A power distribution box operation state monitoring method and system based on multi-source data fusion

Abstract

This invention relates to the field of distribution box technology, specifically disclosing a method and system for monitoring the operating status of distribution boxes based on multi-source data fusion. This invention acquires current signals and thermal imaging video information to construct a multi-dimensional feature analysis system. First, the current signal is decomposed in the time-frequency domain to extract short-time fluctuation features and high-frequency resonance features, accurately capturing transient anomalies caused by latent faults such as intermittent poor contact. Simultaneously, the thermal imaging sequence is dynamically analyzed, and instantaneous temperature features are obtained through heat diffusion values ​​and temperature gradient change maps, enabling real-time monitoring of local overheating. The three types of features are input into a dynamic fusion model to generate comprehensive feature values. By analyzing the feature fusion values, the operating status of the distribution box can be more accurately identified, thereby enabling timely detection and capture of intermittent anomalies and latent faults.

Description

Technical Field

[0001] This invention relates to the field of distribution box technology, and in particular to a method and system for monitoring the operating status of distribution boxes based on multi-source data fusion. Background Technology

[0002] Distribution boxes are key equipment in power systems used to distribute, control, and protect electrical energy. They reduce the voltage of high-voltage current and distribute it to various electrical devices according to different voltage levels, ensuring the stability and security of power supply.

[0003] Existing distribution boxes typically use a single threshold alarm mode (fixed threshold, such as triggering an alarm when the current exceeds a certain value) for status monitoring. However, the single threshold alarm mode cannot capture hidden faults such as intermittent anomalies (not continuously exceeding the threshold, such as poor contact). Therefore, a distribution box operation status monitoring method based on multi-source data fusion is needed to solve the above problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for monitoring the operating status of distribution boxes based on multi-source data fusion, so as to solve the technical problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for monitoring the operational status of a distribution box based on multi-source data fusion, comprising:

[0007] Acquire multi-source data parameters of the distribution box within a preset time period, including current signals and thermal imaging video information;

[0008] The time resolution and frequency resolution of the current signal are obtained, and the length of the time window within a preset time period is obtained based on the time resolution and the frequency resolution. The short-time fluctuation characteristics of the current signal within the time window length are then obtained.

[0009] The corresponding current waveform information is obtained based on the current signal, and the transient time-domain features are obtained based on the current waveform information. The transient time-domain features are then extracted based on the Fourier transform to obtain the high-frequency resonance features.

[0010] The thermal imaging video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple thermal diffusion values ​​are obtained according to the multiple video frame sequence lengths, and a temperature gradient change map is generated according to the multiple thermal diffusion values. Instantaneous temperature characteristics are obtained according to the temperature gradient change map.

[0011] The short-term fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics are input into a dynamic feature fusion model for fusion to obtain feature fusion values;

[0012] The operating status of the distribution box is monitored based on the fusion value of the features.

[0013] Preferably, the step of acquiring the time resolution and frequency resolution of the current signal, acquiring the length of a time window within a preset time period based on the time resolution and frequency resolution, and acquiring the short-time fluctuation characteristics of the current signal within the time window length includes:

[0014] Multiple instantaneous spikes of the current signal within a preset time period are acquired, and multiple time intervals between the multiple instantaneous spikes are acquired according to a preset time sequence. The multiple time intervals are then filtered according to the minimum value to obtain a first time interval, and the first time interval is used as the time resolution.

[0015] The current signal is baseline drift corrected to obtain a standard current signal, and transient event detection is performed on the standard current signal based on the Teager energy operator to obtain a transient signal;

[0016] Frequency components are extracted from transient signals based on a sparse constrained spectrum estimation algorithm to obtain multiple transient frequency components. Multiple transient frequency component differences between the multiple transient frequency components are obtained according to a preset time series. The multiple transient frequency component differences are then filtered according to the minimum value to obtain the first transient frequency component difference. The first transient frequency component difference is used as the frequency resolution.

[0017] Obtain the first sampling interval and the first number of sampling points at the stated time resolution, and calculate the length of the first time window based on the first sampling interval and the first number of sampling points;

[0018] Obtain the second sampling interval and the second number of sampling points at the frequency resolution, and calculate the second time window length based on the second sampling interval and the second number of sampling points;

[0019] The first time window length and the second time window length are merged according to the time sequence within a preset time period to obtain the merged time window length, and the merged time window length is used as the time window length.

[0020] Obtain the current signal fluctuation diagram within the specified time window length, and obtain the short-term fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-term fluctuation duration of the current as the short-term fluctuation feature.

[0021] Preferably, the step of obtaining transient time-domain features based on the current waveform information, and extracting features from the transient time-domain features based on Fourier transform to obtain high-frequency resonance features includes:

[0022] One-dimensional time-series signal for obtaining current waveform information based on a two-stream convolution model;

[0023] The current variation curve is obtained based on the one-dimensional time-series signal;

[0024] The current change curve is transformed into two-dimensional wavelet time-frequency image features based on wavelet transform, and the instantaneous change information of the signal is obtained based on the two-dimensional wavelet time-frequency image features.

[0025] The transient time-domain features are obtained based on the instantaneous change information of the signal;

[0026] Based on the Fourier transform, the transient time-domain features are extracted to obtain instantaneous frequency-domain feature information, and the instantaneous frequency within a preset time period is obtained based on the instantaneous frequency-domain feature information.

[0027] The instantaneous frequency is divided based on a preset frequency division threshold to obtain high-frequency components;

[0028] Based on the spectral analysis of the resonance characteristics in the high-frequency components, the high-frequency resonance characteristics are obtained.

[0029] Preferably, the steps of obtaining multiple thermal diffusion values ​​corresponding to the lengths of the multiple video frame sequences, generating a temperature gradient change map based on the multiple thermal diffusion values, and obtaining instantaneous temperature features based on the temperature gradient change map include:

[0030] The pixel temperature value in each video frame is obtained according to the length of the multiple video frame sequences. The pixel temperature standard deviation of the pixel temperature change between two consecutive frames is calculated based on the multiple pixel temperature values, and the pixel temperature standard deviation is used as the heat diffusion value.

[0031] The video frame sequence direction is obtained based on the video frame sequence length;

[0032] The multiple thermal diffusion values ​​are arranged into a thermal diffusion value sequence according to the video frame sequence direction, and a temperature gradient change map is generated based on the thermal diffusion value sequence;

[0033] Multiple instantaneous temperature values ​​are extracted from the temperature gradient change map, and instantaneous temperature values ​​are obtained based on the multiple instantaneous temperature values. The instantaneous features corresponding to the instantaneous temperature values ​​are used as instantaneous temperature features.

[0034] Preferably, the step of inputting the short-time fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics into a dynamic feature fusion model for fusion to obtain feature fusion values ​​includes:

[0035] Based on the short-term fluctuation characteristics, a short-term fluctuation feature vector is obtained, and the short-term fluctuation feature vector is normalized to obtain a normalized value of the short-term fluctuation feature vector.

[0036] Based on the high-frequency resonance characteristics, a high-frequency resonance feature vector is obtained, and the high-frequency resonance feature vector is normalized to obtain a normalized value of the high-frequency resonance feature vector.

[0037] The instantaneous temperature feature vector is obtained based on the instantaneous temperature features, and the instantaneous temperature feature vector is normalized to obtain the normalized value of the instantaneous temperature feature vector.

[0038] Obtain the bias reference based on the backpropagation algorithm;

[0039] The normalized values ​​of the short-term fluctuation feature vector, the high-frequency resonance feature vector, and the instantaneous temperature feature vector, along with the bias reference, are input into the dynamic feature fusion model for fusion to obtain a comprehensive vector feature fusion value. The function formula of the dynamic feature fusion model is as follows:

[0040] d(r)=σ[D(B)*a+G(X)*b+W(D)*(1-ab)+E];

[0041] Where d(r) represents the fusion value of the comprehensive vector features, σ represents the activation function, such as ReLU or Sigmoid, D(B) represents the normalized value of the short-term fluctuation feature vector, a represents the weight value of the normalized value of the short-term fluctuation feature vector, G(X) represents the normalized value of the high-frequency resonance feature vector, b represents the weight value of the normalized value of the high-frequency resonance feature vector, W(D) represents the normalized value of the instantaneous temperature feature vector, and E represents the bias reference.

[0042] The fused value of the comprehensive vector features is used as the feature fusion value.

[0043] Preferably, the step of monitoring the operating status of the distribution box based on the feature fusion value further includes:

[0044] Multiple feature fusion values ​​within a preset time window are obtained based on the feature fusion values;

[0045] The feature fusion mean is obtained based on the multiple feature fusion values;

[0046] The standard feature fusion difference is obtained based on the mean of the feature fusion and multiple feature fusion values, and the calculation formula is as follows:

[0047] ;

[0048] Where B(Z) represents the standard feature fusion difference, and R(T) iThis represents the fusion value of the i-th feature, where n represents the number of pixel brightness values. Let represent the mean of the feature fusion, and i represent the index of the feature fusion value, where i = 1, 2, 3...n;

[0049] The standard feature fusion value is calculated based on the feature fusion mean and the standard feature fusion difference, wherein the calculation formula is:

[0050] ;

[0051] Where B(R) represents the standard feature fusion value, B(Z) represents the mean of feature fusion, and B(Z) represents the difference of standard feature fusion.

[0052] Determine whether the standard feature fusion value is less than a preset threshold;

[0053] If the standard feature fusion value is greater than or equal to the preset threshold, the operating status of the distribution box is determined to be normal.

[0054] If the standard feature fusion value is less than the preset threshold, the operating status of the distribution box is determined to be an intermittent abnormal state.

[0055] This application also provides a distribution box operation status monitoring system based on multi-source data fusion, including:

[0056] The first acquisition module is used to acquire multi-source data parameters of the distribution box within a preset time period, wherein the multi-source data parameters include current signals and thermal imaging video information;

[0057] The second acquisition module is used to acquire the time resolution and frequency resolution of the current signal, and to acquire the length of the time window within a preset time according to the time resolution and the frequency resolution, and to acquire the short-time fluctuation characteristics of the current signal within the time window length.

[0058] The third acquisition module is used to acquire the corresponding current waveform information based on the current signal, and to acquire transient time-domain features based on the current waveform information. Based on Fourier transform, the transient time-domain features are extracted to obtain high-frequency resonance features.

[0059] The first splitting module is used to split the thermal imaging video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtain multiple corresponding thermal diffusion values ​​according to the multiple video frame sequence lengths, generate a temperature gradient change map according to the multiple thermal diffusion values, and obtain instantaneous temperature characteristics according to the temperature gradient change map.

[0060] The first fusion module is used to input the short-time fluctuation characteristics, high-frequency resonance characteristics and instantaneous temperature characteristics into the dynamic feature fusion model for fusion to obtain feature fusion values;

[0061] The first monitoring module is used to monitor the operating status of the distribution box based on the feature fusion value.

[0062] Preferably, the second acquisition module includes:

[0063] The first acquisition unit is used to acquire multiple instantaneous spikes of the current signal within a preset time period, acquire multiple time intervals between the multiple instantaneous spikes according to a preset time sequence, filter the multiple time intervals according to the minimum value to obtain a first time interval, and use the first time interval as the time resolution.

[0064] The first correction unit is used to perform baseline drift correction on the current signal to obtain a standard current signal, and to perform transient event detection on the standard current signal based on the Teager energy operator to obtain a transient signal.

[0065] The first extraction unit is used to extract frequency components of a transient signal based on a sparse constrained spectrum estimation algorithm, obtain multiple transient frequency components, acquire multiple transient frequency component differences between the multiple transient frequency components according to a preset time series, filter the multiple transient frequency component differences according to the minimum value, obtain a first transient frequency component difference, and use the first transient frequency component difference as the frequency resolution.

[0066] The second acquisition unit is used to acquire the first sampling interval and the first number of sampling points under the time resolution, and to calculate the length of the first time window based on the first sampling interval and the first number of sampling points;

[0067] The third acquisition unit is used to acquire the second sampling interval and the second number of sampling points under the frequency resolution, and to calculate the second time window length based on the second sampling interval and the second number of sampling points;

[0068] The first merging unit is used to merge the first time window length and the second time window length according to the time sequence within a preset time period to obtain the merged time window length, and use the merged time window length as the time window length.

[0069] The fourth acquisition unit is used to acquire the current signal fluctuation diagram within the time window length, acquire the short-time fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-time fluctuation duration of the current as a short-time fluctuation feature.

[0070] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0071] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0072] The beneficial effects of this application are as follows: This invention constructs a multi-dimensional feature analysis system by simultaneously acquiring current signals and thermal imaging video information. First, the current signal is decomposed in the time-frequency domain to extract short-term fluctuation features and high-frequency resonance features, accurately capturing transient anomalies caused by latent faults such as intermittent poor contact. Simultaneously, the thermal imaging sequence is dynamically analyzed, obtaining instantaneous temperature features through heat diffusion values ​​and temperature gradient change maps, achieving real-time monitoring of local overheating. The three types of features are input into a dynamic fusion model to generate comprehensive feature values. Combined with statistical analysis methods, a state assessment system is constructed, effectively distinguishing between normal operation and intermittent abnormal states. This method overcomes the limitations of traditional single-threshold monitoring, improving fault identification accuracy through multi-source data complementarity, especially showing significant detection advantages for latent faults that have not exceeded the threshold, providing reliable technical support for intelligent operation and maintenance of distribution boxes. Attached Figure Description

[0073] Figure 1 This is a schematic diagram of a method flow according to an embodiment of this application.

[0074] Figure 2 This is a schematic diagram of the system structure according to an embodiment of this application.

[0075] Figure 3 This is a schematic diagram of the internal structure of a computer device according to an embodiment of this application.

[0076] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0077] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0078] like Figure 1-3 As shown, this application provides a method for monitoring the operating status of a distribution box based on multi-source data fusion, including:

[0079] S1. Obtain multi-source data parameters of the distribution box within a preset time period, including current signal and thermal imaging video information;

[0080] S2. Obtain the time resolution and frequency resolution of the current signal, and obtain the length of the time window within a preset time according to the time resolution and frequency resolution, and obtain the short-time fluctuation characteristics of the current signal within the time window length;

[0081] S3. Obtain the corresponding current waveform information based on the current signal, and obtain the transient time-domain features based on the current waveform information. Extract the transient time-domain features based on Fourier transform to obtain high-frequency resonance features.

[0082] S4. The thermal imaging video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple thermal diffusion values ​​are obtained according to the multiple video frame sequence lengths. A temperature gradient change map is generated according to the multiple thermal diffusion values. Instantaneous temperature characteristics are obtained according to the temperature gradient change map.

[0083] S5. Input the short-time fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics into the dynamic feature fusion model for fusion to obtain the feature fusion value;

[0084] S6. Monitor the operating status of the distribution box based on the feature fusion value.

[0085] As described in steps S1-S6 above, existing distribution boxes typically use a single threshold alarm mode (fixed threshold, such as triggering an alarm when the current exceeds a certain value) for status monitoring. However, the single threshold alarm mode cannot capture the problem of intermittent abnormalities (not continuously exceeding the threshold, such as poor contact) and hidden faults. Therefore, this invention first acquires multi-source data parameters of the distribution box within a preset time. The multi-source data parameters include current signals and thermal imaging video information. By collecting current signals and thermal imaging video information during the operation of the distribution box, the current signals can reflect the changes in current in the circuit, and the thermal imaging video information can intuitively present the temperature distribution and changes of various parts of the distribution box, providing a data foundation for subsequent analysis and comprehensively understanding the operating status of the distribution box from multiple dimensions. At the same time, the operating status of the distribution box is affected by a variety of factors, and a single data point cannot accurately determine the status. Current and temperature are key indicators reflecting the operating status of distribution boxes. Multi-source data fusion can improve the accuracy and reliability of monitoring, compensate for the shortcomings of traditional single-threshold alarm modes, and comprehensively capture various status information during the operation of the distribution box. Then, the time resolution and frequency resolution of the current signal are acquired, and the length of a time window within a preset time period is obtained based on the time resolution and frequency resolution. The time window length refers to the length of a fixed time period extracted from the continuous time signal, used to extract local features within that time period. In dynamic feature fusion, the time window divides the continuous signal into multiple equal-length segments (e.g., 100ms, 1 second, etc.), and each segment undergoes independent feature extraction (extraction is based on the number of sampling points and the sampling interval) to capture transient or periodic changes, obtaining the short-time fluctuation characteristics of the current signal within the time window length, and determining an appropriate time window length to analyze the short-time fluctuation characteristics of the current signal. Time resolution and frequency resolution reflect the details of changes in the current signal in the time and frequency dimensions, and short-time fluctuation characteristics can reflect the instantaneous changes in current, helping to detect abnormal current fluctuations and providing a basis for judging whether there is a fault in the distribution box. Different faults will cause changes in the time and frequency characteristics of the current signal. By accurately acquiring these resolution and fluctuation characteristics, abnormal changes in current signals can be detected more sensitively. For example, problems such as intermittent poor contact can cause short-term fluctuations in current, thereby improving the monitoring capability for latent faults. Secondly, the corresponding current waveform information is obtained based on the current signal, and transient time-domain features are acquired from this information. High-frequency resonance features are then extracted from these transient time-domain features using Fourier transform. This process of extracting transient time-domain features from current waveform information and then obtaining high-frequency resonance features through Fourier transform is crucial for assessing the stability and health of the circuit within the distribution box. Furthermore, circuit faults often trigger changes in the current waveform, producing abnormal high-frequency components and resonance phenomena.Extracting these features allows for in-depth analysis of the circuit's operating status, timely detection of potential faults, such as damage to electrical components leading to abnormal high-frequency resonance, and early warning of potential malfunctions. Next, the thermal imaging video information is split according to a preset video frame sequence length, resulting in multiple video frame sequence lengths. Multiple heat diffusion values ​​are obtained based on these multiple video frame sequence lengths, and a temperature gradient change map is generated based on these multiple heat diffusion values. Instantaneous temperature characteristics are obtained from the temperature gradient change map. By analyzing the thermal imaging video frame sequence, heat diffusion values ​​and temperature gradient change maps are obtained, thus yielding instantaneous temperature characteristics. These features can intuitively reflect the temperature change trends and instantaneous temperature conditions of various parts of the distribution box. Abnormal temperature increases are often an important indicator of distribution box malfunctions, helping to quickly detect overheating and other problems. Furthermore, temperature is a crucial indicator for measuring the operating status of the distribution box, and heat diffusion values ​​and temperature gradient change maps can more meticulously display the dynamic temperature change process. Compared to traditional single-temperature monitoring, this method can capture abnormal temperature fluctuations more promptly and accurately, effectively monitor the heating status of the distribution box, and prevent malfunctions caused by overheating. The short-term fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics are then input into a dynamic feature fusion model for fusion, resulting in a feature fusion value. This feature fusion value integrates information from multiple aspects, providing a more comprehensive and accurate reflection of the distribution box's operating status, improving monitoring accuracy and reliability. Furthermore, a single feature may not fully reflect the distribution box's operating status; different features are complementary. By fusing multiple features, information from various aspects can be fully utilized, reducing misjudgments and omissions, and more accurately determining the distribution box's operating condition, providing stronger support for subsequent monitoring and decision-making. Finally, the distribution box's operating status is monitored based on the feature fusion value, allowing for the determination of the distribution box's operating status and distinguishing between normal and intermittent abnormal states. Timely detection of abnormalities in distribution boxes provides maintenance personnel with accurate information to take appropriate measures, ensuring the normal operation of distribution boxes and improving the reliability of the power system. Furthermore, traditional single-threshold alarm modes are difficult to detect intermittent anomalies, while the analysis of feature fusion values ​​can more accurately identify the operating status of distribution boxes, thereby enabling timely detection and capture of intermittent anomalies and the discovery of hidden faults.

[0086] In one embodiment, step S2, which involves acquiring the time resolution and frequency resolution of the current signal, obtaining the length of a time window within a preset time period based on the time resolution and frequency resolution, and acquiring the short-time fluctuation characteristics of the current signal within the time window length, includes:

[0087] S201. Acquire multiple instantaneous spikes of the current signal within a preset time period, and acquire multiple time intervals between the multiple instantaneous spikes according to a preset time sequence, and filter the multiple time intervals according to the minimum value to obtain a first time interval, and use the first time interval as the time resolution.

[0088] S202. Baseline drift correction is performed on the current signal to obtain a standard current signal, and transient event detection is performed on the standard current signal based on the Teager energy operator to obtain a transient signal;

[0089] S203. Based on the sparse constrained spectrum estimation algorithm, frequency components are extracted from the transient signal to obtain multiple transient frequency components. Multiple transient frequency component differences between the multiple transient frequency components are obtained according to a preset time series. The multiple transient frequency component differences are filtered according to the minimum value to obtain the first transient frequency component difference. The first transient frequency component difference is used as the frequency resolution.

[0090] S204. Obtain the first sampling interval and the first number of sampling points at the time resolution, and calculate the length of the first time window based on the first sampling interval and the first number of sampling points;

[0091] S205. Obtain the second sampling interval and the second number of sampling points at the frequency resolution, and calculate the second time window length based on the second sampling interval and the second number of sampling points;

[0092] S206. Merge the first time window length and the second time window length according to the time sequence within a preset time period to obtain the merged time window length, and use the merged time window length as the time window length.

[0093] S207. Obtain the current signal fluctuation diagram within the time window length, and obtain the short-time fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-time fluctuation duration of the current as a short-time fluctuation feature.

[0094] As described in steps S201-S207 above, the present invention first acquires multiple instantaneous spikes of the current signal within a preset time period. These instantaneous spikes are extremely short-duration current or voltage signals (in this scheme, they are obtained using current signals, typically in the microsecond to millisecond range). Sudden pulse signals with amplitudes significantly higher than the baseline value in the current information graph are taken as instantaneous spikes. Thus, multiple instantaneous spikes occur within the preset time period. Therefore, multiple time intervals between these instantaneous spikes can be acquired according to a preset time sequence. These time intervals are then filtered according to their minimum values ​​to obtain a first time interval, which is used as the time resolution. By finding the minimum value of the time intervals between adjacent instantaneous spikes of the current signal as the time resolution, the minimum change interval of the current signal in the time dimension can be accurately captured. The time resolution refers to the minimum time interval between two consecutive events (in this scheme, two consecutive instantaneous spikes) that the system can distinguish or measure. It reflects the system's sensitivity to changes in the signal's time dimension and determines its ability to capture details during dynamic processes. This helps to clarify the precision of current signal changes, providing a timescale basis for subsequent analysis. It can keenly detect instantaneous fluctuations in current, assisting in judging the stability of the circuit. At the same time, instantaneous changes in the current signal contain fault information. Determining the time resolution can quantify this change from a time dimension, facilitating accurate analysis of the current signal and providing basic data for monitoring the operating status of the distribution box. Since the actual collected current signal may have baseline drift, which will affect the analysis results, it is necessary to perform baseline drift correction on the current signal to obtain a standard current signal. Then, based on the Teager energy operator, transient event detection is performed on the standard current signal to obtain a transient signal. The Teager Energy Operator (TEO) is a nonlinear signal processing tool mainly used to extract the energy information of the signal and can reflect the instantaneous energy changes of the signal. By standardizing the current signal through baseline drift correction, the baseline drift interference in the signal is removed, making subsequent analysis more accurate. The Teager energy operator detects transient events, highlighting transient changes in signals. These transient changes are often related to anomalies in circuits, helping to identify potential fault signs. Then, based on a sparse constrained spectrum estimation algorithm, frequency components are extracted from the transient signal, resulting in multiple transient frequency components. Multiple transient frequency component differences are obtained according to a preset time series, and these differences are filtered by minimum value to obtain the first transient frequency component difference. This first transient frequency component difference is used as the frequency resolution. By using the sparse constrained spectrum estimation algorithm to extract transient frequency components and finding the minimum difference between adjacent components as the frequency resolution, the minimum variation difference of the current signal in the frequency dimension can be clearly identified, reflecting fine changes in the signal frequency components and assisting in the identification of frequency anomalies in the circuit.Simultaneously, frequency changes can reflect the operating state of the circuit. Obtaining frequency resolution allows for in-depth analysis of the current signal from a frequency perspective, which is of great significance for monitoring distribution box faults. Next, the first sampling interval and the first number of sampling points at the stated time resolution are obtained, and the first time window length is calculated based on these parameters. The first sampling interval and number of points can be determined based on the time resolution, and the first time window length is calculated accordingly. This provides a range for analyzing the characteristics of the current signal at a specific time scale, helping to capture the signal change patterns within that time scale. Furthermore, the first time window length is the foundation for analyzing signal characteristics. The window length determined by the time resolution allows for focusing on short-term changes in the current signal, preparing for subsequent analysis of short-term fluctuation characteristics. Next, the second sampling interval and the second number of sampling points at the stated frequency resolution are obtained, and the second time window length is calculated based on these parameters. Calculating the second time window length based on the frequency resolution allows for supplementary analysis of the current signal from a frequency-related time scale, complementing the first time window length and providing a more comprehensive coverage of signal change characteristics. Then... The first time window length and the second time window length are merged according to a time sequence within a preset time period to obtain a merged time window length. This merged time window length is used as the overall time window length. By merging the two time window lengths, the time scales corresponding to time resolution and frequency resolution are combined to form a unified time window length, which more comprehensively and accurately reflects the changes in the current signal. At the same time, individual time windows have limitations; merging integrates information from different perspectives, providing support for obtaining more comprehensive current signal characteristics. Finally, a current signal fluctuation diagram is obtained within the time window length, and the short-term fluctuation duration of the current is obtained from the current signal fluctuation diagram. This short-term fluctuation duration is used as a short-term fluctuation feature. By drawing a current signal fluctuation diagram and obtaining the short-term fluctuation duration as a short-term fluctuation feature, the fluctuation of the current signal within the merged time window can be intuitively reflected. These fluctuations may be related to abnormal operation of the distribution box, providing a basis for judging the operating status. Furthermore, the short-term fluctuation feature is an important indicator for judging the operating status of the distribution box; directly obtaining this feature can quickly detect abnormal current fluctuations and promptly identify potential faults in the distribution box.

[0095] In one embodiment, step S3, which involves obtaining transient time-domain features based on the current waveform information and extracting features from the transient time-domain features using Fourier transform to obtain high-frequency resonance features, includes:

[0096] S301. A one-dimensional time-series signal for obtaining current waveform information based on a dual-stream convolution model;

[0097] S302. Obtain the current change curve based on the one-dimensional timing signal;

[0098] S303. Based on wavelet transform, the current change curve is converted into two-dimensional wavelet time-frequency image features, and the instantaneous change information of the signal is obtained according to the two-dimensional wavelet time-frequency image features;

[0099] S304. Obtain transient time-domain features based on the instantaneous change information of the signal;

[0100] S305. Based on Fourier transform, the transient time-domain features are extracted to obtain instantaneous frequency-domain feature information, and the instantaneous frequency within a preset time period is obtained based on the instantaneous frequency-domain feature information.

[0101] S306. The instantaneous frequency is divided based on a preset frequency division threshold to obtain high-frequency components;

[0102] S307. Based on the spectral analysis of the resonance characteristics in the high-frequency components, the high-frequency resonance characteristics are obtained.

[0103] As described in steps S301-S307 above, the present invention first obtains a one-dimensional time-series signal of current waveform information based on a two-stream convolution model. The two-stream convolution model can effectively extract the one-dimensional time-series signal from the current waveform information. The one-dimensional time-series signal refers to a discrete or continuous data sequence in which a single variable changes over time. Its core feature is that the data points (current) are arranged in chronological order, and each moment corresponds to a sequential feature of the current changing over time. This current signal retains the sequential feature of the current changing over time, providing a basis for subsequent analysis of the current change trend. It can more clearly show the dynamic changes of the current on the time axis. At the same time, the current waveform information is complex, and the two-stream convolution model can automatically learn and extract the key time-series features. Compared with other methods, it can more efficiently and accurately obtain the basic signal for analyzing current changes. Then, the current change curve is obtained based on the one-dimensional time-series signal. In this way, the one-dimensional time-series signal is transformed into an intuitive current change curve, making the current change trend over time clear at a glance. By observing the shape, slope, and other characteristics of the curve, the stability of the current and the presence of abnormal fluctuations can be quickly determined. Since existing time-domain analysis methods struggle to simultaneously consider both time and frequency information, wavelet transform's time-frequency analysis capabilities can compensate for this deficiency. Therefore, the current variation curve is transformed into a two-dimensional wavelet time-frequency image feature based on wavelet transform, and instantaneous signal variation information is obtained from this two-dimensional wavelet time-frequency image feature. This method allows for the simultaneous observation of changes in the current signal in both time and frequency dimensions, helping to detect frequency component changes at different times and instantaneous abnormal frequencies. Then, transient time-domain features are obtained based on these instantaneous signal variation information, which highlight the current signal's characteristics. The instantaneous changes in the signal can more accurately reflect abnormal current conditions within a short period, which is of great significance for detecting transient faults in circuits and can improve the monitoring capability of transient faults. Next, based on Fourier transform, the transient time-domain features are extracted to obtain instantaneous frequency-domain feature information. Based on this instantaneous frequency-domain feature information, the instantaneous frequency within a preset time period is obtained. The instantaneous frequency can be analyzed from a frequency perspective to understand the frequency distribution at different times, discover potential frequency anomalies, and provide a basis for judging whether there are problems such as resonance in the circuit. Secondly, the instantaneous frequency is segmented based on a preset frequency division threshold to obtain high-frequency components. In the complex frequency components of current signals, the high-frequency part is more reflective of circuit anomalies. Through frequency division threshold segmentation, high-frequency components can be focused on specifically, improving the accuracy and efficiency of fault monitoring. Finally, based on spectral analysis of the resonance characteristics in the high-frequency components, high-frequency resonance characteristics are obtained. Resonance phenomena in the circuit may cause problems such as overvoltage and overcurrent, endangering the normal operation of the distribution box. Obtaining high-frequency resonance characteristics can effectively monitor the resonance in a circuit and promptly detect potential safety hazards. At the same time, high-frequency resonance characteristics are also one of the important parameters for distribution box testing.

[0104] In one embodiment, step S4, which involves obtaining multiple thermal diffusion values ​​corresponding to the lengths of multiple video frame sequences, generating a temperature gradient change map based on the multiple thermal diffusion values, and obtaining instantaneous temperature features based on the temperature gradient change map, includes:

[0105] S401. Obtain the pixel temperature value in each video frame according to the length of the multiple video frame sequences, calculate multiple pixel temperature standard deviations of pixel temperature change between two consecutive frames according to the multiple pixel temperature values, and use the pixel temperature standard deviations as heat diffusion values.

[0106] S402. Obtain the video frame sequence direction based on the video frame sequence length;

[0107] S403. The multiple heat diffusion values ​​are arranged into a heat diffusion value sequence according to the video frame sequence direction, and a temperature gradient change map is generated based on the heat diffusion value sequence.

[0108] S404. Extract multiple instantaneous temperature values ​​based on the temperature gradient change map, obtain instantaneous temperature values ​​based on the multiple instantaneous temperature values, and use the instantaneous features corresponding to the instantaneous temperature values ​​as instantaneous temperature features.

[0109] As described in steps S401-S404 above, this invention obtains the pixel temperature value in each video frame based on the length of the multiple video frame sequences, calculates the standard deviation of pixel temperature change between every two consecutive frames based on the multiple pixel temperature values, and uses the pixel temperature standard deviation as the heat diffusion value. Thus, by calculating the standard deviation of pixel temperature change between every two consecutive frames as the heat diffusion value, the degree of temperature change between video frames can be quantitatively described. A larger heat diffusion value indicates a more drastic temperature change, reflecting frequent heat transfer activity in the corresponding part of the distribution box, potentially indicating abnormal heat dissipation or localized overheating. This provides a quantitative indicator for judging the thermal state of the distribution box. Furthermore, temperature changes in thermal imaging videos are important clues reflecting the operating status of the distribution box. Using the pixel temperature standard deviation to measure heat diffusion converts temperature changes into quantifiable and easily comparable data, facilitating subsequent analysis and judgment. Then, the video frame sequence direction is obtained based on the video frame sequence length, thereby clarifying the video frame sequence direction and providing a spatial reference for subsequent processing of the heat diffusion value. This helps to analyze the changing patterns of thermal diffusion values ​​in the correct order and direction, making the generated temperature gradient change map more accurately reflect the actual temperature distribution trend of the distribution box and enhancing the reliability of the analysis results. Secondly, multiple thermal diffusion values ​​are combined into a thermal diffusion value sequence according to the direction of the video frame sequence, and a temperature gradient change map is generated based on the thermal diffusion value sequence. In this way, the thermal diffusion values ​​are combined into a sequence according to the direction of the video frame sequence and a temperature gradient change map is generated, which intuitively displays the spatial distribution and trend of temperature changes in the distribution box. The graph quickly identifies areas of significant temperature variation, indicating potential sources of abnormal heat generation. This facilitates problem location and analysis for maintenance personnel. Compared to standalone thermal diffusion data, the temperature gradient graph provides a more intuitive and comprehensive view of temperature changes, aligning with human visual perception and aiding in the rapid detection of potential faults. Furthermore, multiple instantaneous temperature values ​​are extracted from the temperature gradient graph. These instantaneous temperature features, corresponding to specific instantaneous temperature values, are then used as instantaneous temperature characteristics. This process of extracting multiple instantaneous temperature values ​​and acquiring instantaneous temperature characteristics from the temperature gradient graph allows for the detection of abnormal temperatures in the distribution box at specific moments. These features serve as crucial evidence for determining the current operating status of the distribution box, enabling timely detection of abnormal temperature fluctuations and providing support for early fault warnings.

[0110] In one embodiment, step S5, which involves inputting the short-time fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics into a dynamic feature fusion model for fusion to obtain feature fusion values, includes:

[0111] S501. Obtain a short-term fluctuation feature vector based on the short-term fluctuation characteristics, and normalize the short-term fluctuation feature vector to obtain a normalized value of the short-term fluctuation feature vector.

[0112] S502. Obtain the high-frequency resonance feature vector based on the high-frequency resonance characteristics, and normalize the high-frequency resonance feature vector to obtain the normalized value of the high-frequency resonance feature vector.

[0113] S503. Obtain the instantaneous temperature feature vector based on the instantaneous temperature feature, and normalize the instantaneous temperature feature vector to obtain the normalized value of the instantaneous temperature feature vector.

[0114] S504. Obtain the bias reference based on the backpropagation algorithm;

[0115] S505. The normalized values ​​of the short-term fluctuation feature vector, the high-frequency resonance feature vector, and the instantaneous temperature feature vector, along with the bias reference, are input into the dynamic feature fusion model for fusion to obtain a comprehensive vector feature fusion value. The function formula of the dynamic feature fusion model is:

[0116] d(r)=σ[D(B)*a+G(X)*b+W(D)*(1-ab)+E];

[0117] Where d(r) represents the fusion value of the comprehensive vector features, σ represents the activation function, such as ReLU or Sigmoid, D(B) represents the normalized value of the short-term fluctuation feature vector, a represents the weight value of the normalized value of the short-term fluctuation feature vector, G(X) represents the normalized value of the high-frequency resonance feature vector, b represents the weight value of the normalized value of the high-frequency resonance feature vector, W(D) represents the normalized value of the instantaneous temperature feature vector, and E represents the bias reference.

[0118] The fused value of the comprehensive vector features is used as the feature fusion value.

[0119] As described in steps S501-S505 above, the present invention first obtains a short-term fluctuation feature vector based on the short-term fluctuation characteristics, and then normalizes the short-term fluctuation feature vector to obtain a normalized value of the short-term fluctuation feature vector. This transforms the short-term fluctuation characteristics into feature vectors, facilitating computer processing and analysis. The normalization process brings different feature vectors to the same dimension, eliminating analytical biases caused by differences in data magnitude, making subsequent fusion calculations more reasonable and accurate, and ensuring that the short-term fluctuation characteristics can reasonably reflect their impact on the operating status of the distribution box during the fusion process. Then, a high-frequency resonance feature vector is obtained based on the high-frequency resonance characteristics, and the high-frequency resonance feature vector is normalized to obtain a normalized value of the high-frequency resonance feature vector. Furthermore, the high-frequency resonance characteristics are represented by vectors and normalized, again for the purpose of facilitating data processing and unifying the dimensions. High-frequency resonance characteristics reflect specific abnormal conditions in the circuit. After normalization, they can compete fairly with other characteristics in the fusion calculation, accurately reflecting their role in judging the operating status of the distribution box. Secondly, an instantaneous temperature feature vector is obtained based on the instantaneous temperature characteristics, and this vector is normalized to obtain a normalized value. This transformation and normalization of the instantaneous temperature characteristics allows them to be better integrated into the overall analysis. Temperature is a key indicator of the distribution box's operating status. After normalization, its importance is reasonably reflected in the fusion model, accurately reflecting the impact of temperature changes on the distribution box's operating status. Next, a bias reference is obtained based on the backpropagation algorithm. Furthermore, the model parameters can be adjusted using the backpropagation algorithm, and the bias reference is used to compensate for the deviation between the model prediction and the actual situation. This helps optimize the dynamic feature fusion model, enabling it to better fit the actual situation when fusing features, thus improving the accuracy and reliability of feature fusion. Finally, the normalized values ​​of the short-term fluctuation feature vector, the high-frequency resonance feature vector, and the instantaneous temperature feature vector, along with the bias reference, are input into the dynamic feature fusion model for fusion to obtain a comprehensive vector feature fusion value. The normalized feature vectors and the bias reference are then input into the dynamic feature fusion model, and the comprehensive vector feature fusion value is calculated using a specific function. This fusion value integrates multiple key feature information, providing a more comprehensive and accurate reflection of the distribution box's operating status, offering strong support for subsequent status monitoring and fault diagnosis. Furthermore, a single feature cannot fully reflect the complex operating status of the distribution box; multi-feature fusion can integrate information from different aspects, and the dynamic feature fusion model can effectively combine this information, improving the performance of the monitoring system and its fault diagnosis capabilities.

[0120] In one embodiment, step S6 of monitoring the operating status of the distribution box based on the feature fusion value further includes:

[0121] S601. Obtain multiple feature fusion values ​​within a preset time window based on the feature fusion values;

[0122] S602. Obtain the feature fusion mean value based on the multiple feature fusion values;

[0123] S603. Obtain the standard feature fusion difference based on the mean of feature fusion and multiple feature fusion values, using the following formula:

[0124] ;

[0125] Where B(Z) represents the standard feature fusion difference, and R(T) i This represents the fusion value of the i-th feature, where n represents the number of pixel brightness values. Let represent the mean of the feature fusion, and i represent the index of the feature fusion value, where i = 1, 2, 3...n;

[0126] S604. Calculate the standard feature fusion value based on the feature fusion mean and the standard feature fusion difference, wherein the calculation formula is:

[0127] ;

[0128] Where B(R) represents the standard feature fusion value, B(Z) represents the mean of feature fusion, and B(Z) represents the difference of standard feature fusion.

[0129] S605. Determine whether the standard feature fusion value is less than a preset threshold;

[0130] If the standard feature fusion value is greater than or equal to the preset threshold, the operating status of the distribution box is determined to be normal.

[0131] If the standard feature fusion value is less than the preset threshold, the operating status of the distribution box is determined to be an intermittent abnormal state.

[0132] As described in steps S601-S605 above, the present invention first obtains multiple feature fusion values ​​within a preset time window based on the feature fusion value. In this way, collecting multiple feature fusion values ​​within the preset time window can reflect the dynamic changes in the operating status of the distribution box over a period of time, avoid errors in judgment due to feature fusion values ​​at a single moment, and increase the reliability and accuracy of the judgment. At the same time, the operating status of the distribution box may fluctuate over time, and the feature fusion value at a single moment cannot fully reflect its true situation. Obtaining multiple values ​​allows for observation of their changing trends from a time series perspective, providing a more accurate grasp of the distribution box's operating status. Then, a feature fusion mean is obtained based on these multiple feature fusion values. Calculating the feature fusion mean provides a representative value of the distribution box's operating status during this period, smoothing out the impact of some random fluctuations and highlighting the overall operating status trend, facilitating subsequent comparison and analysis. Simultaneously, since multiple feature fusion values ​​exhibit some fluctuation, the mean represents the data's central tendency, providing a stable reference benchmark for judging the distribution box's operating status, making the judgment more convincing. Finally, a standard feature fusion difference is obtained based on the feature fusion mean and multiple feature fusion values. This standard feature fusion difference reflects the degree of deviation of each feature fusion value from the mean, measuring the data's dispersion. A larger difference indicates greater fluctuation in the feature fusion value, potentially suggesting unstable operation of the distribution box and potential anomalies. Then, a standard feature fusion value is calculated based on the average feature fusion value and the standard feature fusion difference. This calculation, taking into account both the central tendency and dispersion of the data, forms a more comprehensive quantitative indicator reflecting the distribution box's operating status, providing a more accurate basis for subsequent judgments. Finally, it is determined whether the standard feature fusion value is less than a preset threshold. If the standard feature fusion value is greater than or equal to the preset threshold, the distribution box is considered to be in a normal operating state; if it is less than the preset threshold, it is considered to be in an intermittent abnormal operating state. By comparing the standard feature fusion value with the preset threshold, it is clearly determined whether the distribution box is in a normal or intermittent abnormal state. This allows for rapid understanding of the distribution box's operating status, timely response to anomalies, and the implementation of appropriate measures to ensure the stable operation of the power system.

[0133] This application also provides a distribution box operation status monitoring system based on multi-source data fusion, including:

[0134] The first acquisition module 1 is used to acquire multi-source data parameters of the distribution box within a preset time period, wherein the multi-source data parameters include current signals and thermal imaging video information;

[0135] The second acquisition module 2 is used to acquire the time resolution and frequency resolution of the current signal, and to acquire the length of the time window within a preset time according to the time resolution and the frequency resolution, and to acquire the short-time fluctuation characteristics of the current signal within the time window length.

[0136] The third acquisition module 3 is used to acquire the corresponding current waveform information based on the current signal, acquire transient time-domain features based on the current waveform information, and extract features from the transient time-domain features based on Fourier transform to obtain high-frequency resonance features.

[0137] The first splitting module 4 is used to split the thermal imaging video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtain multiple corresponding thermal diffusion values ​​according to the multiple video frame sequence lengths, generate a temperature gradient change map according to the multiple thermal diffusion values, and obtain instantaneous temperature characteristics according to the temperature gradient change map.

[0138] The first fusion module 5 is used to input the short-time fluctuation characteristics, high-frequency resonance characteristics and instantaneous temperature characteristics into the dynamic feature fusion model for fusion to obtain feature fusion values;

[0139] The first monitoring module 6 is used to monitor the operating status of the distribution box based on the feature fusion value.

[0140] In one embodiment, the second acquisition module includes:

[0141] The first acquisition unit is used to acquire multiple instantaneous spikes of the current signal within a preset time period, acquire multiple time intervals between the multiple instantaneous spikes according to a preset time sequence, filter the multiple time intervals according to the minimum value to obtain a first time interval, and use the first time interval as the time resolution.

[0142] The first correction unit is used to perform baseline drift correction on the current signal to obtain a standard current signal, and to perform transient event detection on the standard current signal based on the Teager energy operator to obtain a transient signal.

[0143] The first extraction unit is used to extract frequency components of a transient signal based on a sparse constrained spectrum estimation algorithm, obtain multiple transient frequency components, acquire multiple transient frequency component differences between the multiple transient frequency components according to a preset time series, filter the multiple transient frequency component differences according to the minimum value, obtain a first transient frequency component difference, and use the first transient frequency component difference as the frequency resolution.

[0144] The second acquisition unit is used to acquire the first sampling interval and the first number of sampling points under the time resolution, and to calculate the length of the first time window based on the first sampling interval and the first number of sampling points;

[0145] The third acquisition unit is used to acquire the second sampling interval and the second number of sampling points under the frequency resolution, and to calculate the second time window length based on the second sampling interval and the second number of sampling points;

[0146] The first merging unit is used to merge the first time window length and the second time window length according to the time sequence within a preset time period to obtain the merged time window length, and use the merged time window length as the time window length.

[0147] The fourth acquisition unit is used to acquire the current signal fluctuation diagram within the time window length, acquire the short-time fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-time fluctuation duration of the current as a short-time fluctuation feature.

[0148] This application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method.

[0149] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0150] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in this application and in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0151] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0152] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A power distribution box operation state monitoring method based on multi-source data fusion, characterized in that, include: Acquire multi-source data parameters of the distribution box within a preset time period, including current signals and thermal imaging video information; Multiple instantaneous spikes of the current signal within a preset time period are acquired, and multiple time intervals between the multiple instantaneous spikes are acquired according to a preset time sequence. The multiple time intervals are then filtered according to the minimum value to obtain a first time interval, and the first time interval is used as the time resolution. The current signal is baseline drift corrected to obtain a standard current signal, and transient event detection is performed on the standard current signal based on the Teager energy operator to obtain a transient signal; Frequency components are extracted from transient signals based on a sparse constrained spectrum estimation algorithm to obtain multiple transient frequency components. Multiple transient frequency component differences between the multiple transient frequency components are obtained according to a preset time series. The multiple transient frequency component differences are then filtered according to the minimum value to obtain the first transient frequency component difference. The first transient frequency component difference is used as the frequency resolution. Obtain the first sampling interval and the first number of sampling points at the stated time resolution, and calculate the length of the first time window based on the first sampling interval and the first number of sampling points; Obtain the second sampling interval and the second number of sampling points at the frequency resolution, and calculate the second time window length based on the second sampling interval and the second number of sampling points; The first time window length and the second time window length are merged according to the time sequence within a preset time period to obtain the merged time window length, and the merged time window length is used as the time window length. Obtain the current signal fluctuation diagram within the time window length, and obtain the short-time fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-time fluctuation duration of the current as the short-time fluctuation feature; The corresponding current waveform information is obtained based on the current signal, and the transient time-domain features are obtained based on the current waveform information. The transient time-domain features are then extracted based on the Fourier transform to obtain the high-frequency resonance features. The thermal imaging video information is split according to a preset video frame sequence length to obtain multiple video frame sequence lengths. Multiple thermal diffusion values ​​are obtained according to the multiple video frame sequence lengths, and a temperature gradient change map is generated according to the multiple thermal diffusion values. Instantaneous temperature characteristics are obtained according to the temperature gradient change map. The short-term fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics are input into a dynamic feature fusion model for fusion to obtain feature fusion values; The operating status of the distribution box is monitored based on the fusion value of the features.

2. The power distribution box operating condition monitoring method based on multi-source data fusion according to claim 1, characterized in that, The step of obtaining transient time-domain features based on the current waveform information, and extracting features from the transient time-domain features based on Fourier transform to obtain high-frequency resonance features includes: One-dimensional time-series signal for obtaining current waveform information based on a two-stream convolution model; The current variation curve is obtained based on the one-dimensional time-series signal; The current change curve is transformed into two-dimensional wavelet time-frequency image features based on wavelet transform, and the instantaneous change information of the signal is obtained based on the two-dimensional wavelet time-frequency image features. The transient time-domain features are obtained based on the instantaneous change information of the signal; Based on the Fourier transform, the transient time-domain features are extracted to obtain instantaneous frequency-domain feature information, and the instantaneous frequency within a preset time period is obtained based on the instantaneous frequency-domain feature information. The instantaneous frequency is divided based on a preset frequency division threshold to obtain high-frequency components; Based on the spectral analysis of the resonance characteristics in the high-frequency components, the high-frequency resonance characteristics are obtained.

3. The power distribution box operating condition monitoring method based on multi-source data fusion according to claim 1, characterized in that, The steps of obtaining multiple thermal diffusion values ​​corresponding to the lengths of multiple video frame sequences, generating a temperature gradient change map based on the multiple thermal diffusion values, and obtaining instantaneous temperature features based on the temperature gradient change map include: The pixel temperature value in each video frame is obtained according to the length of the multiple video frame sequences. The pixel temperature standard deviation of the pixel temperature change between two consecutive frames is calculated based on the multiple pixel temperature values, and the pixel temperature standard deviation is used as the heat diffusion value. The video frame sequence direction is obtained based on the video frame sequence length; The multiple thermal diffusion values ​​are arranged into a thermal diffusion value sequence according to the video frame sequence direction, and a temperature gradient change map is generated based on the thermal diffusion value sequence; Multiple instantaneous temperature values ​​are extracted from the temperature gradient change map, and instantaneous temperature values ​​are obtained based on the multiple instantaneous temperature values. The instantaneous features corresponding to the instantaneous temperature values ​​are used as instantaneous temperature features.

4. The power distribution box operating condition monitoring method based on multi-source data fusion according to claim 1, characterized in that, The step of inputting the short-time fluctuation characteristics, high-frequency resonance characteristics, and instantaneous temperature characteristics into a dynamic feature fusion model for fusion to obtain feature fusion values ​​includes: Based on the short-term fluctuation characteristics, a short-term fluctuation feature vector is obtained, and the short-term fluctuation feature vector is normalized to obtain a normalized value of the short-term fluctuation feature vector. Based on the high-frequency resonance characteristics, a high-frequency resonance feature vector is obtained, and the high-frequency resonance feature vector is normalized to obtain a normalized value of the high-frequency resonance feature vector. The instantaneous temperature feature vector is obtained based on the instantaneous temperature features, and the instantaneous temperature feature vector is normalized to obtain the normalized value of the instantaneous temperature feature vector. The normalized values ​​of the short-term fluctuation feature vector, the high-frequency resonance feature vector, and the instantaneous temperature feature vector are input into the dynamic feature fusion model for fusion to obtain a comprehensive vector feature fusion value, which is then used as the feature fusion value.

5. The power distribution box operating condition monitoring method based on multi-source data fusion according to claim 1, characterized in that, The step of monitoring the operating status of the distribution box based on the feature fusion value further includes: Multiple feature fusion values ​​within a preset time window are obtained based on the feature fusion values; The feature fusion mean is obtained based on the multiple feature fusion values; A standard feature fusion difference is obtained based on the feature fusion mean and multiple feature fusion values; The standard feature fusion value is calculated based on the feature fusion mean and the standard feature fusion difference. Determine whether the standard feature fusion value is less than a preset threshold; If the standard feature fusion value is greater than or equal to the preset threshold, the operating status of the distribution box is determined to be normal. If the standard feature fusion value is less than the preset threshold, the operating status of the distribution box is determined to be an intermittent abnormal state.

6. A power distribution box operation state monitoring system based on multi-source data fusion, characterized in that, include: The first acquisition module is used to acquire multi-source data parameters of the distribution box within a preset time period, wherein the multi-source data parameters include current signals and thermal imaging video information; The second acquisition module is used to acquire multiple instantaneous spikes of the current signal within a preset time period, acquire multiple time intervals between the multiple instantaneous spikes according to a preset time sequence, filter the multiple time intervals according to the minimum value to obtain a first time interval, and use the first time interval as the time resolution. The current signal is baseline drift corrected to obtain a standard current signal, and transient event detection is performed on the standard current signal based on the Teager energy operator to obtain a transient signal; Frequency components are extracted from transient signals based on a sparse constrained spectrum estimation algorithm to obtain multiple transient frequency components. Multiple transient frequency component differences between the multiple transient frequency components are obtained according to a preset time series. The multiple transient frequency component differences are then filtered according to the minimum value to obtain the first transient frequency component difference. The first transient frequency component difference is used as the frequency resolution. Obtain the first sampling interval and the first number of sampling points at the stated time resolution, and calculate the length of the first time window based on the first sampling interval and the first number of sampling points; Obtain the second sampling interval and the second number of sampling points at the frequency resolution, and calculate the second time window length based on the second sampling interval and the second number of sampling points; The first time window length and the second time window length are merged according to the time sequence within a preset time period to obtain the merged time window length, and the merged time window length is used as the time window length. Obtain the current signal fluctuation diagram within the time window length, and obtain the short-time fluctuation duration of the current based on the current signal fluctuation diagram, and use the short-time fluctuation duration of the current as the short-time fluctuation feature; The third acquisition module is used to acquire the corresponding current waveform information based on the current signal, and to acquire transient time-domain features based on the current waveform information. Based on Fourier transform, the transient time-domain features are extracted to obtain high-frequency resonance features. The first splitting module is used to split the thermal imaging video information according to a preset video frame sequence length to obtain multiple video frame sequence lengths, obtain multiple corresponding thermal diffusion values ​​according to the multiple video frame sequence lengths, generate a temperature gradient change map according to the multiple thermal diffusion values, and obtain instantaneous temperature characteristics according to the temperature gradient change map. The first fusion module is used to input the short-time fluctuation characteristics, high-frequency resonance characteristics and instantaneous temperature characteristics into the dynamic feature fusion model for fusion to obtain feature fusion values; The first monitoring module is used to monitor the operating status of the distribution box based on the feature fusion value.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images