A power equipment fault prediction method and system based on multi-source information
By integrating three-phase current, vibration acceleration, and infrared temperature data of power equipment through multi-source information fusion, a resonance mode is constructed and the annihilation risk is assessed. This solves the problem of delayed early warning in existing technologies and enables early warning and timely maintenance of equipment failures.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INNER MONGOLIA GUANGMING ELECTRIC POWER EQUIP INSTALLATION CO LTD
- Filing Date
- 2026-03-17
- Publication Date
- 2026-06-12
Smart Images
Figure FT_1
Abstract
Description
Technical Field
[0001] This application belongs to the field of power equipment fault prediction technology, and specifically relates to a power equipment fault prediction method and system based on multi-source information. Background Technology
[0002] Stable operation of power equipment is the core guarantee for reliable power supply in a power system, and fault prediction technology provides important support for equipment operation and maintenance. Existing fault prediction methods often focus on a single type of monitoring data, such as analyzing changes in electrical parameters through current waveforms, judging the condition of mechanical structures based on vibration data, or identifying thermal anomalies using temperature data.
[0003] Equipment failures are often the result of the combined effects of multiple factors, including electrical, mechanical, and thermal effects. Current methods for predicting power equipment failures tend to overlook early signs of abnormality caused by the interaction of multiple factors and are not sensitive enough to the latent changes in the early stages of failure. This may lead to delayed warning signals, increase the risk of unplanned equipment shutdowns, and affect the continuity and stability of power supply in the power system. Summary of the Invention
[0004] This application provides a power equipment fault prediction method and system based on multi-source information, which effectively solves the problems of early anomalies and delayed early warnings caused by the easy neglect of multiple factors in the existing technology, and improves the timeliness and reliability of early warning.
[0005] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a method for predicting power equipment faults based on multi-source information, including: The raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, are acquired during a continuous monitoring period. The raw data is then preprocessed to obtain a multi-source data matrix.
[0006] The time-frequency domain features are extracted and the resonance mode is constructed to obtain the resonance temperature correlation sequence.
[0007] Recursive quantization analysis was used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and a coupled resonance state abrupt change marker sequence was obtained.
[0008] Based on the coupled resonance state mutation marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain an early warning signal.
[0009] The warning signals are verified to obtain the final fault warning and maintenance recommendations.
[0010] Furthermore, the original data is preprocessed to obtain a multi-source data matrix, including: The original current data, original vibration data, and original temperature data are time-aligned and filtered to obtain current data sequences, vibration data sequences, and temperature data sequences.
[0011] The current data sequence, vibration data sequence, and temperature data sequence are denoised using wavelet soft thresholding, and the denoised data are then fused to obtain a multi-source data matrix.
[0012] Further, by extracting time-frequency domain features and constructing resonance modes, a resonance temperature correlation sequence is obtained, including: Empirical mode decomposition is performed on the current data sequence in the multi-source data matrix to obtain the eigenmode function set of the current data sequence.
[0013] The instantaneous frequencies of the first three eigenmode functions in the set of eigenmode functions at the same time are calculated to obtain the current-dominated instantaneous frequency sequence.
[0014] On the vibration data sequence in the multi-source data matrix, the Hilbert-Huang transform is applied to extract the instantaneous amplitude at the corresponding moment of the current-dominated instantaneous frequency sequence, thus obtaining the resonance amplitude sequence.
[0015] By fusing the resonance amplitude sequence with the temperature data sequence in the multi-source data matrix, the cross-correlation coefficients within the same time window are calculated to obtain the resonance temperature correlation sequence.
[0016] Furthermore, the empirical mode decomposition is specifically achieved by iteratively identifying local extreme points of the data sequence, fitting the upper and lower envelopes, and extracting the mean envelope until the intrinsic mode function condition is met.
[0017] Furthermore, recursive quantization analysis is used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, resulting in a coupled resonance state abrupt change marker sequence, including: Based on the delay time and embedding dimension, the resonant temperature correlation sequence is reconstructed in phase space to generate a high-dimensional phase space trajectory.
[0018] From the recursive graph of the high-dimensional phase space trajectory, the vertical distance distribution of the recursive points relative to the diagonal is calculated to obtain the recursive structure feature vector.
[0019] The skewness and kurtosis of the recursive structure feature vectors are calculated to obtain the recursive distribution morphology index.
[0020] Within the sliding window, mutation point detection is performed on the recursive distribution morphology index to obtain the coupled resonance state mutation marker sequence.
[0021] Furthermore, the phase space reconstruction calculates the delay time using the autocorrelation function method and determines the embedding dimension using the spurious nearest neighbor method, thereby mapping the one-dimensional sequence to a high-dimensional phase space.
[0022] Furthermore, based on the aforementioned coupled resonance state abrupt change marker sequence, the risk of resonance annihilation is assessed using an information geometry method to obtain a warning signal, including: The number of mutation markers in the coupled resonance state mutation marker sequence within a unit time window is counted to obtain the mutation frequency sequence.
[0023] The Bach distance between the probability distribution of the mutation frequency sequence and the probability distribution of the reference healthy state is calculated to obtain the distribution difference sequence.
[0024] The distribution difference sequence is smoothed using the exponentially weighted moving average method to obtain the risk evolution trend sequence.
[0025] The risk evolution trend sequence is compared with preset low and high thresholds. When the value of the risk evolution trend sequence exceeds the low threshold, a yellow warning signal is output, and when it exceeds the high threshold, a red warning signal is output.
[0026] Furthermore, the low threshold is obtained by statistically analyzing the risk evolution trend sequence under historical health conditions, taking the mean and adding 2 times the standard deviation; the high threshold is set to 1.5 times the low threshold.
[0027] Furthermore, the warning signals are verified to obtain final fault warnings and maintenance recommendations, including: When the warning signal is a yellow warning signal, an adaptive adjustment is triggered to shorten the sliding window length for mutation point detection and recalculate the coupled resonance state mutation flag sequence.
[0028] Based on the recalculated coupled resonance state mutation marker sequence, the steps for assessing the risk of resonance annihilation are repeated to obtain new early warning signals.
[0029] If the new warning signal is yellow or red, then the final equipment failure warning and maintenance recommendations will be generated and output.
[0030] Secondly, this application provides a power equipment fault prediction system based on multi-source information, including: Data acquisition and preprocessing module: Acquires raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, within a continuous monitoring period, and preprocesses the raw data to obtain a multi-source data matrix.
[0031] Feature extraction and pattern construction module: Extract time-frequency domain features from the above and construct resonance patterns to obtain resonance temperature correlation sequence.
[0032] Coupled resonance detection module: Recursive quantization analysis is used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and a coupled resonance state abrupt change marker sequence is obtained.
[0033] Risk assessment and early warning module: Based on the coupled resonance state mutation marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain an early warning signal.
[0034] Early warning verification and decision-making module: Verifies the early warning signal to obtain the final fault warning and maintenance suggestions.
[0035] Thirdly, this application provides a power equipment fault prediction device based on multi-source information, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program to implement the steps of the power equipment fault prediction method based on multi-source information.
[0036] Fourthly, this application provides a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the steps of a power equipment fault prediction method based on multi-source information.
[0037] Fifthly, this application provides a computer program product, including a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the steps of implementing a power equipment fault prediction method based on multi-source information are provided.
[0038] The beneficial effects of this application are: This application acquires three-phase current, vibration acceleration, and infrared thermal imaging temperature data of equipment, and after preprocessing, feature extraction, resonance mode construction, and risk assessment, outputs and verifies early warning signals. It effectively solves the problems of early anomalies and delayed early warnings caused by multiple factors that are easily overlooked in existing technologies, and achieves accurate capture of potential equipment failures. It improves the timeliness and reliability of early warnings, provides strong support for equipment operation and maintenance, and reduces the risk of unplanned downtime.
[0039] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures pointed out in the description and the accompanying drawings. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0041] Figure 1 A flowchart illustrating a power equipment fault prediction method based on multi-source information according to this application is shown. Detailed Implementation
[0042] To address the problems raised in the background technology, this application improves the timeliness and reliability of early warning by acquiring three-phase current, vibration acceleration, and infrared thermal image temperature data of the equipment, performing preprocessing, feature extraction, resonance mode construction, and risk assessment, and then outputting and verifying early warning signals.
[0043] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0044] In some embodiments, such as Figure 1 As shown, this application provides a power equipment fault prediction method based on multi-source information, including: S1. Acquire raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, within a continuous monitoring period, and preprocess the raw data to obtain a multi-source data matrix.
[0045] S2. Extract time-frequency domain features and construct resonance modes to obtain the resonance temperature correlation sequence.
[0046] S3. Recursive quantization analysis was used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and a coupled resonance state abrupt change marker sequence was obtained.
[0047] S4. Based on the coupled resonance state mutation marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain early warning signals.
[0048] S5. Verify the warning signals to obtain the final fault warning and maintenance recommendations.
[0049] In some embodiments, the original data is preprocessed to obtain a multi-source data matrix, including: S11. Time alignment is performed on the raw current data, raw vibration data, and raw temperature data. Specifically, all three data streams can be interpolated and resampled using a unified target sampling frequency. The interpolation method can be linear interpolation or cubic spline interpolation.
[0050] Even after time alignment, the data may still contain high-frequency noise. A moving average filter is applied to the three data streams for preliminary smoothing to obtain current data sequences, vibration data sequences, and temperature data sequences.
[0051] S12. Use wavelet transform to decompose each data sequence into multiple layers of coefficients, process the detailed coefficients of each layer through a soft thresholding function, set the small coefficients that are mainly regarded as noise to zero or reduce them, retain the large coefficients that represent the real signal, and then reconstruct them through inverse wavelet transform to obtain the denoised current sequence, vibration sequence and temperature sequence respectively.
[0052] The current sequence, vibration sequence, and temperature sequence are aligned in time order, with each row containing the three observations of current, vibration, and temperature at the same moment, thus forming a data matrix in which rows represent time and columns represent signal types, resulting in a multi-source data matrix.
[0053] In some embodiments, the resonance temperature correlation sequence is obtained by extracting time-frequency domain features and constructing resonance modes, including: S21. Perform empirical mode decomposition on the current data sequence in the multi-source data matrix to obtain the intrinsic mode function set of the current data sequence.
[0054] Specifically, empirical mode decomposition includes the following steps: 1. Identify all local maxima and local minima of the input current sequence x(t).
[0055] 2. Interpolate for all local maxima and local minima to fit the upper and lower envelopes of the original sequence.
[0056] 3. Calculate the mean of the upper and lower envelopes to obtain the mean envelope.
[0057] 4. Subtract the mean envelope from the original sequence to obtain the candidate components. .
[0058] 5. Inspection Does it meet the two conditions of the IMF: the number of extreme points is equal to or differs by at most one from the number of zero crossings throughout the entire data segment; and at any given time, the mean of the upper and lower envelopes defined by the local maxima and local minima is zero. If not, then... As the new x(t), repeat steps 1-4 until the condition is met. At this point, we obtain... This is the first IMF, denoted as .
[0059] 6. Isolate the first IMF from the original sequence to obtain the residual component. .
[0060] 7. As the new original sequence, repeat steps 1-6 to extract the second IMF, denoted as . The third IMF, denoted as ... until the residual components It becomes a monotonic function or a constant value until the IMF can no longer be extracted.
[0061] Finally, the intrinsic mode function set of the current data sequence is obtained.
[0062] S22. Since the first three IMFs usually have higher energy and clearer physical meaning, calculate the instantaneous frequencies of the first three eigenmode functions in the eigenmode function set at the same time and use them as the current-dominated instantaneous frequency sequence.
[0063] S23. From the IMF of the vibration signal, find the IMF component whose instantaneous frequency is statistically or trend-wise closest to the target instantaneous frequency sequence of the input. .right Perform a Hilbert transform to obtain its analytic signal, and then calculate its instantaneous amplitude. That is, the resonance amplitude sequence It characterizes the amplitude intensity of the vibration signal when it changes synchronously with the target current frequency.
[0064] S24. In order to observe the change of the correlation between vibration resonance intensity and temperature over time, a sliding window can be used to calculate the cross-correlation coefficient between the two.
[0065] Specifically, a time window length is set, and at time t, the resonance amplitude subsequence and temperature subsequence within the window are taken, and the Pearson cross-correlation coefficient R(t) of these two subsequences is calculated.
[0066] Slide the window by one time step and repeat the calculation to obtain the cross-correlation coefficient R(t) corresponding to each center time t, which is the resonance temperature correlation sequence R(t). Its magnitude and trend directly reflect the degree of correlation between the intensity of the electro-oscillatory coupling resonance and the temperature change of the equipment. Under healthy conditions, this correlation should remain relatively stable and have a high positive value; when the coupling resonance begins to decay or annihilate, this correlation will decrease or fluctuate drastically.
[0067] In some embodiments, recursive quantization analysis is used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, resulting in a coupled resonance state abrupt change marker sequence, including: S31. Based on the delay time τ and the embedding dimension m, the phase space of the resonance temperature correlation sequence is reconstructed to generate a high-dimensional phase space trajectory.
[0068] Specifically, according to Taken's embedding theorem, a one-dimensional observation sequence of a nonlinear dynamical system can be reconstructed using the time-delay coordinate method to obtain a phase space topologically equivalent to the original system. The autocorrelation function method is used to calculate the autocorrelation function C(τ) of the sequence R(t), choosing the point where the autocorrelation function first decreases to its initial value. The τ value corresponding to the first zero crossing.
[0069] The spurious nearest neighbor method is employed, gradually increasing the embedding dimension m to check whether the nearest neighbor of each point is spurious in the high-dimensional phase space. When the proportion of spurious nearest neighbors falls below a preset threshold, m is considered sufficient.
[0070] Using the obtained τ and m, the one-dimensional sequence R(t) is mapped to a series of vectors in an m-dimensional phase space. : .
[0071] The final series of phase space vectors { This refers to the trajectory in the high-dimensional phase space.
[0072] S32. Calculate the Euclidean distance between all pairwise state vectors in the trajectory. When the distance between a pair of vectors is less than a preset threshold, mark a recursion point at the corresponding position in the two-dimensional matrix to generate a recursion graph that visualizes the reproducibility of the system's dynamics.
[0073] Analyze all recursive points in the recursion graph and calculate the perpendicular distance from the coordinates of each recursive point to the main diagonal. This distance value represents the time interval between state recurrences.
[0074] All vertical distance values are divided into several fixed distance intervals. The proportion of the number of recursive points in each interval to the total number of recursive points is counted to form a probability distribution vector representing the vertical distance distribution, i.e., the recursive structure feature vector, which quantifies the time interval pattern of system state recurrence.
[0075] S33. Vertical distance Treating it as a random variable, its distribution is described by a recursive structure eigenvector. Calculate the skewness and kurtosis of this distribution, where skewness measures the asymmetry of the distribution; zero skewness indicates a symmetrical distribution, positive skewness indicates a long tail on the right (positive skewness), and negative skewness indicates a long tail on the left (negative skewness); kurtosis measures the sharpness and thickness of the tails of the distribution.
[0076] The obtained recursive distribution morphology index contains two numerical tuples: skewness and kurtosis. When the coupled resonance is stable, the state recursion pattern is regular, and the vertical distance distribution may exhibit a specific morphology, such as stable skewness and kurtosis. When the resonance begins to become disordered or annihilate, the time pattern of state recursion is disrupted, and the distribution morphology will change significantly, such as abrupt changes in skewness, a sharp increase or decrease in kurtosis.
[0077] S34. Slide a window on the timeline. Within each window, S31-S33 are performed on the corresponding segment of the resonance temperature correlation sequence to obtain a time-varying skewness sequence S(t) and kurtosis sequence K(t).
[0078] A sliding window combined with statistical tests is used to detect mutation points in the S(t) or K(t) sequence, resulting in a coupled resonance state mutation marker sequence F(t). F(t) is a binary sequence. F(t)=1 indicates that a mutation in the recursive distribution pattern is detected at time t, which is a possible coupled resonance annihilation point; F(t)=0 indicates that no mutation is detected.
[0079] In some embodiments, based on the coupled resonance state abrupt change marker sequence, the risk of resonance annihilation is assessed using an information geometry method to obtain a warning signal, including: S41. Set a statistical time window, length... Slide the window on the time axis t and count the number of times F(t)=1 in each window to obtain the mutation frequency sequence. An increase in mutation frequency means that the dynamic instability of the system is intensifying.
[0080] S42. At time t, take the most recent historical data segment of the mutation frequency sequence and treat it as a sample. Estimate its empirical probability distribution. .
[0081] Using the mutation frequency sequence data of the device during a known health history, the probability distribution Q estimated in the same way is used as a reference health status probability distribution.
[0082] calculate The distance between Q and the Bach is used to obtain the distribution difference sequence D(t). The larger D(t) is, the greater the difference between the statistical characteristics of the current mutation frequency and the healthy baseline, and the higher the risk.
[0083] S43. The risk evolution trend sequence S(t) is obtained by smoothing the distribution difference sequence D(t) using the exponential weighted moving average method.
[0084] S44. Set a low threshold. and high threshold .
[0085] The low threshold is determined by statistically analyzing the risk evolution trend sequence under historical health conditions, taking the mean plus two standard deviations. When the risk trend value exceeds two standard deviations from the health level, abnormal signs are considered to have begun to appear. The high threshold... You can simply set it to a fixed multiple of the low threshold, such as 1.5 times, to indicate that the risk has risen to a more serious level.
[0086] At each time t, compare S(t) with the threshold: if The output is normal; if If so, a yellow warning signal will be output; if If the signal is red, a warning signal will be output.
[0087] In some embodiments, the warning signal is verified to obtain the final fault warning and maintenance recommendations, including: S51. Once the system outputs a yellow warning, it indicates that a potential risk has been detected, which can be verified by increasing the detection sensitivity.
[0088] Specifically, the sliding window length in S34 can be... Reduce the window size, such as by halving it, and using this shorter window, re-execute S31-S34 based on the original resonance temperature correlation sequence to obtain the recalculated coupled resonance state abrupt change marker sequence. .
[0089] S52. Will As new input, S41-S44 are executed repeatedly to obtain a new warning signal.
[0090] S53. If the new warning signal is normal, it indicates that the risk has not been confirmed at a higher sensitivity level. The initial yellow warning may have been a false alarm or an occasional fluctuation. No further action may be taken, or only a minor abnormal event may be recorded.
[0091] If the new warning signal is still a yellow warning, or even upgraded to a red warning, it means that the risk trend has been confirmed to be persistent, the judgment has been verified, and it is confirmed that there is a real risk of coupling resonance annihilation, that is, the equipment has early signs of failure.
[0092] The final equipment failure warning may include relevant information such as: the time period for risk confirmation, the dominant resonant frequency range, the risk evolution trend curve, and preliminary maintenance suggestions given after matching with a common failure mode library.
[0093] In some embodiments, this application provides a power equipment fault prediction system based on multi-source information, including: Data acquisition and preprocessing module: Acquires raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, within a continuous monitoring period, and preprocesses the raw data to obtain a multi-source data matrix.
[0094] Feature extraction and pattern construction module: Extract time-frequency domain features and construct resonance patterns to obtain resonance temperature correlation sequence.
[0095] Coupled resonance detection module: Recursive quantization analysis is used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and the coupled resonance state abrupt change marker sequence is obtained.
[0096] Risk assessment and early warning module: Based on the coupled resonance state mutation marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain early warning signals.
[0097] Early warning verification and decision-making module: Verifies the early warning signals and obtains the final fault warning and maintenance suggestions.
[0098] In some embodiments, this application provides a power equipment fault prediction device based on multi-source information, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program to implement the steps of the power equipment fault prediction method based on multi-source information.
[0099] In some embodiments, this application provides a readable storage medium storing computer program instructions, which are read and executed by a processor to perform the steps of a power equipment fault prediction method based on multi-source information.
[0100] In some embodiments, this application provides a computer program product, including a computer program or instructions, wherein when the computer program or instructions are executed by a processor, the steps of implementing a power equipment fault prediction method based on multi-source information are provided.
[0101] It should be noted that, in this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0102] Any references to memory, storage, database, or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory.
[0103] Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
Claims
1. A method for predicting power equipment faults based on multi-source information, characterized in that, include: The raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, are acquired during the continuous monitoring period. The raw data is then preprocessed to obtain a multi-source data matrix. The time-frequency domain features are extracted and resonance modes are constructed to obtain the resonance temperature correlation sequence; Recursive quantitative analysis was used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and a coupled resonance state abrupt change marker sequence was obtained. Based on the aforementioned coupled resonance state abrupt change marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain an early warning signal; The warning signals are verified to obtain the final fault warning and maintenance recommendations.
2. The method according to claim 1, characterized in that, The original data is preprocessed to obtain a multi-source data matrix, including: The raw current data, raw vibration data, and raw temperature data are time-aligned and filtered to obtain current data sequences, vibration data sequences, and temperature data sequences. The current data sequence, vibration data sequence, and temperature data sequence are denoised using wavelet soft thresholding, and the denoised data are then fused to obtain a multi-source data matrix.
3. The method according to claim 1, characterized in that, The extracted time-frequency domain features are used to construct resonance modes, resulting in a resonance temperature correlation sequence, including: Empirical mode decomposition is performed on the current data sequence in the multi-source data matrix to obtain the eigenmode function set of the current data sequence; Calculate the instantaneous frequencies of the first three eigenmode functions in the set of eigenmode functions at the same time to obtain the current-dominated instantaneous frequency sequence; On the vibration data sequence in the multi-source data matrix, the Hilbert-Huang transform is applied to extract the instantaneous amplitude at the corresponding moment of the current-dominated instantaneous frequency sequence, thus obtaining the resonance amplitude sequence. By fusing the resonance amplitude sequence with the temperature data sequence in the multi-source data matrix, the cross-correlation coefficients within the same time window are calculated to obtain the resonance temperature correlation sequence.
4. The method according to claim 3, characterized in that, The empirical mode decomposition is specifically achieved by iteratively identifying local extreme points of the data sequence, fitting the upper and lower envelopes, and extracting the mean envelope until the intrinsic mode function condition is met.
5. The method according to claim 1, characterized in that, Recursive quantization analysis was used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, resulting in a coupled resonance state abrupt change marker sequence, including: Based on the delay time and embedding dimension, the resonant temperature correlation sequence is reconstructed in phase space to generate a high-dimensional phase space trajectory; From the recursive graph of the high-dimensional phase space trajectory, the vertical distance distribution of the recursive points relative to the diagonal is calculated to obtain the recursive structure feature vector; The skewness and kurtosis of the recursive structure feature vectors are calculated to obtain the recursive distribution morphology index; Within the sliding window, mutation point detection is performed on the recursive distribution morphology index to obtain the coupled resonance state mutation marker sequence.
6. The method according to claim 5, characterized in that, The phase space reconstruction calculates the delay time using the autocorrelation function method and determines the embedding dimension using the spurious nearest neighbor method, thereby mapping the one-dimensional sequence to a high-dimensional phase space.
7. The method according to claim 1, characterized in that, Based on the aforementioned coupled resonance state abrupt change marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain early warning signals, including: The number of mutation markers in the coupled resonance state mutation marker sequence within a unit time window is counted to obtain the mutation frequency sequence; The Bartholomew's distance between the probability distribution of the mutation frequency sequence and the probability distribution of the reference healthy state is calculated to obtain the distribution difference sequence; The distribution difference sequence is smoothed using the exponentially weighted moving average method to obtain the risk evolution trend sequence; The risk evolution trend sequence is compared with preset low and high thresholds. When the value of the risk evolution trend sequence exceeds the low threshold, a yellow warning signal is output, and when it exceeds the high threshold, a red warning signal is output.
8. The method according to claim 7, characterized in that, The low threshold is obtained by statistically analyzing the risk evolution trend sequence under historical health conditions, taking the mean and adding 2 times the standard deviation; the high threshold is set to 1.5 times the low threshold.
9. The method according to claim 1, characterized in that, The warning signals are verified to obtain final fault warnings and maintenance recommendations, including: When the warning signal is a yellow warning signal, adaptive adjustment is triggered to shorten the sliding window length for mutation point detection and recalculate the coupled resonance state mutation flag sequence. Based on the recalculated coupled resonance state mutation marker sequence, the steps for assessing the risk of resonance annihilation are repeated to obtain new early warning signals; If the new warning signal is yellow or red, then the final equipment failure warning and maintenance recommendations will be generated and output.
10. A power equipment fault prediction system based on multi-source information, characterized in that, include: Data acquisition and preprocessing module: Acquires raw data of the target power equipment, including three-phase current waveform data, vibration acceleration data and infrared thermal image temperature data, within a continuous monitoring period, and preprocesses the raw data to obtain a multi-source data matrix; Feature extraction and pattern construction module: Extracts time-frequency domain features and constructs resonance patterns to obtain resonance temperature correlation sequences; Coupled resonance detection module: Recursive quantization analysis is used to detect the coupled resonance annihilation effect in the resonance temperature correlation sequence, and a coupled resonance state abrupt change marker sequence is obtained; Risk assessment and early warning module: Based on the coupled resonance state mutation marker sequence, the risk of resonance annihilation is assessed using information geometry methods to obtain an early warning signal; Early warning verification and decision-making module: Verifies the early warning signal to obtain the final fault warning and maintenance suggestions.