Signal diagnosis method based on multi-parameter signal fusion

Abstract

The invention discloses a signal diagnosis method based on multi-parameter signal fusion, relates to the technical field of signal diagnosis, and solves the problem that the signal diagnosis result is not accurate enough due to the fact that the internal relation of partial discharge signals of parameters and the cooperation effect of the parameters on signal diagnosis are not considered in the prior art. The method comprises the following steps: firstly, performing anti-interference processing on multi-parameter partial discharge signals of metal particles in different existence states based on a wavelet transform method, and extracting time-frequency domain and spatial features of the multi-parameter partial discharge signals to obtain a feature data set of partial discharge key features of the metal particles; a single classifier diagnosis identification algorithm of each parameter partial discharge signal is obtained by using a single classifier; obtaining a classification result of each parameter partial discharge signal; and finally, fusing the diagnosis and identification results of the partial discharge signals of the parameters by using fuzzy integration. In a word, according to a final diagnosis and recognition result, data of the three sensors can be well fused, so that advantage complementation is realized, and the accuracy and reliability of detection are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of signal diagnosis, and in particular to a signal diagnosis method based on multi-parameter signal fusion. Background Art

[0002] When it comes to detecting partial discharge (PD) within GIS, optical, acoustic, and electromagnetic sensors each offer unique advantages, and their synergistic effect is crucial for accurate detection. For example, optical sensors such as silicon photomultipliers (SiPMs) can keenly capture the optical radiation generated by PD. Their high sensitivity, excellent interference resistance, and high-confidence measurement results make them crucial for detection. Ultrasonic sensors detect PD by monitoring ultrasonic signals, performing particularly well in detecting discharges outside GIS equipment and effectively resisting electromagnetic interference. Ultrasonic high-frequency (UHF) sensors specialize in detecting the high-frequency electromagnetic waves generated by PD. Their high sensitivity and interference resistance enable highly accurate localization of discharges within GIS. The combination of the high sensitivity of optical sensors and the interference resistance of ultrasonic sensors can more comprehensively reveal the characteristics of PD, while the high-precision localization capabilities of UHF sensors further enhance detection accuracy. This multi-sensor data fusion approach is extremely important for refined diagnosis and real-time monitoring of PD within GIS.

[0003] However, current discharge detection methods fail to integrate the three sensors in signal diagnosis. Signal diagnosis results are often simply based on the diagnostic results of a single sensor. Alternatively, they are derived based on empirical data from the three sensors alone. This process introduces subjective factors, is time-consuming and labor-intensive, and often overlooks the underlying relationships between the various data points. In either case, neither approach considers the inherent connections between the PD signals of the various parameters and their synergistic effect in signal diagnosis, resulting in inaccurate signal diagnosis results.

[0004] In view of this, a signal diagnosis method based on multi-parameter signal fusion is needed. Summary of the Invention

[0005] In response to the problem that the existing technology does not consider the intrinsic connection between the partial discharge signals of various parameters and their coordinated role in signal diagnosis, resulting in inaccurate signal diagnosis results, the present invention provides a signal diagnosis method based on multi-parameter signal fusion. On the basis of performing signal enhancement and signal diagnosis on the multi-parameter data, the fuzzy integral algorithm is used to process the diagnostic identification results of optical signals, ultrasonic signals and ultra-high frequency signals. The final diagnostic identification result can well fuse the data of these three sensors, thereby achieving complementary advantages and significantly improving the accuracy and reliability of detection. The specific technical solution is as follows:

[0006] A signal diagnosis method based on multi-parameter signal fusion includes the following steps:

[0007] Based on the wavelet transform method, the multi-parameter partial discharge signals of metal particles in different states are processed for anti-interference and their time-frequency and spatial characteristics are extracted to obtain the characteristic data set of the key characteristics of partial discharge of metal particles.

[0008] A single classifier diagnosis and identification algorithm for obtaining partial discharge signals of each parameter is respectively obtained by using a single classifier; and a classification result of the partial discharge signals of each parameter is obtained based on the single classifier diagnosis and identification algorithm;

[0009] Fuzzy measures are defined, and the optimal fuzzy measure combination is obtained based on the fuzzy integral algorithm. The fuzzy integral is then used to fuse the diagnostic identification results of the partial discharge signals of various parameters.

[0010] Preferably, the anti-interference processing is as follows:

[0011] Perform L-layer wavelet decomposition on each signal S(t) to obtain the approximate coefficient cA L and the detail coefficients of each layer cD1,cD2,…,cD L , which represents the detailed features of the signal at different scales, namely:

[0012] {cA L ,cD1,cD2,…,cD L}=wavedec(S(t),L)

[0013] Among them, wavedec() is the wavelet decomposition function;

[0014] Threshold the detail coefficients to remove noise:

[0015] cD' i =sign(cD i )·max(|cD i |-λ,0)

[0016] In the formula, i=1,2,…,L; cD' i is the processed detail coefficient, sign() represents the sign function, and λ is the set threshold;

[0017] The signal is reconstructed using the processed coefficients to obtain the anti-interference signal:

[0018] S'(t)=waverec({cA L ,cD'1,cD'2,…,cD' L},wavelet)

[0019] Where, {cA L,cD'1,cD'2,…,cD' L} represents a set of approximate coefficients and processed detail coefficients, wavelet is the wavelet basis function, which is consistent with that used in wavelet decomposition, and waverec() represents the wavelet reconstruction function, which is used to reconstruct the coefficient set into a signal.

[0020] Preferably, the time-frequency domain features include wavelet coefficient energy, total energy and energy distribution feature vectors.

[0021] Preferably, the spatial features include signal strength distribution and signal propagation direction, wherein the signal strength distribution is represented by calculating the mean and standard deviation of the signal strength at each sensor position, and the signal propagation direction is obtained by analyzing the arrival time difference of the signal between different sensors.

[0022] Preferably, the multi-parameter partial discharge signal includes an optical signal S opt (t), ultrasonic signal S US (t) and UHF signal S UHF (t).

[0023] Preferably, recursive neural networks, convolutional neural networks and vector machine algorithms are used respectively to obtain single classifier diagnostic recognition algorithms for optical signals, ultrasonic signals and ultra-high frequency signals.

[0024] Preferably, the specific steps of fusing the diagnostic identification results of the partial discharge signals of various parameters are as follows:

[0025] Define the fuzzy measure μ opt 、μ US 、μ UHF , and set the initial value, where μ opt 、μ US 、μ UHF They are the fuzzy measures of optical signals, ultrasonic signals and UHF signals respectively;

[0026] A fitness function is constructed to evaluate the performance of the fuzzy measure, and a weighted eigenvalue W is added to the fitness function considering the weight of the signal characteristics. F , and the fitness function also considers the classification accuracy, the fitness function also considers the classification accuracy, which is specifically expressed as follows:

[0027] Fitness(μ)=α·A c +β·W F

[0028] Where, Fitness(μ) represents the fitness function, A c represents the classification accuracy, W F represents the weighted eigenvalue, α and β represent the weight coefficients;

[0029] Among them, the weighted eigenvalue WF The values ​​are as follows:

[0030] W F =μ opt w opt +μ US w US +μ UHF w UHF

[0031] Where w opt 、w US 、w UHF represents the weight assigned according to the signal characteristics;

[0032] Optimizing the Fuzzy Measure μ of Optical, Ultrasonic and UHF Signals Using Genetic Algorithms opt 、μ US 、μ UHF ;

[0033] Based on the optimal fuzzy measure combination of the output, the diagnostic results of the three signal sources are fused using fuzzy integral, as follows:

[0034] F=μ opt ·f opt +μ US ·f US +μ UHF ·f UHF +μ opt∩US min(f opt ,f US )+μ opt∩UHF min(f opt ,f UHF )+

[0035] μ US∩UHF min(f US ,f UHF )+μ opt∩US∩UHF min(f opt ,f US ,f UHF )

[0036] Among them, μ opt∩US =min(μ opt ,μ US ), μ opt∩UHF =min(μ opt ,μ UHF ), μ US∩UHF =min(μ US ,μ UHF ), μ opt∩US∩UHF =min(μ opt ,μ US ,μ UHF), F is the final fusion output result, f opt 、f US 、f UHF , which are the classification results of optical signals, ultrasonic signals and UHF signals respectively.

[0037] Preferably, w opt 、w US 、w UHF The value determination process is defined as follows:

[0038] Collect the diagnostic result data set of optical signal, ultrasonic signal and UHF signal, denoted as D opt 、D US and D UHF ; And standardize the data of each signal source;

[0039] Extract key features from the data of each signal source, including the amplitude mean and variance of the optical signal, the energy distribution of the ultrasonic signal, and the pulse frequency of the UHF signal; and combine these features into a feature matrix X, where each row represents a sample and each column represents a feature;

[0040] The optimization objective function is defined as maximizing the classification accuracy after fusion;

[0041] Use optimization algorithms to find the optimal weight combination.

[0042] A computer-readable storage medium includes a stored program, wherein when the program is run, the device where the computer-readable storage medium is located is controlled to execute the signal diagnosis method based on multi-parameter signal fusion as described above.

[0043] A processor is used to run a program, wherein the program executes the signal diagnosis method based on multi-parameter signal fusion as described above when running.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The present invention first performs anti-interference processing on the multi-parameter partial discharge signals of metal particles in different existence states based on the wavelet transform method and extracts their time-frequency domain and spatial characteristics to obtain a feature data set of key characteristics of partial discharge of metal particles; then uses a single classifier to obtain a single classifier diagnostic identification algorithm for each parameter partial discharge signal; and obtains the classification result of each parameter partial discharge signal based on the single classifier diagnostic identification algorithm; finally, obtains the optimal fuzzy measure combination based on the fuzzy integral algorithm, and then uses the fuzzy integral to fuse the diagnostic identification results of each parameter partial discharge signal. In short, the present invention can use the fuzzy integral algorithm to process the diagnostic identification results of optical signals, ultrasonic signals and ultra-high frequency signals on the basis of performing respective signal enhancement and signal diagnosis on multi-parameter data, so that the final diagnostic identification result can well fuse the data of these three sensors, thereby achieving complementary advantages and significantly improving the accuracy and reliability of detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly describes the drawings required for the specific embodiments or the description of the prior art. Similar elements or parts are generally identified by similar reference numerals throughout the drawings. Elements or parts in the drawings are not necessarily drawn to scale.

[0047] Figure 1 is a flow chart of the method of the present invention;

[0048] Figure 2 Schematic diagram of the process of acquiring feature datasets;

[0049] Figure 3 A schematic diagram of the specific steps for fusing the diagnostic results of three signal sources;

[0050] Figure 4 Schematic diagram of the process of optimizing fuzzy measures using genetic algorithms;

[0051] Figure 5 Schematic diagram of the steps for obtaining the weights of each signal. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0053] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0054] It should also be understood that the terms used in the present specification are only for the purpose of describing particular embodiments and are not intended to limit the present invention. As used in the present specification and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0055] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0056] In one embodiment of the present invention, a signal diagnosis method based on multi-parameter signal fusion is provided. Figure 1 As shown, the following steps are included:

[0057] Step 1: Based on the wavelet transform method, the multi-parameter partial discharge signals of metal particles in different states are processed for anti-interference and their time-frequency domain and spatial characteristics are extracted to obtain a feature data set of key characteristics of partial discharge of metal particles;

[0058] Among them, the multi-parameter partial discharge signal includes the optical signal S opt (t), ultrasonic signal S US (t) and UHF signal S UHF After collecting these signals, they are first preprocessed, including removing DC components and normalizing them, to eliminate dimensional differences and improve the efficiency of subsequent processing.

[0059] like Figure 2 As shown in Figure 2, the process of obtaining the feature dataset is as follows:

[0060] S1: Perform L-layer wavelet decomposition on each signal S(t) to obtain the approximate coefficient cA L and the detail coefficients of each layer cD1,cD2,…,cD L , which represents the detailed features of the signal at different scales, namely:

[0061] {cA L ,cD1,cD2,…,cD L}=wavedec(S(t),L)

[0062] Among them, wavedec() is the wavelet decomposition function.

[0063] S2: Threshold processing is performed on the detail coefficients to remove noise. Let the threshold be λ and use the soft threshold method:

[0064] cD' i =sign(cD i )·max(|cD i |-λ,0)

[0065] In the formula, i=1,2,…,L; cD' i It is the processed detail coefficient, sign() represents the sign function;

[0066] S3: Use the processed coefficients to reconstruct the signal to obtain the anti-interference signal (including the optical signal S' opt (t), ultrasonic signal S' US (t) and UHF signal S' UHF (t)), the process of obtaining the signal after anti-interference is as follows:

[0067] S'(t)=waverec({cA L ,cD'1,cD'2,…,cD' L},wavelet)

[0068] Where, {cA L ,cD'1,cD'2,…,cD' L} represents a set of approximate coefficients and processed detail coefficients, wavelet is the wavelet basis function, which is consistent with that used in wavelet decomposition, and waverec() represents the wavelet reconstruction function, which is used to reconstruct the coefficient set into a signal;

[0069] The wavelet reconstruction function is used to reassemble the coefficients after wavelet decomposition to restore the original signal. This process is also called the Inverse Discrete Wavelet Transform (IDWT). It is the opposite of the wavelet decomposition process. Through upsampling and filtering operations, the decomposed approximate coefficients and detail coefficients are recombined into the original signal.

[0070] The reconstruction steps include upsampling, filtering, and accumulation. Upsampling inserts zeros between each sample to restore the signal's length. Filtering involves filtering the upsampled coefficients using the reconstruction filters associated with the decomposition stage. These filters are carefully designed to eliminate aliasing effects introduced by the decomposition stage. Finally, all filtered coefficient sequences are summed point by point to obtain the reconstructed signal.

[0071] S4: Extract time-frequency domain features, including wavelet coefficient energy, total energy, and energy distribution feature vectors:

[0072] 1. Calculate the energy of each layer's wavelet coefficients as features. For the i-th layer's detail coefficient cD' i , whose energy is:

[0073]

[0074] Among them, N i It's cD' i The length, E i is the energy of the detail coefficient of the i-th layer, cD' i [j] The j-th detail coefficient after processing the i-th layer.

[0075] 2. The total energy of the signal is:

[0076]

[0077] in, is the approximation coefficient cA L energy.

[0078] 3. Normalize the energy of each layer as the eigenvector to obtain the energy distribution eigenvector:

[0079]

[0080] S5: Extract spatial features, including signal strength distribution and signal propagation direction. The signal strength distribution is represented by calculating the mean and standard deviation of the signal strength at each sensor location, and the signal propagation direction is estimated by analyzing the time difference of arrival (TDOA) of the signal between different sensors.

[0081] S6: Construct feature data set: Combine the time-frequency domain features and spatial features into a feature vector to obtain the key feature data set of metal particle partial discharge.

[0082] The feature data set obtained through the above-mentioned signal processing and feature extraction steps based on the wavelet transform method has significant advantages. The multi-resolution analysis capability of the wavelet transform enables it to capture the global trends and local details of the signal at the same time, effectively adapt to the non-stationary characteristics of the signal, and is more flexible and accurate than the traditional single-scale method. Its threshold processing can effectively remove noise while retaining the key features of the signal, avoiding the distortion problems that may be introduced by traditional filtering methods. In addition, the time-frequency domain and spatial features extracted by the wavelet transform highly concentrate the key information of the signal, forming a feature data set with high information density, which can comprehensively describe the physical characteristics of the signal. This feature data set not only has strong dynamic adaptability and can capture mutations and anomalies in the signal, but can also be seamlessly combined with modern machine learning and deep learning technologies to provide high-quality input for subsequent signal classification and diagnosis, thereby significantly improving the efficiency and accuracy of signal processing.

[0083] Step 2: Use recursive neural network, convolutional neural network and vector machine algorithm to obtain single classifier diagnosis and recognition algorithm of optical signal, ultrasonic signal and UHF signal respectively; and obtain the optical signal classification result f based on the single classifier diagnosis and recognition algorithm opt , ultrasonic signal classification result f US , UHF signal classification result f UHF The details are as follows:

[0084] Recurrent Neural Network (RNN) for optical signal classification:

[0085] 1. Divide the optical signal data into time series format, assuming that the length of each sequence is T and the feature dimension is D; and normalize the optical signal;

[0086] 2. Build an RNN model: Use LSTM units, set the number of hidden units to 50, and the input shape to (timestep, number of features). Add two LSTM layers, followed by a Dense layer, with the number of output categories set to C and the activation function set to softmax.

[0087] 3. Set the training parameters (optimizer, loss function, number of training rounds), train the model, and obtain the trained single classifier diagnosis and recognition algorithm model of the optical signal, based on which the optical signal classification result f can be obtained. opt .

[0088] Convolutional Neural Network (CNN) for Ultrasound Signal Classification:

[0089] 1. Input parameters: ultrasonic signal data S US Normalized data;

[0090] 2. Model construction:

[0091] Use the Conv2D convolution layer, set 32 ​​convolution kernels, the convolution kernel size is (3,3), and the activation function is ReLU.

[0092] Add a MaxPooling2D pooling layer with a pooling window size of (2,2).

[0093] The Flatten layer is followed by a Dense layer with 128 hidden units and ReLU activation function.

[0094] The output layer is a Dense layer, the number of output categories is C, and the activation function is softmax.

[0095] 3. Set the training parameters (optimizer, loss function, number of training rounds), train the model, and obtain the trained single classifier diagnostic recognition algorithm model of the ultrasound signal, based on which the optical signal classification result f can be obtained. US.

[0096] Support Vector Machine (SVM) for UHF signal classification:

[0097] 1. Input parameters: eigenvector based on the standardized eigenvector XUHF of the UHF signal;

[0098] 2. Model construction: Use the RBF kernel function, set the penalty coefficient C = 1.0, and the kernel function parameter γ = scale;

[0099] 3. Train the model to obtain a trained single classifier diagnostic recognition algorithm model for UHF signals, which can be used to obtain the optical signal classification result f UHF .

[0100] The other specific algorithm model construction and training of the above three signals are existing methods and will not be elaborated here.

[0101] When fusing the diagnostic results of optical signals, ultrasonic signals, and UHF signals, the recursive neural network (RNN), convolutional neural network (CNN), and support vector machine (SVM) are selected to classify these three signals respectively, based on a comprehensive consideration of the characteristics of each signal and the classification requirements.

[0102] Optical signals exhibit time series characteristics, often manifesting as continuous signals that vary over time, such as modulated signals in optical communications. Recurrent neural networks (RNNs) are particularly well-suited to processing this type of time series data because they leverage their internal recurrent structure to capture dependencies and patterns across the temporal dimension. This capability enables RNNs to effectively process the temporal characteristics of optical signals, thereby improving the accuracy of optical signal classification.

[0103] Ultrasonic signals are typically presented as images or two-dimensional time-frequency plots. For example, in laser ultrasonic testing, the time-frequency spectrum generated by wavelet transform can intuitively reflect the characteristics of defects. Convolutional neural networks (CNNs) excel at processing images and two-dimensional data. They can automatically extract local features in images and capture complex patterns through multi-layer convolution and pooling operations. Therefore, CNNs are well-suited for ultrasonic signal classification, capable of extracting diagnostic features from ultrasound images or time-frequency plots.

[0104] Ultra-high frequency (UHF) signals typically appear as sparse pulses in partial discharge detection. Feature extraction and classification of these signals require a classifier that can effectively handle high-dimensional data and complex boundaries. Support vector machines (SVMs) use nonlinear kernel functions to map data into a high-dimensional space and find the optimal classification boundary within this space. This characteristic enables SVMs to effectively distinguish between different categories when processing UHF signals, demonstrating excellent performance even on small sample datasets.

[0105] In summary, the selection of RNN, CNN, and SVM for classifying optical, ultrasonic, and UHF signals, respectively, is based on a comprehensive consideration of the characteristics of these signals and the classification requirements. This selection fully leverages the strengths of each algorithm, improving the accuracy and reliability of signal classification and thus providing high-quality diagnostic results for subsequent fusion processing.

[0106] Step 3: Fusion processing of the diagnostic recognition results of optical signals, ultrasonic signals and UHF signals based on fuzzy integral algorithm.

[0107] First, let's briefly introduce the diagnostic results characteristics of the three signals:

[0108] The diagnostic results of optical signals are sensitive to changes in signal amplitude, so their significant impact on the overall diagnostic results should be considered during the fusion process. The diagnostic results of ultrasonic signals are generally related to the sound waves generated by mechanical vibration or partial discharge. In practical applications, the detection results of ultrasonic signals may be affected by environmental noise and equipment status, but they have high sensitivity in partial discharge detection. The diagnostic results of ultra-high frequency (UHF) signals are generally related to the electromagnetic pulses generated by partial discharge. UHF signals have good time resolution and frequency characteristics, and can effectively detect the pulse signals of partial discharge.

[0109] Therefore, in the process of optimizing the fuzzy measure, this embodiment takes into account the diagnostic result characteristics of the optical signal, ultrasonic signal and UHF signal, thereby more effectively adjusting the fuzzy measure and improving the fusion effect. Figure 3 As shown, the following are the specific optimization steps:

[0110] S01: Define the fuzzy measure μ opt 、μ US 、μ UHF , the initial value is set to 1 / 3.

[0111] S02: Construct a fitness function to evaluate the performance of the fuzzy measure, and add a weighted eigenvalue W that considers the signal characteristic weight into the fitness function F In addition, the fitness function also considers the classification accuracy, which is specifically expressed as follows:

[0112] Fitness(μ)=α·A c +β·W F

[0113] Where, Fitness(μ) represents the fitness function, A c represents the classification accuracy, W F represents the weighted eigenvalue, and α and β represent the weight coefficients.

[0114] Among them, the weighted eigenvalue WF The values ​​are as follows:

[0115] W F =μ opt w opt +μ US w US +μ UHF w UHF

[0116] Where w opt 、w US 、w UHF Indicates the weight assigned according to the signal characteristics. For example, the optical signal has a greater sensitivity, so w opt ≥w US , w opt ≥w UHF .

[0117] S03: Use genetic algorithm (GA) to optimize the fuzzy measure, such as Figure 4 The specific steps are as follows:

[0118] S301: Initialize the population, including initializing the population size (e.g., 50), variable dimension (e.g., optical signal, ultrasonic signal, ultra-high frequency signal), and variable range ([0,1]);

[0119] S302: Calculate the fitness value of each individual (fuzzy measure combination) using the above fitness function;

[0120] S303: Use roulette wheel selection or other selection strategies to select individuals for reproduction based on fitness probability.

[0121] S304: Perform a crossover operation on the selected individuals to generate new individuals. The crossover rate (pc) can be set to 0.8.

[0122] S305: Randomly mutate the newly generated individuals to introduce a certain degree of randomness. The mutation rate (mu) can be set to 0.1;

[0123] S306: Replace some individuals in the population with the newly generated individuals to form a new population;

[0124] S307: Check whether the maximum number of iterations (maxit) or fitness threshold is reached. If the termination condition is met, the algorithm stops;

[0125] S308: Output optimal fuzzy measure combination μ opt 、μ US 、μ UHF .

[0126] S04: Based on the optimal fuzzy measure combination of the output, the diagnostic results of the three signal sources are fused using fuzzy integrals, as follows:

[0127] F=μ opt ·f opt +μ US ·f US +μ UHF ·f UHF +μ opt∩US min(f opt ,f US )+μ opt∩UHF min(f opt ,f UHF )+

[0128] μ US∩UHF min(f US ,f UHF )+μ opt∩US∩UHF min(f opt ,f US ,f UHF )

[0129] Among them, μ opt∩US =min(μ opt ,μ US ), μ opt∩UHF =min(μ opt ,μ UHF ), μ US∩UHF =min(μ US ,μ UHF ), μ opt∩US∩UHF =min(μ opt ,μ US ,μ UHF ), μ opt 、μ US 、μ UHF are the fuzzy measurements of optical signal, ultrasonic signal and UHF signal respectively, F is the final fusion output result, f opt 、f US 、f UHF These are the classification results of optical signals, ultrasonic signals and UHF signals respectively.

[0130] When fusing diagnostic results from optical, ultrasonic, and UHF signals, the minimum function plays a crucial role in the fuzzy integral. By minimizing the signal source weights, it effectively addresses interactions between signal sources and reduces the impact of single-source anomalies on the fusion results. Specifically, the minimum function uses the minimum value of the signal source weights during the fusion process to comprehensively consider the characteristics of the optical, ultrasonic, and UHF signals, ensuring that the contribution of each signal source does not exceed its minimum weight. This avoids bias in the fusion results caused by noise or uncertainty in any particular signal. This processing approach not only improves the accuracy and robustness of the fusion results but also enhances the fusion method's adaptability to complex signal environments. For example, the minimum function effectively suppresses ambiguity or noise that may occur in optical signals, random interference in ultrasonic signals, and spatial interference in UHF signals, making the fused diagnostic results more reliable.

[0131] Through this optimization process, the fuzzy measure can be dynamically adjusted to achieve optimal performance when fusing diagnostic results from optical, ultrasonic, and UHF signals. The optimized fuzzy measure more accurately reflects the importance of each signal source, thereby improving the accuracy and robustness of the fusion results.

[0132] In one embodiment of the present invention, based on the above embodiment, w opt 、w US 、w UHF The value of is limited as follows: Based on the optimized weight distribution algorithm, the weight distribution of each signal in the multi-parameter partial discharge signal (optical signal, ultrasonic signal and UHF signal) of the metal particles in the current state is calculated. Figure 5 The specific process is as follows:

[0133] S001: Collect the diagnostic result data set of optical signals, ultrasonic signals and UHF signals, denoted as D opt 、D US and D UHF ; and standardize the data from each signal source to ensure they are comparable.

[0134] S002: Extract key features from the data of each signal source, including the amplitude mean and variance of the optical signal, the energy distribution of the ultrasonic signal, and the pulse frequency of the UHF signal; and combine these features into a feature matrix X, where each row represents a sample and each column represents a feature;

[0135] S003: The optimization objective function is defined as maximizing the classification accuracy after fusion;

[0136] S004: Use optimization algorithms (such as genetic algorithm, particle swarm optimization or gradient descent) to find the optimal weight combination; during the optimization process, the weight wopt 、w US 、w UHF is adjusted to maximize the objective function;

[0137] S005: Get the final weight w opt 、w US 、w UHF .

[0138] Those skilled in the art will appreciate that the units of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition of each example has been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0139] In the embodiments provided by the present invention, it should be understood that the division of units is merely a logical function division, and there may be other division methods in actual implementation, for example, multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored, etc.

[0140] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0141] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-0nly Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk, etc., various media that can store program code.

[0142] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some or all of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and they should all be included in the scope of the claims and description of the present invention.

Claims

1. A signal diagnosis method based on multi-parameter signal fusion, characterized in that: The following steps are involved: Based on the wavelet transform method, the multi-parameter partial discharge signals of metal particles in different states are processed for anti-interference and their time-frequency and spatial characteristics are extracted to obtain the characteristic data set of the key characteristics of partial discharge of metal particles. A single classifier diagnosis and identification algorithm for obtaining partial discharge signals of each parameter is respectively obtained by using a single classifier; and a classification result of the partial discharge signals of each parameter is obtained based on the single classifier diagnosis and identification algorithm; Fuzzy measures are defined, and the optimal fuzzy measure combination is obtained based on the fuzzy integral algorithm. The fuzzy integral is then used to fuse the diagnostic identification results of the partial discharge signals of various parameters.

2. The signal diagnosis method based on multi-parameter signal fusion according to claim 1, characterized in that: The anti-interference processing is as follows: Perform L-layer wavelet decomposition on each signal S(t) to obtain the approximate coefficient cA L and the detail coefficients of each layer cD1,cD2,…,cD L , which represents the detailed features of the signal at different scales, namely: {cA L ,cD1,cD2,…,cD L }=wavedec(S(t),L) Among them, wavedec() is the wavelet decomposition function; Threshold the detail coefficients to remove noise: CD' i =sign(cD i )·max(|cD i |-λ,0) In the formula, i=1,2,…,L; cD' i is the processed detail coefficient, sign() represents the sign function, and λ is the set threshold; The signal is reconstructed using the processed coefficients to obtain the anti-interference signal: S'(t)=waverec({cA L ,cD’1,cD'2,…,cD' L },wavelet) Where, {cA L ,cD'1,cD'2,…,cD' L } represents a set of approximate coefficients and processed detail coefficients, wavelet is the wavelet basis function, which is consistent with that used in wavelet decomposition, and waverec() represents the wavelet reconstruction function, which is used to reconstruct the coefficient set into a signal.

3. The signal diagnosis method based on multi-parameter signal fusion according to claim 2, characterized in that: The time-frequency domain features include wavelet coefficient energy, total energy and energy distribution feature vector.

4. The signal diagnosis method based on multi-parameter signal fusion according to claim 2, characterized in that: Spatial features include signal strength distribution and signal propagation direction. The signal strength distribution is represented by calculating the mean and standard deviation of the signal strength at each sensor location, and the signal propagation direction is obtained by analyzing the arrival time difference of the signal between different sensors.

5. A signal diagnosis method based on multi-parameter signal fusion according to claim 3 or 4, characterized in that: Multi-parameter partial discharge signals include optical signal S opt (t), ultrasonic signal S US (t) and UHF signal S UHF (t).

6. The signal diagnosis method based on multi-parameter signal fusion according to claim 5, characterized in that: Recursive neural network, convolutional neural network and vector machine algorithms are used respectively to obtain single classifier diagnostic recognition algorithms for optical signals, ultrasonic signals and ultra-high frequency signals.

7. The signal diagnosis method based on multi-parameter signal fusion according to claim 6, characterized in that: The specific steps for integrating the diagnostic and identification results of the PD signals of various parameters are as follows: Define the fuzzy measure μ opt 、μ US 、μ UHF , and set the initial value, where μ opt 、μ US 、μ UHF They are the fuzzy measures of optical signals, ultrasonic signals and UHF signals respectively; A fitness function is constructed to evaluate the performance of the fuzzy measure, and a weighted eigenvalue W is added to the fitness function considering the weight of the signal characteristics. F , and the fitness function also considers the classification accuracy, the fitness function also considers the classification accuracy, which is specifically expressed as follows: Fitness(μ)=α·A c +β·W F Where, Fitness(μ) represents the fitness function, A c represents the classification accuracy, W F represents the weighted eigenvalue, α and β represent the weight coefficients; Among them, the weighted eigenvalue W F The values ​​are as follows: W F =μ opt ·w opt +m US ·w US +m UHF ·w UHF Where w opt 、w US 、w UHF represents the weight assigned according to the signal characteristics; Optimizing the Fuzzy Measure μ of Optical, Ultrasonic and UHF Signals Using Genetic Algorithms opt 、μ US 、μ UHF ; Based on the optimal fuzzy measure combination of the output, the diagnostic results of the three signal sources are fused using fuzzy integral, as follows: F=μ opt ·f opt +m US ·f US +m UHF ·f UHF +m opt∩US ·min(f opt ,f US )+μ opt∩UHF ·min(f opt ,f UHF )+ m US∩UHF ·min(f US ,f UHF )+μ opt∩US∩UHF ·min(f opt ,f US ,f UHF ) Among them, μ opt∩US =min(μ opt ,μ US ), μ opt∩UHF =min(μ opt ,μ UHF ), μ US∩UHF =min(μ US ,μ UHF ), μ opt∩US∩UHF =min(μ opt ,μ US ,μ UHF ), F is the final fusion output result, f opt 、f US 、f UHF , which are the classification results of optical signals, ultrasonic signals and UHF signals respectively.

8. The signal diagnosis method based on multi-parameter signal fusion according to claim 7, characterized in that: Will w opt 、w US 、w UHF The value determination process is defined as follows: Collect the diagnostic result data set of optical signal, ultrasonic signal and UHF signal, denoted as D opt 、D US and D UHF ; And standardize the data of each signal source; Extract key features from the data of each signal source, including the amplitude mean and variance of the optical signal, the energy distribution of the ultrasonic signal, and the pulse frequency of the UHF signal; and combine these features into a feature matrix X, where each row represents a sample and each column represents a feature; The optimization objective function is defined as maximizing the classification accuracy after fusion; Use optimization algorithms to find the optimal weight combination.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the signal diagnosis method based on multi-parameter signal fusion according to claim 1.

10. A processor, characterized in that: The processor is used to run a program, wherein the signal diagnosis method based on multi-parameter signal fusion according to claim 1 is executed when the program is run.

Images