Island detection method, computer program product, storage medium and equipment
By decomposing the common coupling point voltage and inverter output current signal of the distributed grid system, the energy entropy, fuzzy entropy and sample entropy characteristics are calculated and fused, and inputting them into the neural network, solving the problem of low detection accuracy of single feature The problem has improved the accuracy of island detection.
Patent Information
- Application Number
- CN202410801895.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-20
- Publication Date
- 2024-11-12
AI Technical Summary
When using a single entropy feature for island detection in the prior art, the eigenvalues are not comprehensive enough, resulting in low detection accuracy.
Wavelet packet decomposition is performed by obtaining the common coupling point voltage signal of the distributed grid system and the inverter output current signal, and any two or three of energy entropy, fuzzy entropy, and sample entropy are calculated, and fused into the fusion entropy feature , input it into the trained island detection neural network for judgment.
Multi-scale information is obtained, which reflects fault information more comprehensively than a single feature, improves detection accuracy and avoids detection inaccuracy problems caused by the use of a single entropy value.
Smart Images

Figure CN118937882A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of island detection in power grid systems, and in particular relates to an island detection method, a computer program product, a storage medium, and a device. Background Art
[0002] With the widespread access to inverter-type distributed energy such as photovoltaics, the distribution network may enter an islanded operation state when the external power supply is interrupted. In the islanded state, the connection between the distribution network and the external power grid is cut off, resulting in a loss of control of the internal power supply, which may cause a variety of problems, including power reverse transmission, safety risks for maintenance personnel, degradation of power quality, continuous output of fault current, reduced sensitivity of relay protection and automation equipment, and complication of the distribution network operation mode.
[0003] The detection technology of islanding effect can be divided into two categories according to whether it relies on remote communication: local detection method and remote detection method. The remote detection method relies on the power communication network to perform tasks such as criterion collection, logic analysis and command issuance, including power line carrier communication method, interlocking tripping method and telecontrol release method. Although this method is effective, it needs to rely on the existing power communication infrastructure, so the cost is relatively high. In contrast, the local detection method is divided into passive and active detection methods according to whether interference signals are injected into the power grid. The active detection method identifies the islanding effect by introducing frequency, voltage or phase disturbances. Although this method can effectively identify islands, it may have a negative impact on the power quality of the power grid. The passive detection method identifies the islanding effect by monitoring whether the electrical parameters exceed the preset threshold, such as voltage and frequency anomalies. This method has a stable system, but under different working conditions, it may cause misjudgment or missed judgment due to interference factors, increasing the uncertainty of detection.
[0004] In order to more accurately detect the islanding effect of the distribution network without affecting the normal operation of the distribution network, the Chinese invention patent application with application publication number CN114156955A proposed a detection method combining wavelet packet energy entropy and BP neural network. However, in this method, only the energy entropy is calculated and a single feature quantity is used for detection. The Chinese invention patent application with application publication number CN108108659A proposed a technical solution for islanding detection using singular spectrum entropy. However, these solutions all use a single feature quantity and cannot obtain more comprehensive feature information. Since the feature information input into the neural network is relatively one-sided, it also has a certain impact on the output result of the neural network, resulting in a greater impact on the recognition result; limiting its accuracy in islanding detection. Summary of the invention
[0005] The object of the present invention is to provide an island detection method and a computer program product, a storage medium, and a device to solve the technical problem of low detection accuracy caused by insufficient eigenvalues when using a single entropy for island detection in the prior art.
[0006] In order to solve the above technical problems, the present invention provides an island detection method, comprising the following steps:
[0007] The power data signal of the distributed power grid system is obtained, and whether the distributed power grid system is in an island state is preliminarily determined according to the power data signal; if the preliminary determination result is that it is not in an island state, the common coupling point voltage signal and the inverter output current signal of the distributed power grid system are obtained, and the voltage signal and the current signal are subjected to wavelet packet decomposition; any two or three of the energy entropy, fuzzy entropy and sample entropy are calculated according to the wavelet packet decomposition result, and all the calculated entropies are fused to obtain a fused entropy feature; the fused entropy feature is input into the trained island detection neural network to further determine whether the distributed power grid system is in an island state.
[0008] Furthermore, a method for fusing all the calculated entropies to obtain a fused entropy feature is as follows: all the calculated entropies are input into a trained multi-head self-attention neural network to obtain a fused entropy feature.
[0009] Furthermore, the island detection neural network is a bidirectional long-short term neural network.
[0010] Furthermore, the power data signal is a common coupling point voltage. The method for preliminarily determining whether the distributed power grid system is in an island state based on the power data signal is: determining whether the common coupling point voltage exceeds a corresponding threshold value. If the voltage exceeds the corresponding threshold value, it is determined that the distributed power grid system is in an island state.
[0011] The present invention is an improved invention creation, and its beneficial effects are as follows: the island detection method of the present invention performs wavelet decomposition on the common coupling point voltage and inverter output current signals in the distributed power grid system to be tested, and then calculates any two or three of energy entropy, sample entropy and fuzzy entropy, and fuses the calculated entropies to obtain fused entropy features to obtain multi-scale information. Compared with a single feature, it reflects fault information more comprehensively, can better retain the original feature information, and avoids the problem of inaccurate detection caused by using a single entropy value. At the same time, experiments show that the three entropies of energy entropy, sample entropy and fuzzy entropy have high accuracy in island detection. Therefore, the fused entropy features are input into the trained island detection neural network to determine whether the distributed power grid system to be tested is in an island state, and more comprehensive entropy features can be used to improve the detection accuracy.
[0012] To solve the above technical problem, the present invention further provides a computer program product, including a computer program / instruction, wherein the computer program / instruction is used to implement the method steps described in the island detection method of the present invention when executed by a processor.
[0013] The present invention is an improved invention creation, and its beneficial effects are as follows: when the computer program product of the present invention is executed, the common coupling point voltage and inverter output current signal in the distributed power grid system to be tested are subjected to wavelet decomposition, and any two or three of energy entropy, sample entropy and fuzzy entropy are calculated, and the calculated entropies are fused to obtain fused entropy features to obtain multi-scale information. Compared with a single feature, the fault information is more comprehensive, the original feature information can be better retained, and the problem of inaccurate detection caused by using a single entropy value is avoided. At the same time, experiments show that the three entropies of energy entropy, sample entropy and fuzzy entropy have higher accuracy in island detection. Therefore, the fused entropy features are input into the trained island detection neural network to determine whether the distributed power grid system to be tested is in an island state, and more comprehensive entropy features can be used to improve the detection accuracy.
[0014] To solve the above technical problem, the present invention further provides a computer-readable storage medium having a computer program / instruction stored thereon, wherein the computer program / instruction is used to implement the method steps described in the island detection method of the present invention when executed by a processor.
[0015] The present invention is an improved invention creation, and its beneficial effects are as follows: when the computer program / instruction stored on the computer-readable storage medium of the present invention is executed, the common coupling point voltage and inverter output current signal in the distributed power grid system to be tested can be subjected to wavelet decomposition, and any two or three of energy entropy, sample entropy and fuzzy entropy can be calculated, and the calculated entropies are fused to obtain fused entropy features to obtain multi-scale information. Compared with a single feature, the fault information is more comprehensive, the original feature information can be better retained, and the problem of inaccurate detection caused by using a single entropy value is avoided. At the same time, experiments show that the three entropies of energy entropy, sample entropy and fuzzy entropy have high accuracy in island detection. Therefore, the fused entropy features are input into the trained island detection neural network to determine whether the distributed power grid system to be tested is in an island state, and more comprehensive entropy features can be used to improve the detection accuracy.
[0016] To solve the above technical problem, the present invention further provides a computer device, including a processor, a memory and a computer program stored in the memory, wherein the processor is used to execute the computer program to implement the method steps described in the island detection method of the present invention.
[0017] The present invention is an improved invention creation, and its beneficial effects are as follows: the computer program device of the present invention, when the processor executes the computer program stored in the memory, can perform wavelet decomposition on the common coupling point voltage and the inverter output current signal in the distributed power grid system to be tested, and calculate any two or three of the energy entropy, sample entropy and fuzzy entropy, and fuse the calculated entropies to obtain fused entropy features to obtain multi-scale information. Compared with a single feature, it reflects fault information more comprehensively, can better retain the original feature information, and avoids the problem of inaccurate detection caused by using a single entropy value. At the same time, experiments show that the three entropies of energy entropy, sample entropy and fuzzy entropy have high accuracy in island detection. Therefore, the fused entropy features are input into the trained island detection neural network to determine whether the distributed power grid system to be tested is in an island state, and more comprehensive entropy features can be used to improve the detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 is a flow chart of an island detection method according to a method embodiment of the present invention;
[0019] Figure 2 It is a structural diagram of the LSTM of the method embodiment of the present invention;
[0020] Figure 3 is a BiLSTM structure diagram of a method embodiment of the present invention;
[0021] Figure 4 The figure is a flowchart of the training and recognition of the neural network of the method embodiment of the present invention. DETAILED DESCRIPTION
[0022] The island detection method of the present invention performs wavelet decomposition on the common coupling point voltage and inverter output current signals in the distributed power grid system to be tested, and then calculates any two or three of energy entropy, sample entropy and fuzzy entropy, and fuses the calculated entropies to obtain fused entropy features to obtain multi-scale information. Compared with a single feature, it reflects fault information more comprehensively, can better retain the original feature information, and avoids the problem of inaccurate detection caused by using a single entropy value. At the same time, experiments show that the three entropies of energy entropy, sample entropy and fuzzy entropy have high accuracy in island detection. Therefore, the fused entropy features are input into the trained island detection neural network to determine whether the distributed power grid system to be tested is in an island state, and more comprehensive entropy features can be used to improve the detection accuracy.
[0023] Method Example:
[0024] In this embodiment, a distributed photovoltaic power generation system is taken as an example for explanation. As other implementations, the island detection method based on entropy feature fusion of the present invention can also be applied to island detection of distributed power grid systems such as distributed wind power generation systems and distributed energy storage.
[0025] The island detection method based on entropy feature fusion of the present invention has the following detection process: Figure 1 As shown, the specific steps include:
[0026] Step 1: Preliminarily determine whether it is in an islanding state based on the voltage threshold.
[0027] First, the voltage and output current of the distributed photovoltaic inverter power supply flowing into the common coupling point are collected, and the threshold is used to determine whether it is in an island state; if the voltage at the common coupling point exceeds the set voltage threshold, it is preliminarily determined to be in an island state, and the judgment process is directly terminated without further judgment. When the judgment result is no, although the voltage threshold has not exceeded the limit, it is still possible to be in an island state, so it is necessary to proceed to the subsequent steps to further determine whether it is in an island state. In this embodiment, the common coupling point voltage is used to determine whether it is in an island state. As other implementation methods, other power system data can also be used.
[0028] Step 2: Perform wavelet packet decomposition on the common coupling point voltage and inverter output current, and obtain the fusion entropy feature based on the decomposition results.
[0029] For distributed photovoltaic power generation systems, the voltage at the point of common coupling (PCC) will change before and after the island is generated. Through signal extraction, analysis and processing, the key information in the island and non-island states can be identified, and then the mode can be distinguished. Wavelet transform provides great convenience in signal extraction due to its localized characteristics in both time and frequency domains. However, traditional wavelet transform can only further decompose the low-frequency part of the signal, which may lead to the loss of some key feature information. Wavelet packet decomposition technology can fully capture the transient characteristics of the fault by re-decomposing the high-frequency and low-frequency components at each decomposition level, providing more possibilities for in-depth analysis of fault signals.
[0030] Therefore, in this embodiment, the collected voltage of the common coupling point and the inverter output current of the distributed photovoltaic power generation system are decomposed by wavelet packets. The specific wavelet packet decomposition method is as follows:
[0031] Wavelet packet decomposition decomposes the signal by constructing a pair of low-pass filters L(k) and high-pass filters H(k). At each level of signal decomposition, not only the low-frequency signal is further decomposed, but also the high-frequency signal is further decomposed. The selected wavelet function ψ(t) and scale function φ(t) satisfy the following formula:
[0032]
[0033] Where k is the translation parameter and t is the time.
[0034] The decomposition equation of the wavelet packet is as follows:
[0035]
[0036] In the formula, h k-2i and l k-2i respectively represent the high-pass and low-pass filter coefficients of the wavelet packet decomposition. d is the frequency sub-band of each layer, j is the scale parameter, k is the translation parameter, and n is the frequency parameter. The process of wavelet packet decomposition is essentially to decompose the signal into different frequency bands through a group of high-pass and low-pass filters.
[0037] For the decomposed wavelet signals, reconstruction is performed to obtain the reconstructed signals of each node to achieve lossless recovery and feature extraction of the signals.
[0038] The wavelet reconstruction equation is as follows:
[0039]
[0040] In the formula, h i-2k and l i-2k respectively represent the high-pass and low-pass filter coefficients of the wavelet packet decomposition.
[0041] Using time domain and frequency domain as fault feature quantities is vulnerable to environmental noise. Entropy features can characterize the complexity of signal sequences, have strong anti-interference ability, and can effectively extract fault information in signals. In the present invention, at least two of energy entropy, fuzzy entropy, and sample entropy are used for entropy feature fusion, and multiple parameters that can reflect the characteristic changes before and after the fault signal are used for fusion to enhance the characterization ability of the fault signal and improve the accuracy and efficiency of fault detection.
[0042] As a key reflection of fault information, energy entropy can reveal the degree of change during the fault occurrence process. By performing wavelet packet decomposition on the fault signal and combining the information entropy theory, the energy distributions of the PCC point voltage and the inverter output current at different scales before and after the occurrence of the islanding effect are calculated, thereby obtaining the energy entropy value reflecting the fault characteristics. The energy entropy calculation is as follows:
[0043]
[0044] In the formula, d i is the reconstructed signal obtained at the i-th node; E is the obtained energy entropy.
[0045] Fuzzy entropy is a measure of the complexity of a time series. The more complex the signal time series, the larger its entropy value. Fuzzy entropy has the advantages of small required data volume, good consistency, and strong anti-noise ability. For the N-point time series {x(i): 1 < i < N}, m-bit vectors are established in sequence, and its formula is as follows:
[0046]
[0047] In the formula, r represents the width of the control exponential function boundary, n represents the gradient of the control exponential function boundary, and F m (n,r) represents the relationship dimension under the m dimension.
[0048] Sample entropy measures the complexity of time series by measuring the probability of generating new patterns in the signal. The greater the probability of generating new patterns, the greater the complexity of the sequence. The lower the value of sample entropy, the higher the self-similarity of the sequence; the larger the value of sample entropy, the more complex the sample sequence. For a time series consisting of N data {x(n)}=x(1),x(2),...,x(N), B m (r) is the probability that two sequences match m points under the same tolerance r, A m (r) is the probability that two sequences match m+1 points. Sample entropy is defined as follows:
[0049]
[0050] The energy entropy, fuzzy entropy and sample entropy extracted after the signal wavelet packet processing have the problem of inconsistent scales. This embodiment normalizes different entropy features and uses an entropy feature fusion method with adaptive weights. By treating the entropy features as feature representations of different scales, each feature may capture different information at different scales. For entropy features of different scales, multi-head self-attention technology is used to adjust them to the same spatial resolution. An adaptive weight mechanism is used to learn spatial weights for each entropy feature. Normalization is performed through the softmax function to ensure that the sum of the weights at all positions is 1. According to the learned weights, different entropy features are weightedly fused at each spatial position to generate a fused feature representation, and the fused features are input into the BiLSTM neural network.
[0051] In this embodiment, three entropies are used to obtain a fused entropy feature through an adaptive weight fusion method. As another implementation method, any two of the above three entropies can be selected for calculation, and the two calculated entropies can be fused to obtain a fused entropy feature; a fixed weight fusion method can also be used for fusion.
[0052] The method of adaptive weight entropy feature fusion is as follows:
[0053] First, given the energy entropy E e , fuzzy entropy E f And sample entropy E s . These three entropies are input into the multi-head self-attention neural network.
[0054] Q=E e W Q
[0055] K=E f W K
[0056] V=E s W V
[0057] Among them, W Q , W K and W V The learnable weight matrices corresponding to query (Q), key (K), and value (V) respectively. Split Q, K, V into H heads: Where D is the dimension of the entropy feature, L is the length of the sequence (for each entropy feature), and H is the number of heads in the multi-head attention.
[0058] Calculate the attention weight for each head h and normalize it using the softmax function:
[0059]
[0060] in It is used as a scaling factor to ensure the stability of the model gradient, and then the output of multi-head attention is generated by connecting h different heads in series.
[0061] MultiHead(Q,K,V)=Concat(Head1,...,Head h )W O
[0062] Where W O is the weight corresponding to the concatenation layer.
[0063] The features obtained by processing the three entropy features through the multi-head attention mechanism are MultiHead (E e ), MultiHead(E f )MultiHead(E s ) are fused to obtain the fusion entropy feature:
[0064] F fusion =a e MultiHead(E e )+a f MultiHead(E f )+a s MultiHead(E s )
[0065] Among them, α e ,α f ,α s The adaptive weights corresponding to energy entropy, fuzzy entropy and sample entropy are adaptively adjusted by the multi-head self-attention neural network, while satisfying αe +a f +a s =1; F fusion is the fusion entropy feature.
[0066] Step 3: The obtained fusion entropy features are input into the Bi-directional Long Short-Term Memory (BiLSTM) neural network to further determine whether it is in an island state. If it is not in an island state, it returns to the initial step for cyclic detection to achieve real-time monitoring of the distributed photovoltaic power generation system.
[0067] The fused entropy features are input into the bidirectional long short-term neural network. The bidirectional long short-term neural network (BiLSTM) is a deep learning model widely used in the field of time series prediction. This model aims to overcome the limitations of traditional neural networks in processing time series data, and significantly improves the prediction accuracy by capturing long-term dependencies and contextual information in time series. BiLSTM is composed of long short-term memory (LSTM) units, which have internal states and control gate mechanisms to adaptively store and forget information.
[0068] In the BiLSTM model, the input time series is first converted into a continuous vector through an embedding layer, and then fed into two independent LSTM layers. These two layers process the forward and reverse information of the sequence respectively, and the calculation of each time step is performed independently. This design enables BiLSTM to consider the past and future context of the input sequence at the same time, thereby more comprehensively understanding the input sequence and providing strong support for the accurate detection of the island effect.
[0069] The LSTM network is mainly composed of memory cells and gate cells (i.e., input gate, forget gate, and output gate). The memory cell is used to store and update state information. The input gate determines which information in the input sequence can be passed into the cell. The forget gate selectively forgets part of the hidden state of the previous cell. The output gate filters the information passed into the next cell. The basic structure of LSTM is as follows: Figure 2 As shown, the mathematical expression of the core part of its memory unit is as follows:
[0070] i(t)=σ[W i x(t)+U i h(t-1)+b i ]
[0071] O(t)=σ[W o x(t)+U o h(t-1)+b o ]
[0072] f(t)=σ[W f x(t)+U f h(t-1)+b f ]
[0073] Where: i(t), O(t) and f(t) are the state values of the input gate, output gate and forget gate respectively; σ [ ] is the Sigmoid activation function; W and U are weights; b is the bias term. The state C(t) of the current memory unit at time t is updated by the following formula:
[0074] C temp (t) = tanh[W c x(t)+U c h(t-1)+b c ]
[0075]
[0076] Where: C temp (t) is the memory unit state increment; It means matrix multiplication element by element. After C(t) is filtered by tanh function and output gate O(t), the output result h(t) is obtained:
[0077]
[0078] like Figure 3 As shown in the figure, BiLSTM combines the state information of two different LSTM units at the same time and obtains the final output value y(t):
[0079]
[0080] Where: and Represent the unit outputs of the forward layer and the backward layer respectively.
[0081] Step 4: If the result of step 3 is that the system is not in an island state, the system returns to step 1 for cyclic detection to achieve real-time monitoring of the distributed photovoltaic power generation system.
[0082] Figure 4 The figure shows the flow chart of building an island detection model based on wavelet packet transform and BiLSTM. The process is divided into two main stages: offline model training and online real-time detection.
[0083] In this embodiment, a bidirectional long short-term neural network is used as the island detection neural network, which can simultaneously consider the past and future contexts of the input sequence, thereby more comprehensively understanding the input sequence and improving the accuracy of island detection; as other implementation methods, the BP neural network in the background technology can also be used as the island detection neural network.
[0084] like Figure 4 As shown, the training process of the BiLSTM model is as follows:
[0085] (1) First, the operating parameters of the power grid are defined, and islanding and non-islanding events under different operating conditions are simulated. After each simulation, the voltage at the point of common coupling (PCC) and the output current of the inverter are collected, and the training data set is constructed with ΔT as the sampling time interval;
[0086] (2) Apply wavelet packet decomposition to the training data, use entropy feature fusion processing technology, and perform normalization to standardize data features.
[0087] (3) For each training feature matrix, there is a corresponding category label. Since it is a binary classification problem, y =
[10] is used to represent islanding events, and y = [0 1] is used to represent power grid disturbance events, that is, non-islanding events.
[0088] (4) After obtaining the training data set, select a suitable BiLSTM structure training model and adjust the training parameters so that the BiLSTM model meets the requirements of actual applications.
[0089] In summary, the island detection method of the present invention, after performing wavelet packet decomposition on the voltage and output current of the common coupling point, calculates at least two of the energy entropy, fuzzy entropy and sample entropy, and then fuses them to form a more comprehensive fusion entropy feature to obtain multi-scale information. Compared with a single feature, it reflects the fault information more comprehensively, can better retain the original feature information, increase the recognition accuracy, improve the recognition ability, enhance the fault signal characterization ability, and improve the accuracy of island detection.
[0090] Computer program product embodiment:
[0091] The computer program product of the present invention includes a computer program / instruction, and the computer program / instruction is used to implement the method steps described in the method embodiment when executed by a processor. The specific steps and beneficial effects of the method have been described in detail in the method implementation column, and will not be repeated in this embodiment.
[0092] Computer readable storage medium embodiment:
[0093] The computer program product of the present invention stores a computing program / instruction thereon, wherein the computer program / instruction is used to implement the method steps described in the method embodiment when executed by a processor. The specific steps and beneficial effects of the method have been described in detail in the method embodiment, and will not be repeated in this embodiment.
[0094] Computer equipment example:
[0095] The computer device of the present invention includes a processor, a memory and a computer program stored in the memory, and the processor is used to execute the computer program to implement the method steps described in the method embodiment. The specific steps and beneficial effects of the method have been described in detail in the method implementation column, and will not be repeated in this embodiment.
[0096] The processor may be a microprocessor MCU, a programmable logic device FPGA or other processing device; the memory may be various memories, such as a hard disk, a floppy disk, etc.
Claims
1. An island detection method, characterized in that: The following steps are involved: Acquire a power data signal of a distributed power grid system, and preliminarily determine whether the distributed power grid system is in an island state according to the power data signal; if the preliminary determination result is that the distributed power grid system is not in an island state, obtain a common coupling point voltage signal and an inverter output current signal of the distributed power grid system, and perform wavelet packet decomposition on the voltage signal and the current signal; calculate any two or three of energy entropy, fuzzy entropy and sample entropy according to the wavelet packet decomposition result, and fuse all the calculated entropies to obtain a fused entropy feature; The fused entropy features are input into the trained island detection neural network to further determine whether the distributed power grid system is in an island state.
2. The island detection method according to claim 1, characterized in that: The method of fusing all the calculated entropies to obtain the fused entropy feature is as follows: all the calculated entropies are input into the trained multi-head self-attention neural network to obtain the fused entropy feature.
3. The island detection method according to claim 1, characterized in that: The island detection neural network is a bidirectional long-term and short-term neural network.
4. The island detection method according to any one of claims 1 to 3, characterized in that: The power data signal is the common coupling point voltage. The method for preliminarily determining whether the distributed power grid system is in an island state based on the power data signal is: determining whether the common coupling point voltage exceeds the corresponding threshold value. If the voltage exceeds the corresponding threshold value, it is determined that the distributed power grid system is in an island state.
5. A computer program product comprising a computer program / instructions, characterized in that The computer program / instructions are used to implement the method steps according to any one of claims 1 to 4 when executed by a processor.
6. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: The computer program / instructions are used to implement the method steps according to any one of claims 1 to 4 when executed by a processor.
7. A computer device comprising a processor, a memory and a computer program stored in the memory, characterized in that: The processor is configured to execute the computer program to implement the method steps according to any one of claims 1 to 4.
Citation Information
Cited By
Photovoltaic grid-connected inverter island detection method, system and device based on wavelet transformation
CN119846290A
