LSTM typhoon prediction method and system based on two-dimensional sliding energy segmentation preprocessing

Abstract

The invention discloses an LSTM typhoon prediction method and system based on two-dimensional sliding energy segmentation preprocessing. The method comprises the steps of obtaining optical phase data; performing time domain segmentation on the optical phase data according to a time window to obtain segmented phase signals; performing GST conversion on the segmented single-period phase signal to obtain a time-frequency spectrum; calculating a power spectrum of the segmented single-period phase signal, searching a maximum value of the power spectrum and setting a threshold parameter so as to determine a frequency range of a low-frequency signal; setting a local sliding window in the frequency direction of the time-frequency spectrum, recording the frequency corresponding to the maximum value position of the module value of the time-frequency spectrum, and setting the rest frequencies to zero to obtain the instantaneous frequency of the low-frequency-band typhoon signal; and inputting the instantaneous frequency of the low-frequency-band typhoon signal as a data set into the LSTM network, and performing typhoon weather prediction by using the trained LSTM network. The typhoon prediction method solves the problem of low typhoon prediction precision caused by typhoon vibration signal multi-factor coupling and difficulty in low and medium frequency feature extraction in existing typhoon prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of meteorological forecasting, and in particular relates to an LSTM typhoon prediction method and system based on two-dimensional sliding energy segmentation preprocessing. Background Art

[0002] Typhoons are common natural disasters in my country's coastal areas. When they occur, they can easily destroy houses and power facilities, cause falling objects and collapsed trees, and the extreme wind speeds that accompany them may damage the structure of offshore wind turbines, causing tower buckling or nacelle overturning, and even trigger secondary disasters such as floods, mudslides, and urban waterlogging. Therefore, real-time observation and prediction of typhoon weather is of great significance.

[0003] Currently, high-resolution satellite remote sensing technology and weather radar are generally used to monitor typhoons' location, intensity, movement path and other information in real time. High-performance computers and numerical weather forecast models are used, combined with atmospheric motion simulations, to predict the future frequency of typhoons. These models are the core tools for typhoon path prediction. However, existing numerical forecast models still have shortcomings in typhoon frequency prediction. Direct ocean observation data is scarce, the density of ground meteorological stations is insufficient, and observation capabilities are limited, especially in remote areas and islands. This makes it difficult to achieve refined forecasts of typhoon frequency. In addition, although remote sensing technologies such as satellites and radars have improved detection capabilities, their resolution and coverage in typhoon weather are still limited. Meteorological satellites have "looking down" blind spots, and radar detection range is limited and susceptible to signal attenuation.

[0004] Fiber-optic sensing technology, through Rayleigh, Brillouin, and Raman scattering, can simultaneously detect changes in multiple physical quantities, such as temperature, vibration, and strain. Distributed sensing is easily implemented, thus meeting the needs for multi-parameter sensing and long-distance monitoring of typhoon weather, while also addressing the limited spatial resolution of traditional technologies. Fiber-optic sensing networks can cover long distances (up to 100 km for a single channel) without blind spots. For example, deploying a distributed fiber-optic system along a coastline can provide continuous spatial monitoring. However, while fiber-optic sensing systems can capture a wealth of meteorological data, typhoon vibration signals exhibit unique physical characteristics. Their generation mechanism is closely related to the coupling of multiple factors, including strong winds, waves, and structural resonance. Typhoon-induced vibration signals, in particular, are primarily concentrated in the low- and medium-frequency bands. As a typhoon approaches, the dominant frequency of the vibration signal gradually decreases, returning to normal levels after the typhoon passes. Therefore, analyzing fiber-optic vibration signals is a key technology in typhoon forecasting.

[0005] Chinese patent publication number CN116822716A discloses a typhoon prediction method, system, device and medium based on spatiotemporal attention. The method obtains sea surface variable characteristics by inputting the typhoon data to be predicted into a preset two-dimensional convolutional network to learn the spatial characteristics of sea surface variables; inputs the typhoon data to be predicted into a preset three-dimensional convolutional network to learn the spatial characteristics of atmospheric variables to obtain atmospheric variable characteristics; wherein the three-dimensional convolutional network includes multiple cascaded three-dimensional convolutional layers, and a three-dimensional spatiotemporal attention module combining spatial attention mechanism and channel attention mechanism is provided between at least one pair of adjacent three-dimensional convolutional layers; merges and connects the sea surface variable characteristics and the atmospheric variable characteristics to obtain merged characteristics; inputs the merged characteristics into a preset long-short-term memory network for feature learning to obtain typhoon time series characteristics; and predicts the intensity of the typhoon based on the typhoon time series characteristics. The method described in this invention is based on sea surface and atmospheric variable data, and it is difficult to solve the problem of limited prediction accuracy caused by the scarcity of observation data in the ocean and remote areas; and the method does not design a preprocessing mechanism for the multi-factor coupled vibration signals unique to typhoons (such as medium and low frequency characteristics), and the spatiotemporal attention module mentioned in the method cannot effectively separate noise and typhoon key frequency components. Summary of the Invention

[0006] The purpose of the present invention is to provide an LSTM typhoon prediction method and system based on two-dimensional sliding energy segmentation preprocessing to solve the problem of low typhoon prediction accuracy in existing typhoon prediction technology due to insufficient observation data in oceans and remote areas, multi-factor coupling of typhoon vibration signals, and difficulty in extracting medium and low-frequency features.

[0007] The technical solutions of the present invention are as follows:

[0008] In one aspect, the present invention provides an LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing, comprising the following steps:

[0009] Obtain optical phase data caused by external field vibration based on distributed optical fiber sensing array;

[0010] Select a segmentation period based on the time period of the microseismic signal, and segment the optical phase data into multiple time periods according to the time window to obtain the segmented phase signal;

[0011] Perform GST transformation on the segmented single-period phase signal to obtain the time-frequency spectrum;

[0012] Calculate the power spectrum of the segmented single-period phase signal, find the maximum value of the power spectrum and set the threshold parameter, and retain the single-period phase signal whose power spectrum is less than the threshold parameter multiplied by the power spectrum to determine the frequency range of the low-frequency signal distribution;

[0013] The frequency range of the low-frequency signal distribution is defined as the sliding window width. Based on the sliding window width, a local sliding window is set in the frequency direction of the time-frequency spectrum. The maximum value position of the time-frequency spectrum modulus value within the local sliding window is searched at each moment. The corresponding frequency of the maximum value position of the time-frequency spectrum modulus value is recorded and the remaining frequencies are set to zero to obtain the instantaneous frequency of the low-frequency typhoon signal.

[0014] The instantaneous frequency of low-frequency typhoon signals in different frequency bands is input into the LSTM network as a data set for training to obtain a trained LSTM network, which is then used to perform typhoon weather forecasting.

[0015] Preferably, the GST transformation of the segmented single-period phase signal is expressed as:

[0016]

[0017] Where S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [T N-1 ,T N ] is any single period of time segmentation; For the first [T N-1 ,T N ] Phase signal within a single period; t is the time within a single period; e is a natural constant, e -i2πft is the correction factor; i is the imaginary unit; θ is the power parameter, and u is the parameter coefficient, which together adjust the width and height of the Gaussian window.

[0018] Preferably, the power spectrum of the segmented single-period phase signal is calculated as follows:

[0019]

[0020] Where P(f) is the power spectrum; d(n) is the sliding window function; μ is the sampling rate; N-1 is the time period number of the segment, [T N-1 ,T N ] is any single time period of time division; i=1,2,…,L, the first [T N-1 ,T N The phase signal in a single time period is split into L segments, each segment contains M discrete data points, n = 1, 2, ..., M; U is the normalization factor obtained by processing d(n); For the first [T N-1 ,T N ]The phase signal of the nth point in the i-th segment within a single time period.

[0021] Preferably, finding the maximum value of the power spectrum is expressed as:

[0022] P(f) max =max([P(f)])

[0023] Where P(f) max is the maximum value of the time spectrum modulus in the local sliding window at each moment; max() is the maximum value function; P(f) is the power spectrum; [P(f)] is the modulus value of the power spectrum;

[0024] Set the threshold parameter b, when P(f)≥b*P(f) max When P(f) is zeroed, the high-frequency noise corresponding to P(f) is retained. <b*P(f) max The single-period phase signal of the low-frequency signal is used to determine the frequency range in which the low-frequency signal is distributed.

[0025] Preferably, the maximum position of the time-frequency spectrum modulus value in the local sliding window at each moment is searched, the frequency corresponding to the maximum position of the time-frequency spectrum modulus value is recorded, and the remaining frequencies are set to zero, so as to obtain the instantaneous frequency of the low-frequency typhoon signal:

[0026]

[0027] Where, f tN is the instantaneous frequency of the low-frequency typhoon signal; argmax() is the maximum value function; S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [S g (τ,f)] is the time-frequency spectrum modulus; f0 is the preset frequency constant; a is the sliding window width parameter.

[0028] Preferably, the instantaneous frequencies of low-frequency typhoon signals in different frequency bands are input into the LSTM network as data sets to perform typhoon weather forecasting, and the typhoon weather forecast output result is specifically as follows:

[0029]

[0030] Where c t is the cell state of the LSTM network; K f For the forget gate; c t-1 is the cell state of the LSTM network at the previous moment; K i is the input gate; K is the new input data obtained by multiplying the instantaneous frequency of the input low-frequency typhoon signal by the weight matrix W through the activation function; h t is the hidden state of the LSTM network at the previous moment; K o is the output gate; tanh() is the hyperbolic tangent activation function; y t is the typhoon weather forecast output result; σ is the Sigmoid activation function; W is the weight matrix.

[0031] On the other hand, the present invention provides an LSTM typhoon prediction system based on two-dimensional sliding energy segmentation preprocessing, including a data acquisition module, a time domain segmentation module, a time-frequency analysis module, a power spectrum analysis module, an instantaneous frequency extraction module and an LSTM network training and prediction module.

[0032] The data acquisition module is used to obtain optical phase data caused by external field vibration based on a distributed optical fiber sensor array.

[0033] The time domain segmentation module is used to select a segmentation period according to the time period of the microseismic signal, and to segment the optical phase data into multiple time periods according to the time window to obtain segmented phase signals.

[0034] The time-frequency analysis module is used to perform GST transformation on the segmented single-period phase signal to obtain a time-frequency spectrum.

[0035] The power spectrum analysis module is used to calculate the power spectrum of the segmented single-period phase signal, find the maximum value of the power spectrum and set the threshold parameter, and retain the single-period phase signal whose power spectrum is less than the threshold parameter multiplied by the power spectrum to determine the frequency range of the low-frequency signal distribution.

[0036] The instantaneous frequency extraction module is used to define the frequency range in which the low-frequency signal is distributed as the sliding window width, set a local sliding window in the frequency direction of the time-frequency spectrum based on the sliding window width, search for the maximum position of the time-frequency spectrum modulus value within the local sliding window at each moment, record the frequency corresponding to the maximum position of the time-frequency spectrum modulus value and set the remaining frequencies to zero to obtain the instantaneous frequency of the low-frequency typhoon signal.

[0037] The LSTM network training and prediction module is used to input the instantaneous frequency of low-frequency typhoon signals in different frequency bands as a data set into the LSTM network for training, obtain a trained LSTM network, and use the trained LSTM network to perform typhoon weather forecasts.

[0038] On the other hand, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored on the memory and runnable on the processor, wherein when the processor executes the computer program, it implements the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in any embodiment of the present invention.

[0039] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in any embodiment of the present invention.

[0040] Compared with the prior art, the present invention has the following technical effects:

[0041] This method accurately extracts low-frequency typhoon signals from fiber-optic sensor data, combining time-frequency analysis with power spectrum analysis to effectively identify key typhoon-related features from complex vibration signals. By inputting instantaneous frequencies of different frequency bands into an LSTM network for training, the method can accurately capture the dynamic changes of typhoons, thereby achieving efficient prediction of typhoon weather. The innovative data processing and feature extraction methods of this invention significantly improve the accuracy and reliability of the prediction model, providing more precise technical support for typhoon prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is the overall flow chart of the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to the present invention;

[0043] Figure 2 This is a time-frequency analysis diagram of the optical phase signal according to the present invention. DETAILED DESCRIPTION

[0044] In order to make the objectives, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in combination with specific embodiments of the present application and with reference to the accompanying drawings.

[0045] Example 1

[0046] Distributed fiber-optic vibration sensor arrays can sense the external environment over long distances and over a wide range. The vibration signals caused by typhoons appear as continuous, "spindle-shaped" disturbances in the time domain (lasting hours to days). In the frequency domain, typhoon-induced vibration signals are primarily concentrated in the low- and medium-frequency bands. Seismograph and gravimeter observations show that the dominant frequency band for microseismic signals is 0.05–0.5 Hz (period 2–20 seconds), with energy peaks occurring primarily at 0.1–0.3 Hz (short-period dual-frequency microseisms) and 0.05–0.15 Hz (long-period dual-frequency microseisms). Therefore, analyzing vibration signals measured solely in the frequency or time domain by fiber-optic sensors will result in significant fluctuations and lead to significant errors. Therefore, this embodiment provides an LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing. By utilizing fiber-optic sensing technology to accurately extract typhoon-induced low-frequency vibration signals and combining two-dimensional sliding energy segmentation preprocessing with an LSTM network, the accuracy of typhoon prediction is improved, overcoming the resolution and coverage limitations of traditional remote sensing technology in typhoon weather. See Figure 1 As shown, the method described in this embodiment includes the following steps:

[0047] The optical phase data caused by external field vibrations is obtained based on a distributed fiber optic sensor array. Specifically, due to the long duration of typhoon weather, according to current research status, the signal can be approximated as a long-term non-stationary random signal, which can be used to collect the phase signal of the distributed fiber optic vibration sensor caused by external field environmental vibrations.

[0048] The segmentation period is selected according to the time period of the microseismic signal (a time segmentation period greater than the time period T of the microseismic signal is selected), and the optical phase data is divided into multiple time periods according to the time window, which can be recorded as [T1, T2], [T2, T3]…[T N-1 ,T N ], and obtain the segmented phase signal.

[0049] Perform GST transformation on the segmented single-period phase signal to obtain the time-frequency spectrum.

[0050] As a preferred implementation of this embodiment, the GST transformation of the segmented single-period phase signal is expressed as:

[0051]

[0052] Where S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [T N-1 ,T N ] is any single period of time segmentation; For the first [T N-1 ,T N ] Phase signal within a single period; t is the time within a single period; e is a natural constant, e -i2πft is the correction factor; i is the imaginary unit; θ is the power parameter, and u is the parameter coefficient, which together adjust the width and height of the Gaussian window.

[0053] The power spectrum of the segmented single-period phase signal is calculated, the maximum value of the power spectrum is found and the threshold parameter is set. The single-period phase signal with a power spectrum less than the threshold parameter multiplied by the power spectrum is retained to determine the frequency range in which the low-frequency signal is distributed.

[0054] As a preferred implementation of this embodiment, in order to more reasonably select the sliding window width, this embodiment quantifies the frequency dependence of the signal to be measured by power spectrum density and divides the typhoon signal frequency band components by calculating the power spectrum characteristic curve. The power spectrum of the segmented single-period phase signal is expressed as:

[0055]

[0056] Where P(f) is the power spectrum; d(n) is the sliding window function; μ is the sampling rate; N-1 is the time period number of the segment, [T N-1 ,T N] is any single time period of time division; i=1,2,…,L, the first [T N-1 ,T N The phase signal in a single time period is split into L segments, each segment contains M discrete data points, n = 1, 2, ..., M; U is the normalization factor obtained by processing d(n); For the first [T N-1 ,T N ]The phase signal of the nth point in the i-th segment within a single time period.

[0057] As a preferred implementation of this embodiment, finding the maximum value of the power spectrum is expressed as:

[0058] P(f) max =max([P(f)])

[0059] Where P(f) max is the maximum value of the time-frequency spectrum modulus within the local sliding window at each moment; max() is the maximum value function; P(f) is the power spectrum; [P(f)] is the modulus value of the power spectrum.

[0060] Set the threshold parameter b, when P(f)≥b*P(f) max When P(f) is zeroed, the high-frequency noise corresponding to P(f) is retained. <b*P(f) max The single-period phase signal of the low-frequency signal is used to determine the frequency range in which the low-frequency signal is distributed.

[0061] To reduce the impact of noise, the high-frequency noise components are reset to zero. In this embodiment, the frequency range in which the low-frequency signal is distributed is defined as the sliding window width. Based on the sliding window width, a local sliding window is set in the frequency direction of the time-frequency spectrum. The maximum position of the time-frequency modulus value within the local sliding window at each moment is searched, and the corresponding frequency of the maximum position of the time-frequency modulus value is recorded and the remaining frequencies are set to zero to obtain the instantaneous frequency of the low-frequency typhoon signal.

[0062] As a preferred implementation of this embodiment, the maximum position of the time-frequency modulus value in the local sliding window at each moment is searched, the corresponding frequency of the maximum position of the time-frequency modulus value is recorded, and the remaining frequencies are set to zero to reconstruct the frequency band. The instantaneous frequency of the low-frequency typhoon signal is obtained as follows:

[0063]

[0064] Where, f tN is the instantaneous frequency of the low-frequency typhoon signal; argmax() is the maximum value function; S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [S g (τ,f)] is the time-frequency spectrum modulus; f0 is the preset frequency constant; a is the sliding window width parameter.

[0065] The instantaneous frequency of low-frequency typhoon signals in different frequency bands is input into the LSTM network as a data set for training to obtain a trained LSTM network, which is then used to perform typhoon weather forecasting.

[0066] As a preferred implementation of this embodiment, the LSTM network has two transmission conversion states, namely the cell state and the hidden state. After the instantaneous frequency of low-frequency typhoon signals in different frequency bands is input into the LSTM network as a data set, four states are obtained, which are expressed as follows:

[0067]

[0068] Where K f , K i , K o are the forget gate, input gate, and output gate respectively; K is the new input data obtained by multiplying the instantaneous frequency of the input low-frequency typhoon signal by the weight matrix W and converting it into a value between (-1, 1) through the Tanh activation function; σ is the Sigmoid activation function; W f 、W i 、W o , W are K f , K i , K o , the weight matrix corresponding to K; h t-1 is the hidden state of the LSTM network at the previous moment; f tN is the instantaneous frequency of low-frequency typhoon signals in different frequency bands; b f 、b i 、b o and b are K f , K i , K o , the bias term corresponding to K.

[0069] Furthermore, the LSTM network performs typhoon weather forecasting to obtain the following typhoon weather forecast output:

[0070]

[0071] Where c t is the cell state of the LSTM network; K f For the forget gate; c t-1 is the cell state of the LSTM network at the previous moment; K i is the input gate; K is the new input data obtained by multiplying the instantaneous frequency of the input low-frequency typhoon signal by the weight matrix W through the activation function; h t is the hidden state of the LSTM network at the previous moment; K o is the output gate; tank() is the hyperbolic tangent activation function; yt The output results of typhoon weather forecast include typhoon frequency signals at future moments; σ is the Sigmoid activation function; W is the weight matrix.

[0072] To verify the effectiveness and superiority of the method provided in this embodiment, some specific cases are provided below:

[0073] A period of typhoon monitoring data is selected for analysis. First, the data is segmented into time domains with a period of 1000s, and then frequency segmented. The energy of the second-segmented frequency bands can be obtained in the time domain segments, such as Figure 2 The figure shows the time domain results obtained by dividing the sliding window into different frequency bands. It can be seen that different frequencies have different signal amplitude energies.

[0074] An LSTM network was trained on the instantaneous frequencies of low-frequency typhoon signals in different frequency bands. This example uses a gradient descent algorithm, with a maximum number of iterations set to 1000 and a learning rate set to 0.002. The dataset was further fragmented after each training session. After iteration, it was found that the training error converged rapidly and eventually stabilized. The training cycle lasted 1000 times, with nearly 80,000 iterations per cycle. The lowest error point was reached at Epoch 4, with a minimum root mean square error of 0.04735.

[0075] Example 2

[0076] Accordingly, this embodiment provides an LSTM typhoon prediction system based on two-dimensional sliding energy segmentation preprocessing. The system is used to implement the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in the embodiment of the present invention, including a data acquisition module, a time domain segmentation module, a time-frequency analysis module, a power spectrum analysis module, an instantaneous frequency extraction module and an LSTM network training and prediction module.

[0077] The data acquisition module is used to obtain optical phase data caused by external field vibration based on a distributed optical fiber sensor array.

[0078] The time domain segmentation module is used to select a segmentation period according to the time period of the microseismic signal, and to segment the optical phase data into multiple time periods according to the time window to obtain segmented phase signals.

[0079] The time-frequency analysis module is used to perform GST transformation on the segmented single-period phase signal to obtain a time-frequency spectrum.

[0080] The power spectrum analysis module is used to calculate the power spectrum of the segmented single-period phase signal, find the maximum value of the power spectrum and set the threshold parameter, and retain the single-period phase signal whose power spectrum is less than the threshold parameter multiplied by the power spectrum to determine the frequency range of the low-frequency signal distribution.

[0081] The instantaneous frequency extraction module is used to define the frequency range in which the low-frequency signal is distributed as the sliding window width, set a local sliding window in the frequency direction of the time-frequency spectrum based on the sliding window width, search for the maximum position of the time-frequency spectrum modulus value within the local sliding window at each moment, record the frequency corresponding to the maximum position of the time-frequency spectrum modulus value and set the remaining frequencies to zero to obtain the instantaneous frequency of the low-frequency typhoon signal.

[0082] The LSTM network training and prediction module is used to input the instantaneous frequency of low-frequency typhoon signals in different frequency bands as a data set into the LSTM network for training, obtain a trained LSTM network, and use the trained LSTM network to perform typhoon weather forecasts.

[0083] Example 3

[0084] This embodiment provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in Example 1 of the present invention is implemented.

[0085] Example 4

[0086] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in the first embodiment of the present invention is implemented.

[0087] In the embodiments of the present application, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can represent: a, b, c, a and b, a and c, b and c or a and b and c, where a, b, c can be single or multiple.

[0088] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented using a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0089] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0090] In the several embodiments provided in this application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of this application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory; hereinafter referred to as: ROM), random access memory (Random Access Memory; hereinafter referred to as: RAM), magnetic disk or optical disk, and other media that can store program code.

[0091] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention's description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. An LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing, characterized in that: The following steps are involved: Obtain optical phase data caused by external field vibration based on distributed optical fiber sensing array; Select a segmentation period based on the time period of the microseismic signal, and segment the optical phase data into multiple time periods according to the time window to obtain the segmented phase signal; Perform GST transformation on the segmented single-period phase signal to obtain the time-frequency spectrum; Calculate the power spectrum of the segmented single-period phase signal, find the maximum value of the power spectrum and set the threshold parameter, and retain the single-period phase signal whose power spectrum is less than the threshold parameter multiplied by the power spectrum to determine the frequency range of the low-frequency signal distribution; The frequency range of the low-frequency signal distribution is defined as the sliding window width. Based on the sliding window width, a local sliding window is set in the frequency direction of the time-frequency spectrum. The maximum value position of the time-frequency spectrum modulus value within the local sliding window is searched at each moment. The corresponding frequency of the maximum value position of the time-frequency spectrum modulus value is recorded and the remaining frequencies are set to zero to obtain the instantaneous frequency of the low-frequency typhoon signal. The instantaneous frequency of low-frequency typhoon signals in different frequency bands is input into the LSTM network as a data set for training to obtain a trained LSTM network, which is then used to perform typhoon weather forecasting.

2. The LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to claim 1 is characterized in that: The GST transformation of the segmented single-period phase signal is expressed as: Where S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [T N-1 ,T N ] is any single period of time segmentation; For the first [T N-1 ,T N ] Phase signal within a single period; t is the time within a single period; e is a natural constant, e -i2πft is the correction factor; i is the imaginary unit; θ is the power parameter, and u is the parameter coefficient, which together adjust the width and height of the Gaussian window.

3. The LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to claim 1 is characterized in that: The power spectrum of the segmented single-period phase signal is expressed as: Where P(f) is the power spectrum; d(n) is the sliding window function; μ is the sampling rate; N-1 is the time period number of the segment, [T N-1 ,T N ] is any single time period of time division; i=1,2,…,L, the first [T N-1 ,T N The phase signal in a single time period is split into L segments, each segment contains m discrete data points, n = 1, 2, ..., m; U is the normalization factor obtained by processing d(n); For the first [T N-1 ,T N ]The phase signal of the nth point in the i-th segment within a single time period.

4. The LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to claim 1 is characterized in that: Finding the maximum value of the power spectrum is expressed as: P(f) max =max([P(f)]) Where P(f) max is the maximum value of the time spectrum modulus in the local sliding window at each moment; max() is the maximum value function; P(f) is the power spectrum; [P(f)] is the modulus value of the power spectrum; Set the threshold parameter b, when P(f)≥b*P(f) max When P(f) is zeroed, the high-frequency noise corresponding to P(f) is retained. <b*P(f) max The single-period phase signal of the low-frequency signal is used to determine the frequency range in which the low-frequency signal is distributed.

5. The LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to claim 1 is characterized in that: Search for the maximum position of the time-frequency modulus value within the local sliding window at each moment, record the corresponding frequency of the maximum position of the time-frequency modulus value and set the remaining frequencies to zero, and obtain the instantaneous frequency of the low-frequency typhoon signal: Where, f tN is the instantaneous frequency of the low-frequency typhoon signal; argmax() is the maximum value function; S g (τ,f) is the time spectrum; τ is the time shift factor; f is the frequency; [S g (τ,f)] is the time-frequency spectrum modulus value; f0 is the preset frequency constant; a is the sliding window width parameter.

6. The LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to claim 1 is characterized in that: The instantaneous frequencies of low-frequency typhoon signals in different frequency bands are input into the LSTM network as data sets to perform typhoon weather forecasting. The typhoon weather forecast output results are as follows: Where c t is the cell state of the LSTM network; K f For the forget gate; c t-1 is the cell state of the LSTM network at the previous moment; K i is the input gate; K is the new input data obtained by multiplying the instantaneous frequency of the input low-frequency typhoon signal by the weight matrix W through the activation function; h t is the hidden state of the LSTM network at the previous moment; K o is the output gate; tanh() is the hyperbolic tangent activation function; y t is the typhoon weather forecast output result; σ is the Sigmoid activation function; W is the weight matrix.

7. An LSTM typhoon prediction system based on two-dimensional sliding energy segmentation preprocessing, characterized in that: The system is used to implement the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to any one of claims 1 to 6, comprising a data acquisition module, a time domain segmentation module, a time-frequency analysis module, a power spectrum analysis module, an instantaneous frequency extraction module, and an LSTM network training and prediction module; A data acquisition module is used to obtain optical phase data caused by external field vibration based on a distributed optical fiber sensor array; The time domain segmentation module is used to select a segmentation period according to the time period of the microseismic signal, and to segment the optical phase data into multiple time periods according to the time window to obtain the segmented phase signal; The time-frequency analysis module is used to perform GST transformation on the segmented single-period phase signal to obtain the time-frequency spectrum; The power spectrum analysis module is used to calculate the power spectrum of the segmented single-period phase signal, find the maximum value of the power spectrum and set the threshold parameter, and retain the single-period phase signal whose power spectrum is less than the threshold parameter multiplied by the power spectrum to determine the frequency range of the low-frequency signal distribution; The instantaneous frequency extraction module is used to define the frequency range of the low-frequency signal as the sliding window width, set a local sliding window in the frequency direction of the time-frequency spectrum based on the sliding window width, search for the maximum value position of the time-frequency spectrum modulus value within the local sliding window at each moment, record the corresponding frequency of the maximum value position of the time-frequency spectrum modulus value and set the remaining frequencies to zero to obtain the instantaneous frequency of the low-frequency typhoon signal; The LSTM network training and prediction module is used to input the instantaneous frequency of low-frequency typhoon signals in different frequency bands as a data set into the LSTM network for training, obtain a trained LSTM network, and use the trained LSTM network to perform typhoon weather forecasts.

8. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing as described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the LSTM typhoon prediction method based on two-dimensional sliding energy segmentation preprocessing according to any one of claims 1 to 6 is implemented.

Images