A method and device for monitoring the operating state of a distribution box

Abstract

The application discloses a power distribution box operation state monitoring method and device, and relates to the intelligent operation and maintenance field of a power system. The steps are as follows: first, processing collected power distribution box environment, load and vibration acceleration data to obtain standardized multi-dimensional time series data; then, extracting monitoring feature parameters therefrom, and obtaining short-term and long-term multi-scale feature vectors through sliding window statistics; finally, inputting the multi-scale feature vectors into an attention mechanism feature fusion model to obtain a risk anomaly score, setting a dynamic threshold in combination with the statistical distribution of the historical normal operation score of the power distribution box, and judging a fault trend type according to the dynamic threshold. The scheme has the following advantages: multi-source data fusion avoids the limitation of a single data source, and improves the accuracy of anomaly identification; multi-scale analysis can capture single window information and long-term trends; the dynamic threshold is more accurate than a fixed threshold in adapting to different equipment and environments; and hierarchical evaluation can provide clear maintenance suggestions and help operation and maintenance decision-making.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for power systems, specifically to a method and device for monitoring the operating status of a distribution box. Background Technology

[0002] As a core device at the end of the power system, the distribution box undertakes the critical functions of power distribution, load control, and safety protection. Its operational stability directly affects the continuity and safety of electricity supply for industrial production, commercial operations, and residential use. Therefore, the need for distribution box operational status monitoring technology is increasingly urgent. Traditional methods of monitoring distribution box operational status rely on manual inspections. These inspections are time-consuming, difficult to cover sudden faults, and rely on tools such as multimeters and infrared thermometers, which cannot reflect overall operational trends. This approach not only results in poor timeliness and insufficient accuracy but also increases inspection costs due to the large amount of manpower required. Therefore, developing a technology capable of intelligently monitoring the operational status of distribution boxes is particularly important.

[0003] In recent years, with the development of sensor technology, IoT technology, and big data analytics, intelligent monitoring of the operating status of distribution boxes has become an effective way to solve the problem of distribution box inspection. By collecting environmental data, load data, and vibration acceleration during the operation of the distribution box, and processing them, the data is input into a feature fusion model based on an attention mechanism to obtain a risk score that can comprehensively reflect the operating status of the distribution box.

[0004] In the prior art, CN106208390A discloses a monitoring system and method for the operating status of a distribution box. The method involves collecting the current at the main circuit air circuit breaker and the temperature at the air circuit breakers of each branch circuit. Thresholds are set for the output current of the main circuit air circuit breaker and the temperature of each branch air circuit breaker. An early warning module compares the collected current and temperature data. If the current or temperature reaches the set threshold, it is determined to be an abnormal circuit, and an early warning message containing the air circuit breaker code of the abnormal circuit is generated. This method comprehensively considers the current of the main circuit and the temperature data of each branch, resulting in highly targeted monitoring and clear fault location. However, this solution does not consider humidity, air quality, or vibration monitoring; it only considers current and temperature factors, making the monitoring parameters singular and failing to cover all possible risks.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and device for monitoring the operating status of a distribution box, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for monitoring the operating status of a distribution box, comprising the following steps:

[0009] Within multiple preset monitoring cycles, environmental data, electrical load data, and vibration acceleration of the distribution box are collected. The collected environmental data, load data, and vibration acceleration are processed according to the time of the monitoring cycle to obtain standardized multidimensional time series data.

[0010] For each monitoring cycle, the rate of change of environmental data, the fluctuation amplitude of electrical load data, and the abrupt change amplitude of vibration acceleration are calculated to obtain monitoring characteristic parameters reflecting the operating status of the distribution box in each monitoring cycle. The monitoring characteristic parameters are then statistically analyzed in a sliding window over the time domain to obtain a multi-scale monitoring characteristic vector reflecting short-term and long-term operating trends.

[0011] The multi-scale monitoring feature vector is input into the feature fusion model based on the attention mechanism. By combining the sensitivity weights of different feature parameters, dynamic weighted fusion of environmental, load and vibration multi-source monitoring data is achieved to obtain an abnormal risk score that reflects the current comprehensive operating status of the distribution box.

[0012] Based on the statistical distribution of abnormal risk scores during the historical normal operation phase of the distribution box, a dynamic threshold is set. The abnormal risk score of the current monitoring period is compared with the dynamic threshold. When the number of monitoring periods in which the abnormal risk score continuously exceeds the dynamic threshold reaches a preset threshold, it is determined that the distribution box has an abnormal operation risk.

[0013] After determining that the distribution box has an abnormal risk, cluster analysis is performed on the monitoring characteristic parameters of each cycle, and the type of fault trend of the distribution box is determined by combining the time-series change trend of the monitoring characteristic parameters. Furthermore, the preset multiple monitoring cycles are pre-set continuous time intervals of equal length, with the end time of each monitoring cycle being the same as the start time of the next monitoring cycle. All monitoring cycles are arranged continuously on the time axis. The distribution box environmental data includes temperature data inside the distribution box, the electrical load data includes the total load data inside the distribution box, and the vibration acceleration refers to the vibration acceleration of the distribution box body. The distribution box environmental data, electrical load data, and vibration acceleration are collected continuously during each monitoring cycle.

[0014] The collected data of each type are grouped according to their respective sampling timestamps and time-aligned in units of monitoring cycles. Within each monitoring cycle, each type of data is normalized to form multidimensional time series data arranged according to the monitoring cycle. The multidimensional time series data includes temperature time series data, load time series data and vibration time series data.

[0015] Furthermore, the method for calculating the rate of change of ambient temperature is as follows: the rate of change of ambient temperature is obtained by taking the first derivative of the ambient temperature.

[0016] The method for calculating the fluctuation amplitude of electrical load data is as follows: the fluctuation amplitude of electrical load data is obtained by calculating the standard deviation of the electrical load data;

[0017] The method for calculating the amplitude of sudden changes in vibration acceleration is as follows: the amplitude of sudden changes in vibration acceleration is obtained by calculating the difference between adjacent vibration acceleration values;

[0018] The environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude obtained above are combined into a set and used as the monitoring characteristic parameters for this monitoring period.

[0019] Furthermore, after obtaining the monitoring characteristics of the environmental data change rate, load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within each monitoring cycle, the lengths of multiple sliding windows are pre-set, including short-term and long-term windows. Each sliding window slides sequentially on the feature parameter sequence with a fixed step size. For each slide, the window contains feature parameters from several consecutive monitoring cycles. The length of the short-term window is... The length of each monitoring cycle, and the length of the long-term window are... The length of each monitoring cycle ;

[0020] Within each sliding window, the monitoring characteristic parameters of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude contained in the window are statistically processed to obtain representative statistics within that window.

[0021] Furthermore, the method for obtaining sliding window statistics reflecting short-term operating trends is as follows: The short-term mean statistics are obtained by calculating the average values ​​of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window; the short-term standard deviation statistics are obtained by calculating the standard deviation values ​​of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window; the short-term maximum statistics are obtained by inputting the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window into the MAX function; and the short-term minimum statistics are obtained by inputting the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window into the MIN function.

[0022] The method for obtaining sliding window statistics reflecting long-term operating trends is as follows: The long-term mean statistic is obtained by calculating the average values ​​of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window; the long-term standard deviation statistic is obtained by calculating the standard deviation of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window; the long-term maximum statistic is obtained by inputting the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window into the MAX function; and the long-term minimum statistic is obtained by inputting the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude within the window into the MIN function.

[0023] Furthermore, two multi-scale monitoring feature vectors, one short-term and one long-term, are constructed using the sliding window statistical method. The short-term feature vector reflects the most recent... The operating status of each cycle, and the short-term multi-scale monitoring feature vector. For one The vectors are: row vectors representing environmental data, load data, and vibration acceleration, and column vectors representing the average, maximum, minimum, and standard deviation of the monitored characteristic parameters.

[0024] Long-term multi-scale eigenvectors reflect the most recent The operational status of each monitoring cycle, and the long-term multi-scale monitoring feature vector. For one The vectors are: row vectors representing environmental data, load data, and vibration acceleration, and column vectors representing the average, maximum, minimum, and standard deviation of the monitored characteristic parameters.

[0025] Furthermore, the feature vectors from the two time periods are combined into a complete multi-scale feature matrix: ;

[0026] in, For short-term multi-scale monitoring feature vectors, For long-term multi-scale monitoring feature vectors;

[0027] The feature matrix is ​​divided into three groups according to physical attributes: environmental feature group, load feature group, and vibration feature group. The three sets of feature vectors are input into an attention-based feature fusion model to obtain sensitivity weight vectors for environmental, load, and vibration features. These weight vectors are then fused using multi-head attention to obtain a global weight vector. ;

[0028] The global weight vector is input into a multi-source feature fusion model based on attention weights to obtain the comprehensive feature values: ;

[0029] The comprehensive eigenvalues ​​are transformed through a fully connected network and then... The mapping function yields the mapping score. The mapping score is processed to obtain a final anomaly risk score with a value range of 0-100: .

[0030] Furthermore, by inputting the historical normal scores into a binomial distribution, it was found that when the risk score is 0-20, the distribution box is in normal operation; when the risk score is 21-50, the distribution box is in slightly abnormal operation; when the risk score is 51-80, the distribution box is in moderately abnormal operation; and when the risk score is 81-100, the distribution box is in severely abnormal operation.

[0031] The present invention also provides a distribution box operation status monitoring device, which is used to implement the above-mentioned distribution box operation status monitoring method, including:

[0032] The environmental data monitoring module is used to collect the temperature and humidity inside the distribution box and process them according to the time of the monitoring cycle to obtain standardized multi-dimensional time series data;

[0033] The electrical load monitoring module is used to collect real-time three-phase current and obtain three-phase unbalance after processing. The three-phase current and three-phase unbalance are aligned according to the time of the monitoring cycle to obtain standardized multi-dimensional time series data.

[0034] The vibration monitoring module is used to collect the vibration acceleration of the distribution box, align the data according to the time of the monitoring period, and obtain standardized multi-dimensional time series data.

[0035] The data processing module processes environmental data, electrical load data, and vibration acceleration to obtain monitoring characteristic parameters. It then performs sliding window statistics on the monitoring characteristic parameters in the time domain to obtain multi-scale monitoring feature vectors.

[0036] The monitoring module processes multi-scale monitoring feature vectors to obtain risk scores, and then uses statistical distribution to obtain risk score thresholds to determine the operational status.

[0037] Compared with the prior art, the beneficial effects of the present invention are:

[0038] This invention overcomes the limitations of a single data source by integrating multi-source data and time alignment processing, thereby improving the comprehensiveness and accuracy of anomaly identification. At the same time, it takes into account both short-term and long-term trends by combining multi-scale feature extraction with attention fusion mechanism, thereby enhancing the identification of anomaly features. In addition, it adapts to multiple scenarios and provides clear fault location and maintenance suggestions by using dynamic threshold determination and cluster analysis of fault trends.

[0039] This invention also sets thresholds based on the statistical distribution of historical normal data, which can adapt to different equipment characteristics and operating environments and is more accurate than fixed thresholds. At the same time, it provides clear maintenance suggestions based on hierarchical anomaly assessment, which facilitates operation and maintenance decisions. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0041] Figure 2 This is a graph showing the changes in temperature change rate, vibration abrupt change amplitude, load fluctuation amplitude, and abnormal risk score of the present invention.

[0042] Figure 3 This is a structural block diagram of the system of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0044] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0045] Example:

[0046] Please see Figures 1 to 2 The present invention provides a technical solution:

[0047] A method for monitoring the operating status of a distribution box, comprising the following steps:

[0048] Step 1: Collect environmental data, electrical load data, and vibration acceleration of the distribution box within multiple preset monitoring cycles. Align the collected environmental data, load data, and vibration acceleration according to the time of the monitoring cycle to obtain standardized multidimensional time series data.

[0049] The preset monitoring cycles are pre-defined continuous time intervals of equal length. The end time of each monitoring cycle is the same as the start time of the next monitoring cycle. All monitoring cycles are connected end to end on the time axis and arranged continuously. The distribution box environmental data is the temperature data inside the distribution box, the electrical load data is the total load data inside the distribution box, and the vibration acceleration refers to the vibration acceleration of the distribution box body. When collecting distribution box environmental data, electrical load data, and vibration acceleration in each monitoring cycle, the data is collected continuously. Each type of collected data is grouped according to its own sampling timestamp and time-aligned by monitoring cycle. In each monitoring cycle, each type of data is normalized to form multidimensional time series data arranged by monitoring cycle. The multidimensional time series data includes temperature time series data, load time series data, and vibration time series data.

[0050] Environmental data is directly related to the operational stability of components within the distribution box. Abnormal temperatures within the distribution box may originate from poor component contact, short-circuit precursors, or heat dissipation failures. Without environmental data, non-electrical fault causes are easily overlooked, leading to missed fault diagnosis. Electrical load data reflects the rationality of power distribution and load operation within the distribution box. Excessive load fluctuations may be due to overload, overheating of the load circuit, localized overheating of components, or increased equipment wear. Ignoring load data makes it impossible to locate electrical circuit faults. Vibration acceleration reflects mechanical faults and installation abnormalities within the distribution box. Changes in vibration acceleration can identify mechanical faults such as loose box housings and detached internal components, which cannot be reflected by environmental or load data. Collecting environmental data, electrical load data, and vibration acceleration is the core prerequisite for this solution to overcome the limitations of traditional single data parameters. These three types of data construct a monitoring system for the operating status of the distribution box from three aspects: environmental stability, electrical safety, and mechanical reliability.

[0051] Step 2: For each monitoring cycle, calculate the mean and rate of change of environmental data, the mean and fluctuation amplitude of electrical load data, and the vibration acceleration and abrupt change amplitude of vibration acceleration to obtain monitoring characteristic parameters reflecting the operating status of the distribution box in each monitoring cycle. Then, perform sliding window statistics on the monitoring characteristic parameters in the time domain to obtain a multi-scale monitoring characteristic vector reflecting short-term and medium-to-long-term operating trends.

[0052] First, calculate the rate of change of environmental data within a monitoring period. Fluctuation range of electrical load data Amplitude of sudden changes in vibration acceleration The method is as follows:

[0053] Rate of change of environmental data:

[0054] Electrical load fluctuation range:

[0055] Amplitude of sudden changes in vibration acceleration:

[0056] Secondly, after obtaining the environmental data mean and rate of change, load mean and fluctuation amplitude, vibration acceleration and sudden change amplitude monitoring characteristics within each monitoring period, the lengths of multiple sliding windows are preset, with short-term windows (5 monitoring periods) and long-term windows (100 monitoring periods) set respectively. Each sliding window slides sequentially on the feature parameter sequence with a fixed step size (1 monitoring period). For each slide, the window contains the feature parameters of the most recent several consecutive monitoring periods.

[0057] The rate of environmental change represents the rate of temperature change. If the absolute temperature value does not exceed the threshold but the rate of change suddenly increases, it may be a signal of poor component contact or an impending internal short circuit. Under normal circumstances, the rate of temperature change is gradual when the load increases. If the load is stable but the rate of temperature change suddenly increases, non-load anomalies can be quickly located, narrowing down the scope of fault investigation. The amplitude of electrical load fluctuation quantifies the operational stability of the electrical circuit. Excessive load fluctuation may indicate overload of the load circuit or abnormality of the load equipment, allowing for early intervention to prevent circuit breaker tripping and equipment damage. The amplitude of sudden changes in vibration acceleration is used to capture mechanical faults, filling the monitoring gaps that environmental and load data cannot cover. A sudden increase in the amplitude of sudden changes in vibration acceleration may indicate mechanical faults such as loose mounting bolts or detached internal components. If the amplitude of sudden changes in vibration acceleration is abnormal but environmental and load data are normal, mechanical faults can be directly located, avoiding confusion with environmental overheating or load anomalies. If all three are abnormal simultaneously, it can be identified as a compound fault, providing a basis for accurate maintenance.

[0058] Within each sliding window, the monitoring characteristic parameters of the environmental change rate, electrical load data fluctuation amplitude, and vibration acceleration abrupt change amplitude contained in the window are statistically processed to obtain representative statistics within that window.

[0059] The methods for representative statistics within the short-term and long-term windows are as follows:

[0060] Mean: ,in The window length is set to 5 for the short-term window and 100 for the long-term window. To monitor characteristic parameters;

[0061] Maximum value: Among them, short-term window The value is 4, and the long-term window value is 99. These are the monitoring characteristic parameters for each period within the window;

[0062] Minimum value: , including the short-term window The value is 4, and the long-term window value is 99. These are the monitoring characteristic parameters for each period within the window;

[0063] Standard deviation: ,in The window length is set to 5 for the short-term window and 100 for the long-term window. To monitor characteristic parameters, To monitor the mean value of characteristic parameters.

[0064] Finally, two multi-scale monitoring feature vectors, one for short-term and one for long-term, are constructed using the sliding window statistical method. The short-term feature vector reflects the operating status of the most recent five cycles and focuses on capturing instantaneous fluctuations and sudden anomalies. ;

[0065] in, For short-term ambient temperature change rate, For short-term load fluctuations, This refers to the amplitude of short-term vibrational abrupt changes. This represents the mean. Indicates the maximum value. This represents the minimum value. Indicates standard deviation;

[0066] Long-term multi-scale feature vectors reflect the operational status over the most recent 100 monitoring periods, capturing long-term operational patterns and slow changing trends: ;

[0067] in, The rate of change of ambient temperature over a long period of time. For long-term load fluctuation range, This represents the amplitude of long-term vibration abrupt changes. This represents the mean. Indicates the maximum value. This represents the minimum value. It represents the standard deviation.

[0068] The mean of statistics within a short-term window quantifies the average level of characteristic parameters across multiple monitoring periods within that window. Its core function is to eliminate random biases in single-period data, accurately reflect the true overall trend of characteristic parameters within the short-term dimension, and avoid misjudgment or missed detection of hidden anomalies. The maximum value of statistics within a short-term window focuses on extreme values ​​of characteristic parameters within the window. Its core function is to accurately capture instantaneous sudden anomalies in single-period data that may be masked by the mean, avoiding the omission of high signals. The minimum value of statistics within a short-term window quantifies the lower limit level of characteristic parameters within the window, providing evidence for the reliability of monitoring data and the identification of special anomalies. The mean of statistics within a long-term window quantifies the central tendency of characteristic parameters across a large number of continuous monitoring periods within the window. Its core function is to filter out short-term fluctuations and accurately capture gradual abnormal changes in characteristic parameters within the long-term dimension—changes that are difficult to detect within a single period or short-term window. However, it may be an early signal of equipment aging and the accumulation of hidden dangers. The maximum value of the statistic within the long-term window focuses on the extreme values ​​of characteristic parameters within the long-term period. Its core function is to identify the trend of increasing frequency of occasional high-risk events. Extreme values ​​within a single period or short-term window may be accidental interference, but the increase in the maximum value or the increase in the frequency of occurrence within the long-term window often means the accumulation of high-risk hidden dangers. The minimum value of the statistic within the long-term window is a quantification of the lower limit level of characteristic parameters within the long-term dimension. Its main functions are reflected in two aspects: providing support for the reliability of long-term monitoring and the identification of special anomalies. The standard deviation of the statistic within the long-term window is a quantification of the dispersion of characteristic parameters within the long-term period. Its core function is to assess the decline in long-term operational stability. The short-term standard deviation reflects instantaneous fluctuations, while the increase in the long-term standard deviation often means that the "stability of equipment operation is continuously declining," which is an important signal of the accumulation of hidden dangers.

[0069] Step 3: Combine the feature vectors from the two time periods into a complete multi-scale feature.

[0070] matrix: ;

[0071] in, For short-term multi-scale monitoring feature vectors, For long-term multi-scale monitoring feature vectors;

[0072] The feature matrix is ​​divided into three groups according to physical attributes: environmental feature group, load feature group, and vibration feature group. The three sets of feature vectors are input into an attention-based feature fusion model to obtain sensitivity weight vectors for environmental, load, and vibration features. These weight vectors are then fused using multi-head attention to obtain a global weight vector. ;

[0073] The global weight vector is input into a multi-source feature fusion model based on attention weights to obtain the comprehensive feature values: ;

[0074] The comprehensive eigenvalues ​​are transformed through a fully connected network and then... The mapping function yields the mapping score. The mapping score is processed to obtain a final anomaly risk score with a value range of 0-100: .

[0075] Step 4: Based on the statistical distribution of abnormal risk scores during the historical normal operation phase of the distribution box, set a dynamic threshold, compare the abnormal risk score of the current monitoring period with the dynamic threshold, and determine that the distribution box has an abnormal operation risk when the number of monitoring periods in which the abnormal risk score continuously exceeds the dynamic threshold reaches a preset threshold.

[0076] After determining that there is an abnormal risk in the distribution box, cluster analysis is performed on the monitoring characteristic parameters of each period, and the type of fault trend of the distribution box is determined by combining the time series change trend of the monitoring characteristic parameters.

[0077] By inputting historical normal scores into a binomial distribution, it can be concluded that when the risk score is 0-20, the distribution box is in normal operation; when the risk score is 21-50, the distribution box is in slightly abnormal operation; when the risk score is 51-80, the distribution box is in moderately abnormal operation; and when the risk score is 81-100, the distribution box is in severely abnormal operation.

[0078] As shown in Table 1,

[0079] Table 1 shows the operating status of the distribution boxes during some monitoring periods.

[0080] Table 1: Operating Status of Some Distribution Boxes

[0081]

[0082] Table 1 shows a clear correlation between the abnormal risk score and the dynamic threshold, with the threshold increasing as the abnormal state escalates. For example, during monitoring periods 1-3, the distribution box operated normally with an abnormal risk score below 20; during monitoring periods 6-8, the distribution box operated slightly abnormally with an abnormal risk score between 20 and 50; during monitoring periods 21-23, the distribution box operated moderately abnormally with an abnormal risk score between 50 and 80; and during monitoring periods 31-33, the distribution box operated severely abnormally with an abnormal risk score above 80. This indicates that the division between the abnormal risk score and the dynamic threshold is reasonable.

[0083] As shown in Table 2, Table 2 presents the short-term and long-term feature vectors of the distribution box operating status during some monitoring periods.

[0084] Table 2: Short- and Long-Term Feature Vectors of Distribution Box Operating Status During Some Monitoring Periods

[0085]

[0086] As shown in Table 2, the long-term and short-term feature vectors have small numerical differences and consistent trends. This indicates that the short-term fluctuations in the monitoring data have high stability in the long-term dimension. It also shows that the short-term feature vector can be used for real-time early warning, such as sudden temperature rises, while the long-term feature vector can be used for trend judgment, such as continuous temperature rises. The combination of the two greatly improves the monitoring accuracy.

[0087] Please see Figure 3 The present invention also provides a distribution box operation status monitoring device, which is used to implement the above-mentioned distribution box operation status monitoring method, including:

[0088] The environmental data monitoring module is used to collect the temperature and humidity inside the distribution box and process them according to the time of the monitoring cycle to obtain standardized multi-dimensional time series data;

[0089] The electrical load monitoring module is used to collect real-time three-phase current and obtain three-phase unbalance after processing. The three-phase current and three-phase unbalance are aligned according to the time of the monitoring cycle to obtain standardized multi-dimensional time series data.

[0090] The vibration monitoring module is used to collect the vibration acceleration of the distribution box, align the data according to the time of the monitoring period, and obtain standardized multi-dimensional time series data.

[0091] The data processing module processes environmental data, electrical load data, and vibration acceleration to obtain monitoring characteristic parameters. It then performs sliding window statistics on the monitoring characteristic parameters in the time domain to obtain multi-scale monitoring feature vectors.

[0092] The monitoring module processes multi-scale monitoring feature vectors to obtain risk scores, and then uses statistical distribution to obtain risk score thresholds to determine the operational status.

[0093] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0094] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0095] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0096] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method of monitoring the operating state of an electrical distribution box, characterized in that, The specific steps include: In a plurality of preset monitoring periods, environment data, electrical load data and vibration acceleration of the distribution box are collected, and the collected environment data, load data and vibration acceleration are respectively aligned in time according to the monitoring period to obtain standardized multi-dimensional time series data; For each monitoring period, the change rate of the environment data, the fluctuation amplitude of the electrical load data and the mutation amplitude of the vibration acceleration are calculated respectively to obtain monitoring characteristic parameters reflecting the running state of the distribution box in each monitoring period, and the monitoring characteristic parameters are statistically processed in a sliding window in the time domain to obtain a multi-scale monitoring characteristic vector reflecting short-term and long-term running trends; The multi-scale monitoring characteristic vector is input into a feature fusion model based on an attention mechanism, and the sensitivity weights of different characteristic parameters are combined to realize dynamic weighted fusion of multi-source monitoring data of the environment, load and vibration, so as to obtain an abnormal risk score for reflecting the current comprehensive running state of the distribution box; According to the statistical distribution of the abnormal risk score in the historical normal running stage of the distribution box, a dynamic threshold is set, the abnormal risk score of the current monitoring period is compared with the dynamic threshold, and when the number of monitoring periods in which the abnormal risk score continuously exceeds the dynamic threshold reaches a preset threshold, it is determined that the distribution box has an abnormal running risk; After it is determined that the distribution box has an abnormal risk, clustering analysis is performed on the monitoring characteristic parameters of each period, and the type of the fault trend of the distribution box is judged in combination with the time sequence change trend of the monitoring characteristic parameters; The two time length feature vectors are combined into a complete multi-scale feature matrix: ; wherein, is a short-term multiscale monitoring feature vector, is a long-term multiscale monitoring feature vector; The feature matrix is divided into three groups according to physical properties, i.e., an environmental feature group, a load feature group and a vibration feature group: The three groups of feature vectors are input into a feature fusion model based on an attention mechanism to obtain a sensitivity weight vector of the environmental, load and vibration features, and a global weight vector is obtained through multi-head attention fusion . The global weight vector is input into a multi-source feature fusion model based on attention weight to obtain a comprehensive feature value: ; The comprehensive feature values are converted through a full connection network and mapped through a mapping function to obtain mapping scores The mapping scores are processed to obtain a final abnormal risk score with a value range of 0-100: .​ 2. The method of claim 1, wherein: The plurality of preset monitoring periods are equal-length continuous time intervals, the end time of each monitoring period is the same as the start time of the next monitoring period, all monitoring periods are arranged in series with the first and last ends connected, the environment data of the distribution box is temperature data in the distribution box, the electrical load data is the total load data inside the distribution box, and the vibration acceleration refers to the vibration acceleration of the distribution box body. When the environment data, electrical load data and vibration acceleration of the distribution box are collected in each monitoring period, the collection is continuous; Each type of collected data is grouped according to the respective sampling time stamp, and time alignment processing is performed in units of monitoring periods. In each monitoring period, each type of data is subjected to normalization processing to form multi-dimensional time series data arranged in monitoring periods. The multi-dimensional time series data includes temperature time series data, load time series data and vibration time series data.

3. The method of claim 2, wherein: The method for calculating the change rate of the environment temperature is to obtain the change rate of the environment temperature by first-order derivation of the environment temperature; The method for calculating the fluctuation amplitude of the electrical load data is to obtain the fluctuation amplitude of the electrical load data by calculating the standard deviation of the electrical load data; The method for calculating the mutation amplitude of the vibration acceleration is to obtain the mutation amplitude of the vibration acceleration by calculating the adjacent difference value of the vibration acceleration; The above-obtained environment change rate, electrical load data fluctuation amplitude and vibration acceleration mutation amplitude form a set as the monitoring characteristic parameters of the monitoring period.

4. The method of claim 1, wherein: After obtaining the environmental data change rate, load data fluctuation amplitude and vibration acceleration mutation amplitude monitoring features in each monitoring period, a plurality of sliding window lengths are pre-set, a short-term window and a long-term window are respectively set, each sliding window slides on the feature parameter sequence in a fixed step, for each sliding, the window contains a plurality of continuous monitoring period feature parameters, the length of the short-term window is monitoring period, the length of the long-term window is monitoring period, ; In each of the obtained sliding window, the monitoring characteristic parameters of the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude contained in the window are respectively statistically processed to obtain the representative statistics of the window.

5. The method of claim 4, wherein: The method for obtaining the sliding window statistics reflecting the short-term operation trend is: the mean value of the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window is obtained to obtain the short-term mean value statistics; the standard deviation of the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window is obtained to obtain the short-term standard deviation statistics; the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window are respectively input into the MAX function to obtain the short-term maximum value statistics; the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window are respectively input into the MIN function value to obtain the short-term minimum value statistics. The method for obtaining the sliding window statistics reflecting the long-term operation trend is: the mean value of the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window is obtained to obtain the long-term mean value statistics; the standard deviation of the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window is obtained to obtain the long-term standard deviation statistics; the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window are respectively input into the MAX function to obtain the long-term maximum value statistics; the environmental change rate, the electrical load data fluctuation amplitude and the vibration acceleration mutation amplitude in the window are respectively input into the MIN function value to obtain the long-term minimum value statistics.

6. The method of claim 1, wherein: The short-term and long-term two multi-scale monitoring feature vectors are respectively constructed by the sliding window statistical method, the short-term feature vector reflects the operation state of the recent period, the short-term multi-scale monitoring feature vector is a vector, and the row vectors are environment data, load data and vibration acceleration, and the column vectors are the average value, maximum value, minimum value and standard deviation of the monitoring feature parameters. Long-term multi-scale feature vector reacts to recent operating status of a monitoring period, long-term multi-scale monitoring feature vector is a vector , the row vectors are environmental data, load data and vibration acceleration, and the column vectors are the average value, maximum value, minimum value and standard deviation of the monitoring feature parameters.

7. The method of claim 1, wherein: The historical normal score is input into the binomial distribution to obtain that when the risk score is 0-20, the distribution box is in a normal operation state, when the risk score is 21-50, the distribution box is in a mild abnormal operation state, when the risk score is 51-80, the distribution box is in a moderate abnormal operation state, and when the risk score is 81-100, the distribution box is in a serious abnormal operation state.

8. A distribution box operation state monitoring device, the device is used to realize the distribution box operation state monitoring method of any one of claims 1-7, characterized in that: an environmental data monitoring module is used to collect the temperature and humidity inside the distribution box and align the processing according to the time of the monitoring period to obtain standardized multi-dimensional time series data; an electrical load monitoring module is used to collect real-time three-phase current and obtain three-phase unbalance degree after processing, and align the processing of the three-phase current and three-phase unbalance degree according to the time of the monitoring period to obtain standardized multi-dimensional time series data; a vibration monitoring module is used to collect the vibration acceleration of the distribution box and align the processing according to the time of the monitoring period to obtain standardized multi-dimensional time series data; a data processing module is used to process the environmental data, electrical load data and vibration acceleration to obtain monitoring characteristic parameters, and perform sliding window statistics on the monitoring characteristic parameters in the time domain to obtain a multi-scale monitoring characteristic vector; a monitoring module is used to process the multi-scale monitoring characteristic vector to obtain a risk score, and obtain a risk score threshold value through statistical distribution to determine the operation state.

Images