Distribution box bus temperature rise early warning method and system

Abstract

The present application relates to the technical field of distribution box bus temperature rise early warning, and particularly relates to a distribution box bus temperature rise early warning method and system. Temperature and current monitoring data of the distribution box bus are collected, and sustained temperature drift characteristics are extracted from the temperature data after wavelet decomposition; cross-correlation operation and attenuation fitting are performed on the current and temperature data within multiple time scale sliding windows to obtain heat dissipation path distortion indexes; the temperature evolution rate and the heat dissipation path distortion indexes are converted into degradation risk representation quantities; the features of adjacent nodes are used to construct a node feature matrix, and node fusion features are obtained through graph attention multi-hop neighborhood aggregation; time series recursive calculation is performed based on the node fusion features and the current data to output a temperature rise prediction trajectory, and when the trajectory is monotonously rising and the distortion index exceeds a threshold value, an early warning signal is output, thereby realizing accurate early warning of temperature rise abnormalities.

Description

Technical Field

[0001] This invention relates to the field of early warning technology for busbar temperature rise in distribution boxes, and particularly to a method and system for early warning of busbar temperature rise in distribution boxes. Background Technology

[0002] Busbar temperature rise early warning is a common requirement in power equipment maintenance. The conventional approach involves installing temperature sensors on the busbar surface or at connection points to collect temperature data at fixed intervals and pre-setting alarm thresholds. When the measured temperature exceeds the threshold, the system issues an audible and visual alarm. To mitigate the impact of ambient temperature and load levels, more sophisticated systems also simultaneously collect current data, using heat balance formulas or empirical temperature rise curves to calculate the upper limit of the allowable temperature under the current load, and comparing the measured value with a dynamic benchmark. This method, based on threshold comparison and simplified thermal models, is easy to install and flexible in deployment, and is therefore widely used in medium and low voltage power distribution systems.

[0003] However, the aforementioned conventional methods have significant drawbacks. The deterioration process of the busbar heat dissipation path is often slow. Dust accumulation, metal oxidation, and loose connections all contribute to a gradual increase in thermal resistance, which is reflected in a long-term, continuous, small-amplitude temperature drift. Fixed thresholds and simple comparison mechanisms are insensitive to such signals, and alarms are only triggered when the temperature rise accumulates to a significant level exceeding the limit. As a result, by the time the fault is discovered, it has already entered the middle or late stages. Conventional methods also have extremely limited utilization of the dynamic correlation between current and temperature, as well as the thermal coupling relationship between adjacent nodes. They are usually based on static comparisons of a single cross-section or a short time window, processing each measuring point independently. Temperature lag caused by load changes can easily lead to false alarms, and the lack of node correlation information makes it difficult to detect local heat dissipation anomalies in a timely manner, thus increasing the risk of missed alarms. The timeliness and accuracy of conventional early warning methods are therefore significantly limited. Summary of the Invention

[0004] The present invention provides a method and system for early warning of temperature rise in distribution box busbars, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for early warning of temperature rise on a distribution box busbar, comprising: Temperature and current monitoring data of the distribution box busbar are collected, and wavelet decomposition is performed on the temperature monitoring data to extract continuous temperature drift features. Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is performed on the multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The temperature evolution rate is obtained by taking the time-domain derivative of the continuous temperature drift characteristic, and the temperature evolution rate is converted into a degradation risk characterization quantity by mapping the heat dissipation path distortion index. The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are used to construct a node feature matrix. The node feature matrix is ​​then aggregated into a multi-hop neighborhood using a graph attention propagation mechanism to obtain node fusion features. Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output.

[0006] Wavelet decomposition was performed on the temperature monitoring data to extract persistent temperature drift features, including: The temperature monitoring data is subjected to multi-level wavelet decomposition to obtain wavelet coefficients at multiple scale levels. The energy distribution ratio of the wavelet coefficients at each scale level is calculated, and the scale level where the energy distribution ratio changes abruptly is identified as the separation scale. Extract the wavelet coefficients corresponding to the separation scale, calculate the amplitude change rate of the wavelet coefficients in the time domain, filter the wavelet coefficients according to the distribution characteristics of the amplitude change rate, and reconstruct the filtered wavelet coefficients using wavelet reconstruction to obtain the initial drift signal; The initial drift signal is subjected to multi-segment linear fitting, and the slope change between adjacent fitting segments is calculated. When the signs of the slope change are consistent, the corresponding fitting segments are merged to obtain the continuous temperature drift characteristics.

[0007] Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is then performed on each of these multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between these multi-scale attenuation parameters is calculated to obtain heat dissipation path distortion indices, including: The fluctuation amplitude of the current monitoring data is statistically analyzed. A reference time scale is set according to the fluctuation amplitude. Multiple sliding windows with different time scales are set according to the multiple relationship of the reference time scale. In each sliding window, the current monitoring data and temperature monitoring data are cross-correlated from zero with time delay to obtain the multi-scale cross-correlation sequence corresponding to each time scale. The peak positions of each multi-scale cross-correlation sequence are extracted as the thermal response time delay. The cross-correlation coefficient segments that decrease with increasing time delay after the peak position are extracted as the decay segments. The decay curves are obtained by fitting each decay segment with an exponential function. The decay rate of each decay curve is extracted as the multi-scale decay parameter. The difference between the thermal response delays corresponding to adjacent time scales is calculated as the temporal gradient. When the temporal gradient changes from a positive value to a negative value, the corresponding time scale is marked as the heat dissipation degradation scale. The multi-scale decay parameters corresponding to the heat dissipation degradation scale are extracted as the degradation decay parameters. The magnitude difference of the temporal gradient changing from a positive value to a negative value is recorded as the temporal gradient reversal magnitude. The deviation between the degradation decay parameter and the mean of the multi-scale decay parameter across all time scales is calculated. The deviation is then weighted and fused with the temporal gradient reversal magnitude to calculate the difference between the multi-scale decay parameters, thus obtaining the heat dissipation path distortion index.

[0008] The temperature evolution rate is obtained by time-domain differentiation of the sustained temperature drift characteristics. The temperature evolution rate and the heat dissipation path distortion index are then converted into a degradation risk characterization quantity through a mapping relationship, including: The temperature evolution rate is obtained by time-domain differentiation of the continuous temperature drift characteristics. The causal relationship between the heat dissipation path distortion index and the temperature evolution rate is tested. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transformation signal is output. Using the moment of causal chain breakage as the dividing point, the sensitivity coefficients of temperature evolution rate to heat dissipation path distortion index are extracted for the time period before and after the dividing point. The ratio of the sensitivity coefficient after the dividing point to the sensitivity coefficient before the dividing point is calculated. When the ratio exceeds the self-excitation judgment threshold, it is judged as the self-excitation stage. The duration and intensity of the self-excitation stage are statistically analyzed as a measure of degradation acceleration. A dynamic coupling network of temperature evolution rate and heat dissipation path distortion index is constructed. The ratio of positive feedback loop strength to negative feedback loop strength of the dynamic coupling network is calculated. When the ratio exceeds a preset critical threshold, it is determined to be an unstable critical state. The evolution rate of the unstable critical state is extracted. A two-dimensional mapping space is established between the evolution rate and the degradation acceleration measure. The current evolution rate and degradation acceleration measure are located as feature coordinate points. The risk level is matched according to the region where the feature coordinate points are located as the degradation risk characterization quantity.

[0009] A causal relationship test is performed on the heat dissipation path distortion index and the temperature evolution rate. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transition signal is output, including: A dynamic Bayesian network containing heat dissipation path distortion index, temperature evolution rate and ambient temperature is constructed. The directed edge weight matrix of the dynamic Bayesian network is learned. The causal edge weight sequence pointing from the heat dissipation path distortion index to the temperature evolution rate is extracted. When the temporal gradient of the causal edge weight sequence is negative and the absolute value exceeds the structural mutation threshold, the start time of the causal topology change is recorded. After the initial moment, the in-degree and out-degree of the temperature evolution rate node are counted. When the difference between the in-degree and out-degree changes from a positive value to a negative value, the time period of role change is recorded. Extract the first-order difference sequence of temperature evolution rate during the role transition period. When the longest duration of consecutive positive values ​​in the first-order difference sequence exceeds the monotonicity determination window, output a continuously rising indicator. Extract the sum of out-degree weights of the heat dissipation path distortion index nodes during the role transition period. When the relative deviation of the sum of out-degree weights from the historical stable period benchmark value is less than the synchronization change threshold, output the asynchronous change flag. When both the continuously rising indicator and the asynchronous change indicator exist, the start time and the absolute value of the temporal gradient are encapsulated as a degradation mechanism transition signal.

[0010] The continuous temperature drift characteristics and degradation risk representations of adjacent bus nodes are used to construct a node feature matrix. This node feature matrix is ​​then aggregated using a graph attention propagation mechanism through multi-hop neighborhood aggregation to obtain the node fusion features, which include: The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are obtained and spliced ​​to form a node feature matrix; Construct the electrical connection adjacency matrix of the bus topology network, extract the neighbor node set of each bus node, calculate the difference in degradation risk representation between the current bus node and each neighbor node in the neighbor node set in the node feature matrix, generate graph attention weights based on the difference in degradation risk representation, and obtain the neighborhood aggregation feature by weighted summation of the feature vectors of the neighbor node set. The coupling degree between the continuous temperature drift feature component and the degradation risk characterization component in the neighborhood aggregation feature is calculated. When the coupling degree exceeds the preset coupling threshold, an inhibition factor is applied to the neighborhood aggregation feature to obtain the modified neighborhood aggregation feature. The modified neighborhood aggregation feature is concatenated with the feature vector of the current bus node in the node feature matrix to form an extended feature vector and the node feature matrix is ​​updated. The set of neighboring nodes of the neighboring node set is used as the next hop neighboring node set. Based on the updated node feature matrix, multi-hop neighborhood aggregation is iteratively performed until the preset number of hops is reached. During the iteration process, the difference in the risk representation of each jump is accumulated to form the risk propagation path feature. The risk propagation path feature is then fused with the extended feature vector after the iteration is completed to obtain the node fusion feature.

[0011] Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output, including: The node fusion features are mapped to the initial hidden state vector, the current current monitoring data is divided into current time series segments according to time steps, and the risk propagation path feature components in the node fusion features are extracted as risk modulation vectors. Based on the initial hidden state vector, risk modulation vector and current time sequence segments of each time step, a time-series recursive calculation is performed. The coupling degree between the risk modulation vector and the recursive state of each time step is calculated. According to the coupling degree, a risk gain factor is applied to the temperature rise prediction value of the corresponding time step to obtain the temperature rise prediction value of each time step and connect to output the temperature rise prediction trajectory. The temperature rise prediction trajectory is obtained by performing a first-order difference to obtain the temperature rise rate of change sequence. When the continuous length of positive values ​​in the temperature rise rate of change sequence exceeds the monotonicity judgment length, the temperature rise prediction trajectory is determined to be monotonically increasing. When the predicted temperature rise trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, the risk propagation path feature component is fused with the heat dissipation path distortion index to generate a warning signal and the warning signal is output.

[0012] A second aspect of the present invention provides a distribution box busbar temperature rise early warning system, comprising: The data acquisition unit is used to collect temperature monitoring data and current monitoring data of the distribution box bus, and to perform wavelet decomposition on the temperature monitoring data to extract continuous temperature drift features. The wavelet decomposition unit is used to perform cross-correlation operations on the current monitoring data and temperature monitoring data in sliding windows of multiple different time scales to obtain a multi-scale cross-correlation sequence. The multi-scale cross-correlation sequence is then subjected to attenuation fitting to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The mapping unit is used to perform time-domain differentiation on the continuous temperature drift characteristics to obtain the temperature evolution rate, and convert the temperature evolution rate and the heat dissipation path distortion index into a degradation risk characterization quantity through a mapping relationship. The distortion calculation unit is used to construct a node feature matrix by combining the continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes, and to perform multi-hop neighborhood aggregation on the node feature matrix through a graph attention propagation mechanism to obtain node fusion features. The mapping and transformation unit is used to perform time-series recursive calculations based on the node fusion features and the current current monitoring data, and output the temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, a warning signal is output.

[0013] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0015] It can effectively suppress transient fluctuations in temperature monitoring data, extract continuous temperature drift features through wavelet decomposition, accurately reflect the long-term heat accumulation trend of the busbar, and provide a stable and reliable feature basis for temperature rise early warning. Compared with simple threshold judgment, it has stronger robustness to short-term load fluctuations and sensor noise, and significantly reduces the false alarm rate caused by instantaneous temperature jumps.

[0016] It can simultaneously capture the correlation between current and temperature changes across multiple time scales, quantifying the dynamic distortion of the heat dissipation path through attenuation fitting. This distortion index can accurately identify heat dissipation performance degradation caused by loose connections, dust accumulation, insulation aging, etc., issuing signs of degradation before the temperature rise becomes significant, thus gaining valuable time for maintenance. By combining the mapping transformation between temperature evolution rate and heat dissipation path distortion index, the degree of degradation risk can be quantified, making early warning decisions more precise. The calculation of multi-scale difference effectively distinguishes between normal load fluctuations and abnormal heat dissipation, avoiding misjudging temperature rises caused by increased current as faults.

[0017] It can utilize graph attention propagation mechanism to perform multi-hop aggregation of temperature characteristics and degradation risks of adjacent bus nodes, fully explore the spatial correlation in the bus array, and enable the abnormal signs of a single node to be corroborated by adjacent nodes, significantly improving the reliability and coverage of early warning. Time-series recursive calculation combined with current current data dynamically extrapolates the temperature rise trajectory, accurately identifies the continuous deterioration trend, and provides timely warnings when the trajectory rises monotonically and the heat dissipation distortion exceeds the limit, effectively reducing the risk of missed detections. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the method for early warning of temperature rise on the busbar of a distribution box; Figure 2 Flowchart for the implementation of degradation mechanism transformation and risk assessment; Figure 3 This is a schematic diagram of a two-dimensional risk mapping space; Figure 4 This is a schematic diagram illustrating the sensitivity coefficient and the determination of the self-excitation stage. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0021] Figure 1 This is a flowchart illustrating the method for early warning of temperature rise in the distribution box busbar according to an embodiment of the present invention.

[0022] A method for early warning of temperature rise in distribution box busbars includes: Temperature and current monitoring data of the distribution box busbar are collected, and wavelet decomposition is performed on the temperature monitoring data to extract continuous temperature drift features. Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is performed on the multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The temperature evolution rate is obtained by taking the time-domain derivative of the continuous temperature drift characteristic, and the temperature evolution rate is converted into a degradation risk characterization quantity by mapping the heat dissipation path distortion index. The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are used to construct a node feature matrix. The node feature matrix is ​​then aggregated into a multi-hop neighborhood using a graph attention propagation mechanism to obtain node fusion features. Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output.

[0023] Wavelet decomposition was performed on the temperature monitoring data to extract persistent temperature drift features, including: The temperature monitoring data is subjected to multi-level wavelet decomposition to obtain wavelet coefficients at multiple scale levels. The energy distribution ratio of the wavelet coefficients at each scale level is calculated, and the scale level where the energy distribution ratio changes abruptly is identified as the separation scale. Extract the wavelet coefficients corresponding to the separation scale, calculate the amplitude change rate of the wavelet coefficients in the time domain, filter the wavelet coefficients according to the distribution characteristics of the amplitude change rate, and reconstruct the filtered wavelet coefficients using wavelet reconstruction to obtain the initial drift signal; The initial drift signal is subjected to multi-segment linear fitting, and the slope change between adjacent fitting segments is calculated. When the signs of the slope change are consistent, the corresponding fitting segments are merged to obtain the continuous temperature drift characteristics.

[0024] When performing multi-level wavelet decomposition on temperature monitoring data, wavelet basis functions with tight support and symmetry (such as the db or sym series wavelets) are selected to decompose the acquired bus temperature time series layer by layer into wavelet coefficients on different frequency sub-bands. The number of decomposition layers is determined based on the sampling rate and the target analysis frequency band, typically set to 5 to 8 layers to ensure that low-frequency drift components can be sufficiently separated to a deeper scale level. Each decomposition layer generates corresponding approximation coefficients and detail coefficients, where the approximation coefficients carry low-frequency trend information, and the detail coefficients reflect the local fluctuations of each frequency band.

[0025] After obtaining the wavelet coefficients at each scale level, the energy distribution ratio of the wavelet coefficients at each scale level is calculated. Let the first wavelet coefficient be... Layer wavelet coefficient sequence is The energy of this layer is The sum of the energies of each layer is denoted as Then the first The energy distribution ratio of the layer is .right Calculate the difference in proportions between adjacent levels along the hierarchical direction. ,when When the preset mutation judgment threshold is exceeded, the corresponding level This is identified as the separation scale. The physical significance of the separation scale lies in the fact that a significant jump in the energy distribution occurs at this level, indicating that the slowly varying drift components and the rapidly fluctuating components in the bus temperature signal have achieved effective frequency separation at this point. If multiple candidate abrupt change levels exist, the level with the largest abrupt change amplitude is selected as the final separation scale to ensure the stability of subsequent drift feature extraction.

[0026] After extracting the wavelet coefficients corresponding to the separation scale, it is necessary to further filter out the noise interference components mixed in to avoid misjudging short-term disturbances as continuous drift. To this end, the amplitude change rate of the wavelet coefficients at this layer in the time domain is calculated. Let the length of the coefficient sequence at the separation scale be... , No. The coefficients are The rate of change of amplitude between adjacent coefficients is defined as ,in To prevent extremely small positive numbers with a denominator of zero. Statistical analysis was performed on the overall distribution, and its mean was calculated. with standard deviation , will satisfy The coefficient positions are marked as abnormal fluctuation points, where The sensitivity coefficient is adjustable, typically ranging from 1.5 to 2.5. Coefficients at points of abnormal fluctuation are set to zero or replaced with linear interpolations of adjacent coefficients to complete coefficient filtering. The filtered coefficients retain the drift body with stable amplitude changes and eliminate spike-pulse interference. The filtered wavelet coefficients are then fed into the wavelet reconstruction process, where the time-domain signal is recovered through inverse wavelet transform to obtain the initial drift signal. During reconstruction, only coefficients from the separation scale and above (lower frequency) levels are used in the synthesis to ensure that the reconstructed signal focuses on the low-frequency drift trend without introducing high-frequency components.

[0027] Although the initial drift signal reflects the slow trend of temperature change, its local morphology may still exhibit subtle fluctuations, making it difficult to directly determine the direction and amplitude of the drift. Multi-segment linear fitting is performed on the initial drift signal, dividing the signal into several segments of equal or adaptive length along the time axis. Least-squares linear regression is then performed on each segment to obtain the fitting slope for each segment. Let the... The fitted slope of the segment is The change in slope between two adjacent segments is defined as .right The signs of the slope changes are evaluated segment by segment: if the signs of the slope changes in several consecutive segments are consistent (all positive or all negative), it indicates that the drift direction described by these fitted segments has not reversed and belongs to the same continuous drift process. These segments are then merged into a single fitted segment, and the combined slope is recalculated. The merging operation uses a weighted average method, with the time span of each sub-segment as the weight, to ensure that longer sub-segments contribute more to the combined slope.

[0028] When the sign of the slope change changes, it indicates a turning point in the drift direction. This point serves as the boundary for segmenting the fitted segment and is not merged. After the above merging and segmentation processes, the initial drift signal is represented as a set of line segments with a clear and continuous direction. Each line segment corresponds to a monotonic drift process, and its slope, start and end times, and amplitude range together constitute a quantitative description of the continuous temperature drift characteristics. For merged segments with slope absolute values ​​lower than the preset minimum drift slope threshold, they can be marked as temperature stable intervals and not included in subsequent drift feature analysis to reduce the interference of small fluctuations caused by measurement noise on the drift identification results.

[0029] The extracted results of the persistent temperature drift features are stored in a structured manner. Each drift record includes the start and end timestamps, the overall slope, the cumulative drift amplitude, and the corresponding separation scale level number. In subsequent processing, these features will be fed into the time-domain derivative stage to obtain the temperature evolution rate, and will participate in the calculation of the degradation risk characterization quantity together with the heat dissipation path distortion index. Since the persistent temperature drift features have undergone two processing steps—frequency separation and coefficient screening—based on wavelet decomposition, their anti-interference ability against noise and short-term disturbances is significantly better than the scheme of directly extracting trends from the original temperature sequence. This allows them to more accurately reflect the true evolution trend of the bus thermal state, providing a reliable basic feature input for subsequent temperature rise early warning analysis.

[0030] Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is then performed on each of these multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between these multi-scale attenuation parameters is calculated to obtain heat dissipation path distortion indices, including: The fluctuation amplitude of the current monitoring data is statistically analyzed. A reference time scale is set according to the fluctuation amplitude. Multiple sliding windows with different time scales are set according to the multiple relationship of the reference time scale. In each sliding window, the current monitoring data and temperature monitoring data are cross-correlated from zero with time delay to obtain the multi-scale cross-correlation sequence corresponding to each time scale. The peak positions of each multi-scale cross-correlation sequence are extracted as the thermal response time delay. The cross-correlation coefficient segments that decrease with increasing time delay after the peak position are extracted as the decay segments. The decay curves are obtained by fitting each decay segment with an exponential function. The decay rate of each decay curve is extracted as the multi-scale decay parameter. The difference between the thermal response delays corresponding to adjacent time scales is calculated as the temporal gradient. When the temporal gradient changes from a positive value to a negative value, the corresponding time scale is marked as the heat dissipation degradation scale. The multi-scale decay parameters corresponding to the heat dissipation degradation scale are extracted as the degradation decay parameters. The magnitude difference of the temporal gradient changing from a positive value to a negative value is recorded as the temporal gradient reversal magnitude. The deviation between the degradation decay parameter and the mean of the multi-scale decay parameter across all time scales is calculated. The deviation is then weighted and fused with the temporal gradient reversal magnitude to calculate the difference between the multi-scale decay parameters, thus obtaining the heat dissipation path distortion index.

[0031] After collecting current monitoring data from the distribution box busbar, the fluctuation amplitude is first statistically analyzed. Specifically, the peak-to-peak value or standard deviation of the current signal is calculated within a complete historical time window, and the statistically obtained fluctuation amplitude is used as the basis for setting the reference time scale. When the current fluctuation amplitude is large, it indicates drastic load changes, requiring a relatively short reference time scale to capture rapid thermal response processes. When the fluctuation amplitude is small, the reference time scale is appropriately extended to ensure sufficient statistical stability in the cross-correlation calculation. After the reference time scale is determined, it is expanded sequentially according to its integer multiples, such as 1, 2, 4, 8 times, etc., forming multiple sliding windows with different time scales. Each sliding window scrolls forward on the time axis with a fixed step size, covering the complete monitoring data segment, thereby extracting the dynamic correlation information between current and temperature at different time granularities.

[0032] Within each sliding window, cross-correlation calculations are performed on the current monitoring data and temperature monitoring data, incrementing from zero with a time delay. Increasing the time delay from zero means gradually shifting the temperature sequence relative to the current sequence. At each time delay value, the normalized cross-correlation coefficient is calculated, resulting in a cross-correlation curve with time delay on the horizontal axis and the cross-correlation coefficient on the vertical axis—a multi-scale cross-correlation sequence corresponding to that time scale. Due to the physical lag in the bus's thermal response, temperature changes caused by current changes typically require a period of heat conduction. Therefore, the cross-correlation sequence will exhibit a peak at a certain time delay, and the time delay position corresponding to this peak reflects the thermal response delay between current excitation and temperature response. Performing the above calculations on all time scales yields a set of multi-scale cross-correlation sequences, each corresponding to a specific time observation granularity.

[0033] The peak positions in each multi-scale cross-correlation sequence are extracted and recorded as the thermal response delay at the corresponding time scale. ,in The index is used for time scale numbering. After the peak position, the cross-correlation coefficients typically show a decreasing trend as time delay continues to increase; this decreasing segment reflects the decay characteristics of the thermal response. The segment of cross-correlation coefficients that continuously decreases after the peak position is extracted as the decay segment. An exponential function is then applied to this decay segment, with the fitting form being: ,in For delay The number of cross-relationships at each location To fit the amplitude coefficients, For the first The decay rate corresponding to each time scale is called the multi-scale decay parameter. The physical meaning lies in characterizing how quickly the thermal response decays from its peak to the background level on that timescale: The larger the value, the faster the decay, indicating that the heat dissipation path has a higher heat transfer efficiency at this scale; The smaller the value, the slower the decay, indicating that the heat dissipation capacity is suppressed.

[0034] The difference between the thermal response delays at adjacent time scales is calculated as the temporal gradient. Under normal heat dissipation conditions, the thermal response delay typically exhibits a monotonically increasing trend with increasing time scale, i.e. Maintaining a positive value reflects the hierarchical delay characteristics of the heat conduction path. At a certain time scale... When a value changes from positive to negative, it means that the thermal response delay is shortened at that scale. This anomaly indicates a structural change in the heat dissipation path around this time scale, possibly due to increased contact resistance, deterioration of the heat dissipation medium, or local blockage of the heat dissipation channel, leading to distortion of the heat conduction path. The time scale corresponding to the change from positive to negative value is marked as the heat dissipation degradation scale, denoted as . At the same time, record The magnitude difference between the positive and negative values, i.e. (in , ),Will Defined as the magnitude of temporal gradient reversal. The larger the value, the more drastic the abrupt change in thermal response delay at the heat dissipation degradation scale, and the more severe the distortion of the heat dissipation path.

[0035] Extracting heat dissipation degradation scale Corresponding multi-scale attenuation parameters As a degradation decay parameter, the mean of the multi-scale decay parameter is calculated across all time scales. ,in The total number of time scales. The deviation of the degradation parameter from the mean is defined as... Deviation The larger the value, the more significant the difference between the thermal response decay behavior at the heat dissipation degradation scale and the overall average level, and the more prominent the local anomaly of the heat dissipation path at that scale.

[0036] Deviation With time-series gradient inversion magnitude Weighted fusion is performed to obtain the heat dissipation path distortion index. The calculation method is as follows ,in and The fusion weights for deviation and temporal gradient reversal magnitude are respectively, satisfying... , As a reference normalized baseline value, it is used to map the temporal gradient reversal magnitude to a range of the same dimensions as the deviation, so as to facilitate the weighted fusion of the two. and The value can be calibrated based on the actual bus type and historical fault data. When the heat dissipation path distortion is mainly manifested in abnormal attenuation behavior, the value should be appropriately increased. When the distortion is mainly manifested in abrupt changes in thermal response time delay, appropriately increase... .

[0037] Heat dissipation path distortion index This comprehensively reflects two dimensions of heat dissipation anomaly information: first, the deviation at the decay parameter level, capturing the deviation of the thermal response decay rate from the overall pattern at the degradation scale; and second, the gradient reversal magnitude at the time delay level, reflecting the abrupt change in the heat conduction path during time scale transitions. The weighted fusion of these two aspects enables… It has a strong comprehensive characterization ability for heat dissipation path distortion, and can show significant numerical changes in the early stages such as poor busbar contact and heat dissipation aging, providing a reliable quantitative basis for subsequent temperature rise early warning. When When the preset warning threshold is exceeded, a warning signal is triggered based on the monotonous upward trend of the temperature rise prediction trajectory, prompting maintenance personnel to check and handle the heat dissipation status of the corresponding bus node.

[0038] like Figure 2 As shown, Figure 2 This is a flowchart illustrating the operational process of degradation mechanism transformation and risk assessment in this invention.

[0039] The temperature evolution rate is obtained by time-domain differentiation of the sustained temperature drift characteristics. The temperature evolution rate and the heat dissipation path distortion index are then converted into a degradation risk characterization quantity through a mapping relationship, including: The temperature evolution rate is obtained by time-domain differentiation of the continuous temperature drift characteristics. The causal relationship between the heat dissipation path distortion index and the temperature evolution rate is tested. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transformation signal is output. Using the moment of causal chain breakage as the dividing point, the sensitivity coefficients of temperature evolution rate to heat dissipation path distortion index are extracted for the time period before and after the dividing point. The ratio of the sensitivity coefficient after the dividing point to the sensitivity coefficient before the dividing point is calculated. When the ratio exceeds the self-excitation judgment threshold, it is judged as the self-excitation stage. The duration and intensity of the self-excitation stage are statistically analyzed as a measure of degradation acceleration. A dynamic coupling network of temperature evolution rate and heat dissipation path distortion index is constructed. The ratio of positive feedback loop strength to negative feedback loop strength of the dynamic coupling network is calculated. When the ratio exceeds a preset critical threshold, it is determined to be an unstable critical state. The evolution rate of the unstable critical state is extracted. A two-dimensional mapping space is established between the evolution rate and the degradation acceleration measure. The current evolution rate and degradation acceleration measure are located as feature coordinate points. The risk level is matched according to the region where the feature coordinate points are located as the degradation risk characterization quantity.

[0040] When performing time-domain differentiation on the continuous temperature drift characteristics, a discrete-time series with equally spaced sampling is used as the basis. The drift characteristic values ​​of adjacent sampling points are differentially processed, and the difference value is divided by the corresponding sampling time interval to obtain the temperature evolution rate at each moment. This rate series reflects how fast the bus temperature changes over time. When the bus contact resistance increases or the heat dissipation channel is blocked, the temperature evolution rate often shows a continuous positive value and a gradual increasing trend. After obtaining the temperature evolution rate series, a causal relationship test is performed on the heat dissipation path distortion index and the temperature evolution rate. Specifically, a nonlinear causal test method based on hysteresis mutual information is adopted: the heat dissipation path distortion index series and the temperature evolution rate series are used as conditional variables, and the conditional mutual information of the two under different hysteresis step sizes is calculated. If the mutual information is significantly higher than the random permutation benchmark within a certain hysteresis range, the two are considered to have a causal relationship. If the temperature evolution rate continues to rise while the mutual information of the heat dissipation path distortion index in the corresponding hysteresis window drops below the significance threshold, the causal chain between the two is determined to be broken, the break time is recorded, and a degradation mechanism transition signal is output to the subsequent processing stage. The causal chain break means that the continuous rise in temperature is no longer driven by measurable distortion of the heat dissipation path, but may be dominated by deep degradation mechanisms such as insulation aging and micro-melting of the contact surface, requiring a change in the analysis framework.

[0041] Using the moment of causal chain breakage as the dividing point, the temperature evolution rate sequence and heat dissipation path distortion index sequence are extracted for the time periods before and after the breakage. Linear regression models are established for both time periods, with temperature evolution rate as the independent variable and heat dissipation path distortion index as the dependent variable. The regression slope for each time period is obtained and defined as the sensitivity coefficient for that time period. Let the sensitivity coefficient for the time period before the breakage be... The sensitivity coefficient for the time period after fracture is Calculate the ratio of the two. .when When the preset self-excitation threshold is exceeded, it indicates that the driving force of the temperature evolution rate after fracture on the heat dissipation path distortion index is significantly enhanced, meaning the system enters the self-excitation stage—the temperature rise no longer requires significant deterioration of the external heat dissipation path to self-reinforce. After determining that the system has entered the self-excitation stage, the duration of this stage is recorded. The integral cumulative amount of the temperature evolution rate within the stage ,Will and The combination of these is defined as a degradation acceleration metric, where This reflects the time span of the self-excitation process. Both reflect the cumulative intensity of the self-excitation process and together describe the degree of accelerated degradation.

[0042] When constructing the dynamic coupling network between the temperature evolution rate and the heat dissipation path distortion index, the interaction between the two variables within the sliding time window is modeled as a directed weighted graph. The graph contains two nodes, corresponding to the temperature evolution rate and the heat dissipation path distortion index, respectively. The weights of the edges are determined by the Granger causality test statistics in each direction: the edge weight from the heat dissipation path distortion index to the temperature evolution rate represents the driving force of heat dissipation degradation on temperature rise, denoted as […]. (Positive feedback loop weights); The edge weights in the direction of temperature evolution rate to heat dissipation path distortion index characterize the reverse driving strength of temperature rise on further deterioration of the heat dissipation path, denoted as... (Negative feedback loop weight). Under normal operating conditions, Typically small values ​​indicate that the temperature increase is insufficient to significantly accelerate the distortion of the heat dissipation path; when the busbar enters a state of deep degradation, This will increase significantly, forming a positive feedback closed loop. Calculate the ratio of the positive feedback loop strength to the negative feedback loop strength. ,when When a preset critical threshold is exceeded, the system is determined to have entered a critical instability state. At this point, the mutual driving relationship between the two variables has lost its balance, and the temperature rise will evolve in an accelerated manner. After the critical instability state is determined, the slope of the change of each edge weight in the coupled network over time is extracted, and the weighted average of the slopes of the edge weight changes in the two directions is taken as the evolution rate of the critical instability state. This indicator reflects how quickly the system migrates towards complete instability.

[0043] Establish based on evolution rate A two-dimensional mapping space is defined with the horizontal axis and the degradation acceleration metric as the vertical axis. This space is then divided into several risk regions. Specifically, using... The low, medium, and high intervals intersect with the low, medium, and high intervals of the degradation acceleration metric, forming nine sub-regions. Each sub-region is assigned a corresponding risk level label based on the statistical distribution of historical fault samples, with risk levels ranging from low to high as Normal, Attention, Warning, and Alarm. The current evolution rate... With degradation acceleration measure , The combined values ​​constitute a feature coordinate point. This coordinate point is located in a two-dimensional mapping space, and the risk level corresponding to its sub-region is read as the final output degradation risk characterization quantity. This degradation risk characterization quantity integrates the acceleration of temperature evolution and the evolution trend of system coupling instability. It can identify high-risk degradation states in advance before the bus temperature rise reaches the hardware protection threshold, providing a reliable risk quantification basis for subsequent temperature rise prediction and early warning signal output. In practical applications, the regional boundaries of the two-dimensional mapping space can be adaptively adjusted according to historical data of different voltage levels and different bus materials to improve the accuracy and applicability of risk level matching.

[0044] like Figure 3 As shown, Figure 3 This is a schematic diagram of a two-dimensional risk mapping space. The horizontal axis represents the critical evolution rate of instability, reflecting how quickly the system migrates towards complete instability; the vertical axis represents the degradation acceleration measure, characterized by the duration and intensity of the self-excitation phase. The space is divided into four regions: normal, watchful, early warning, and alarm, with boundaries determined based on historical fault statistics. The current feature coordinates fall into the "early warning" region, indicating a risk level of 2, suggesting that the temperature rise has accelerated and distortion indicators have exceeded limits, and a check is recommended within 30 minutes. This mapping transforms the complex output of the dynamically coupled network into an intuitive level, supporting rapid location of high-risk nodes and prediction of risk evolution direction, thus improving the timeliness of early warning decisions.

[0045] A causal relationship test is performed on the heat dissipation path distortion index and the temperature evolution rate. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transition signal is output, including: A dynamic Bayesian network containing heat dissipation path distortion index, temperature evolution rate and ambient temperature is constructed. The directed edge weight matrix of the dynamic Bayesian network is learned. The causal edge weight sequence pointing from the heat dissipation path distortion index to the temperature evolution rate is extracted. When the temporal gradient of the causal edge weight sequence is negative and the absolute value exceeds the structural mutation threshold, the start time of the causal topology change is recorded. After the initial moment, the in-degree and out-degree of the temperature evolution rate node are counted. When the difference between the in-degree and out-degree changes from a positive value to a negative value, the time period of role change is recorded. Extract the first-order difference sequence of temperature evolution rate during the role transition period. When the longest duration of consecutive positive values ​​in the first-order difference sequence exceeds the monotonicity determination window, output a continuously rising indicator. Extract the sum of out-degree weights of the heat dissipation path distortion index nodes during the role transition period. When the relative deviation of the sum of out-degree weights from the historical stable period benchmark value is less than the synchronization change threshold, output the asynchronous change flag. When both the continuously rising indicator and the asynchronous change indicator exist, the start time and the absolute value of the temporal gradient are encapsulated as a degradation mechanism transition signal.

[0046] After obtaining time-series data on heat dissipation path distortion indices and temperature evolution rates, it is necessary to quantitatively test the causal relationship between the two to determine whether temperature degradation has shifted from being dominated by heat dissipation deterioration to being dominated by other mechanisms. To this end, a dynamic Bayesian network is constructed, comprising nodes of three types: heat dissipation path distortion indices, temperature evolution rates, and ambient temperature. The ambient temperature node is introduced as an exogenous variable to eliminate the interference of environmental disturbances on causal inference and avoid misjudging temperature rise caused by changes in ambient temperature as a result of heat dissipation path deterioration. The dynamic Bayesian network maintains the conditional probability distribution of node states at each time step and captures the temporal causal dependencies between variables through the transition structure between adjacent time steps.

[0047] During the learning phase of the dynamic Bayesian network, the expectation-maximization algorithm is used to iteratively estimate the directed edge weight matrix. Each element in the directed edge weight matrix corresponds to the conditional influence strength of a directed edge, reflecting the causal driving ability of the source node on the target node. The focus is on extracting the directed edge weights that point to the temperature evolution rate as an indicator of heat dissipation path distortion, arranging them on the time axis to form a causal edge weight sequence, denoted as... ,in This serves as the time step index. The causal edge weight sequence reflects the real-time causal contribution of the heat dissipation path distortion index to the temperature evolution rate. During the normal degradation phase, the continuous deterioration of the heat dissipation path drives the temperature evolution rate to increase; therefore, the causal edge weight sequence should be maintained at a high level and remain relatively stable.

[0048] For causal edge weight sequence The temporal gradient is calculated and defined as the weight difference between adjacent time steps. .when The absolute value is negative. Exceeding the preset structural mutation threshold At this point, the causal driving force of the heat dissipation path distortion index on the temperature evolution rate was structurally weakened, and this moment was recorded as the starting point of the causal topological structure change. Structural mutation threshold The gradient of the causal edge weight sequence is determined based on the statistical distribution of the gradient during the historical stable operation period. Usually, the mean of the absolute values ​​of the historical gradients is taken plus several times the standard deviation to ensure that the recording is triggered only when the weights drop significantly.

[0049] At the start time Next, the in-degree and out-degree of the temperature evolution rate nodes in the dynamic Bayesian network are statistically analyzed. The in-degree is defined as the sum of the weights of all directed edges pointing to the temperature evolution rate node within the current time step, and the out-degree is defined as the sum of the weights of all directed edges originating from the temperature evolution rate node. In the normal degradation state, the temperature evolution rate primarily acts as the driven node, with its in-degree greater than its out-degree. The difference between the in-degree and out-degree is... It is a positive value. When the causal structure changes, and the temperature evolution rate node gradually changes from a passive response role to an active driving role, its out-degree begins to increase, while its in-degree relatively decreases. The value then changes from positive to negative. Record. The period during which the positive and negative transformations are completed is the role transition period. This period marks a shift in the dominant mechanism of warming degradation.

[0050] During the period of role transition Within, the first-order difference sequence of temperature evolution rate is extracted. ,in , Indicates the first Temperature evolution rate at each time step. Statistical analysis of the longest duration of consecutive positive values ​​in a first-order difference sequence. That is, continuously satisfying The maximum number of consecutive time steps. When Exceeding the monotonicity determination window When the rate of temperature evolution shows a continuous upward trend during the role transition period, a continuously rising indicator is output. Monotonicity determination window The value of needs to be set in combination with the sampling frequency and the actual thermal response time constant. Too short a window will lead to misjudgment under noise interference, while too long a window will delay the recognition of continuous rise. Usually, a certain proportion of the number of sampling points corresponding to the thermal response time constant is taken as a reference benchmark.

[0051] Meanwhile, during the period of role transition The sum of the out-degree weights of the nodes extracting the heat dissipation path distortion index is denoted as... This is the sum of the weights of all directed edges originating from the node representing the distortion index of the heat dissipation path. (Calculation) Compared with historical stable period benchmark value The relative deviation, defined as .when Less than the synchronous change threshold If the heat dissipation path distortion index does not change synchronously during the role transition period, an asynchronous change flag is output. Historical stable period baseline value. The moving average of the total weighted values ​​of the heat dissipation path distortion index during normal equipment operation is taken to reflect the stable driving level of the heat dissipation path on the temperature evolution rate under normal conditions.

[0052] When both the continuously rising indicator and the asynchronous change indicator exist simultaneously, it is determined that the continuous increase in the temperature evolution rate has deviated from the causal drive of the heat dissipation path distortion index, and the causal chain is in its initial state. A fracture occurred at this point, and the degradation mechanism has shifted from being dominated by heat dissipation degradation to being dominated by other mechanisms. At this point, the starting time will be... The absolute value of the temporal gradient at the corresponding time point The degradation mechanism transition signal is encapsulated and output. (Starting moment) It provides temporal location information for the transformation of degradation mechanisms, facilitating subsequent source tracing analysis; absolute value of temporal gradient. It reflects the severity of the abrupt change in the causal structure. The larger the value, the more rapidly the causal driving force weakens, the more sudden the change in the degradation mechanism, and the higher the urgency required for early warning response.

[0053] The output of degradation mechanism transition signals provides crucial mechanism-level information for subsequent early warning decisions. Before the causal chain breaks, temperature rise early warning primarily relies on exceeding the threshold of heat dissipation path distortion indicators. After the causal chain breaks, temperature rise early warning needs to consider the continuous upward trend of the temperature evolution rate itself, because at this point, the temperature rise is partially driven by internal positive feedback mechanisms, and relying solely on heat dissipation path distortion indicators may underestimate the actual risk. By jointly analyzing degradation mechanism transition signals and temperature rise prediction trajectories, it is possible to more accurately distinguish between gradual temperature rises caused by heat dissipation degradation and accelerated temperature rises driven by internal positive feedback. This allows for differentiated early warning strategies at different degradation stages, improving the accuracy and timeliness of temperature rise early warnings for distribution box busbars.

[0054] like Figure 4 As shown, Figure 4 This diagram illustrates the sensitivity coefficient and self-excitation stage determination of the present invention. It shows the change in the sensitivity coefficient of the temperature evolution rate to the heat dissipation path distortion index before and after the causal chain breaks. Before the break (approximately 14:50-15:04), the sensitivity coefficient is stable at 0.82, indicating a stable causal relationship between the temperature rise rate and heat dissipation distortion. After the break, the coefficient jumps to 1.57, exceeding the self-excitation threshold, indicating that the system has entered the self-excitation stage—the temperature rise no longer depends on external heat dissipation degradation but is self-reinforcing through internal positive feedback. This stage lasts for 12 minutes and has an intensity of 0.43, constituting a measure of accelerated degradation. This determination mechanism effectively distinguishes between gradual degradation and accelerated loss of control, and is the core logic of the early warning system in different degradation stages.

[0055] The continuous temperature drift characteristics and degradation risk representations of adjacent bus nodes are used to construct a node feature matrix. This node feature matrix is ​​then aggregated using a graph attention propagation mechanism through multi-hop neighborhood aggregation to obtain the node fusion features, which include: The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are obtained and spliced ​​to form a node feature matrix; Construct the electrical connection adjacency matrix of the bus topology network, extract the neighbor node set of each bus node, calculate the difference in degradation risk representation between the current bus node and each neighbor node in the neighbor node set in the node feature matrix, generate graph attention weights based on the difference in degradation risk representation, and obtain the neighborhood aggregation feature by weighted summation of the feature vectors of the neighbor node set. The coupling degree between the continuous temperature drift feature component and the degradation risk characterization component in the neighborhood aggregation feature is calculated. When the coupling degree exceeds the preset coupling threshold, an inhibition factor is applied to the neighborhood aggregation feature to obtain the modified neighborhood aggregation feature. The modified neighborhood aggregation feature is concatenated with the feature vector of the current bus node in the node feature matrix to form an extended feature vector and the node feature matrix is ​​updated. The set of neighboring nodes of the neighboring node set is used as the next hop neighboring node set. Based on the updated node feature matrix, multi-hop neighborhood aggregation is iteratively performed until the preset number of hops is reached. During the iteration process, the difference in the risk representation of each jump is accumulated to form the risk propagation path feature. The risk propagation path feature is then fused with the extended feature vector after the iteration is completed to obtain the node fusion feature.

[0056] After obtaining the continuous temperature drift characteristics and degradation risk characterization quantities of adjacent busbar nodes in the distribution box, the two are concatenated in the column direction to form a node feature matrix. For the first in the network Each bus node has its eigenvector. Characteristic components of continuous temperature drift Components of degradation risk characterization Together constitute, that is Node feature matrix Each row in the matrix corresponds to the eigenvector of a bus node, and the number of rows in the matrix equals the total number of bus nodes. This stitching method preserves the independence of information on both temperature drift and degradation risk, while providing complete multi-dimensional feature input for subsequent attention mechanisms.

[0057] Construct the electrical connection adjacency matrix of the bus topology network based on the actual electrical connection relationships of the distribution boxes. Elements in the matrix A value of 1 indicates a node With nodes An electrical connection exists; a value of 0 indicates no connection. For the current bus node... Extract its value in the adjacency matrix The non-zero column index of the corresponding row is used to obtain the set of neighboring nodes. Calculate the current node set of neighboring nodes Each neighbor node The difference in degradation risk characterization This difference reflects the relative degree of degradation between adjacent nodes. The larger the difference in degradation risk representation, the higher the degree of degradation of the current node relative to its neighbors, and the greater the corresponding attention weight should be, so as to absorb more information from high-risk nodes during the aggregation process.

[0058] Based on the difference in degradation risk characterization measures When generating graph attention weights, a normalized exponential function is used to transform the difference. For the neighbor node set... Each neighbor node in Attention weight Calculate using the following method: The raw attention score is obtained by inputting the activation function, and then... Normalize the original scores of all neighboring nodes, so that Normalization ensures that the sum of the weights of each neighbor node's contribution during the aggregation process is 1, avoiding inconsistencies in feature magnitudes due to different numbers of neighbors. For the neighbor node set... The feature vectors of all neighboring nodes are summed by the attention weights to obtain the neighborhood aggregated feature. This feature integrates temperature drift and degradation risk information of each node in the neighborhood, and the neighboring nodes corresponding to high risk differences contribute more.

[0059] Calculate neighborhood aggregation features Mid-term sustained temperature drift characteristic components Components of degradation risk characterization Coupling between The degree of coupling reflects the linear correlation between two types of features after aggregation, and can be measured by calculating the Pearson correlation coefficient or the normalized inner product of the two components. Exceeding the preset coupling threshold This indicates that the temperature drift information and degradation risk information in the neighborhood aggregation features are over-coupled, which may lead to feature redundancy and affect the accuracy of subsequent predictions. In this case, an inhibition factor should be applied to the neighborhood aggregation features. The range of values ​​for the inhibition factor is: Its size is positively correlated with the degree to which the coupling exceeds the threshold, that is... ,in The suppression intensity coefficient is used. The corrected neighborhood aggregation feature is... By suppressing over-coupling components, it ensures that the two types of features retain their independent information contributions during the fusion process.

[0060] Correct neighborhood aggregation features With node feature matrix Current bus node The original feature vector The features are concatenated to form an extended feature vector. The extended feature vector contains both the local features of the current node and aggregated information from its neighborhood, thus representing the relative degradation state of the node in the network topology. The extended feature vectors of all nodes are then recombine to update the node feature matrix. This expands the feature dimension of each node in each row of the matrix from the original 2D to 4D. In the next hop iteration, the current set of neighboring nodes is... The set of neighboring nodes is used as the set of next-hop neighboring nodes. Based on the updated node feature matrix, the above neighborhood aggregation, coupling suppression, and feature concatenation operations are repeated until the preset number of hops is reached. After each iteration, the dimension of the node feature matrix expands, enabling the final expanded feature vector to perceive topological degradation propagation information within the multi-hop range.

[0061] During the multi-hop iteration process, the sequence of differences in degradation risk representation between the current node and its neighboring nodes in each hop is accumulated to form the risk propagation path characteristics. . Specifically, no. The current node in the jump iteration Its first skip neighbor node set Middle node The difference is denoted as By concatenating the differences of all hops in hop order, the feature vector of the risk propagation path is obtained. This feature vector describes the gradient distribution of degradation risk spreading outward from the current node along the electrical connection path, reflecting the propagation direction and intensity of degradation risk in the bus network. If the difference in the risk propagation path feature of a certain hop is significantly greater than that of other hops, it indicates that the degradation risk exhibits a significant concentrated propagation phenomenon at the topological level corresponding to that hop, which needs to be given special attention in early warning analysis.

[0062] Characteristics of risk transmission paths Complete all Extended feature vector after skip iteration By performing fusion, node fusion features are obtained. The fusion method employs linear transformation followed by splicing, i.e. ,in For learnable fusion weight matrix, These are bias vectors, obtained through supervised training using historical sample data. Node fusion features. By integrating the local temperature drift and degradation risk status of the current node, the aggregation information of multi-hop neighborhoods, and the propagation path characteristics of degradation risk in the topological network, a sufficient spatial-state joint representation is provided for the subsequent output of the temperature rise prediction trajectory based on time-series recursive calculation. Through this multi-hop graph attention propagation mechanism, the spatial diffusion pattern of degradation risk in the bus network can be effectively captured, improving the accuracy and lead time of temperature rise early warning.

[0063] Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output, including: The node fusion features are mapped to the initial hidden state vector, the current current monitoring data is divided into current time series segments according to time steps, and the risk propagation path feature components in the node fusion features are extracted as risk modulation vectors. Based on the initial hidden state vector, risk modulation vector and current time sequence segments of each time step, a time-series recursive calculation is performed. The coupling degree between the risk modulation vector and the recursive state of each time step is calculated. According to the coupling degree, a risk gain factor is applied to the temperature rise prediction value of the corresponding time step to obtain the temperature rise prediction value of each time step and connect to output the temperature rise prediction trajectory. The temperature rise prediction trajectory is obtained by performing a first-order difference to obtain the temperature rise rate of change sequence. When the continuous length of positive values ​​in the temperature rise rate of change sequence exceeds the monotonicity judgment length, the temperature rise prediction trajectory is determined to be monotonically increasing. When the predicted temperature rise trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, the risk propagation path feature component is fused with the heat dissipation path distortion index to generate a warning signal and the warning signal is output.

[0064] After linear projection transformation, the node fusion features are mapped to an initial hidden state vector. This vector carries the spatial topological risk distribution information and multi-hop neighborhood aggregation results of each bus node, serving as the initial state input for time-series recursive computation. Simultaneously, risk propagation path feature components are extracted from the node fusion features and used as a risk modulation vector. This vector reflects the risk propagation direction and intensity of each node in the electrical network at the current moment. The current monitoring data is sliced ​​according to a fixed time step, forming a current time-series segment sequence. Each time step corresponds to a current amplitude or current feature vector, used to drive subsequent recursive state updates.

[0065] In the time-series recursive computation phase, the initial hidden state vector is used as the initial hidden state of the recursive network. The risk modulation vector and the current time-series segments of each time step are used as input, and the recursive computation is advanced step by step. At each time step, the recursive unit receives the current time-series segment of the current time step and the hidden state of the previous time step, updates the current hidden state, and outputs the intermediate temperature rise prediction value of the current time step. The risk modulation vector participates in the calculation at each time step, and its role is to continuously inject the path information of heat dissipation degradation into the recursive process, so that the prediction result can perceive the distortion degree of the current heat dissipation channel.

[0066] When calculating the coupling degree between the risk modulation vector and the recursive state at each time step, the risk modulation vector is... With the Recursive hidden state of time step Perform inner product normalization to obtain the coupling degree at that time step. The calculation method is as follows ,in To prevent extremely small positive numbers with a denominator of zero, coupling degree is used. This reflects the current distortion state of the heat dissipation path on the first The degree of influence of the recursive state at the time step; the higher the coupling, the more significant the driving force of heat dissipation degradation on the temperature rise at that time step.

[0067] A risk gain factor is applied to the predicted temperature rise at the corresponding time step based on the coupling degree. Let the first... The base temperature rise prediction value for the time step is Risk gain factor Due to coupling degree After nonlinear activation, it is obtained as follows: ,in It is the sigmoid activation function. This is the gain adjustment coefficient. The final... Temperature rise prediction at time step It is obtained by multiplying the base temperature rise forecast by the risk gain factor, i.e. Predicted temperature rise values ​​for all time steps By connecting the data in chronological order, a complete temperature rise prediction trajectory is formed. This mechanism ensures that the predicted temperature rise value can be appropriately amplified at time steps where the heat dissipation path distortion is high and the recursive state and risk vector are highly coupled, thus more realistically reflecting the actual temperature rise trend of the bus under heat dissipation degradation conditions.

[0068] The temperature rise prediction trajectory is processed using the first-order difference method to obtain the temperature rise rate sequence. Let the temperature rise prediction trajectory be at the [missing information - likely a specific value or sequence]. The predicted value of the time step Then the first Rate of change of temperature at time step Defined as The temperature rise rate of change sequence is scanned stepwise, and the duration of consecutive positive values ​​in the sequence is counted, i.e., the duration of consecutive time steps. The longest continuous segment length .when Exceeding the preset monotonicity determination length At that time, the predicted temperature rise trajectory was determined to exhibit a monotonic upward trend. Monotonicity determination length. The calibration is performed based on the actual bus thermal response time constant, typically taking a value corresponding to a time window range of several minutes to more than ten minutes, in order to avoid false monotonic judgments caused by short-term current fluctuations.

[0069] When the temperature rise prediction trajectory is determined to be monotonically increasing, and the heat dissipation path distortion index is... Exceeding the preset warning threshold At that time, the early warning signal generation process begins. The risk propagation path characteristic components are then... With heat dissipation path distortion index The signals are merged to generate early warning signals. Specifically, After linear transformation and scalar The weighted summaries are used to obtain the early warning feature vector, which is then mapped to an early warning signal through the output layer. The early warning signal contains risk level coding and node location information. The risk level is determined by the amplitude of the fused early warning feature vector, and the node location information is determined by the contribution weight of each node in the risk propagation path feature component, thereby achieving accurate location of the bus node where an abnormal temperature rise occurs.

[0070] After the warning signal is issued, the risk level can be dynamically adjusted by combining the slope estimate of the predicted temperature rise trajectory. If the mean of the temperature rise rate sequence continues to increase, it indicates a significant accelerating trend in temperature rise, and the risk level of the warning signal should be adjusted upward accordingly. Warning Threshold It can be adaptively calibrated based on the rated current carrying capacity of the busbar, the temperature resistance rating of the insulation material, and historical fault data, maintaining a reasonable balance between early warning sensitivity and false alarm rate under different operating conditions. The entire temperature rise prediction and early warning output process forms a complete closed loop from feature extraction, time-series recursion, monotonicity judgment to early warning generation, ensuring that the distribution box busbar can be detected and alarmed in the early stage of heat dissipation degradation.

[0071] A second aspect of the present invention provides a distribution box busbar temperature rise early warning system, comprising: The data acquisition unit is used to collect temperature monitoring data and current monitoring data of the distribution box bus, and to perform wavelet decomposition on the temperature monitoring data to extract continuous temperature drift features. The wavelet decomposition unit is used to perform cross-correlation operations on the current monitoring data and temperature monitoring data in sliding windows of multiple different time scales to obtain a multi-scale cross-correlation sequence. The multi-scale cross-correlation sequence is then subjected to attenuation fitting to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The mapping unit is used to perform time-domain differentiation on the continuous temperature drift characteristics to obtain the temperature evolution rate, and convert the temperature evolution rate and the heat dissipation path distortion index into a degradation risk characterization quantity through a mapping relationship. The distortion calculation unit is used to construct a node feature matrix by combining the continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes, and to perform multi-hop neighborhood aggregation on the node feature matrix through a graph attention propagation mechanism to obtain node fusion features. The mapping and transformation unit is used to perform time-series recursive calculations based on the node fusion features and the current current monitoring data, and output the temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, a warning signal is output.

[0072] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0073] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0074] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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 or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of temperature rise on the busbar of a distribution box, characterized in that, include: Temperature and current monitoring data of the distribution box busbar are collected, and wavelet decomposition is performed on the temperature monitoring data to extract continuous temperature drift features. Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is performed on the multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The temperature evolution rate is obtained by taking the time-domain derivative of the continuous temperature drift characteristic, and the temperature evolution rate is converted into a degradation risk characterization quantity by mapping the heat dissipation path distortion index. The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are used to construct a node feature matrix. The node feature matrix is ​​then aggregated into a multi-hop neighborhood using a graph attention propagation mechanism to obtain node fusion features. Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output.

2. The method according to claim 1, characterized in that, Wavelet decomposition was performed on the temperature monitoring data to extract persistent temperature drift features, including: The temperature monitoring data is subjected to multi-level wavelet decomposition to obtain wavelet coefficients at multiple scale levels. The energy distribution ratio of the wavelet coefficients at each scale level is calculated, and the scale level where the energy distribution ratio changes abruptly is identified as the separation scale. Extract the wavelet coefficients corresponding to the separation scale, calculate the amplitude change rate of the wavelet coefficients in the time domain, filter the wavelet coefficients according to the distribution characteristics of the amplitude change rate, and reconstruct the filtered wavelet coefficients using wavelet reconstruction to obtain the initial drift signal; The initial drift signal is subjected to multi-segment linear fitting, and the slope change between adjacent fitting segments is calculated. When the signs of the slope change are consistent, the corresponding fitting segments are merged to obtain the continuous temperature drift characteristics.

3. The method according to claim 1, characterized in that, Cross-correlation calculations are performed on the current monitoring data and temperature monitoring data within multiple sliding windows at different time scales to obtain multi-scale cross-correlation sequences. Attenuation fitting is then performed on each of these multi-scale cross-correlation sequences to obtain multi-scale attenuation parameters. The difference between these multi-scale attenuation parameters is calculated to obtain heat dissipation path distortion indices, including: The fluctuation amplitude of the current monitoring data is statistically analyzed. A reference time scale is set according to the fluctuation amplitude. Multiple sliding windows with different time scales are set according to the multiple relationship of the reference time scale. In each sliding window, the current monitoring data and temperature monitoring data are cross-correlated from zero with time delay to obtain the multi-scale cross-correlation sequence corresponding to each time scale. The peak positions of each multi-scale cross-correlation sequence are extracted as the thermal response time delay. The cross-correlation coefficient segments that decrease with increasing time delay after the peak position are extracted as the decay segments. The decay curves are obtained by fitting each decay segment with an exponential function. The decay rate of each decay curve is extracted as the multi-scale decay parameter. The difference between the thermal response delays corresponding to adjacent time scales is calculated as the temporal gradient. When the temporal gradient changes from a positive value to a negative value, the corresponding time scale is marked as the heat dissipation degradation scale. The multi-scale decay parameters corresponding to the heat dissipation degradation scale are extracted as the degradation decay parameters. The magnitude difference of the temporal gradient changing from a positive value to a negative value is recorded as the temporal gradient reversal magnitude. The deviation between the degradation decay parameter and the mean of the multi-scale decay parameter across all time scales is calculated. The deviation is then weighted and fused with the temporal gradient reversal magnitude to calculate the difference between the multi-scale decay parameters, thus obtaining the heat dissipation path distortion index.

4. The method according to claim 1, characterized in that, The temperature evolution rate is obtained by time-domain differentiation of the sustained temperature drift characteristics. The temperature evolution rate and the heat dissipation path distortion index are then converted into a degradation risk characterization quantity through a mapping relationship, including: The temperature evolution rate is obtained by time-domain differentiation of the continuous temperature drift characteristics. The causal relationship between the heat dissipation path distortion index and the temperature evolution rate is tested. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transformation signal is output. Using the moment of causal chain breakage as the dividing point, the sensitivity coefficients of temperature evolution rate to heat dissipation path distortion index are extracted for the time period before and after the dividing point. The ratio of the sensitivity coefficient after the dividing point to the sensitivity coefficient before the dividing point is calculated. When the ratio exceeds the self-excitation judgment threshold, it is judged as the self-excitation stage. The duration and intensity of the self-excitation stage are statistically analyzed as a measure of degradation acceleration. A dynamic coupling network of temperature evolution rate and heat dissipation path distortion index is constructed. The ratio of positive feedback loop strength to negative feedback loop strength of the dynamic coupling network is calculated. When the ratio exceeds a preset critical threshold, it is determined to be an unstable critical state. The evolution rate of the unstable critical state is extracted. A two-dimensional mapping space is established between the evolution rate and the degradation acceleration measure. The current evolution rate and degradation acceleration measure are located as feature coordinate points. The risk level is matched according to the region where the feature coordinate points are located as the degradation risk characterization quantity.

5. The method according to claim 4, characterized in that, A causal relationship test is performed on the heat dissipation path distortion index and the temperature evolution rate. When the temperature evolution rate continues to rise but the heat dissipation path distortion index does not change synchronously, the causal chain is determined to be broken, and the degradation mechanism transition signal is output, including: A dynamic Bayesian network containing heat dissipation path distortion index, temperature evolution rate and ambient temperature is constructed. The directed edge weight matrix of the dynamic Bayesian network is learned. The causal edge weight sequence pointing from the heat dissipation path distortion index to the temperature evolution rate is extracted. When the temporal gradient of the causal edge weight sequence is negative and the absolute value exceeds the structural mutation threshold, the start time of the causal topology change is recorded. After the initial moment, the in-degree and out-degree of the temperature evolution rate node are counted. When the difference between the in-degree and out-degree changes from a positive value to a negative value, the time period of role change is recorded. Extract the first-order difference sequence of temperature evolution rate during the role transition period. When the longest duration of consecutive positive values ​​in the first-order difference sequence exceeds the monotonicity determination window, output a continuously rising indicator. Extract the sum of out-degree weights of the heat dissipation path distortion index nodes during the role transition period. When the relative deviation of the sum of out-degree weights from the historical stable period benchmark value is less than the synchronization change threshold, output the asynchronous change flag. When both the continuously rising indicator and the asynchronous change indicator exist, the start time and the absolute value of the temporal gradient are encapsulated as a degradation mechanism transition signal.

6. The method according to claim 1, characterized in that, The continuous temperature drift characteristics and degradation risk representations of adjacent bus nodes are used to construct a node feature matrix. This node feature matrix is ​​then aggregated using a graph attention propagation mechanism through multi-hop neighborhood aggregation to obtain the node fusion features, which include: The continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes are obtained and spliced ​​to form a node feature matrix; Construct the electrical connection adjacency matrix of the bus topology network, extract the neighbor node set of each bus node, calculate the difference in degradation risk representation between the current bus node and each neighbor node in the neighbor node set in the node feature matrix, generate graph attention weights based on the difference in degradation risk representation, and obtain the neighborhood aggregation feature by weighted summation of the feature vectors of the neighbor node set. The coupling degree between the continuous temperature drift feature component and the degradation risk characterization component in the neighborhood aggregation feature is calculated. When the coupling degree exceeds the preset coupling threshold, an inhibition factor is applied to the neighborhood aggregation feature to obtain the modified neighborhood aggregation feature. The modified neighborhood aggregation feature is concatenated with the feature vector of the current bus node in the node feature matrix to form an extended feature vector and the node feature matrix is ​​updated. The set of neighboring nodes of the neighboring node set is used as the next hop neighboring node set. Based on the updated node feature matrix, multi-hop neighborhood aggregation is iteratively performed until the preset number of hops is reached. During the iteration process, the difference in the risk representation of each jump is accumulated to form the risk propagation path feature. The risk propagation path feature is then fused with the extended feature vector after the iteration is completed to obtain the node fusion feature.

7. The method according to claim 1, characterized in that, Based on the node fusion characteristics and current current monitoring data, a time-series recursive calculation is performed to output a temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds a preset warning threshold, a warning signal is output, including: The node fusion features are mapped to the initial hidden state vector, the current current monitoring data is divided into current time series segments according to time steps, and the risk propagation path feature components in the node fusion features are extracted as risk modulation vectors. Based on the initial hidden state vector, risk modulation vector and current time sequence segments of each time step, a time-series recursive calculation is performed. The coupling degree between the risk modulation vector and the recursive state of each time step is calculated. According to the coupling degree, a risk gain factor is applied to the temperature rise prediction value of the corresponding time step to obtain the temperature rise prediction value of each time step and connect to output the temperature rise prediction trajectory. The temperature rise prediction trajectory is obtained by performing a first-order difference to obtain the temperature rise rate of change sequence. When the continuous length of positive values ​​in the temperature rise rate of change sequence exceeds the monotonicity judgment length, the temperature rise prediction trajectory is determined to be monotonically increasing. When the predicted temperature rise trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, the risk propagation path feature component is fused with the heat dissipation path distortion index to generate a warning signal and the warning signal is output.

8. A distribution box busbar temperature rise early warning system, used to implement the method as described in any one of claims 1-7, characterized in that, include: The data acquisition unit is used to collect temperature monitoring data and current monitoring data of the distribution box bus, and to perform wavelet decomposition on the temperature monitoring data to extract continuous temperature drift features. The wavelet decomposition unit is used to perform cross-correlation operations on the current monitoring data and temperature monitoring data in sliding windows of multiple different time scales to obtain a multi-scale cross-correlation sequence. The multi-scale cross-correlation sequence is then subjected to attenuation fitting to obtain multi-scale attenuation parameters. The difference between the multi-scale attenuation parameters is calculated to obtain the heat dissipation path distortion index. The mapping unit is used to perform time-domain differentiation on the continuous temperature drift characteristics to obtain the temperature evolution rate, and convert the temperature evolution rate and the heat dissipation path distortion index into a degradation risk characterization quantity through a mapping relationship. The distortion calculation unit is used to construct a node feature matrix by combining the continuous temperature drift characteristics and degradation risk characterization of adjacent bus nodes, and to perform multi-hop neighborhood aggregation on the node feature matrix through a graph attention propagation mechanism to obtain node fusion features. The mapping and transformation unit is used to perform time-series recursive calculations based on the node fusion features and the current current monitoring data, and output the temperature rise prediction trajectory. When the temperature rise prediction trajectory shows a monotonically increasing trend and the heat dissipation path distortion index exceeds the preset warning threshold, a warning signal is output.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images