Gravitational wave signal parameter search model training method and device and electronic equipment

Abstract

The invention provides a gravitational wave signal parameter search model training method and device and electronic equipment, relates to the technical field of data processing, in particular to the fields of astronomy and universe research, gravitational wave detection, signal processing and the like, and can be used for application scenes such as gravitational wave signal processing and the like. According to the specific implementation scheme, an initial gravitational wave parameter set is generated according to a prior range and a preset signal-to-noise ratio threshold value; according to a target gravitational wave parameter in the initial gravitational wave parameter set, generating a time domain gravitational wave signal based on a preset time parameter and a response parameter; performing normalization processing on the time domain gravitational wave signal to generate a gravitational wave signal; constructing a training sample according to the target gravitational wave parameter, the time parameter and the gravitational wave signal; and training the initial neural network by using the training sample to obtain a gravitational wave signal parameter search model. According to the scheme, the training sample is constructed and the model is trained, so that the gravitational wave signal parameter searching efficiency and accuracy can be improved.

Description

Technical Field

[0001] The present disclosure relates to the field of data processing technology, in particular to the fields of astronomy and cosmology research, gravitational wave detection and signal processing, and can be used in application scenarios such as gravitational wave signal processing, and specifically to a gravitational wave signal parameter search model training method, device, and electronic equipment. Background Art

[0002] Existing techniques for analyzing extreme mass ratio precession gravitational wave signals have limitations, primarily due to low computational efficiency, difficulty sampling high-dimensional parameter spaces, and information loss associated with long-term signal observations. Furthermore, traditional methods cannot effectively integrate machine learning models to model the posterior distribution, resulting in insufficient parameter estimation accuracy. Summary of the Invention

[0003] The present disclosure provides a gravitational wave signal parameter search model training method, device and electronic equipment.

[0004] According to a first aspect of the present disclosure, a method for training a gravitational wave signal parameter search model is provided, comprising: generating an initial gravitational wave parameter set according to a priori range and a preset signal-to-noise ratio threshold; generating a time-domain gravitational wave signal according to target gravitational wave parameters in the initial gravitational wave parameter set based on preset time parameters and response parameters; normalizing the time-domain gravitational wave signal to generate a gravitational wave signal; constructing training samples according to the target gravitational wave parameters, time parameters, and gravitational wave signal; and training an initial neural network using the training samples to obtain a gravitational wave signal parameter search model.

[0005] According to a second aspect of the present disclosure, a gravitational wave signal parameter search method is provided, comprising: acquiring observation data from a gravitational wave detector; inputting the observation data into a gravitational wave signal parameter search model, so that the gravitational wave signal parameter search model performs parameter estimation to obtain a posterior distribution of the gravitational wave signal parameters; the gravitational wave signal parameter search model is trained by the method of the first aspect described above.

[0006] According to a third aspect of the present disclosure, a gravitational wave signal parameter search model training device is provided, comprising: a parameter generation module for generating an initial gravitational wave parameter set based on a priori range and a preset signal-to-noise ratio threshold; a signal generation module for generating a time-domain gravitational wave signal based on preset time parameters and response parameters according to target gravitational wave parameters in the initial gravitational wave parameter set; a signal processing module for normalizing the time-domain gravitational wave signal to generate a gravitational wave signal; a sample construction module for constructing training samples based on the target gravitational wave parameters, time parameters, and gravitational wave signals; and a model training module for training an initial neural network using the training samples to obtain a gravitational wave signal parameter search model.

[0007] According to a fourth aspect of the present disclosure, a gravitational wave signal parameter search device is provided, comprising: a data acquisition module for acquiring observation data from a gravitational wave detector; a model estimation module for inputting the observation data into a gravitational wave signal parameter search model so that the gravitational wave signal parameter search model performs parameter estimation to obtain a posterior distribution of the gravitational wave signal parameters; the gravitational wave signal parameter search model is trained using the method of the first aspect described above.

[0008] According to a fifth aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute any method in the embodiments of the present disclosure.

[0009] According to a sixth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable the computer to execute any method according to the embodiments of the present disclosure.

[0010] According to a seventh aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements any method according to the embodiments of the present disclosure when executed by a processor.

[0011] By adopting the solution disclosed in the present invention, the efficiency and accuracy of gravitational wave signal parameter search can be improved by constructing training samples and training the model.

[0012] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a flowchart of a gravitational wave signal parameter search model training method according to an embodiment of the present disclosure;

[0015] Figure 2 is a flowchart of a gravitational wave signal generation process according to an embodiment of the present disclosure;

[0016] Figure 3 is a flowchart of a model training process according to an embodiment of the present disclosure;

[0017] Figure 4 is a flow chart of a method for searching gravitational wave signal parameters according to an embodiment of the present disclosure;

[0018] Figure 5 is another flowchart of a method for searching for gravitational wave signal parameters according to an embodiment of the present disclosure;

[0019] Figure 6 is a schematic diagram of the posterior distribution results of the gravitational wave signal parameter search method according to an embodiment of the present disclosure;

[0020] Figure 7 It is a schematic diagram of the posterior distribution result based on the Markov chain Monte Carlo method in the prior art;

[0021] Figure 8 is a structural diagram of a gravitational wave signal parameter search model training device according to an embodiment of the present disclosure;

[0022] Figure 9 is a schematic structural diagram of a gravitational wave signal parameter search device according to an embodiment of the present disclosure;

[0023] Figure 10 1 is a schematic diagram of a scenario of a gravitational wave signal parameter search model training method according to an embodiment of the present disclosure;

[0024] Figure 11 1 is a schematic diagram of a scenario of a method for searching gravitational wave signal parameters according to an embodiment of the present disclosure;

[0025] Figure 12 3 is a structural diagram of an electronic device used to implement the gravitational wave signal parameter search model training method and / or the gravitational wave signal parameter search method of the embodiments of the present disclosure. DETAILED DESCRIPTION

[0026] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0027] The term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. The term "at least one" in this article means any combination of at least two of any one or more of a plurality of. For example, including at least one of A, B, and C, can mean including any one or more elements selected from the set consisting of A, B, and C. The terms "first" and "second" in this article refer to multiple similar technical terms and distinguish them, and do not mean to limit the order or to limit to only two. For example, the first feature and the second feature refer to two categories / two features. The first feature can be one or more, and the second feature can also be one or more.

[0028] In addition, numerous specific details are provided in the following detailed description to better illustrate the present disclosure. Those skilled in the art will appreciate that the present disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main points of the present disclosure.

[0029] Before introducing the technical solutions of the embodiments of the present disclosure, the following technical terms that may be used in the present disclosure are further explained:

[0030] Gravitational waves: ripples in the curvature of space-time that propagate outward from the radiation source in the form of waves. These waves transmit energy in the form of gravitational radiation.

[0031] Extreme Mass Ratio Inspiral (EMRI) refers to a special type of celestial system in the universe. When a small, compact object happens to approach a supermassive black hole at the center of a galaxy, it may be captured by the black hole's gravitational field, resulting in the compact object moving around the black hole in a new orbit. In theory, this motion would release gravitational waves, causing a gradual loss of kinetic energy in the system. This would cause the object's orbit to shrink at a very slow rate, eventually causing it to fall into the black hole.

[0032] Among related technologies, the analysis of EMRI gravitational wave signals currently primarily employs a Bayesian inference framework, exploring high-dimensional parameter spaces through Markov Chain Monte Carlo methods. However, this technique suffers from significant drawbacks: first, computational efficiency is low, with single waveform generation taking an extremely long time and requiring significant computational resources. Second, the sampling efficiency and accuracy of high-dimensional space cannot meet practical requirements, resulting in limited parameter estimation results. Furthermore, traditional methods are susceptible to initial conditions, becoming trapped in local optima, and the low signal-to-noise ratio of long-term observations leads to loss of frequency domain information, further reducing analysis accuracy. Existing machine learning techniques have also failed to overcome these issues, providing only limited point estimates under ideal conditions and lacking the ability to model posterior distributions, making them unsuitable for practical applications.

[0033] To at least partially address one or more of the aforementioned and other potential issues, the present disclosure proposes a gravitational wave signal parameter search model training method that can efficiently generate a joint posterior distribution in a high-dimensional parameter space under a strong noise background, supporting near-real-time data processing for space gravitational wave detection missions.

[0034] The present disclosure provides a method for training a gravitational wave signal parameter search model. Figure 1 It is a flow chart of a gravitational wave signal parameter search model training method according to an embodiment of the present disclosure, and the gravitational wave signal parameter search model training method can be applied to a gravitational wave signal parameter search model training device. The gravitational wave signal parameter search model training device is located in an electronic device. The electronic device includes but is not limited to fixed devices and / or mobile devices. For example, fixed devices include but are not limited to servers, and servers can be cloud servers or ordinary servers. For example, mobile devices include but are not limited to gravitational wave signal processing devices (corresponding to the application scenario), and gravitational wave signal processing devices can be mobile phones, tablet computers, etc. In some possible implementations, the gravitational wave signal parameter search model training method can also be implemented by a processor calling computer-readable instructions stored in a memory. For example Figure 1 As shown, the gravitational wave signal parameter search model training method includes:

[0035] S101. Generate an initial gravitational wave parameter set based on a priori range and a preset signal-to-noise ratio threshold.

[0036] S102. Generate a time-domain gravitational wave signal according to the target gravitational wave parameters in the initial gravitational wave parameter set and based on preset time parameters and response parameters.

[0037] S103. Normalize the time-domain gravitational wave signal to generate a gravitational wave signal.

[0038] S104. Construct training samples based on target gravitational wave parameters, time parameters, and gravitational wave signals.

[0039] S105. Use the training samples to train the initial neural network to obtain a gravitational wave signal parameter search model.

[0040] The a priori range refers to a preset range of reasonable values ​​for gravitational wave parameters. In the disclosed embodiments, the a priori range can be based on theoretical models and historical data empirical values ​​to ensure that the generated data covers all possible physical phenomena.

[0041] The signal-to-noise ratio (SNR) is a measure of the ratio of signal strength to background noise strength. In the disclosed embodiment, a preset SNR threshold is used to filter data to ensure that the quality of the signal in the training set is sufficiently high.

[0042] The initial gravitational wave parameter set is a set of gravitational wave physical parameters generated by an algorithm. In the disclosed embodiments, the initial gravitational wave parameter set can be used for subsequent gravitational wave signal simulations.

[0043] In the disclosed embodiments, the prior ranges of various parameters can first be determined based on the physical model and observational data. Subsequently, a parameter set can be generated randomly or based on a specific distribution. For example, this can be based on a uniform distribution, a normal distribution, or a Gaussian distribution. Finally, the generated parameters can be screened for signal-to-noise ratio, retaining those that meet a threshold condition. The above is merely illustrative and does not constitute a complete list of all possible scenarios for generating the initial gravitational wave parameter set. This list is not exhaustive.

[0044] The target gravitational wave parameters refer to a set of physical parameters specifically used to generate gravitational wave signals. In the disclosed embodiment, the gravitational wave parameters may include at least mass, spin, orbital parameters, etc.

[0045] The time parameter is time information related to signal generation. In the embodiment of the present disclosure, the time parameter may include a sampling interval, a signal duration, and the like.

[0046] The response parameter refers to the sensitivity parameter of the detector. In the disclosed embodiment, the response parameter may include frequency response, directionality, etc., which can be used to describe the characteristics of the detector's ability to capture gravitational wave signals.

[0047] The time-domain gravitational wave signal is a simulated gravitational wave signal. In the disclosed embodiments, the time-domain gravitational wave signal can be used to describe the changes in gravitational waves within a specific time range and can be represented in the form of a time-domain waveform.

[0048] In the disclosed embodiments, a mathematical representation of the gravitational wave signal can first be generated using a physical model based on the target gravitational wave parameters. For example, the physical model can be an existing model such as the equations of general relativity. Subsequently, the signal can be discretized based on the time parameter to generate time-domain data. Finally, the detector response parameters can be used to adjust the signal strength and waveform, simulating the detector's capture of the gravitational wave signal. The above description is merely illustrative and does not constitute a complete list of all possible scenarios for generating time-domain gravitational wave signals. This list is not exhaustive.

[0049] The gravitational wave signal is normalized time-domain gravitational wave data. In the disclosed embodiment, the gravitational wave signal can be represented in a signal form suitable for subsequent training and analysis based on actual conditions.

[0050] In the disclosed embodiments, time-domain gravitational wave data is normalized to ensure that the data amplitude is compatible with the training model. For example, normalization can be performed using a standardization method or within a specific range. Furthermore, noise data can be added or removed based on practical circumstances. The above description is merely illustrative and does not constitute a complete list of all possible scenarios for generating gravitational wave signals. This list is not exhaustive.

[0051] The training samples refer to data for training a neural network. In the disclosed embodiment, the training samples may include gravitational wave signals and their corresponding physical parameter labels.

[0052] The initial neural network is an untrained neural network model. In the embodiment of the present disclosure, the initial neural network can be constructed by using flow matching and continuous normalization flow.

[0053] In the disclosed embodiments, the normalized gravitational wave signal can be paired with the target gravitational wave parameters to generate training samples. The above is merely an example and does not limit all possible scenarios for constructing training samples. This is not intended to be exhaustive.

[0054] In the disclosed embodiments, the neural network structure can be first defined and the network parameters initialized. Subsequently, iterative training can be performed using training samples to optimize the network parameters. Finally, after training, the optimized model can be saved as the gravitational wave signal parameter search model. The above description is merely illustrative and does not limit all possible scenarios for obtaining a gravitational wave signal parameter search model. This is not intended to be exhaustive.

[0055] The technical solutions of the disclosed embodiments, by generating an initial set of gravitational wave parameters, ensure that the parameters cover the physical range where gravitational waves may occur, enhancing the generalization capability of the model. Furthermore, through signal-to-noise ratio screening, the authenticity and effectiveness of the training samples are improved. By generating time-domain gravitational wave signals, real physical processes can be simulated while taking into account the detector's response characteristics, improving the relevance of model training and its practical applicability. By generating gravitational wave signals through normalization processing, data uniformity can be ensured while eliminating anomalies or imbalances in the data, enhancing the generalization capability of the model. By constructing training samples and training the model, the efficiency and accuracy of gravitational wave signal parameter searches can be improved. A reusable model is constructed, reducing human intervention, improving gravitational wave data processing capabilities, and enabling application in real-time gravitational wave detection missions.

[0056] In some embodiments, an initial gravitational wave parameter set is generated based on a priori range and a preset signal-to-noise ratio threshold, including: generating a candidate gravitational wave parameter set based on the priori range; and screening the candidate gravitational wave parameter set based on the signal-to-noise ratio threshold to obtain the initial gravitational wave parameter set.

[0057] The candidate gravitational wave parameter set is a set of potential gravitational wave parameter sets generated based on a priori ranges. In the disclosed embodiments, the candidate gravitational wave parameter set has not been further screened and may include some parameters that do not meet the signal-to-noise ratio threshold or physical plausibility.

[0058] In the disclosed embodiments, parameter ranges can first be set based on theoretical models, historical observational data, and physical laws. For example, the mass range of two celestial bodies can be set to 10-30 solar masses, the spin range can be set to 0-1, and so on. Other parameter ranges can be set similarly. Subsequently, parameter values ​​can be randomly generated while ensuring that the generated parameters conform to the physical laws within the a priori range. For example, Monte Carlo methods or Latin hypercube sampling can be used to generate parameter values. The above is merely illustrative and does not constitute a complete list of all possible scenarios for generating candidate gravitational wave parameter sets; however, this list is not exhaustive.

[0059] In the disclosed embodiments, theoretical gravitational wave signals can first be simulated based on candidate gravitational wave parameters, and the signal-to-noise ratio (SNR) of each signal can then be calculated. For example, the SNR can be represented by the ratio of signal intensity to noise intensity. In particular, the noise intensity can be determined based on the detector sensitivity curve. Furthermore, a SNR threshold can be pre-set to filter candidate parameters that meet the preset criteria and eliminate samples with low SNRs. Preferably, the SNR threshold can be set to 60. The above is merely an example and does not constitute a complete list of all possible scenarios for obtaining the initial set of gravitational wave parameters. This is simply not an exhaustive list.

[0060] By generating a set of candidate gravitational wave parameters in this way, we can ensure that the generated set covers the possible physical range of gravitational wave signals. Furthermore, by adjusting the generation rules, we can adapt to different gravitational wave events. By screening the candidate gravitational wave parameter set, we can remove noise-dominated samples, ensuring that the generated parameter set produces high-quality signals while avoiding performance degradation caused by excessive low-quality samples during model training.

[0061] In some embodiments, the response parameters include a universal time domain response function and a detector response function; according to the target gravitational wave parameters in the initial gravitational wave parameter set, based on preset time parameters and response parameters, a time domain gravitational wave signal is generated, including: according to the target gravitational wave parameters and time parameters, using a theoretical waveform model to obtain a theoretical gravitational wave waveform; according to the universal time domain response function, the theoretical gravitational wave waveform is projected onto the detector response function to generate a simulated observation signal; according to the simulated observation signal, a time domain gravitational wave signal is generated using time delay interferometry.

[0062] The universal time-domain response function (UTRF) is a theoretical function used to describe the fundamental interaction between gravitational waves and the observation system. In the disclosed embodiments, the UTF is a universal response characteristic of gravitational wave signals, which is related to the polarization and propagation direction of the gravitational waves.

[0063] The detector response function is a mathematical model for a specific detector, used to describe the detector's sensitivity to gravitational wave signals. In the disclosed embodiments, the detector response function includes the detector's directivity, frequency response, and system noise. Specifically, the detector response function can be selected based on the actual detector being used.

[0064] The theoretical waveform model is a mathematical representation of a gravitational wave signal derived from physical theories such as general relativity and astrophysical processes. In the disclosed embodiments, the theoretical waveform model can be used to generate an ideal gravitational wave signal for a specific astrophysical event.

[0065] The theoretical gravitational wave waveform is a time-domain signal generated by a theoretical waveform model. In the disclosed embodiments, the theoretical gravitational wave waveform can describe how the gravitational wave signal changes over time under idealized conditions.

[0066] In the disclosed embodiments, the target gravitational wave parameters and time parameters can be first input into a theoretical model. The model then calculates the time-dependent changes in the frequency, amplitude, and phase of the gravitational waves, thereby generating a theoretical gravitational wave waveform. For example, the theoretical model can be a numerical relativity model, a fitted model, or the like. The above description is merely illustrative and does not limit all possible scenarios for obtaining a theoretical gravitational wave waveform. This is simply not an exhaustive list.

[0067] The simulated observation signal is a signal generated by processing the theoretical gravitational wave waveform through a universal time-domain response function and a detector response function. In the disclosed embodiments, the simulated observation signal can simulate the gravitational wave signal that the detector should observe under actual conditions.

[0068] In the disclosed embodiments, a theoretical waveform can first be processed using a universal time-domain response function to adjust the signal's polarization and directionality. Subsequently, the adjusted waveform can be further processed using a detector response function. For example, a detector sensitivity curve can be superimposed on the adjusted waveform, or a detector noise model can be incorporated. The above is merely illustrative and does not constitute a complete list of all possible scenarios for generating simulated observation signals, though this is not intended to be exhaustive.

[0069] Time-delay interferometry (TDI) is a signal processing technique commonly used in space detectors. In the disclosed embodiments, TDI can eliminate noise and recover gravitational wave signals by correcting the time delays of multiple detector signals.

[0070] In the disclosed embodiments, time delay information can first be calculated based on the actual distance and direction of the detectors. Subsequently, time delay matching can be performed on multiple signals to eliminate system noise between detectors. Finally, signal superposition and filtering can be used to eliminate noise and enhance the gravitational wave signal, thereby generating a time-domain gravitational wave signal. In particular, the generated time-domain gravitational wave signal can be output as a second-generation TDI variable. The above description is merely illustrative and does not constitute a complete list of all possible scenarios for generating time-domain gravitational wave signals, but it is not intended to be exhaustive.

[0071] In this way, by generating a theoretical gravitational wave waveform, a reasonable initial signal can be obtained with high precision, accurately reflecting the physical properties of the gravitational wave source. By generating a simulated observation signal, the ideal signal can be converted into a signal observable by the detector, while simulating the real observation environment and improving the realism of signal processing. By generating a time-domain gravitational wave signal, a full waveform signal can be synthesized, preserving key physical features such as harmonic coherence, enabling the model to learn the complete phase evolution law dominated by general relativity effects.

[0072] In some embodiments, a time-domain gravitational wave signal is normalized to generate a gravitational wave signal, including: generating a frequency-domain gravitational wave signal using Fourier transform based on the time-domain gravitational wave signal; determining a power spectral density based on the frequency-domain gravitational wave signal; performing whitening on the frequency-domain gravitational wave signal based on the power spectral density to obtain a whitened frequency-domain gravitational wave signal; and generating a gravitational wave signal using an inverse Fourier transform based on the whitened frequency-domain gravitational wave signal.

[0073] The frequency-domain gravitational wave signal is a Fourier transform of the time-domain gravitational wave signal. In the disclosed embodiments, the frequency-domain gravitational wave signal emphasizes the frequency component of the signal rather than its change over time.

[0074] In the disclosed embodiments, a Fast Fourier Transform (FFT) can be applied to the time-domain gravitational wave signal to convert it into a frequency-domain signal, thereby generating a frequency-domain gravitational wave signal. The above description is merely illustrative and does not limit all possible scenarios for generating frequency-domain gravitational wave signals. This list is not exhaustive.

[0075] The power spectral density (PSD) is the energy distribution of a signal in the frequency domain, describing the intensity of the signal power at each frequency component. For gravitational wave detection, the power spectral density reflects the distribution characteristics of the signal and noise in the frequency domain.

[0076] In the disclosed embodiments, the PSD can be calculated using the square modulus of the frequency domain signal. Specifically, the power spectral density can be estimated using a multi-segment averaging method, thereby reducing the impact of random fluctuations. The above description is merely illustrative and does not limit all possible scenarios for determining the power spectral density. This is not intended to be exhaustive.

[0077] The whitened frequency-domain gravitational wave signal is a signal representation that has been processed to make the noise distribution of each frequency component uniform. In the disclosed embodiment, the influence of signal noise in the whitened frequency-domain gravitational wave signal is equalized.

[0078] In the disclosed embodiments, frequency-domain gravitational wave information can be normalized based on the PSD, thereby adjusting the frequency distribution of the noise to a uniform distribution and removing the dominant effect of the noise. In particular, the PSD can be smoothed to avoid overfitting the noise fluctuations. The above description is merely illustrative and does not limit all possible scenarios for obtaining a whitened frequency-domain gravitational wave signal. This is not intended to be exhaustive.

[0079] In the disclosed embodiments, an inverse Fourier transform (IFFT) can be applied to the whitened frequency-domain gravitational wave signal to convert it back into a time-domain signal. The resulting time-domain signal is then used as the gravitational wave signal. The above description is merely illustrative and does not limit all possible scenarios for generating gravitational wave signals. This is not intended to be an exhaustive list.

[0080] Generating a frequency-domain gravitational wave signal in this way provides information about the signal's frequency components, facilitating analysis of noise characteristics. Determining the power spectral density accurately describes the signal's frequency characteristics and highlights the noise contribution. Whitening eliminates frequency-dependent noise distribution, making the signal more uniform in the frequency domain. Generating gravitational waves allows the whitened frequency-domain signal to be re-expressed as a time-domain signal, resulting in an optimized signal representation.

[0081] In some embodiments, a training sample is constructed based on target gravitational wave parameters, time parameters and gravitational wave signals, including: concatenating the target gravitational wave parameters and time parameters to generate a parameter time series vector; based on the gravitational wave signal, using an embedding network to perform feature extraction to obtain a gravitational wave signal feature vector; and using the parameter time series vector and the gravitational wave signal feature vector to construct a training sample.

[0082] The parameter time series vector is a sequence or vector formed by concatenating the target gravitational wave parameters and the time parameter. In the disclosed embodiment, the parameter time series vector is a combined data structure containing target event characteristics and time-related information.

[0083] In this disclosed embodiment, the target gravitational wave parameters and time parameters are concatenated in a predetermined order to form a unified time series vector. Specifically, the data type and scale can be unified to avoid conflicts between different data features. The above description is merely illustrative and does not limit all possible scenarios for generating parameter time series vectors. This is not intended to be an exhaustive list.

[0084] The embedding network is a neural network structure that performs dimensionality reduction or feature extraction on input data through nonlinear transformations, generating a low-dimensional but information-dense feature representation. In the disclosed embodiments, the embedding network can be a convolutional neural network, including convolutional or fully connected layers, capable of extracting deep features from gravitational wave signals.

[0085] The gravitational wave signal feature vector is a low-dimensional feature representation generated by the embedding network after feature extraction of the gravitational wave signal. In the disclosed embodiments, the gravitational wave signal feature vector may include core information such as the frequency, amplitude, and phase of the gravitational wave signal.

[0086] In the disclosed embodiments, a gravitational wave signal can first be input into an embedding network. The embedding network can then be used to reduce the signal's dimensionality and extract features through multiple layers of convolution, pooling, or recurrent units. Finally, the embedding network can output a low-dimensional feature vector. The above description is merely illustrative and does not limit all possible scenarios for obtaining a gravitational wave signal's feature vector; however, this is not intended to be exhaustive.

[0087] In the disclosed embodiments, gravitational wave signal feature vectors can be used as samples, and parameter time series vectors as labels to establish a corresponding relationship, thereby generating training samples. The above description is merely illustrative and does not limit all possible scenarios for constructing training samples. This is simply not an exhaustive list.

[0088] By generating parameter time series vectors, we can integrate target physical parameters and time information to provide a complete description of the event. Extracting gravitational wave signal feature vectors can reduce signal dimensionality, reduce noise and redundancy, and improve training efficiency. By constructing training samples, we can integrate parameter information and signal features into a unified sample, facilitating model learning.

[0089] In some embodiments, the initial neural network is trained using training samples to obtain a gravitational wave signal parameter search model, including: inputting the training samples into the initial neural network; based on the training samples, the initial neural network is trained using a flow matching loss function to obtain the gravitational wave signal parameter search model.

[0090] The flow matching loss function is a specific loss function used for training the Continuous Normalizing Flow (CNF) model. In the disclosed embodiments, the flow matching loss function can guide model optimization by defining the difference between the data distribution and the model distribution.

[0091] The gravitational wave signal parameter search model is a machine learning-based model used to extract parameters of target astrophysical events from signals received by gravitational wave detectors. In the disclosed embodiments, the gravitational wave signal parameter search model can be a continuous normalized flow model.

[0092] In particular, the continuous normalized flow model is a probability model based on ordinary differential equations (ODEs). In the disclosed embodiments, the continuous normalized flow model can map a Gaussian distribution to a posterior distribution by learning a continuous-time probability transformation.

[0093] In the disclosed embodiments, after inputting the training samples into the initial neural network, a flow matching loss function can be first defined. For example, this flow matching loss function can be based on maximum likelihood estimation or Kullback-Leibler (KL) divergence, measuring the distance between the data distribution and the model distribution. Subsequently, gradient descent or adaptive methods can be used to optimize the neural network parameters, and the network weights can be continuously adjusted to match the distribution generated by the model with the training sample distribution. The resulting continuous normalized flow model is used as the gravitational wave signal parameter search model. The above is merely an example and does not limit all possible scenarios for obtaining a gravitational wave signal parameter search model. This is simply not an exhaustive list.

[0094] In this way, by training the neural network using the flow matching loss function, the flow model can be optimized so that it can accurately fit the distribution of training samples.

[0095] In some embodiments, a gravitational wave signal parameter search model training method further includes: discarding training samples, removing target gravitational wave parameters from an initial gravitational wave parameter set, and generating a new gravitational wave parameter set; the number of gravitational wave parameters in the new gravitational wave parameter set is less than the number of gravitational wave parameters in the initial gravitational wave parameter set; and reconstructing new training samples based on any new target gravitational wave parameter in the new gravitational wave parameter set to train the gravitational wave signal parameter search model to obtain a new gravitational wave signal parameter search model, until the new gravitational wave parameter set is an empty set, and using the new gravitational wave signal parameter search model generated last time as the final gravitational wave signal parameter search model.

[0096] In the disclosed embodiment, the current training sample can be deleted from the system memory, and the current target gravitational wave parameters can be removed from the initial gravitational wave parameter set. Subsequently, after the removal is complete, the gravitational wave parameter set without the current target gravitational wave parameters is used as the new gravitational wave parameter set. The above description is merely illustrative and does not limit all possible scenarios for generating a new gravitational wave parameter set. This is not intended to be an exhaustive list.

[0097] In the disclosed embodiment, a target gravitational wave parameter is selected from a new gravitational wave parameter set, and a time-domain gravitational wave signal and a gravitational wave signal are regenerated to construct new training samples. Subsequently, a new round of iterative training is performed on the gravitational wave signal parameter search model generated in the previous training using the new training samples. Specifically, each time a target gravitational wave parameter is removed, the size of the new gravitational wave parameter set is reduced by one parameter. This removal process is repeated until the gravitational wave parameter set is empty, meaning that all gravitational wave parameters have been processed. Finally, the newly generated gravitational wave signal parameter search model can be used as the final gravitational wave signal parameter search model. The above description is merely illustrative and does not constitute a complete list of all possible scenarios for obtaining the final gravitational wave signal parameter search model; however, this is not intended to be an exhaustive list.

[0098] By discarding training samples and removing target gravitational wave parameters, we can avoid retraining already processed parameters, improving training efficiency. We can also dynamically update the parameter set, process gravitational wave parameters in stages, and reduce the complexity of the training task. By reconstructing new training samples and training the model, we can eliminate the need to store large amounts of gravitational wave signal data, saving only parameters with a high signal-to-noise ratio for training.

[0099] In some embodiments, a random sampling can be performed from the prior parameter range of the EMRI hypothesis to select 500,000 EMRI signal parameters with a signal-to-noise ratio greater than 60 to form a training set, and another 100,000 parameter sets meeting the same signal-to-noise ratio threshold are selected to form a test set.

[0100] Furthermore, the response function of the latest Laser Interferometer Space Antenna (LISA) can be used to generate second-generation TDI variables. Simultaneously, based on the EMRI signal parameters in the training set, a high-precision EMRI waveform can be generated using an analytic kludge model (AAK) that simulates the gravitational wave signal. Next, a universal time-domain response function can be used to project the high-precision EMRI waveform onto the LISA detector response. The resulting EMRI signal is output as a second-generation TDI variable, yielding a time-domain gravitational wave signal. Subsequently, the time-domain gravitational wave signal can be Fourier transformed to obtain a frequency-domain gravitational wave signal. This signal is then whitened and noise is added based on the corresponding power spectral density, yielding the final gravitational wave signal used for training. Specifically, this data generation method can be embedded in a machine learning framework, essentially reconstructing the dataset portion of a deep learning framework. This allows the generation of corresponding EMRI signal data based on the EMRI parameters stored in the training set.

[0101] Furthermore, a CNF model that continuously evolves from a Gaussian distribution to a posterior distribution can be constructed. Specifically, the evolution process can be controlled by a time parameter t∈[0,1]. When the target distribution is the posterior distribution, CNF models the time-varying vector field v through a neural network. t,x , which describes the sample trajectory φ t The derivative characteristics from the base distribution to the target distribution. The process of CNF to achieve distribution transformation can be expressed by the following ODE:

[0102]

[0103] Where, φ t,x (θ) represents the state of parameter θ at time t; v t,x (φ t,x (θ)) represents the time t and position φ t,x (θ); t represents the time parameter; θ represents the input parameter, i.e., the variable sampled from the initial Gaussian distribution; θ0 represents the initial state of the parameter, corresponding to the sample of the initial Gaussian distribution; φ t＝0,x (θ) = θ0 represents the initial state of the parameter when t = 0; θ1 represents the target state of the parameter, corresponding to the sample of the target posterior distribution; φ t＝1,x (θ)=θ1 represents the final state of the parameters when t=1.

[0104] Subsequently, the ODE in the CNF model can be solved numerically, that is, by numerically integrating the sample q from the basis distribution t＝0,x (θ)~q0(θ) generate target distribution samples. t＝0,x (θ) represents the distribution of samples at time t = 0, that is, samples sampled from the initial Gaussian distribution. q0(θ) represents the probability density function of the initial distribution. This process can be viewed as moving samples from the base distribution toward the target distribution along a time-varying vector field.

[0105] In particular, the initial condition of the ODE is φ in the basis distribution t＝0,x (θ) = θ0, and solve the equation to transform the base distribution samples into target distribution samples. The target density can be expressed by the following continuity equation:

[0106]

[0107] Where q t,x (θ) represents the probability density distribution of parameter θ at time t; v t,x (θ) represents the vector field defined at time t and position θ; div(q t,x (θ)v t,x (θ)) represents the vector field q t,x (θ)v t,xThe probability density distribution q can be described by the continuity equation t,x (θ) evolves with time t.

[0108] Furthermore, gravitational wave signals can be used to train the initial neural network model. The goal of the training process is to learn the data distribution and the generative process. Learning the data distribution means that the model needs to approximate the true distribution of the gravitational wave signal so that the generated samples have characteristics similar to the gravitational wave signal. Learning the generative process involves using a continuous normalized flow model to learn how to generate a posterior distribution from a Gaussian distribution through continuous time evolution.

[0109] Specifically, the flow matching framework can be used to directly regress the training vector field. The training process can be expressed by the following formula:

[0110] u t (θ|θ1)=θ1-θ

[0111] Where u t (θ|θ1) represents a vector field that describes the direction and velocity from the current sample θ to the target sample θ1.

[0112] In particular, during the time evolution process, the distribution of samples is a Gaussian distribution that changes over time and eventually converges to the target state. The evolution process can be expressed by the following Gaussian probability path formula:

[0113]

[0114] Where p t (θ|θ1) represents the conditional probability distribution of parameter θ at time t; represents Gaussian distribution; I d Represents the d-dimensional identity matrix, which is used to indicate that each dimension is independent and has the same variance, that is, the covariance matrix of the Gaussian distribution.

[0115] The loss function can be used to optimize the model's vector field. By minimizing the loss, the model can learn how to generate samples of the target distribution from the initial distribution. The loss function can be expressed as follows:

[0116]

[0117] Where, L FM represents the loss function; E t～p(t) It means to calculate the weighted expectation of the distribution p(t) at time t; Indicates the weighted expectation of the distribution p(θ) of the target distribution sample θ1; Indicates the weighted expectation of the conditional distribution p(x|θ1) of the conditional sample x; Represents the sample θ at time tt the distribution p t (θ t |θ1) is weighted and expected;||r|| 2 represents the square of the residual.

[0118] In particular, after each iteration of training, the training data is directly discarded to generate new data. The entire method does not need to save large amounts of EMRI signal data, only the parameters of the EMRI signal with high signal-to-noise ratio need to be saved for training.

[0119] Figure 2 A flowchart of a gravitational wave signal generation process of an embodiment of the present disclosure is shown in FIG. 2, which includes the following steps: Figure 2

[0120] S201, determining the prior range of the EMRI wave source.

[0121] S202, sampling from the prior range to generate EMRI gravitational wave parameters.

[0122] S203, determining the AAK model, the detector response, and the TDI variable.

[0123] S204, generating the time-domain gravitational wave signal.

[0124] S205, using Fourier transform to convert the time-domain gravitational wave signal to a frequency-domain gravitational wave signal.

[0125] S206, whitening the frequency-domain gravitational wave signal according to the corresponding power spectral density.

[0126] S207, converting the whitened frequency-domain gravitational wave signal to a time-domain signal, and adding noise to obtain a gravitational wave signal for training.

[0127] Figure 3 A flowchart of a model training process of an embodiment of the present disclosure is shown in FIG. 3, which includes the following steps: Figure 3

[0128] S301a, vector splicing the time parameter and the gravitational wave parameter to generate a parameter time sequence vector.

[0129] S301b, inputting the gravitational wave signal into an embedding network to extract features and generate a gravitational wave signal feature vector.

[0130] S302, inputting the parameter time sequence vector and the gravitational wave signal feature vector into the model for training.

[0131] S303, obtaining the trained CNF model.

[0132] It should be understood that Figures 2 to 3 ​​The schematic diagram shown is only exemplary and not restrictive, and it is scalable, and those skilled in the art can Figures 2 to 3 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0133] The present disclosure provides a method for searching gravitational wave signal parameters. Figure 4 It is a flow chart of a gravitational wave signal parameter search method according to an embodiment of the present disclosure, and the gravitational wave signal parameter search method can be applied to a gravitational wave signal parameter search device. The gravitational wave signal parameter search device is located in an electronic device. The electronic device includes but is not limited to fixed devices and / or mobile devices. For example, fixed devices include but are not limited to servers, and servers can be cloud servers or ordinary servers. For example, mobile devices include but are not limited to gravitational wave signal processing devices (corresponding to the application scenario), and gravitational wave signal processing devices can be mobile phones, tablet computers, vehicle-mounted terminals, etc. In some possible implementations, the gravitational wave signal parameter search method can also be implemented by a processor calling computer-readable instructions stored in a memory. For example Figure 4 As shown, the gravitational wave signal parameter search method includes:

[0134] S401. Obtain observation data from a gravitational wave detector;

[0135] S402. Input the observation data into the gravitational wave signal parameter search model so that the gravitational wave signal parameter search model performs parameter estimation to obtain the posterior distribution of the gravitational wave signal parameters.

[0136] The observation data is the raw data or preliminarily processed signal data recorded by the gravitational wave detector during the detection process. In the disclosed embodiment, the observation data may include the interferometer vibration signal recorded by the gravitational wave detector, including time domain waveform data.

[0137] In the disclosed embodiments, a gravitational wave detector records time-domain vibration signals using an interferometer and stores the data at a high sampling rate. Observational data can be directly obtained from the information transmitted back by the gravitational wave detector. The above description is merely illustrative and does not limit all possible scenarios for obtaining observational data. This list is not exhaustive.

[0138] Among them, the posterior distribution is a probability distribution based on Bayesian inference, which is used to describe the uncertainty of event parameters.

[0139] In the disclosed embodiments, observational data can first be input into a gravitational wave signal parameter search model. Subsequently, the gravitational wave signal parameter search model can calculate a posterior distribution through time integration. For example, an embedding network can be used to extract features from the input data to generate a signal feature vector. The dynamic system of the gravitational wave signal parameter search model can then be used to map the feature vector into a probability distribution in parameter space. The posterior distribution can then be generated using the model's built-in probability calculation module. Furthermore, the observational data can be preprocessed based on actual circumstances. The above description is merely illustrative and does not constitute a complete list of all possible gravitational wave signal parameter search methods. This is simply not intended to be an exhaustive list.

[0140] The technical solutions of the disclosed embodiments enable rapid parameter estimation through inference based on models and posterior distributions, improving work efficiency. Furthermore, the posterior distribution provides not only parameter estimates but also the uncertainty range of the estimates, providing astronomers with more complete and reliable information.

[0141] In some embodiments, parameter estimation is performed using a gravitational wave signal parameter search model, including: based on observation data, using a solver to time-integrate the gravitational wave signal parameter search model to obtain a posterior distribution of the gravitational wave signal parameters.

[0142] The solver is a numerical method for solving the ODE in the continuous normalized flow model. In the embodiment of the present disclosure, the solver can calculate the output of the flow model by time integration, and a numerical integration algorithm can be used.

[0143] In the disclosed embodiments, the continuous normalized flow model is a dynamic system based on ordinary differential equations, describing the evolution of data from an initial state to a target state. Therefore, numerical integration can be used to solve this evolutionary process, calculating the model's time evolution and using the resulting result as the posterior distribution of the gravitational wave signal parameters. The above description is merely illustrative and does not limit all possible scenarios for deriving the posterior distribution of gravitational wave signal parameters; however, this is not intended to be an exhaustive list.

[0144] In this way, using the solver to integrate the model over time can capture the dynamic characteristics of gravitational wave signals, support complex signal modeling, and rapidly complete the numerical solution of complex differential equations, enabling real-time gravitational wave signal analysis. By generating a posterior distribution, not only does it provide point estimates of the parameters, but it also describes their probability distribution, combining observational data with prior knowledge to enhance the confidence of the parameter estimates.

[0145] Figure 5 Another flow chart of the method for searching gravitational wave signal parameters according to an embodiment of the present disclosure is shown in FIG. Figure 5 Shown, including:

[0146] S501. Obtain observation data from a gravitational wave detector.

[0147] S502: Input the observation data into the CNF model.

[0148] S503. Based on the observation data, the ODE solver performs time integration on the CNF model to generate a posterior distribution of the gravitational wave signal parameters.

[0149] It should be understood that Figure 5 The schematic diagram shown is only exemplary and not restrictive, and it is scalable, and those skilled in the art can Figure 5 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0150] Figure 6 Figure 2 shows a schematic diagram of the posterior distribution results of the gravitational wave signal parameter search method according to an embodiment of the present disclosure. The shaded area is the posterior distribution obtained by the gravitational wave signal parameter search method according to an embodiment of the present disclosure, and the black line is the parameter of the real EMRI injected. Figure 6 As shown, the parameters of the real EMRI all fall within the posterior distribution range obtained by the gravitational wave signal parameter search method of the embodiment of the present disclosure, that is, the gravitational wave signal parameter search method of the embodiment of the present disclosure can successfully search out the EMRI signal and give the posterior distribution.

[0151] In contrast, the same EMRI signal and a priori range assumption can be used to apply the Markov Chain Monte Carlo method in the prior art. Figure 7 The figure shows the posterior distribution result of the Markov chain Monte Carlo method in the prior art. The shaded area is the posterior distribution obtained by the prior art method, and the black line is the parameter of the real EMRI injected. Figure 7 As shown in FIG, the parameters of the real EMRI do not fall within the posterior distribution range obtained by the Markov Chain Monte Carlo method in the prior art. That is, the Markov Chain Monte Carlo method in the prior art cannot successfully search out the EMRI signal and give the posterior distribution, resulting in serious distortion.

[0152] The present disclosure provides a gravitational wave signal parameter search model training device, such as Figure 8As shown, the apparatus may include: a parameter generation module 801 for generating an initial gravitational wave parameter set according to a priori range and a preset signal-to-noise ratio threshold; a signal generation module 802 for generating a time-domain gravitational wave signal according to target gravitational wave parameters in the initial gravitational wave parameter set, based on preset time parameters and response parameters; a signal processing module 803 for normalizing the time-domain gravitational wave signal to generate a gravitational wave signal; a sample construction module 804 for constructing training samples according to the target gravitational wave parameters, time parameters and gravitational wave signals; and a model training module 805 for training the initial neural network using the training samples to obtain a gravitational wave signal parameter search model.

[0153] In some embodiments, the parameter generation module 801 includes: a candidate parameter generation submodule for generating a set of candidate gravitational wave parameters based on a priori ranges; and a candidate parameter screening submodule for screening the set of candidate gravitational wave parameters based on a signal-to-noise ratio threshold to obtain an initial set of gravitational wave parameters.

[0154] In some embodiments, the signal generation module 802 includes: a theoretical waveform generation submodule, which is used to obtain a theoretical gravitational wave waveform based on target gravitational wave parameters and time parameters using a theoretical waveform model; a simulation signal generation submodule, which is used to project the theoretical gravitational wave waveform onto the detector response function based on a universal time domain response function to generate a simulated observation signal; and a time delay interferometry submodule, which is used to generate a time domain gravitational wave signal based on the simulated observation signal using time delay interferometry.

[0155] In some embodiments, the signal processing module 803 includes: a Fourier transform submodule, used to generate a frequency-domain gravitational wave signal using Fourier transform based on the time-domain gravitational wave signal; a power spectrum density sub-determination module, used to determine the power spectrum density based on the frequency-domain gravitational wave signal; a whitening processing submodule, used to whiten the frequency-domain gravitational wave signal based on the power spectrum density to obtain a whitened frequency-domain gravitational wave signal; and a gravitational wave signal generation submodule, used to generate a gravitational wave signal using an inverse Fourier transform based on the whitened frequency-domain gravitational wave signal.

[0156] In some embodiments, the sample construction module 804 includes: a parameter splicing submodule, which is used to splice the target gravitational wave parameters and time parameters to generate a parameter time series vector; a feature extraction submodule, which is used to extract features based on the gravitational wave signal using an embedded network to obtain a gravitational wave signal feature vector; and a sample generation submodule, which is used to construct a training sample using the parameter time series vector and the gravitational wave signal feature vector.

[0157] In some embodiments, the model training module 805 includes a sample input submodule for inputting training samples into the initial neural network; a model training submodule for training the initial neural network based on the training samples using a flow matching loss function to generate a continuous normalized flow model; and an integral solution submodule for performing time integration on the continuous normalized flow model using a solver to obtain a gravitational wave signal parameter search model.

[0158] In some embodiments, a gravitational wave signal parameter search model training device further includes: a sample discarding module ( Figure 8 (not shown in the figure), used to discard training samples, remove the target gravitational wave parameters from the initial gravitational wave parameter set, and generate a new gravitational wave parameter set; the number of gravitational wave parameters in the new gravitational wave parameter set is less than the number of gravitational wave parameters in the initial gravitational wave parameter set. Iterative training module ( Figure 8 (not shown) is used to reconstruct a new training sample to train the gravitational wave signal parameter search model according to any new target gravitational wave parameter in the new gravitational wave parameter set, to obtain a new gravitational wave signal parameter search model, until the new gravitational wave parameter set is an empty set, and the new gravitational wave signal parameter search model generated last time is used as the final gravitational wave signal parameter search model.

[0159] For the description of specific functions and examples of each module and submodule of the device in the embodiment of the present disclosure, please refer to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0160] The gravitational wave signal parameter search model training device in the disclosed embodiments generates an initial gravitational wave parameter set, ensuring that the parameters cover the physical range where gravitational waves may occur, enhancing the model's generalization capabilities. Signal-to-noise ratio screening also improves the authenticity and effectiveness of training samples. By generating time-domain gravitational wave signals, it simulates real-world physical processes while taking into account the detector's response characteristics, improving the relevance of model training and its practical applicability. Normalization of gravitational wave signals ensures data uniformity, eliminates anomalies or imbalances in the data, and enhances the model's generalization capabilities. By constructing training samples and training the model, the efficiency and accuracy of gravitational wave signal parameter searches are improved. A reusable model is constructed, reducing human intervention, improving gravitational wave data processing capabilities, and enabling application in real-time gravitational wave detection missions.

[0161] The present disclosure provides a gravitational wave signal parameter search model training device, such as Figure 9As shown, the device may include: a data acquisition module 901, used to acquire observation data from a gravitational wave detector; a model estimation module 902, used to input the observation data into a gravitational wave signal parameter search model, so that the gravitational wave signal parameter search model performs parameter estimation to obtain a posterior distribution of the gravitational wave signal parameters; the gravitational wave signal parameter search model is trained by any one of the gravitational wave signal parameter search model training methods.

[0162] In some embodiments, the model estimation module 902 includes: a solver integration submodule, which is used to use a solver to perform time integration on the gravitational wave signal parameter search model based on the observation data to obtain the posterior distribution of the gravitational wave signal parameters.

[0163] For the description of specific functions and examples of each module and submodule of the device in the embodiment of the present disclosure, please refer to the relevant description of the corresponding steps in the above method embodiment, which will not be repeated here.

[0164] The gravitational wave signal parameter search device in the disclosed embodiments can rapidly estimate parameters through inference based on a model and a posterior distribution, improving work efficiency. Furthermore, the posterior distribution provides not only the estimated value of the parameter but also the uncertainty range of the estimate, providing astronomers with more complete and reliable information.

[0165] The embodiment of the present disclosure provides a scenario diagram of a gravitational wave signal parameter search model training method, such as Figure 10 shown.

[0166] As previously mentioned, the gravitational wave signal parameter search model training method provided in the embodiments of the present disclosure is applied to electronic devices. The term "electronic device" is intended to refer to various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0167] Specifically, the electronic device can perform the following operations:

[0168] An initial gravitational wave parameter set is generated based on a priori range and a preset signal-to-noise ratio threshold. A time-domain gravitational wave signal is generated based on the target gravitational wave parameters in the initial gravitational wave parameter set and preset time parameters and response parameters. The time-domain gravitational wave signal is normalized to generate a gravitational wave signal. A training sample is constructed based on the target gravitational wave parameters, time parameters, and gravitational wave signal. The training sample is used to train the initial neural network to obtain a gravitational wave signal parameter search model.

[0169] It should be understood that Figure 10 The scene diagram shown is only illustrative and not restrictive. Those skilled in the art can Figure 10Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0170] The embodiment of the present disclosure provides a scenario diagram of a method for searching parameters of a gravitational wave signal. Figure 11 shown.

[0171] As previously mentioned, the gravitational wave signal parameter search method provided in the embodiments of the present disclosure is applied to electronic devices. The term "electronic device" is intended to refer to various forms of digital computers, such as laptops, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0172] Specifically, the electronic device can perform the following operations:

[0173] Observation data from a gravitational wave detector is obtained; and the observation data is input into a gravitational wave signal parameter search model so as to perform parameter estimation using the gravitational wave signal parameter search model to obtain a posterior distribution of gravitational wave signal parameters.

[0174] It should be understood that Figure 11 The scene diagram shown is only illustrative and not restrictive. Those skilled in the art can Figure 11 Various obvious changes and / or substitutions can be made to the examples, and the resulting technical solutions still fall within the scope of the disclosure of the embodiments of the present disclosure.

[0175] In the technical solutions disclosed herein, the acquisition, storage, and application of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0176] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0177] Figure 12 A schematic block diagram of an example electronic device 1200 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0178] like Figure 12As shown, device 1200 includes a computing unit 1201, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1202 or a computer program loaded from a storage unit 1208 into a random access memory (RAM) 1203. Various programs and data required for the operation of device 1200 can also be stored in RAM 1203. Computing unit 1201, ROM 1202, and RAM 1203 are connected to each other via a bus 1204. An input / output (I / O) interface 1205 is also connected to bus 1204.

[0179] Various components in device 1200 are connected to I / O interface 1205, including an input unit 1206, such as a keyboard and mouse; an output unit 1207, such as various types of displays and speakers; a storage unit 1208, such as a magnetic disk and optical disk; and a communication unit 1209, such as a network card, a modem, a wireless communication transceiver, etc. Communication unit 1209 allows device 1200 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0180] Computing unit 1201 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of computing unit 1201 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Computing unit 1201 performs the various methods and processes described above, such as the gravitational wave signal parameter search model training method and / or the gravitational wave signal parameter search method. For example, in some embodiments, the gravitational wave signal parameter search model training method and / or the gravitational wave signal parameter search method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as storage unit 1208. In some embodiments, part or all of the computer program can be loaded and / or installed onto device 1200 via ROM 1202 and / or communication unit 1209. When the computer program is loaded into RAM 1203 and executed by computing unit 1201, one or more steps of the gravitational wave signal parameter search model training method and / or gravitational wave signal parameter search method described above can be performed. Alternatively, in other embodiments, computing unit 1201 can be configured to perform the gravitational wave signal parameter search model training method and / or gravitational wave signal parameter search method in any other suitable manner (e.g., via firmware).

[0181] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0182] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0183] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory, a read-only memory, an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0184] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or a liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0185] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0186] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0187] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not limited herein.

[0188] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A gravitational wave signal parameter search model training method, comprising: Generate an initial set of gravitational wave parameters based on a priori range and a preset signal-to-noise ratio threshold; generating a time-domain gravitational wave signal according to the target gravitational wave parameters in the initial gravitational wave parameter set and based on preset time parameters and response parameters; performing normalization processing on the time-domain gravitational wave signal to generate a gravitational wave signal; constructing a training sample according to the target gravitational wave parameter, the time parameter, and the gravitational wave signal; The initial neural network is trained using the training samples to obtain a gravitational wave signal parameter search model.

2. The method according to claim 1, wherein Generating an initial gravitational wave parameter set according to a priori range and a preset signal-to-noise ratio threshold includes: generating a set of candidate gravitational wave parameters according to the a priori range; The candidate gravitational wave parameter set is screened according to the signal-to-noise ratio threshold to obtain the initial gravitational wave parameter set.

3. The method according to claim 1, wherein The response parameters include the universal time domain response function and the detector response function σ Generating a time-domain gravitational wave signal according to the target gravitational wave parameters in the initial gravitational wave parameter set based on preset time parameters and response parameters includes: Obtaining a theoretical gravitational wave waveform using a theoretical waveform model according to the target gravitational wave parameters and the time parameter; According to the universal time-domain response function, the theoretical gravitational wave waveform is projected onto the detector response function to generate a simulated observation signal; The time-domain gravitational wave signal is generated according to the simulated observation signal using time-delay interferometry.

4. The method according to claim 1, wherein The normalizing process of the time-domain gravitational wave signal to generate a gravitational wave signal includes: Generating a frequency-domain gravitational wave signal by Fourier transform according to the time-domain gravitational wave signal; determining a power spectral density according to the frequency-domain gravitational wave signal; performing whitening processing on the frequency-domain gravitational wave signal according to the power spectral density to obtain a whitened frequency-domain gravitational wave signal; The gravitational wave signal is generated by inverse Fourier transform according to the whitened frequency-domain gravitational wave signal.

5. The method according to claim 1, wherein The constructing of a training sample according to the target gravitational wave parameter, the time parameter, and the gravitational wave signal includes: splicing the target gravitational wave parameter and the time parameter to generate a parameter time series vector; Based on the gravitational wave signal, extract features using an embedding network to obtain a gravitational wave signal feature vector; The training sample is constructed using the parameter time series vector and the gravitational wave signal feature vector.

6. The method according to claim 1, wherein The method of training an initial neural network using the training samples to obtain a gravitational wave signal parameter search model includes: Inputting the training samples into the initial neural network; Based on the training samples, the initial neural network is trained using a flow matching loss function to obtain the gravitational wave signal parameter search model.

7. The method according to claim 1, wherein The method further comprises: discarding the training sample, removing the target gravitational wave parameter from the initial gravitational wave parameter set, and generating a new gravitational wave parameter set; wherein the number of gravitational wave parameters in the new gravitational wave parameter set is less than the number of gravitational wave parameters in the initial gravitational wave parameter set; According to any new target gravitational wave parameter in the new gravitational wave parameter set, a new training sample is reconstructed to train the gravitational wave signal parameter search model to obtain a new gravitational wave signal parameter search model, until the new gravitational wave parameter set is an empty set, and the new gravitational wave signal parameter search model generated last time is used as the final gravitational wave signal parameter search model.

8. A method for searching for gravitational wave signal parameters, comprising: Obtain observation data from gravitational wave detectors; The observation data is input into a gravitational wave signal parameter search model so that the gravitational wave signal parameter search model performs parameter estimation to obtain a posterior distribution of gravitational wave signal parameters; the gravitational wave signal parameter search model is trained by the method described in any one of claims 1 to 7.

9. The method according to claim 8, wherein The performing parameter estimation using the gravitational wave signal parameter search model includes: According to the observation data, a solver is used to perform time integration on the gravitational wave signal parameter search model to obtain the posterior distribution of the gravitational wave signal parameters.

10. A gravitational wave signal parameter search model training device, comprising: A parameter generation module is used to generate an initial set of gravitational wave parameters based on a priori range and a preset signal-to-noise ratio threshold; a signal generation module, configured to generate a time-domain gravitational wave signal according to the target gravitational wave parameters in the initial gravitational wave parameter set, based on preset time parameters and response parameters; a signal processing module, configured to perform normalization processing on the time-domain gravitational wave signal to generate a gravitational wave signal; a sample construction module, configured to construct a training sample according to the target gravitational wave parameter, the time parameter, and the gravitational wave signal; A model training module is used to train the initial neural network using the training samples to obtain a gravitational wave signal parameter search model.

11. A gravitational wave signal parameter search device, comprising: A data acquisition module, used to obtain observation data from gravitational wave detectors; A model estimation module, configured to input the observation data into a gravitational wave signal parameter search model so that the gravitational wave signal parameter search model performs parameter estimation to obtain a posterior distribution of gravitational wave signal parameters; the gravitational wave signal parameter search model is trained by the method according to any one of claims 1 to 7.

12. An electronic device comprising: at least one processor; as well as a memory communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 9.

13. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are for causing a computer to execute the method according to any one of claims 1-9.

14. A computer program product comprising a computer program stored on a storage medium, the computer program implementing the method according to any one of claims 1 to 9 when executed by a processor.

Images