A method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage

By collecting and integrating sound data, meteorological conditions and reservoir data, the prediction model of feature extractors and classifiers is constructed, which solves the problem of insufficient prediction accuracy and adaptability of rainfall in the existing technology, and achieves more accurate and economical rainfall estimates.

CN118797447BActive Publication Date: 2024-12-06FUJIAN SHUIKOU POWER GENERATION GROUP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411282470.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-12-06
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

In the rainfall prediction, the measurement tools are expensive and sparsely distributed and cannot be popularized. The existing methods fail to effectively combine meteorological and reservoir data, reducing prediction accuracy and adaptability.

Method used

Sound wave rain increase sound data is collected through the pickup, and a database is built based on meteorological conditions and reservoir data. After using wavelet transform to denoise, a sonic wave rain increase rainfall prediction model of feature extractor, gradient inversion layer and main classifier is constructed to integrate sound, Meteorological and reservoir characteristics are predicted.

Benefits of technology

It improves the accuracy and adaptability of rainfall prediction, enhances the generalization ability and robustness of the model, and provides a economical and practical rainfall estimation method.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118797447B_ABST
    Figure CN118797447B_ABST
Patent Text Reader

Abstract

The present invention is a method for estimating the rainfall corresponding to acoustic rainfall enhancement for reservoir storage. An acoustic rainfall enhancement database is constructed through acoustic rainfall enhancement sound data and related data collected by a microphone. First, the collected acoustic rainfall enhancement sound data is subjected to denoising by applying wavelet transform to obtain clear acoustic rainfall enhancement data; then, a training set and a test set are constructed by combining the denoised data and the information in the database to train an acoustic rainfall enhancement rainfall prediction model, wherein the model includes a feature extractor, a gradient inversion layer, a domain classifier and a main classifier, and the acoustic rainfall enhancement corresponding rainfall can be accurately predicted by training the model; finally, the acoustic rainfall enhancement rainfall is estimated by using the trained model to formulate a reasonable reservoir storage plan, thereby realizing the effective management and utilization of water resources.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of signal processing, and mainly to a method for estimating rainfall corresponding to acoustic rainfall enhancement for reservoir storage. Background Art

[0002] Acoustic rainfall enhancement is a method of inducing rainfall through sound wave technology. In order to optimize the timing of rainfall induction, improve resource utilization efficiency, ensure safety management, and rationally dispatch water resources, accurate rainfall prediction is required. Currently, more accurate rainfall intensity measurement relies on professional measurement tools, such as rain gauges. Although rain gauges can provide accurate data, they are expensive and are usually only installed in hydrological stations and meteorological stations, resulting in sparse and uneven distribution of measurement points. In addition, although rain radars can cover a rainfall intensity range of tens to hundreds of kilometers, their prices are also high and difficult to popularize. Therefore, an economical and practical means of rainfall prediction is urgently needed.

[0003] For example, CN114626417A "A method and system for rainfall detection based on acoustic signal feature analysis" discloses "A method and system for rainfall detection based on acoustic signal feature analysis. First, a rain sound signal library is established; then, a threshold noise reduction method is used to perform noise reduction analysis on the rain sound signal to obtain the wavelet basis function and decomposition layer number for optimal noise reduction; then, a BP neural network classification algorithm is used to select a training set and a test set from the rain sound signal library, and the BP neural network is trained by the selected training set to establish a rain sound classification model; finally, the classification model is used to predict the classification label of the test set." For another example, CN111276157B "A method and device for rainfall intensity recognition and model training based on rain sound" discloses "The rainfall intensity recognition device based on rain sound includes: a rainfall audio signal acquisition module for acquiring rainfall audio signals collected by a mobile terminal device; environmental data An acquisition module is used to acquire environmental data of the mobile terminal device; a rain sound feature extraction module is used to extract rain sound features according to rainfall audio signals; and a rainfall intensity recognition module is used to input rain sound features and environmental data into a rainfall intensity recognition model based on rain sound to obtain rainfall intensity. By implementing the present invention, the rainfall intensity at the user's current location can be acquired through the rainfall audio signal collected by the mobile terminal and the environmental data of the mobile terminal device. The above methods are all based on rain sound to predict rainfall. The rain sound acquisition device is simple and highly practical. However, CN114626417A mainly relies on the sound wave signal itself for feature extraction and classification without combining other relevant information, which will reduce the accuracy and reliability of the prediction. Although CN111276157B includes environmental data and rainfall audio signals, it does not mention how to deal with the adaptability of the model under different environments or conditions. Therefore, there is an urgent need for a rainfall estimation method with simple data acquisition equipment, more comprehensive data and strong adaptability. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present application provides a method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage.

[0005] The technical solution of this application is as follows:

[0006] A method for estimating the rainfall corresponding to acoustic rainfall enhancement for reservoir storage, the method comprising:

[0007] Collecting acoustic rain-enhancing sound data through a microphone, and obtaining corresponding acoustic rain-enhancing related data, wherein the acoustic rain-enhancing related data includes meteorological condition data and reservoir data, and constructing an acoustic rain-enhancing database using the acoustic rain-enhancing sound data and the acoustic rain-enhancing related data;

[0008] The collected acoustic rain enhancement sound data are denoised by using wavelet transform to obtain the denoised acoustic rain enhancement sound data, and a training set and a test set are constructed based on the acoustic rain enhancement database and the denoised acoustic rain enhancement sound data;

[0009] Constructing an acoustic wave rainfall enhancement rainfall prediction model, the acoustic wave rainfall enhancement rainfall prediction model includes a feature extractor, a gradient reversal layer, a domain classifier and a main classifier, obtaining comprehensive features including sound features, meteorological features and reservoir features through the feature extractor, inputting the comprehensive features into the gradient reversal layer and the domain classifier to complete the weight update of the feature extractor, and inputting the comprehensive features output by the feature extractor after the weight update into the main classifier to complete the acoustic wave rainfall enhancement rainfall prediction; using the training set to train the acoustic wave rainfall enhancement rainfall prediction model to obtain a trained acoustic wave rainfall enhancement rainfall prediction model;

[0010] The trained acoustic rainfall prediction model is used to estimate the acoustic rainfall required by users to increase reservoir storage, and a reservoir storage plan is formulated based on the estimated acoustic rainfall data.

[0011] As a preferred implementation of the present application, when collecting the acoustic wave rain-enhancing sound data through the microphone, a method of simultaneously recording at multiple collection points is adopted, and the recorded data at the multiple collection points are processed to obtain the acoustic wave rain-enhancing sound data.

[0012] As a preferred implementation mode of the present application, the constructed acoustic wave rainfall enhancement database includes a reservoir information table, an acoustic wave rainfall enhancement sound data table and a meteorological condition table, wherein the reservoir information table includes a reservoir ID, a reservoir name, topographic data and a reservoir area, wherein the topographic data includes a slope and an altitude; the acoustic wave rainfall enhancement sound data table includes an acoustic wave rainfall enhancement sound data ID, an associated reservoir ID, a collection time, an acoustic wave rainfall enhancement sound data, a microphone height and an orientation; the meteorological condition table includes an associated acoustic wave rainfall enhancement sound data ID, an acoustic wave rainfall enhancement intensity, a wind speed, a wind direction, a season and a time, wherein the acoustic wave rainfall enhancement intensity includes heavy rain, moderate rain and light rain, the season includes spring, summer, autumn and winter, and the time includes day and night.

[0013] As a preferred implementation of the present application, the collected acoustic wave rain enhancement sound data is denoised using wavelet transform, and the denoised acoustic wave rain enhancement sound data is obtained as follows:

[0014] The wavelet transform adopts a Daubechies wavelet filter, and the Daubechies wavelet filter includes a low-pass filter and a high-pass filter;

[0015] Impulse responses using low-pass and high-pass filters respectively and Sound signal for acoustic rainfall enhancement Perform convolution, expressed as:

[0016] ;

[0017] ;

[0018] In the formula, is the approximate coefficient, is the detail coefficient;

[0019] The convolved acoustic rain enhancement sound signal is downsampled to obtain the components of the approximate coefficient and the detail coefficient, which constitute the wavelet coefficient. , expressed as:

[0020] ;

[0021] In the formula, is the approximate coefficient component, is the detail coefficient component; For the Tier wavelet coefficients;

[0022] The hard threshold processing of wavelet coefficients is expressed as follows:

[0023] ;

[0024] In the formula, is the wavelet coefficient after hard threshold processing, including the approximate coefficient after hard threshold processing and detail factor , is the threshold value;

[0025] The detail coefficient is processed by soft threshold, which is expressed as follows:

[0026] ;

[0027] ;

[0028] In the formula, is the detail coefficient after soft threshold processing, is the symbolic function, is the threshold, where if the standard deviation of the wavelet coefficients is ,but , is an empirical constant;

[0029] According to the wavelet coefficients after hard thresholding and detail coefficient after soft threshold processing Get the wavelet coefficients after threshold processing , using the wavelet coefficients Perform inverse wavelet transform to obtain the denoised acoustic rain-enhancing sound signal , expressed as:

[0030] ;

[0031] In the formula, is the inverse wavelet transform, specifically the wavelet coefficients Up-sampling is performed, and the up-sampled wavelet coefficients are filtered and combined using a wavelet reconstruction filter to obtain the denoised acoustic rain-enhancing sound signal.

[0032] As a preferred embodiment of the present application, the feature extractor includes a sound feature extraction module, a reservoir feature extraction module and a meteorological feature extraction module. The sound feature extraction module is used to extract frequency domain and time domain features from the denoised acoustic rain-enhancing sound signal to obtain comprehensive sound features; the reservoir feature extraction module is used to extract features from reservoir area, slope and altitude data and obtain comprehensive reservoir features; the meteorological feature extraction module is used to extract features from wind speed, wind direction, season and time and obtain comprehensive meteorological features; the feature extractor performs feature fusion on the comprehensive sound features, comprehensive reservoir features and comprehensive meteorological features to obtain comprehensive features.

[0033] As a preferred embodiment of the present application, the frequency domain features extracted by the sound feature extraction module include The frequency cepstrum coefficients and spectrum flatness, the extraction steps are as follows:

[0034] The denoised acoustic rain-enhancing sound signal is converted from a time domain signal to a frequency domain signal using the Fast Fourier Transform (FFT). Filter Bank Scale conversion to obtain Spectrum, yes The spectrum is logarithmically transformed to obtain Spectrum, from Computing in the Spectrum Frequency cepstrum coefficients, and spectrum flatness is obtained by measuring the flatness of the spectrum;

[0035] The time domain features extracted by the sound feature extraction module include the zero crossing rate and the first normalized autocorrelation coefficient. The formula is:

[0036] ;

[0037] In the formula, is the denoised acoustic rain enhancement sound signal, is the total number of sampling points of the acoustic rain enhancement sound signal;

[0038] The first normalized autocorrelation coefficient extraction step comprises:

[0039] Calculate the autocorrelation function , expressed as:

[0040] ;

[0041] In the formula, is the discrete delay amount, Indicates that the acoustic signal of the acoustic rain enhancement moves in time Self-similarity after sampling points;

[0042] The normalized autocorrelation coefficient is the normalized form of the autocorrelation function, expressed as:

[0043] ;

[0044] In the formula, for The normalized form of ;

[0045] The first normalized autocorrelation coefficient is The normalized autocorrelation coefficient at is expressed as:

[0046] .

[0047] As a preferred embodiment of the present application, the reservoir feature extraction module is a fully connected neural network, including an input layer, a hidden layer and a comprehensive feature extraction layer, wherein:

[0048] The input layer is used to receive the numerical values ​​of the reservoir area, slope and altitude. The hidden layer includes two fully connected layers using the ReLU activation function, each with 128 neurons, which are used to extract deep features of the reservoir area, slope and altitude; the comprehensive feature extraction layer is a fully connected layer using the ReLU activation function, containing 64 neurons, which is used to extract comprehensive reservoir features.

[0049] As a preferred implementation of the present application, the meteorological feature extraction module includes an input layer, a wind speed layer, a wind direction layer, a season layer, a time layer and a feature fusion layer, wherein:

[0050] The input layer includes four input layers: continuous value input of wind speed, categorical value input of wind direction, custom one-hot encoding input of season, and binary value input of time;

[0051] Different types of features are processed through the wind speed layer, wind direction layer, season layer and time layer. Specifically, the wind speed layer is a fully connected layer, which maps the wind speed value to a hidden layer with 16 neurons to obtain the wind speed feature; the wind direction layer maps the wind direction value to an embedding space with 8 dimensions, and flattens the output of the embedding layer into one dimension to obtain the wind direction feature; the season layer is a fully connected layer, which maps the unique hot encoding of the season to a hidden layer with 8 neurons to obtain the season feature; the time layer is a fully connected layer, which maps the binary input of time to a hidden layer with 4 neurons to obtain the time feature;

[0052] The wind speed features, wind direction features, season features and time features are input into the feature fusion layer for comprehensive feature extraction. The feature fusion layer includes two fully connected layers, which are mapped to hidden layers of 64 and 32 neurons respectively to obtain comprehensive meteorological features.

[0053] As a preferred implementation of the present application, the acoustic rain enhancement rainfall prediction model sends the comprehensive features output by the feature extractor to the gradient reversal layer, and the comprehensive features are input to the domain classifier through the gradient reversal layer. The domain classifier judges the domain based on the comprehensive features after the gradient reversal. The gradient of the loss function of the domain classifier is reversed through the gradient reversal layer during the back propagation process and transmitted to the feature extractor. The weight of the feature extractor is updated according to the inverted gradient. The domain classifier is composed of three fully connected layers. The comprehensive features pass through the first fully connected layer, apply the ReLU activation function to introduce nonlinearity, and then pass through the second fully connected layer. The ReLU activation function is also applied to the final output layer to map the output of the hidden layer to the corresponding domain; the loss function of the domain classifier is the cross entropy function.

[0054] As a preferred implementation of the present application, the comprehensive features output by the weighted feature extractor are used as the input of the main classifier, and the main classifier includes an input layer, a convolution layer 1, a pooling layer 1, a convolution layer 2, a pooling layer 2, a convolution layer 3, a pooling layer 3, a flattening layer, a fully connected layer 1, a fully connected layer 2 and an output layer, wherein:

[0055] The input latitude of the input layer is , is the number of samples input into the main classifier;

[0056] The number of convolution kernels in the convolution layer 1, convolution layer 2, and convolution layer 3 are 32, 64, and 128 respectively, and the size of the convolution kernel is 1 3, the activation function is ReLU function, the stride is 1; the pooling type of pooling layer 1, pooling layer 2 and pooling layer 3 is maximum pooling, and the pooling kernel size is 1 2, the stride is 2; the number of neurons in the fully connected layer 1 and the fully connected layer 2 are 512 and 256 respectively, and the activation function is the ReLU function; the number of neurons in the output layer is 3, corresponding to the heavy rain, moderate rain and light rain of the acoustic rain enhancement intensity respectively, and the activation function is the softmax function;

[0057] The main classifier uses convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, and flattening layer to extract features, classifies through fully connected layer 1 and fully connected layer 2, and finally outputs the acoustic wave rainfall enhancement intensity.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. The present invention provides a method for estimating the corresponding rainfall of acoustic rain enhancement for reservoir storage, which collects acoustic rain enhancement sound data and uses wavelet transform to denoise, more accurately extracts information from the acoustic rain enhancement sound, and improves the prediction accuracy of rainfall;

[0060] 2. The present invention provides a method for estimating the corresponding rainfall of acoustic rain enhancement for reservoir storage, which integrates acoustic rain enhancement sound data, meteorological condition data and reservoir characteristic data. The fusion of multiple data sources considers various factors more comprehensively and improves the accuracy of the prediction model;

[0061] 3. The present invention provides a method for estimating the rainfall corresponding to acoustic rainfall enhancement for reservoir storage. The acoustic rainfall enhancement rainfall prediction model includes a feature extractor, a gradient reversal layer, a domain classifier and a main classifier. The generalization ability and robustness of the model are significantly improved by sending the comprehensive features output by the feature extractor to the gradient reversal layer and then passing through the domain classifier; the gradient reversal layer enables the model to ignore domain specificity during training and focus on extracting common features, thereby reducing overfitting and sensitivity to noise; the domain classifier helps the model to better adapt to different data environments, enhances the model's adaptability to data under different domain conditions, and improves the accuracy and reliability of rainfall prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a method flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0064] The present invention provides the following technical solution: a method for estimating the rainfall corresponding to acoustic rainfall enhancement for reservoir storage.

[0065] Example 1

[0066] Acoustic rainfall enhancement is a method of inducing rainfall through sound wave technology. In order to optimize the implementation timing of rainfall induction, improve resource utilization efficiency, ensure safe management, and rationally dispatch water resources, there is an urgent need for a rainfall estimation method corresponding to acoustic rainfall enhancement for reservoir storage. Accurate rainfall forecasts can help effectively reduce costs, avoid flood risks, and at the same time improve the water storage capacity and overall management level of reservoirs.

[0067] like Figure 1 As shown, this embodiment provides a method for estimating the rainfall corresponding to acoustic rainfall enhancement for reservoir storage, and the specific steps include:

[0068] S1. Collecting sound data of acoustic wave rainfall enhancement through a microphone, and obtaining corresponding acoustic wave rainfall enhancement related data, and constructing an acoustic wave rainfall enhancement database;

[0069] S11. The method described in this embodiment is applied to the prediction of acoustic rainfall corresponding to reservoir storage. The different reservoir terrain data will affect the propagation and reflection of the acoustic rainfall. Therefore, when arranging microphones in different reservoir environments, it is necessary to consider the orientation and height of the microphones to avoid placing the microphones in locations where sound reflection is obvious.

[0070] This embodiment collects acoustic rain enhancement sound data in different scenarios at different reservoir locations. The different scenarios are acoustic rain enhancement related data, covering meteorological conditions. Furthermore, the meteorological conditions include rainfall intensity, wind speed, wind direction, season and time, and the rainfall intensity is divided into heavy rain, moderate rain and light rain. The acoustic rain enhancement intensity affects the loudness and frequency components of the sound, and the wind speed and wind direction affect the propagation and hearing of the sound; the seasons are divided into spring, summer, autumn and winter, and the air temperature and humidity in different seasons are different, which affects the effect of sound wave propagation and the state of clouds; the time is divided into day and night, and the sound propagation characteristics of different time periods are also different.

[0071] Furthermore, the size and topographic data of the reservoir will also affect the propagation of sound. For example, the sound of rainfall over a large area of ​​water is different from that over a small reservoir. Therefore, the area size and topographic data of the reservoir are also collected as one of the acoustic rainfall enhancement related data. Preferably, the area size and topographic data can be obtained through a geographic information system (GIS).

[0072] According to the above analysis, this embodiment collects the sound data of acoustic rainfall enhancement in different reservoirs through a microphone such as a high-sensitivity microphone or an acoustic sensor, and obtains the corresponding acoustic rainfall enhancement related data. Different microphones are used to collect data in different reservoirs. The microphones should be able to accurately capture the sound of acoustic rainfall enhancement and have a sufficient frequency response range. When collecting data, try to choose an environment with clear rainfall sounds to reduce the interference of background noise. Preferably, the acoustic rainfall enhancement technology uses sound waves to stimulate the formation of water droplets in the clouds, and the sound data collected by the microphone in this embodiment is mainly to analyze the characteristics of sound in nature, and then infer the precipitation situation. In theory, the microphone is far away from the acoustic rainfall enhancement source, and the rainfall is under a certain protection. The sound wave is controlled within a certain range and will not affect the collection of sound data. However, if there is strong sound wave activity near the sound data collection point, or the sound wave instrument is very close to the source point of the sound wave rainfall enhancement, the sound wave may have a certain impact on the collection of sound data. Therefore, in order to ensure the accuracy of the sound data collected, a method of simultaneous recording of multiple collection points is adopted, and the uncertainty of single-point data is reduced by processing the data of multiple collection points. The processing includes but is not limited to average value calculation, weighted average, median processing, spatial interpolation processing and data fusion. In this embodiment, the uncertainty of single-point data is reduced by selecting the recorded average value as the final collected sound data;

[0073] S12, establish a database table structure to organize and store data;

[0074] In this embodiment, the acoustic wave rainfall enhancement database includes a reservoir information table, an acoustic wave rainfall enhancement sound data table and a meteorological condition table, wherein the reservoir information table includes a reservoir ID, a reservoir name, terrain data and a reservoir area, wherein the terrain data includes a slope and an altitude; the acoustic wave rainfall enhancement sound data table includes an acoustic wave rainfall enhancement sound data ID, an associated reservoir ID, a collection time, an acoustic wave rainfall enhancement sound data, a pickup height and an orientation; the meteorological condition table includes an associated acoustic wave rainfall enhancement sound data ID, an acoustic wave rainfall enhancement intensity, a wind speed, a wind direction, a season and a time;

[0075] According to the reservoir information table, the acoustic rain enhancement sound data table and the meteorological condition table and the relationship between them, the data on reservoirs, acoustic rain enhancement sounds and meteorological conditions can be organized and queried and subsequent analysis can be carried out;

[0076] S2, denoising the collected acoustic wave rainfall enhancement sound data to obtain denoised acoustic wave rainfall enhancement sound data;

[0077] The acoustic data of rain enhancement is a continuous signal when it is sampled. , for subsequent analysis, in continuous time Select a specific discrete time point , the continuous signal Convert to discrete signal , expressed as:

[0078] ;

[0079] In the formula, is the sampling period, n is the sampling point, which is an integer;

[0080] S21, denoising the collected acoustic rain-enhancing sound data by wavelet transform;

[0081] The acoustic rain enhancement sound data is converted from the time domain to the wavelet domain. Wavelet transforms include discrete wavelet transform (DWT) and continuous wavelet transform (CWT). In this embodiment, since the acoustic rain enhancement sound is a signal with mutation and non-stationary characteristics, DWT can effectively capture the instantaneous characteristics in the signal, so the selected wavelet transform is DWT;

[0082] Further, a suitable wavelet function is selected, and the wavelet function includes Haar, Daubechies, Symlet, etc., and the selection depends on the signal characteristics and the denoising target. In this embodiment, because the noise of the sound wave rain enhancement sound is often complex in time and frequency, and the Daubechies wavelet has compact support and good frequency localization characteristics, it can better maintain the main features of the signal in the denoising process and remove the noise at the same time. Therefore, the Daubechies wavelet is selected in the denoising process of the sound wave rain enhancement sound signal in this embodiment;

[0083] S211, select Daubechies wavelet for acoustic rain enhancement sound signal Discrete wavelet transform is performed and expressed as:

[0084] ;

[0085] In the formula, For the Tier wavelet coefficients, is the Daubechies wavelet function;

[0086] The wavelet coefficients are obtained by performing wavelet decomposition on the signal, including approximate coefficients and detail coefficients. The applicability of different Daubechies wavelets depends on specific application requirements and performance requirements. DB6 and DB8 provide higher resolution and finer detail representation, but the computational complexity is correspondingly higher. DB4 has a lower computational complexity and is sufficient to handle most noises. In this embodiment, the Daubechies 4 wavelet (DB4) filter is used for denoising. Specifically:

[0087] For acoustic rainfall enhancement sound signal , using the impulse response of the low-pass filter, i.e. the filter coefficients Perform convolution, expressed as:

[0088] ;

[0089] in:

[0090] The coefficients of the db4 low-pass filter are expressed as:

[0091] ;

[0092] The numerical approximation is ;

[0093] So the low pass filter coefficients The formula is:

[0094] ;

[0095] In the formula, is the approximate coefficient, represents convolution, is a continuous approximation of the basis function;

[0096] Impulse response using a high pass filter Perform convolution, expressed as:

[0097] ;

[0098] in:

[0099] The coefficients of the db4 high-pass filter are expressed as:

[0100] ;

[0101] The numerical approximation is ;

[0102] So the high pass filter The formula is:

[0103] ;

[0104] In the formula, is the detail coefficient;

[0105] The convolved acoustic rain enhancement sound signal is downsampled to obtain the approximate coefficient and detail coefficient component. In this embodiment, the downsampling is to take one sample every two samples. The obtained approximate coefficient and detail coefficient component are expressed as and , ;

[0106] S212, in the wavelet domain, threshold processing is performed on the wavelet coefficients to remove noise. The main purpose of threshold processing is to compress or set the noise coefficient to zero while retaining the main features of the signal. The threshold processing includes hard threshold and soft threshold, where:

[0107] The hard threshold is expressed as:

[0108] ;

[0109] In the formula, is the wavelet coefficient after hard threshold processing, including and , is the threshold value;

[0110] Since the approximate coefficient contains the main information of the acoustic signal of the acoustic rain enhancement, the approximate coefficient remains unchanged, and the detail coefficient is subjected to soft threshold processing, which is expressed as follows:

[0111] ;

[0112] In the formula, is the detail coefficient after soft threshold processing, is the symbolic function, is the threshold, where if the standard deviation of the wavelet coefficients is ,but , is an empirical constant, and in this embodiment, it is taken as 2; ;

[0113] S213, wavelet coefficients after hard threshold processing and detail coefficient after soft threshold processing Get the wavelet coefficients after threshold processing , using the wavelet coefficients Perform inverse wavelet transform to obtain the denoised acoustic rain-enhancing sound signal, which can be expressed as:

[0114] ;

[0115] In the formula, is the inverse discrete wavelet transform, specifically:

[0116] S2131, up-sample the approximate coefficients and the detail coefficients, insert zeros between each sample in the coefficients, expressed as:

[0117] ;

[0118] ;

[0119] That is, the original sample or or copy to a new sample location , and insert zeros in each sample;

[0120] S2132, using a wavelet reconstruction filter to filter the upsampled wavelet coefficients, expressed as:

[0121] ;

[0122] ;

[0123] in, and are the impulse responses of the low-pass filter and high-pass filter corresponding to the wavelet reconstruction filter, and the reconstruction filter coefficients are the mirror image and / or inversion operation of the decomposition filter coefficients, which will not be repeated here. and Specific value;

[0124] S2133, combining the wavelet coefficients after reconstruction and filtering to obtain a restored signal, that is, a denoised acoustic rain-enhancing sound signal , expressed as:

[0125] ;

[0126] It is worth noting that the method described in this embodiment takes db4 as an example, and does not limit the denoising selection of this application to only db4. Different Daubechies wavelets have different smoothness and resolution. Therefore, the user can make a choice based on the characteristics of the sound wave rain enhancement sound and the denoising requirements; each dbN wavelet has a corresponding decomposition filter and reconstruction filter. Although the coefficients may be different, they are similar in structure. Combined with the denoising steps and corresponding filter coefficients of the above db4, the corresponding denoising operation can be performed according to the denoising requirements of the user; the filter coefficients can be generated by using the function daubechies in the WaveletToolbox of MATLAB. Wavelet filter coefficients of Daubechies wavelets can also be obtained through the PyWavelets library in Python;

[0127] S3, constructing an acoustic wave rainfall enhancement rainfall prediction model, the acoustic wave rainfall enhancement rainfall prediction model comprising a feature extractor, a gradient inversion layer, a domain classifier and a main classifier; constructing a training set and a test set according to the acoustic wave rainfall enhancement database combined with the denoised acoustic wave rainfall enhancement sound data, training the acoustic wave rainfall enhancement rainfall prediction model, and obtaining a trained acoustic wave rainfall enhancement rainfall prediction model;

[0128] S31, the feature extractor is used to extract useful features from the input data so that the subsequent classifier can perform effective classification; in this embodiment, the input data includes sound data, reservoir data and meteorological data, and the sound data is the denoised acoustic rain-increasing sound signal , the reservoir data is terrain data and reservoir area, and the meteorological data is wind speed, wind direction, season and time; in this embodiment, the feature extractor includes a sound feature extraction module, a reservoir feature extraction module and a meteorological feature extraction module, wherein:

[0129] S311, sound feature extraction module;

[0130] The sound feature extraction module is used to extract frequency domain and time domain features from the denoised acoustic rain enhancement sound signal to describe the characteristics of the sound;

[0131] S3111, frequency domain feature extraction;

[0132] Use Fast Fourier Transform FFT to convert the time domain signal into the frequency domain signal. The filter bank performs the spectrum Scale conversion, yes The spectrum is logarithmically transformed to obtain Spectrum, from Computing in the Spectrum Frequency cepstrum coefficients, and spectrum flatness is obtained by measuring the flatness of the spectrum;

[0133] Furthermore, the time domain signal is converted into a frequency domain signal using the fast Fourier transform FFT and is expressed as follows:

[0134] ;

[0135] In the formula, is the frequency domain signal, is the total number of sampling points of the acoustic rain enhancement sound signal, is the frequency index;

[0136] Applying the magnitude spectrum filter bank to convert it from a linear frequency scale Frequency scale, each The filter response is a triangular window function. The frequency scale covers a frequency range. The scale is a nonlinear frequency scale based on human auditory perception and is expressed as:

[0137] ;

[0138] In the formula, for Frequency index; for Frequency on scale;

[0139] right The spectrum is logarithmically transformed to obtain The spectrum enhances the distinguishability of the signal and makes the subsequent features more stable, which can be expressed as:

[0140] ;

[0141] In the formula, For the indivual The energy of the filter bank output;

[0142] from Computing in the Spectrum Frequency cepstrum coefficients, specifically, The spectrum is calculated by discrete cosine transform The frequency cepstrum coefficient is expressed as:

[0143] ;

[0144] In the formula, For the indivual Frequency cepstral coefficients; for The spectrum The frequency index;

[0145] The spectrum flatness is obtained by measuring the flatness of the spectrum. The spectrum flatness is an indicator to measure the smoothness of the spectrum and reflects the noise level of the signal. The measurement steps include:

[0146] Calculate the geometric mean of the spectrum , expressed as:

[0147] ;

[0148] In the formula, The spectrum value for each frequency point The logarithm of ;

[0149] Calculate the arithmetic mean of the spectrum , expressed as:

[0150] ;

[0151] The spectrum flatness is obtained by calculating the ratio of the arithmetic mean and the geometric mean. , expressed as:

[0152] ;

[0153] The closer the value of spectrum flatness is to 1, the flatter the spectrum is, and the smaller the value is, the sharper the spectrum is.

[0154] S3112, time domain feature extraction;

[0155] The time domain features include a zero-crossing rate and a first normalized autocorrelation coefficient, wherein:

[0156] The zero crossing rate Refers to the number of times the acoustic signal of rain-enhancing sound waves crosses the zero axis per unit time. It is used to describe the noise component of the acoustic signal of rain-enhancing sound waves and the stability of the signal. It is expressed by the formula:

[0157] ;

[0158] ;

[0159] The first normalized autocorrelation coefficient is used to measure the similarity of the acoustic rain enhancement sound signal at different time delays. First, the autocorrelation function is calculated. , expressed as:

[0160] ;

[0161] In the formula, is the discrete delay amount, Indicates that the acoustic signal of the acoustic rain enhancement moves in time Self-similarity after sampling points;

[0162] The normalized autocorrelation coefficient is the normalized form of the autocorrelation function, expressed as:

[0163] ;

[0164] In the formula, for The normalized form of ;

[0165] The first normalized autocorrelation coefficient is The normalized autocorrelation coefficient at is expressed as:

[0166] ;

[0167] The frequency domain and time domain features of the acoustic rain enhancement sound signal are extracted through steps S3111 and S3112, and the frequency domain features and the time domain features are fused to obtain comprehensive sound features. In this embodiment, the feature fusion is splicing fusion, which directly splices all features into a long vector.

[0168] S312, reservoir feature extraction module;

[0169] The reservoir feature extraction module is a fully connected neural network, including an input layer, a hidden layer and a comprehensive feature extraction layer. The input layer is used to receive the numerical values ​​of the reservoir area, slope and altitude. The hidden layer includes two fully connected layers, each with 128 neurons, and uses the ReLU activation function to learn the complex relationship of the features and extract the deep features of the reservoir area, slope and altitude. The comprehensive feature extraction layer is a fully connected layer, including 64 neurons, which is used to extract comprehensive features and also uses the ReLU activation function; the comprehensive reservoir features are obtained through the reservoir feature extraction module;

[0170] S313, meteorological feature extraction module;

[0171] The meteorological feature extraction module includes an input layer, a wind speed layer, a wind direction layer, a season layer, a time layer and a feature fusion layer, wherein:

[0172] In this embodiment, four input layers are defined, namely, continuous value input of wind speed, categorical value input of wind direction, one-hot encoding input of season, and binary value input of time. The one-hot encoding of season is defined as follows in this embodiment: spring: [1, 0, 0, 0]; summer: [0, 1, 0, 0]; autumn: [0, 0, 1, 0]; winter: [0, 0, 0, 1];

[0173] Different types of features are processed through the wind speed layer, wind direction layer, season layer and time layer. Specifically, the wind speed layer is a fully connected layer, which maps the wind speed value to a hidden layer with 16 neurons to obtain the wind speed feature; the wind direction layer maps the wind direction value (assuming the maximum angle is 360°) to an embedding space with 8 dimensions, and flattens the output of the embedding layer into one dimension to obtain the wind direction feature; the season layer is a fully connected layer, which maps the unique hot encoding of the season to a hidden layer with 8 neurons to obtain the season feature; the time layer is a fully connected layer, which maps the binary input of time to a hidden layer with 4 neurons to obtain the time feature;

[0174] The above features are input into the feature fusion layer for comprehensive feature extraction. The feature fusion layer includes two fully connected layers, which are mapped to hidden layers of 64 and 32 neurons respectively to obtain comprehensive meteorological features;

[0175] According to the steps of S311, S312 and S313, the comprehensive sound features, the comprehensive reservoir features and the comprehensive meteorological features are obtained, and the feature extractor also includes a comprehensive feature fusion module, and the comprehensive feature fusion module is used to fusion the comprehensive sound features. , Comprehensive reservoir characteristics and comprehensive meteorological characteristics Perform standardization such as zero mean unit variance standardization (Z-score standardization), and fuse the standardized features to obtain comprehensive features In this implementation, the fusion adopts weighted average fusion, which is expressed as follows:

[0176] ;

[0177] In the formula, , and are preset weights of comprehensive sound features, comprehensive reservoir features and comprehensive meteorological features. In this embodiment, since acoustic rainfall enhancement will cause changes in sound features such as water flow sound, wave sound and water level change in the reservoir, the actual condition of the reservoir can be obtained more directly by analyzing the sound features, thereby more accurately predicting the reservoir storage and rainfall intensity. The main feature of this embodiment is the sound feature, so the preset is high weight;

[0178] S32, gradient reversal layer and domain classifier;

[0179] In the data collection of step S1, the acoustic data of rainfall enhancement in different scenarios at different reservoir locations are collected, that is, there are multi-reservoir data. In the case of multi-reservoir data, the data of each reservoir can be regarded as an independent field, and the data of each field has different feature distributions;

[0180] For the applicability of the acoustic rainfall enhancement rainfall prediction model, this embodiment uses domain adaptation to enable the acoustic rainfall enhancement rainfall prediction model to be effectively migrated and adapted between the source domain (such as the data of reservoir A) and the target domain (such as the data of reservoirs B and C). Specifically:

[0181] The comprehensive features output by the feature extractor are sent to the gradient reversal layer. The function of the gradient reversal layer is to multiply the gradient of the domain classifier by a negative coefficient during back propagation. Therefore, the comprehensive features are not changed when they are forward propagated to the domain classifier. When the comprehensive features are transmitted to the domain classifier, the domain classifier determines the domain of the data source based on the comprehensive features. The domain classifier calculates the loss based on the predicted results and the true domain label. The gradient of its loss function is processed by the gradient reversal layer, so that the gradient of back propagation is reversed; the reversed gradient is then passed to the domain classifier;

[0182] Through back propagation, the feature extractor adjusts the weights according to the inverted gradient, so that it can learn features that are difficult for the domain classifier to distinguish, thereby achieving domain adaptation;

[0183] Furthermore, the domain classifier consists of three fully connected layers. The input comprehensive features pass through the first fully connected layer, apply the ReLU activation function to introduce nonlinearity, pass through the second fully connected layer, and also apply the ReLU activation function to the final output layer to map the output of the hidden layer to the corresponding domain.

[0184] The training process of the above steps includes forward propagation, loss calculation and back propagation, where:

[0185] The forward propagation specifically obtains comprehensive features through the feature extractor and passes through the gradient reversal layer, which is expressed as follows:

[0186] ;

[0187] In the formula, Represents the gradient reversal layer Directly transmit the comprehensive features during forward propagation. Preferably, the back propagation is specifically a gradient reversal layer that multiplies the gradient by a negative coefficient, such as -1, so as to prompt the feature extractor to learn features that are difficult to distinguish for the domain classifier;

[0188] The domain prediction is performed through the domain classifier, which is expressed as follows:

[0189] ;

[0190] In the formula, The domain classifier predicts the domain label;

[0191] The loss in the calculation loss is the cross entropy loss. The domain classification loss is used to evaluate the gap between the prediction result of the domain classifier and the true domain label, which is expressed as:

[0192] ;

[0193] In the formula, is the number of samples, For the domain classifier The predicted probability of samples;

[0194] The goal of the domain classifier is to minimize the domain classification loss, and the weights of the domain classifier are updated by a gradient descent algorithm such as Adam or SGD; and because of the gradient reversal layer, the optimization goal of the feature extractor is to maximize the domain classification loss;

[0195] S33, main classifier;

[0196] The comprehensive features output by the feature extractor after weight adjustment are used as the input of the main classifier, and the output of the main classifier is the rainfall intensity;

[0197] Wherein, in step S2, denoised acoustic wave rainfall enhancement sound data is obtained, and according to the acoustic wave rainfall enhancement database, the denoised acoustic wave rainfall enhancement sound data is combined with the corresponding reservoir data, meteorological data and the acoustic wave rainfall enhancement intensity in the meteorological condition table to construct a training set and a test set, wherein the denoised acoustic wave rainfall enhancement sound data, the reservoir data and the meteorological data are used as input data, and the acoustic wave rainfall enhancement intensity is used as output data;

[0198] In this embodiment, the total feature dimension of the comprehensive feature is 11. The comprehensive feature is regarded as one-dimensional, and the input dimension of the input layer is ,in, is the number of samples input into the main classifier, and the main classifier further includes convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, flattening layer, fully connected layer 1, fully connected layer 2 and output layer, wherein: the number of convolution kernels of convolution layer 1, convolution layer 2 and convolution layer 3 are 32, 64 and 128 respectively, and the size of convolution kernel is 1 3, the activation function is ReLU function, the stride is 1; the pooling type of pooling layer 1, pooling layer 2 and pooling layer 3 is maximum pooling, and the pooling kernel size is 1 2, the stride is 2; the number of neurons in the fully connected layer 1 and the fully connected layer 2 are 512 and 256 respectively, and the activation function is the ReLU function; the number of neurons in the output layer is 3, corresponding to the heavy rain, moderate rain and light rain of the acoustic rain enhancement intensity respectively, and the activation function is the softmax function;

[0199] The main classifier uses convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, and flattening layer to extract features, classifies through fully connected layer 1 and fully connected layer 2, and finally outputs the category of acoustic wave rainfall enhancement intensity;

[0200] S4. Use the trained acoustic rainfall prediction model to estimate the acoustic rainfall that users need to perform acoustic rainfall to increase reservoir storage, and formulate a reasonable reservoir storage plan based on the estimated acoustic rainfall data to ensure the rational use of water resources;

[0201] Furthermore, the acoustic wave rainfall enhancement rainfall prediction model outputs the acoustic wave rainfall enhancement intensity, and the acoustic wave rainfall enhancement intensity includes heavy rain, moderate rain and light rain. In the present embodiment, the heavy rain corresponds to a rainfall of more than 7.7 mm / hour; the moderate rain corresponds to a rainfall of 2.6 mm / hour to 7.6 mm / hour; the light rain corresponds to a rainfall of 0.1 mm / hour to 2.5 mm / hour. Preferably, the classification standard described in the present embodiment is only an example, and users can divide it according to the standards of local meteorological agencies.

[0202] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0203] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0204] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0205] 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 specification 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. A method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage, characterized in that: The method comprises: Collecting acoustic rain-enhancing sound data through a microphone, and obtaining corresponding acoustic rain-enhancing related data, wherein the acoustic rain-enhancing related data includes meteorological condition data and reservoir data, and constructing an acoustic rain-enhancing database using the acoustic rain-enhancing sound data and the acoustic rain-enhancing related data; The collected acoustic rain enhancement sound data are denoised by using wavelet transform to obtain the denoised acoustic rain enhancement sound data, and a training set and a test set are constructed based on the acoustic rain enhancement database and the denoised acoustic rain enhancement sound data; Constructing an acoustic wave rainfall enhancement rainfall prediction model, the acoustic wave rainfall enhancement rainfall prediction model includes a feature extractor, a gradient reversal layer, a domain classifier and a main classifier, obtaining comprehensive features including sound features, meteorological features and reservoir features through the feature extractor, inputting the comprehensive features into the gradient reversal layer and the domain classifier to complete the weight update of the feature extractor, and inputting the comprehensive features output by the feature extractor after the weight update into the main classifier to complete the acoustic wave rainfall enhancement rainfall prediction; using the training set to train the acoustic wave rainfall enhancement rainfall prediction model to obtain a trained acoustic wave rainfall enhancement rainfall prediction model; The trained acoustic rainfall prediction model is used to estimate the acoustic rainfall that users need to perform acoustic rainfall enhancement to achieve reservoir storage, and a reservoir storage plan is formulated based on the estimated acoustic rainfall data. Wherein, the wavelet transform adopts Daubechies wavelet filter, and the Daubechies wavelet filter includes a low-pass filter and a high-pass filter; Impulse responses using low-pass and high-pass filters respectively and Acoustic Rainfall Enhancement Sound Signal Convolution is performed, which is expressed as: ; ; In the formula, is the approximate coefficient, is the detail coefficient; The convolved acoustic rain enhancement sound signal is downsampled to obtain the components of the approximate coefficient and the detail coefficient, which constitute the wavelet coefficient. , expressed as: ; In the formula, is the approximate coefficient component, is the detail coefficient component; For the Tier wavelet coefficients; The hard threshold processing of wavelet coefficients is expressed as follows: ; In the formula, is the wavelet coefficient after hard threshold processing, including the approximate coefficient after hard threshold processing and detail factor , is the threshold value; The detail coefficient is processed by soft threshold, which is expressed as follows: ; ; In the formula, is the detail coefficient after soft threshold processing, is the symbolic function, is the threshold, where if the standard deviation of the wavelet coefficients is ,but , is an empirical constant; According to the wavelet coefficients after hard thresholding And the detail coefficient after soft threshold processing Get the wavelet coefficients after threshold processing , using the wavelet coefficients Perform inverse wavelet transform to obtain the denoised acoustic rain-enhancing sound signal , expressed as: ; In the formula, is the inverse wavelet transform, specifically the wavelet coefficients Up-sampling is performed, and the up-sampled wavelet coefficients are filtered and combined using a wavelet reconstruction filter to obtain the denoised acoustic rain-enhancing sound signal.

2. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 1 is characterized in that: When collecting the acoustic wave rainfall enhancement sound data through the microphone, a method of simultaneously recording at multiple collection points is adopted, and data processing is performed on the recorded data at the multiple collection points to obtain the acoustic wave rainfall enhancement sound data.

3. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 1 is characterized in that: The constructed acoustic wave rainfall enhancement database includes a reservoir information table, an acoustic wave rainfall enhancement sound data table and a meteorological condition table, wherein the reservoir information table includes a reservoir ID, a reservoir name, topographic data and a reservoir area, wherein the topographic data includes a slope and an altitude; the acoustic wave rainfall enhancement sound data table includes an acoustic wave rainfall enhancement sound data ID, an associated reservoir ID, a collection time, an acoustic wave rainfall enhancement sound data, a microphone height and an orientation; the meteorological condition table includes an associated acoustic wave rainfall enhancement sound data ID, an acoustic wave rainfall enhancement intensity, a wind speed, a wind direction, a season and a time, wherein the acoustic wave rainfall enhancement intensity includes heavy rain, moderate rain and light rain, the season includes spring, summer, autumn and winter, and the time includes day and night.

4. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 1 is characterized in that: The feature extractor includes a sound feature extraction module, a reservoir feature extraction module and a meteorological feature extraction module. The sound feature extraction module is used to extract frequency domain and time domain features from the denoised acoustic rain-increasing sound signal to obtain comprehensive sound features; the reservoir feature extraction module is used to extract features from reservoir area, slope and altitude data and obtain comprehensive reservoir features; the meteorological feature extraction module is used to extract features from wind speed, wind direction, season and time and obtain comprehensive meteorological features; The feature extractor fuses the comprehensive sound features, comprehensive reservoir features and comprehensive meteorological features to obtain the comprehensive features.

5. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 4 is characterized in that: The frequency domain features extracted by the sound feature extraction module include The frequency cepstrum coefficients and spectrum flatness, the extraction steps are as follows: The denoised acoustic rain-enhancing sound signal is converted from a time domain signal to a frequency domain signal using the Fast Fourier Transform (FFT). Filter Bank Scale conversion to obtain Spectrum, yes The spectrum is logarithmically transformed to obtain Spectrum, from Computing in the Spectrum Frequency cepstrum coefficients, and spectrum flatness is obtained by measuring the flatness of the spectrum; The time domain features extracted by the sound feature extraction module include the zero crossing rate and the first normalized autocorrelation coefficient. The formula is: ; In the formula, is the denoised acoustic rain-enhancing sound signal, is the total number of sampling points of the acoustic rain enhancement sound signal; The first normalized autocorrelation coefficient extraction step comprises: Calculate the autocorrelation function , expressed as: ; In the formula, is the discrete delay amount, Indicates that the acoustic signal of the acoustic rain enhancement moves in time Self-similarity after sampling points; The normalized autocorrelation coefficient is the normalized form of the autocorrelation function, expressed as: ; In the formula, for The normalized form of ; The first normalized autocorrelation coefficient is The normalized autocorrelation coefficient at is expressed as: 。 6. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 4 is characterized in that: The reservoir feature extraction module is a fully connected neural network, including an input layer, a hidden layer and a comprehensive feature extraction layer, wherein: The input layer is used to receive the numerical values ​​of the reservoir area, slope and altitude. The hidden layer includes two fully connected layers using the ReLU activation function, each with 128 neurons, which are used to extract deep features of the reservoir area, slope and altitude; the comprehensive feature extraction layer is a fully connected layer using the ReLU activation function, containing 64 neurons, which is used to extract comprehensive reservoir features.

7. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 4 is characterized in that: The meteorological feature extraction module includes an input layer, a wind speed layer, a wind direction layer, a season layer, a time layer and a feature fusion layer, wherein: The input layer includes four input layers: continuous value input of wind speed, categorical value input of wind direction, custom one-hot encoding input of season, and binary value input of time; Different types of features are processed through the wind speed layer, wind direction layer, season layer and time layer. Specifically, the wind speed layer is a fully connected layer, which maps the wind speed value to a hidden layer with 16 neurons to obtain the wind speed feature; the wind direction layer maps the wind direction value to an embedding space with 8 dimensions, and flattens the output of the embedding layer into one dimension to obtain the wind direction feature; the season layer is a fully connected layer, which maps the unique hot encoding of the season to a hidden layer with 8 neurons to obtain the season feature; the time layer is a fully connected layer, which maps the binary input of time to a hidden layer with 4 neurons to obtain the time feature; The wind speed features, wind direction features, season features and time features are input into the feature fusion layer for comprehensive feature extraction. The feature fusion layer includes two fully connected layers, which are mapped to hidden layers of 64 and 32 neurons respectively to obtain comprehensive meteorological features.

8. The method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 4 is characterized in that: The acoustic rain enhancement rainfall prediction model sends the comprehensive features output by the feature extractor to the gradient reversal layer, and the comprehensive features are input to the domain classifier through the gradient reversal layer. The domain classifier judges the domain according to the comprehensive features after the gradient reversal. The gradient of the loss function of the domain classifier is reversed through the gradient reversal layer during the back propagation process and transmitted to the feature extractor. The weight of the feature extractor is updated according to the inverted gradient. The domain classifier is composed of three fully connected layers. The comprehensive features pass through the first fully connected layer, apply the ReLU activation function to introduce nonlinearity, and then pass through the second fully connected layer. The ReLU activation function is also applied to the final output layer to map the output of the hidden layer to the corresponding domain; the loss function of the domain classifier is the cross entropy function.

9. A method for estimating the corresponding rainfall amount of acoustic rainfall enhancement for reservoir storage according to claim 8, characterized in that: The comprehensive features output by the weighted feature extractor are used as the input of the main classifier, which includes an input layer, a convolution layer 1, a pooling layer 1, a convolution layer 2, a pooling layer 2, a convolution layer 3, a pooling layer 3, a flattening layer, a fully connected layer 1, a fully connected layer 2, and an output layer, wherein: The input latitude of the input layer is , is the number of samples input into the main classifier; The number of convolution kernels in the convolution layer 1, convolution layer 2, and convolution layer 3 are 32, 64, and 128 respectively, and the size of the convolution kernel is 1 3, the activation function is ReLU function, the stride is 1; the pooling type of pooling layer 1, pooling layer 2 and pooling layer 3 is maximum pooling, and the pooling kernel size is 1 2, the stride is 2; the number of neurons in the fully connected layer 1 and the fully connected layer 2 are 512 and 256 respectively, and the activation function is the ReLU function; the number of neurons in the output layer is 3, corresponding to the heavy rain, moderate rain and light rain of the acoustic rain enhancement intensity respectively, and the activation function is the softmax function; The main classifier uses convolution layer 1, pooling layer 1, convolution layer 2, pooling layer 2, convolution layer 3, pooling layer 3, and flattening layer to extract features, classifies through fully connected layer 1 and fully connected layer 2, and finally outputs the acoustic wave rainfall enhancement intensity.