 st moment, and the input of the nth low frequency LSTM network comprises: the nth high frequency intervenes the low frequency fusion component and the low frequency component at the nth moment; the nth high-frequency intervention low-frequency fusion component is a component obtained by performing intervention operation on the output of the (n-1) th low-frequency LSTM network based on the output of all the high-frequency LSTM networks in the high-frequency network;
in the high frequency network, the input of the 1 st high frequency LSTM network is the high frequency component of the 1 st moment, and the input of the nth high frequency LSTM network includes: the nth low frequency intervenes the high frequency fusion component and the high frequency component at the nth moment; the nth low-frequency intervention high-frequency fusion component is a component obtained by performing intervention operation on the output of the (n-1) th high-frequency LSTM network based on the output of all the low-frequency LSTM networks in the low-frequency network; n is a positive integer greater than 1 and less than or equal to N.
Both the high frequency LSTM network and the low frequency LSTM network are LSTM network structures as shown in fig. 2.
Intervening on the output of the n-1 th low frequency LSTM network based on the outputs of all of the high frequency LSTM networks in the high frequency network, comprising:
respectively connecting the output of the (N-1) th low-frequency LSTM network with the output of each high-frequency LSTM network to obtain N first connection components;
performing convolution dimension reduction operation on the first connection component to obtain a first convolution component; the dimension of the first convolution component is the same as the dimension of the output of the n-1 th low frequency LSTM network; n first convolution components are correspondingly obtained by the N first connection components;
performing connection operation on the N first convolution components to obtain convolution connection components;
performing convolution dimension reduction operation on the convolution connection component to obtain an nth low-frequency interference high-frequency fusion component; the dimension of the nth low frequency intervening high frequency fusion component is the same as the dimension of the output of the nth-1 low frequency LSTM network.
The first LSTM network, the third LSTM network and the fifth LSTM network are low-frequency LSTM networks; the second LSTM network, the fourth LSTM network, and the sixth LSTM network are high frequency LSTM networks.
The interaction operation and the interaction operation between the high-frequency signal and the low-frequency signal in the high-frequency and low-frequency interference model will be described with reference to fig. 6 and 7.
For low frequency networks:
for time t-1, the input includes x t-1 ,x t-1 Indicating the low frequency component at time t-1.
For the t time, t is a positive integer greater than 2The input of the nth low-frequency LSTM network corresponding to the t moment is the low-frequency component x of the t moment t And output of high frequency LSTM network of high frequency network to output h of n-1 th low frequency LSTM network t-1 Results after intervention operations t I.e. in l t Instead of h in formulas (1) to (6) t-1 。l t The obtaining mode of l1 is as shown in any one of formulas (7) to (9).
For the t+1 time, then the input of the n+1 low frequency LSTM network corresponding to the t+1 time is the low frequency component x of the t+1 time t+1 And output of high frequency LSTM network of high frequency network to output h of nth low frequency LSTM network t Results after intervention operations t+1 I.e. in l t+1 Instead of h in formulas (1) to (6) t-1 。l t+1 The obtaining method of (2) is the obtaining method of l1 shown in any one of formulas (7) to (11).
For high frequency networks:
for time t-1, the input includes y t-1 ,y t-1 Representing the high frequency component at time t-1.
For the t moment, t is a positive integer greater than 2, then the input of the nth high-frequency LSTM network corresponding to the t moment is the high-frequency component y of the t moment t And output of low frequency LSTM network of low frequency network to output z of n-1 high frequency LSTM network t-1 Results g after intervention operations t In g t Instead of h in formulas (1) to (6) t-1 。g t The obtaining mode of l1 is as shown in any one of formulas (7) to (9).
For the t+1 time, then the input of the n+1-th high-frequency LSTM network corresponding to the t+1 time is the high-frequency component y of the t+1 time t+1 And output of low frequency LSTM network of low frequency network to output z of nth high frequency LSTM network t Results g after intervention operations t In g t Instead of h in formulas (1) to (6) t-1 。g t The obtaining method of (2) is the obtaining method of l1 shown in any one of formulas (7) to (11).
In summary, the technical schemes shown in fig. 6 and fig. 7 fully demonstrate the interaction between the high-frequency signal and the low-frequency signal, and the high-frequency interference model and the low-frequency interference model are trained through the interaction of the high-frequency component and the low-frequency component, so that the training process of the high-frequency or low-frequency knowledge-driven deep learning model is realized, the performance of the original data-driven deep learning method is improved, and the obtained high-frequency output component and low-frequency output component can accurately represent the signal characteristics.
In the embodiment of the invention, the high-frequency network and the low-frequency network are trained simultaneously, namely, the interaction between the high-frequency component and the low-frequency component, the high-frequency network and the low-frequency network are performed simultaneously until the output of the high-frequency network and the low-frequency network are converged, and then the network training is finished.
Example 2
Based on the method for representing the data-driven inter-frequency intelligent intervention signal provided by the embodiment, the embodiment of the invention provides a system for representing the data-driven inter-frequency intelligent intervention signal, which is used for executing the method and comprises the following steps:
the obtaining module is used for obtaining the time-frequency characteristics of the frequency-converted signals.
And the extraction module is used for extracting the low-frequency component and the high-frequency component in the time-frequency characteristic.
And the interaction module is used for carrying out interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component.
And the splicing module is used for carrying out splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal.
And the recovery module is used for carrying out convolution operation on the spliced signals to obtain recovered time-frequency domain signals.
The dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency characteristic. The high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component. The high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
The detailed manner in which the respective modules perform the operations has been described in detail in relation to the above-described embodiments of the method, and will not be explained in detail here.
The embodiment of the invention also provides a complex signal multi-component interaction characteristic signal processing system, as shown in fig. 8, which comprises a memory 504, a processor 502 and a computer program stored in the memory 504 and capable of running on the processor 502, wherein the processor 502 implements the steps of any one of the data-driven inter-frequency intelligent intervention signal representation methods when executing the program.
Where in FIG. 8 a bus architecture (represented by bus 500), bus 500 may include any number of interconnected buses and bridges, with bus 500 linking together various circuits, including one or more processors, represented by processor 502, and memory, represented by memory 504. Bus 500 may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., as are well known in the art and, therefore, will not be described further herein. Bus interface 505 provides an interface between bus 500 and receiver 501 and transmitter 503. The receiver 501 and the transmitter 503 may be the same element, i.e. a transceiver, providing a means for communicating with various other apparatus over a transmission medium. The processor 502 is responsible for managing the bus 500 and general processing, while the memory 504 may be used to store data used by the processor 502 in performing operations.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, realizes the steps of any one of the above-mentioned data-driven mutual frequency intelligent intervention signal representation methods.
The algorithms and displays presented herein are not inherently related to any particular computer, virtual system, or other apparatus. Various general-purpose systems may also be used with the teachings herein. The required structure for a construction of such a system is apparent from the description above. In addition, the present invention is not directed to any particular programming language. It will be appreciated that the teachings of the present invention described herein may be implemented in a variety of programming languages, and the above description of specific languages is provided for disclosure of enablement and best mode of the present invention.
In the description provided herein, numerous specific details are set forth. However, it is understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.
Similarly, it should be appreciated that in the foregoing description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof for the purpose of streamlining the disclosure and aiding in the understanding of one or more of the various inventive aspects. However, the disclosed method should not be construed as reflecting the intention that: i.e., the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of this invention.
Those skilled in the art will appreciate that the modules in the apparatus of the embodiments may be adaptively changed and disposed in one or more apparatuses different from the embodiments. The modules or units or components of the embodiments may be combined into one module or unit or component and, furthermore, they may be divided into a plurality of sub-modules or sub-units or sub-components. Any combination of all features disclosed in this specification (including any accompanying claims, abstract and drawings), and all of the processes or units of any method or apparatus so disclosed, may be used in combination, except insofar as at least some of such features and/or processes or units are mutually exclusive. Each feature disclosed in this specification (including any accompanying claims, abstract and drawings), may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.
Furthermore, those skilled in the art will appreciate that while some embodiments herein include some features but not others included in other embodiments, combinations of features of different embodiments are meant to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.
Claims (10)
1. A method for representing a mutual-frequency intelligent intervention signal based on data driving, which is characterized by comprising the following steps:
obtaining the time-frequency characteristics of the variable frequency signals;
extracting a low-frequency component and a high-frequency component in the time-frequency characteristic;
performing interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component;
the high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
2. The data-driven mutual-frequency intelligent intervention signal representation method according to claim 1, wherein the high-frequency network comprises 2 high-frequency units, and the low-frequency network comprises 2 low-frequency units;
The 2 low frequency units are a first LSTM network and a third LSTM network respectively; the 2 high frequency units are a second LSTM network and a fourth LSTM network respectively;
the input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the second LSTM network and/or the output of the fourth LSTM network on the output of the first LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network.
3. The method for expressing a data-driven mutual-frequency intelligent intervention signal according to claim 2, wherein the step of obtaining a first fusion component based on the intervention operation of the output of the second LSTM network to the output of the first LSTM network comprises the steps of:
performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component;
and carrying out convolution operation on the first connection component to obtain a first fusion component.
4. The data-driven inter-frequency intelligent intervention signal representation method of claim 2, wherein the input of the third LSTM network comprises a first fused component after an intervention operation on the output of the first LSTM network based on the output of the second LSTM network and the output of the fourth LSTM network, comprising:
Performing connection operation on the output of the first LSTM network and the output of the second LSTM network to obtain a first connection component;
performing connection operation on the output of the first LSTM network and the output of the fourth LSTM network to obtain a second connection component;
performing connection operation on the first connection component and the second connection component to obtain a third connection component;
performing convolution operation on the third connection component to obtain a first fusion component; the dimensions of the first fused component are the same as the dimensions of the low frequency component.
5. The data-driven mutual-frequency intelligent intervention signal representation method according to claim 1, wherein the high-frequency network comprises 3 high-frequency units, and the low-frequency network comprises 3 low-frequency units;
the 3 low frequency units are a first LSTM network, a third LSTM network and a fifth LSTM network respectively; the 3 high frequency units are a second LSTM network, a fourth LSTM network and a sixth LSTM network, respectively;
the input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the first LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network;
The input of the fourth LSTM network comprises a second fusion component obtained after the intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network;
the input of the fifth LSTM network comprises a third fusion component obtained after the intervention operation of the output of the third LSTM network based on the output of the second LSTM network and/or the output of the fourth LSTM network and/or the output of the sixth LSTM network;
the input of the sixth LSTM network comprises a fourth fusion component obtained by performing intervention operation on the output of the fourth LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network and/or the output of the fifth LSTM network.
6. The method for representing the data-driven mutual-frequency intelligent intervention signal according to claim 1, wherein after the interaction processing is performed on the low-frequency component and the high-frequency component through the pre-trained high-low frequency intervention model to obtain the high-frequency output component and the low-frequency output component, the method for representing the data-driven mutual-frequency intelligent intervention signal further comprises:
performing splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal;
Performing convolution operation on the spliced signals to obtain recovered time-frequency domain signals;
the dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency characteristic.
7. A data-driven inter-frequency intelligent intervention signal representation system, the system comprising:
the acquisition module is used for acquiring the time-frequency characteristics of the variable frequency signals;
the extraction module is used for extracting a low-frequency component and a high-frequency component in the time-frequency characteristic;
the interaction module is used for carrying out interaction processing on the low-frequency component and the high-frequency component through a pre-trained high-low frequency intervention model to obtain a high-frequency output component and a low-frequency output component;
the high-low frequency intervention model comprises a high-frequency network and a low-frequency network, wherein the input of the high-frequency network is a high-frequency component, the input of the low-frequency network is a low-frequency component, the output of the high-frequency network is a high-frequency output component, and the output of the low-frequency network is a low-frequency output component; the high-frequency network comprises a plurality of high-frequency units, the low-frequency network comprises a plurality of low-frequency units, the input of one or more low-frequency units is subjected to intervention operation through the output of one or more high-frequency units, and the input of one or more high-frequency units is subjected to intervention operation through the output of one or more low-frequency units, so that the interaction processing of the low-frequency components and the high-frequency components by the high-frequency and low-frequency intervention model is realized.
8. The data-driven mutual-frequency intelligent intervention signal representation system of claim 7, wherein the high-frequency network comprises 2 high-frequency units and the low-frequency network comprises 2 low-frequency units;
the 2 low frequency units are a first LSTM network and a third LSTM network respectively; the 2 high frequency units are a second LSTM network and a fourth LSTM network respectively;
the input of the first LSTM network is the low frequency component and the input of the second LSTM network is the high frequency component;
the input of the third LSTM network comprises a first fusion component obtained after the intervention operation of the output of the second LSTM network and/or the output of the fourth LSTM network on the output of the first LSTM network;
the input of the fourth LSTM network comprises a second fusion component obtained after intervention operation of the output of the second LSTM network based on the output of the first LSTM network and/or the output of the third LSTM network.
9. The data-driven, inter-frequency intelligent intervention signal representation system of claim 8, wherein the system further comprises:
the splicing module is used for carrying out splicing operation on the high-frequency output component and the low-frequency output component to obtain a spliced signal;
the recovery module is used for carrying out convolution operation on the spliced signals to obtain recovered time-frequency domain signals;
The dimension of the recovered time-frequency domain signal is the same as the dimension of the time-frequency characteristic.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the method according to any one of claims 1-6 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211137836.7A CN116016068B (en) | 2022-09-19 | 2022-09-19 | Data-driven-based mutual-frequency intelligent intervention signal representation method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211137836.7A CN116016068B (en) | 2022-09-19 | 2022-09-19 | Data-driven-based mutual-frequency intelligent intervention signal representation method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116016068A CN116016068A (en) | 2023-04-25 |
| CN116016068B true CN116016068B (en) | 2023-08-11 |
Family
ID=86036018
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211137836.7A Active CN116016068B (en) | 2022-09-19 | 2022-09-19 | Data-driven-based mutual-frequency intelligent intervention signal representation method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116016068B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950759A (en) * | 2020-05-18 | 2020-11-17 | 武汉科技大学 | Short-term wind speed prediction method based on two-stage decomposition, LSTM and AT |
| CN112818891A (en) * | 2021-02-10 | 2021-05-18 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Intelligent identification method for communication interference signal type |
| CN114547795A (en) * | 2022-02-21 | 2022-05-27 | 苏州科技大学 | Rolling bearing residual life prediction method based on data driving |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11521299B2 (en) * | 2020-10-16 | 2022-12-06 | Adobe Inc. | Retouching digital images utilizing separate deep-learning neural networks |
-
2022
- 2022-09-19 CN CN202211137836.7A patent/CN116016068B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111950759A (en) * | 2020-05-18 | 2020-11-17 | 武汉科技大学 | Short-term wind speed prediction method based on two-stage decomposition, LSTM and AT |
| CN112818891A (en) * | 2021-02-10 | 2021-05-18 | 西南电子技术研究所（中国电子科技集团公司第十研究所） | Intelligent identification method for communication interference signal type |
| CN114547795A (en) * | 2022-02-21 | 2022-05-27 | 苏州科技大学 | Rolling bearing residual life prediction method based on data driving |
Non-Patent Citations (1)
| Title |
|---|
| 基于改进长短期记忆网络的时间序列预测研究;陈孝文等;《武汉理工大学学报》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116016068A (en) | 2023-04-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Weak in the NEES?: Auto-tuning Kalman filters with Bayesian optimization | |
| Li et al. | A novel nonlinear optimization method for fitting a noisy Gaussian activation function | |
| CN111310828B (en) | Target detection model fine-tuning method and device for ADAS scene | |
| CN104331635A (en) | Predication method for electric power optical fiber communication optical power | |
| Castellano et al. | Reinforcement learning with almost sure constraints | |
| CN116016068B (en) | Data-driven-based mutual-frequency intelligent intervention signal representation method and system | |
| CN116628468A (en) | Electromagnetic signal sensing method capable of being driven by interpretation of sample attack defense effect | |
| Tollner et al. | Demonstrating a new evaluation method on ReLU based Neural Networks for classification problems | |
| DE112022005940T5 (en) | IDENTIFICATION OF TARGETS FOR INDIRECT FUNCTION CALLS IN SOFTWARE | |
| US20230004799A1 (en) | One-dimensional-convolution-based signal classifier | |
| Kafka et al. | An empirical study into finding optima in stochastic optimization of neural networks | |
| CN114925618A (en) | Vehicle speed prediction method, vehicle speed prediction device, computer equipment and storage medium | |
| Schön et al. | Bayesian approach to temporal logic control of uncertain systems | |
| CN114611990A (en) | Method and device for evaluating contribution rate of element system of network information system | |
| Peng et al. | A fast algorithm for sparse support vector machines for mobile computing applications | |
| CN118462050B (en) | Vehicle-mounted reverse circulation drilling machine and method suitable for sampling loose stratum in shallow coverage area | |
| Qin | Estimation and identification of nonlinear parameter of motion index based on least squares algorithm | |
| CN114019944B (en) | Joint interval estimation method for networked control system state and fault under FDI attack | |
| CN117093853B (en) | Time sequence data processing method and device, computer readable medium and electronic equipment | |
| Krivec et al. | Deep Neural Input-Filtered Gaussian Process Model | |
| CN118534862B (en) | Oil field auxiliary agent production control system and method based on full automation | |
| CN118364270B (en) | Domain generalization modeling method, device, equipment and medium for industrial equipment health prediction | |
| Galtier et al. | Relative entropy minimizing noisy non-linear neural network to approximate stochastic processes | |
| CN117951200A (en) | Space-time sequence data complement method, equipment and medium | |
| Selvarajan et al. | Process Model Inversion in the Data-Driven Engineering Context for Improved Parameter Sensitivities. Processes 2022, 10, 1764 |
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 |