Distributed energy-oriented power distribution network fault diagnosis method

Abstract

The invention provides a distributed energy-oriented power distribution network fault diagnosis method, and relates to the technical field of power grid data diagnosis, and the method comprises the steps: collecting the multi-source heterogeneous data of distributed energy, carrying out the layering of a collection process through a layered data collection preprocessing architecture, and carrying out the fault diagnosis of the distributed energy after the collection preprocessing process is completed. Establishing a dynamic characteristic modeling stage of the power distribution network; and extracting power distribution network fault characteristics to complete diagnosis. According to the method, the accuracy and efficiency of power distribution network fault diagnosis are improved, comprehensive acquisition and efficient screening of multi-source heterogeneous data such as voltage, current and temperature are realized, and the identification capability of abnormal data is enhanced, so that the change rule of the operation state of the power distribution network is comprehensively captured, the accuracy and reliability of fault characteristic extraction and diagnosis are improved, and the fault diagnosis efficiency is improved. Potential risks are found in advance, alarm is triggered in time, and powerful technical guarantee is provided for safe and stable operation of the power distribution network.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid data diagnosis, and more particularly, to a power distribution network fault diagnosis method for distributed energy. BACKGROUND

[0002] With the rapid development of society, distributed energy is increasingly widely used in power distribution networks, which marks the transition of energy management from the traditional centralized mode to a more flexible and efficient distributed mode. Under this background, the demand for energy data processing in the fields of administration, business, finance and management is increasingly urgent. In order to realize effective supervision and prediction of distributed energy, a data processing system or method specially applicable to the above-mentioned fields needs to be constructed. The system should not only be able to monitor the running state of the power distribution network in real time, but also be able to support decision optimization through data analysis, so as to improve energy utilization efficiency and system reliability. Therefore, under the current trend of social development, it is particularly important to transform the power distribution network fault diagnosis into a data-driven management mode.

[0003] However, there are some cold but key defects in the prior art, which limit the accuracy and efficiency of power distribution network fault diagnosis. For example, most of the current fault diagnosis methods rely on traditional signal analysis techniques, ignoring the influence of dynamic characteristics of distributed energy on data quality. In particular, in the aspect of multi-source heterogeneous data fusion, existing methods often lack effective processing capability for unstructured data (such as sensor noise, environmental interference, etc.), which may cause deviation in the diagnosis result. In addition, traditional algorithms have difficulty in quickly locating fault points and accurately classifying when facing complex network topologies, which greatly reduces the prediction performance of the system. In view of these rare technical bottlenecks, it is urgent to develop a new type of data processing method to improve the intelligent level of power distribution network fault diagnosis. SUMMARY

[0004] In order to solve the above technical problems, the present application provides a power distribution network fault diagnosis method for distributed energy.

[0005] According to one aspect of the present application, a power distribution network fault diagnosis method for distributed energy is provided, which comprises: Collecting multi-source heterogeneous data of distributed energy, using a layered data collection preprocessing architecture to layer the collection process and complete the preprocessing process of collection; After completing the preprocessing process of collection, establishing a dynamic characteristic modeling stage of the power distribution network; Decomposing the running state change law of the power distribution network in the established dynamic characteristic modeling stage of the power distribution network, extracting the fault characteristics of the power distribution network to complete diagnosis.

[0006] Further, the data collection preprocessing architecture includes a bottom layer architecture, a middle layer architecture and a top layer architecture, the bottom layer architecture includes collecting physical quantities of voltage, current and temperature in the power distribution network by using a distributed sensor network, the middle layer architecture includes an anomaly detection mechanism using an adaptive learning mechanism to filter raw data that has not been diagnosed by power grid data, and the top layer architecture includes converting the filtered raw data into a unified format through a feature extraction operation, and introducing a timestamp labeling mechanism to establish time dimension information for the unified format raw data. The filtering of raw data that has not been diagnosed by power grid data includes real-time analysis of raw data using a sliding window technique, calculating statistical indicators in the current time period, dynamically generating a threshold range for anomaly detection according to the calculated statistical indicators, and defining the upper and lower threshold ranges for anomaly detection according to the calculated statistical indicators.

[0007] Further, the conversion of the filtered raw data into a unified format includes normalizing the numerical values of the physical quantities of voltage, current and temperature to the same dimension range, and concatenating the feature vectors of voltage, current and temperature in time sequence to form a multi-dimensional feature matrix. The temperature includes introducing a sliding time window technique to analyze the temperature data distribution in the current time period, and deploying a distributed temperature sensor network at the nodes of the power distribution network, the distributed temperature sensor network includes using infrared temperature measurement technology to obtain the surface temperature of the power distribution network, and monitoring the surface temperature of the power distribution network.

[0008] Further, the infrared temperature measurement technology for obtaining the surface temperature of the power distribution network includes internal temperature and external temperature, the internal temperature includes measuring the transformer winding temperature of the distributed temperature sensor through an embedded sensor; and the external temperature includes collecting temperature changes around the distributed temperature sensor using an outdoor environment temperature sensor. The monitoring of the surface temperature of the power distribution network includes non-contact temperature monitoring of the running state of the embedded sensor and the environment temperature sensor according to the infrared thermal imaging technology, and capturing temperature data of local hot spot areas. The concatenation of the feature vectors of voltage, current and temperature in time sequence includes analyzing the temperature data distribution in the current time period according to the sliding time window technique, and dynamically updating and , wherein, represents the maximum value of the temperature data, represents the minimum value of the temperature data. The temperature data distribution includes smoothness, periodicity and abruptness, and the abnormal values of different temperature data distributions are identified according to the time series characteristics of the temperature data.

[0009] Further, the time dimension information established for the unified format raw data comprises capturing abnormal values of different temperature data distribution, adding the abnormal values to the time series format organized, and designing a sliding time window adjustment strategy based on a feedback mechanism according to fluctuation characteristics of the power distribution network operation state. When the power distribution network load is detected to be increased, the time length of the sliding window is shortened to improve the time response speed; otherwise, the time length of the sliding window is lengthened to reduce the time response speed.

[0010] Further, the dynamic characteristic modeling stage of the power distribution network comprises using a data hierarchical analysis method of the time dimension information and the data acquisition preprocessing architecture to construct a power distribution network dynamic characteristic framework reflecting the change law of the power distribution network operation state. The data hierarchical analysis method comprises layering and analyzing the preprocessed multi-source heterogeneous data according to time scales, and extracting dynamic characteristics at different levels, wherein the dynamic characteristics comprise a low-frequency trend layer, a medium-frequency fluctuation layer and a high-frequency disturbance layer. The low-frequency trend layer comprises using an exponential smoothing algorithm of a sliding window to extract physical quantity trend components of voltage, current and temperature of the low-frequency trend layer of the bottom layer architecture, and capture the change law of the power distribution network operation state. The medium-frequency fluctuation layer comprises using an empirical mode decomposition technology to decompose the multi-source heterogeneous data into N intrinsic mode functions, and capture the change law of the power distribution network operation state. The high-frequency disturbance layer comprises using a time resolution analysis method to determine the operation state sudden disturbance components of the power distribution network, and capture the change law of the power distribution network operation state.

[0011] Further, the change law of the power distribution network operation state in the dynamic characteristic modeling stage of the power distribution network is decomposed by using an empirical mode decomposition and a variational mode decomposition, and through the empirical mode decomposition and the variational mode decomposition process, the change law component characteristics of the power distribution network operation state of the low-frequency trend layer, the change law component characteristics of the power distribution network operation state of the medium-frequency fluctuation layer and the change law component characteristics of the power distribution network operation state of the high-frequency disturbance layer are extracted, and the power distribution network fault characteristic extraction is completed.

[0012] According to another aspect of the present application, a power distribution network fault diagnosis system for distributed energy is provided, which comprises: a hierarchical acquisition module which uses a hierarchical data acquisition preprocessing architecture to layer the acquisition process, and completes the preprocessing process of acquisition; a dynamic characteristic modeling stage module which, after the preprocessing process of acquisition is completed, establishes a dynamic characteristic modeling stage of the power distribution network; and a power distribution network fault characteristic diagnosis module which decomposes the change law of the power distribution network operation state in the dynamic characteristic modeling stage of the power distribution network established, and extracts the power distribution network fault characteristics to complete diagnosis.

[0013] Compared with the prior art, the application significantly improves the accuracy and efficiency of power distribution network fault diagnosis through innovative technical means. Firstly, in the data acquisition and preprocessing stage, a hierarchical architecture combined with a distributed sensor network and an adaptive learning mechanism is adopted, which not only realizes comprehensive acquisition and efficient screening of multi-source heterogeneous data such as voltage, current and temperature, but also enhances the identification ability of abnormal data, and ensures the integrity of the time dimension information of the data through the time stamp marking mechanism. Secondly, in the dynamic characteristic modeling stage, the application introduces a multi-scale decomposition method (including low-frequency trend layer, medium-frequency fluctuation layer and high-frequency disturbance layer), uses advanced technologies such as sliding window exponential smoothing algorithm, empirical mode decomposition (EMD) and variational mode decomposition (VMD), accurately extracts and analyzes the dynamic characteristics in different frequency ranges, thereby comprehensively capturing the change law of the power distribution network operation state. In addition, by constructing a multi-dimensional feature matrix and applying infrared thermal imaging technology, the accuracy and reliability of fault characteristic extraction and diagnosis are further improved. In terms of prediction and early warning mechanism, the application uses joint threshold setting and time series prediction model based on ARIMA or LSTM, which can discover potential risks in advance and trigger alarms in time. At the same time, by dynamically adjusting the sliding window length, the response speed and monitoring accuracy are optimized. Finally, the modular design supports the high expansibility of the system, and the millisecond-level timestamp accuracy ensures the real-time and accuracy under the high-frequency monitoring requirement. In summary, the application has significant advantages in improving the comprehensiveness, accuracy and response speed of power distribution network fault diagnosis, and provides strong technical support for the safe and stable operation of power distribution network. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings described below are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor. In the drawings: Figure 1 The specific flow chart of the power distribution network fault diagnosis method for distributed energy sources according to the present application. DETAILED DESCRIPTION

[0015] In the following, the example embodiments according to the present application will be described in detail with reference to the drawings. Obviously, the described embodiments are only some of the embodiments of the present application, not all embodiments of the present application, and it should be understood that the present application is not limited to the example embodiments described here.

[0016] As described in the above background, with the rapid development of society, distributed energy is increasingly widely used in power distribution networks, which marks the transition of energy management from traditional centralized mode to more flexible and efficient distributed mode. Under this background, the demand for processing energy data in the fields of administration, business, finance and management is increasingly urgent. In order to realize effective supervision and prediction of distributed energy, it is necessary to build a data processing system or method specially applicable to the above fields. The system not only can monitor the running state of power distribution network in real time, but also can support decision optimization through data analysis, so as to improve energy utilization efficiency and system reliability. Therefore, under the current social development trend, it is particularly important to transform power distribution network fault diagnosis into a data-driven management mode.

[0017] However, there are some cold but key defects in the prior art, which limit the accuracy and efficiency of power distribution network fault diagnosis. For example, most of the current fault diagnosis methods rely on traditional signal analysis technology, ignoring the influence of distributed energy dynamic characteristics on data quality. Especially in the aspect of multi-source heterogeneous data fusion, the existing methods often lack effective processing capability for unstructured data (such as sensor noise, environmental interference, etc.), which may cause deviation in the diagnosis result. In addition, the traditional algorithm is difficult to quickly locate the fault point and accurately classify when facing complex network topology, which greatly reduces the prediction performance of the system. In view of these rare technical bottlenecks, it is urgent to develop a new type of data processing method to improve the intelligent level of power distribution network fault diagnosis.

[0018] In the embodiment of the present application, S1: collecting multi-source heterogeneous data of distributed energy, using a hierarchical data collection preprocessing architecture to layer the collection process and complete the preprocessing process of collection.

[0019] The data collection preprocessing architecture includes a bottom layer architecture, a middle layer architecture and a top layer architecture. The bottom layer architecture includes collecting physical quantities such as voltage, current and temperature in the power distribution network by using a distributed sensor network. The middle layer architecture includes an anomaly detection mechanism using an adaptive learning mechanism to filter the raw data of the power grid data that has not been diagnosed. The top layer architecture includes converting the filtered raw data into a unified format through a feature extraction operation and establishing time dimension information for the unified format raw data by introducing a timestamp marking mechanism.

[0020] Filtering the raw data of the power grid data that has not been diagnosed includes using a sliding window technology to analyze the raw data in real time, calculating statistical indicators (such as mean, standard deviation) in the current time period, and dynamically generating a threshold range for anomaly detection according to the calculated statistical indicators.

[0021] According to the calculated statistical indicators, the upper and lower threshold ranges for anomaly detection are defined as: Upper threshold: ; Lower threshold: ; in, represents the mean, represents the standard deviation, Indicates the multiple of the standard deviation used to define how far a data point of the power grid deviates from the mean and whether it is considered abnormal.

[0022] Furthermore, calculating the statistical indicators within the current time period includes defining the time of the sliding window (such as the data of the past 5 minutes or 30 minutes), extracting data fragments from the physical quantities of voltage, current and temperature collected by the distributed sensor network, performing statistical analysis on the data within the time window, and calculating the voltage data. Time window data set ; Calculate the mean: , in, represents the mean, Indicates the number of samples of time window data, Indicates the data point number that defines the sliding window time, Indicates the specific value of the data point number; Calculate the standard deviation: , in, represents the standard deviation, The square of the deviation between the specific value of the data point number and the mean.

[0023] Preferably, according to the calculated statistical indicators, when defining the upper and lower threshold range of anomaly detection, first determine the time length of a sliding window, for example, the data of the past 5 minutes or 30 minutes, in this time window, extract data segments from the voltage, current and temperature data collected by the distributed sensor network, and perform statistical analysis on the data in this time period. Calculate the average value of all data points in this period as the baseline level, then calculate the average value of the square deviation between each data point and the average value, and take the square root to get the standard deviation, which measures the dispersion of the data. Based on these statistical results, set a reasonable standard deviation multiple to define the upper and lower threshold range to identify which data points deviate from the average to the extent that they can be considered abnormal. Specifically, if the value of a certain data point is higher than the average value plus several times the standard deviation, or lower than the average value minus the same multiple of the standard deviation, the data point is marked as abnormal. This dynamically generated threshold range can be self-adjusted according to the fluctuations of the actual data, ensuring that both real abnormal situations and false positives caused by normal fluctuations can be captured, so that by monitoring and analyzing key physical quantities in the power grid in real time, potential problems can be effectively screened out, providing reliable basis for subsequent fault diagnosis and preventive measures.

[0024] In the embodiment of the application, S1.1: converting the screened raw data into a unified format includes normalizing the numerical values of the physical quantities of voltage, current and temperature into the same dimension range, and concatenating the feature vectors of voltage, current and temperature in time sequence to form a multi-dimensional feature matrix; Temperature includes introducing a sliding time window technique to analyze the temperature data distribution in the current time period, deploying a distributed temperature sensor network at the nodes of the distribution network, and the distributed temperature sensor network includes using infrared temperature measurement technology to obtain and monitor the surface temperature of the distribution network.

[0025] In the embodiment of the application, S1.2: the surface temperature of the distribution network includes internal temperature and external temperature, the internal temperature includes measuring the transformer winding temperature of the distributed temperature sensor through the embedded sensor; the external temperature includes collecting the temperature changes of the environment around the distributed temperature sensor using the outdoor environmental temperature sensor; Monitoring includes non-contact temperature monitoring of the running state of the embedded sensor and the environmental temperature sensor according to the infrared thermal imaging technology, capturing the temperature data of the local hot spot area; Concatenating the feature vectors of voltage, current and temperature in time sequence includes analyzing the temperature data distribution in the current time period according to the sliding time window technique, dynamically updating and , wherein, represents the maximum value of the temperature data, represents the minimum value of the temperature data; The temperature data distribution includes smoothness, periodicity and abruptness, and the abnormal values of different temperature data distributions are identified according to the time sequence characteristics of the temperature data.

[0026] Further, a temperature data normalization formula for capturing local hot spot areas is defined, mapping is performed on the normalized data, and an adaptive learning dynamic normalization mapping interval adjustment mechanism is established. The temperature data normalization formula is: ; wherein, represents the normalized temperature value, represents the original temperature data, represents the minimum value of the temperature data, represents the maximum value of the temperature data; and By historical data statistics, a fixed range of -20° to 150° is set, and the normalized temperature value range is [0, 1]; Mapping the temperature data to the temperature value range [0, 1] is: ; wherein, represents the mapped normalized data.

[0027] When the temperature data distribution is periodic, the temperature data refers to the continuous and slow trend of temperature change in a short time without sharp fluctuations, which usually occurs when the device runs stably and the environmental conditions change little.

[0028] The abnormal value identification of smoothness is based on the difference analysis of the sliding window, a short time window (such as 5 minutes or 10 minutes) is defined, the difference between adjacent data points in the window is calculated, and when the difference of a certain point significantly deviates from the average difference in the window (for example, more than 3 times the standard deviation), it is marked as an abnormal value. An adaptive difference threshold mechanism is introduced, which dynamically adjusts the tolerance range of the difference according to historical data to adapt to different device running states.

[0029] When the temperature data distribution is smooth, the temperature change presents a repeated regular pattern, which is usually related to external environment, such as diurnal temperature difference, seasonal change or device running cycle.

[0030] Based on the frequency domain analysis of Fourier transform, the temperature time series data is converted to frequency domain, and the main periodic component is extracted, when the energy distribution of a certain data point deviates significantly from the main periodic component, such as frequency shift or amplitude anomaly, it is marked as an abnormal value, and sparse Fourier transform is introduced to reduce the calculation complexity while preserving high frequency abnormal information.

[0031] When the temperature data distribution is in the mutation, the mutation of the temperature data refers to that the temperature changes sharply in a short time, which is usually related to equipment failure, environmental interference or operation error.

[0032] Based on the multi-scale transform analysis, the temperature time series data is decomposed into detail information and trend information on different time scales to capture the mutation characteristics in the data.

[0033] In the embodiment of the present application, S1.3: establishing time dimension information for the unified format raw data includes capturing abnormal values of different temperature data distributions, adding the abnormal values to the organized time series format, and designing a sliding time window adjustment strategy based on a feedback mechanism according to the fluctuation characteristics of the power distribution network operation state. When the power distribution network load is detected to be increased, the time length of the sliding window is shortened to improve the time response speed; otherwise, the time length of the sliding window is lengthened to reduce the time response speed.

[0034] Further, the feature vector of each abnormal value is extended as , wherein, represents the extracted feature value, represents the time stamp, the time stamp adopts the international standard time format ISO8601, supports millisecond level precision, ensures that the time resolution meets the high frequency monitoring demand, and organizes the data with time stamp mark into time series format.

[0035] In the embodiment of the present application, S2: after completing the preprocessing process of the collection, a dynamic characteristic modeling stage of the power distribution network is established.

[0036] The dynamic characteristic modeling stage of the power distribution network includes using a data hierarchical analysis method of the time dimension information and the data collection preprocessing architecture to construct a power distribution network dynamic characteristic framework reflecting the change law of the power distribution network operation state; The data hierarchical analysis method includes layering and analyzing the preprocessed multi-source heterogeneous data according to the time scale, extracting dynamic characteristics at different levels, and the dynamic characteristics include low frequency trend layer, medium frequency fluctuation layer and high frequency disturbance layer; The low frequency trend layer includes using an exponential smoothing algorithm of a sliding window to extract the trend components of the physical quantities of voltage, current and temperature in the low frequency trend layer of the bottom layer architecture, and capture the change law of the power distribution network operation state; The medium frequency fluctuation layer includes using an empirical mode decomposition technology to decompose the multi-source heterogeneous data into N intrinsic mode functions, and capture the change law of the power distribution network operation state; The high frequency disturbance layer includes using a time resolution analysis method to determine the running state disturbance components of the power distribution network, and capture the change law of the power distribution network operation state.

[0037] Furthermore, the high-frequency disturbance layer reflects the characteristics of drastic changes that occur in the distribution network in a short period of time. These characteristics are usually closely related to equipment failures and external interference such as lightning strikes or load switching. By accurately extracting and analyzing the high-frequency disturbance components, it can provide key basis for early fault diagnosis and operating status assessment.

[0038] When capturing the changing laws of the distribution network's operating status based on the low-frequency trend layer, the sliding window parameters are first defined and initialized, and the smoothing parameters of the initialized sliding window parameters are calculated to capture the changing laws of the operating status.

[0039] Defining sliding window parameters means dynamically adjusting the length of the sliding window based on the operating characteristics of the distribution network and the data sampling frequency. In high-frequency sampling scenarios (such as once per second), the window length can be set to 30 minutes; in low-frequency sampling scenarios (such as once per minute), the window length can be set to 1 hour. In order to improve capture accuracy, the overlap ratio between adjacent windows is set (such as 50%) to avoid information loss due to window switching.

[0040] Initializing the sliding window parameters means determining the smoothing parameters The initial value is set based on experience And gradually adjust through subsequent optimization, introduce adaptive mechanism, dynamically adjust based on the fluctuation characteristics of the data in the current window When the data fluctuates greatly, increase to increase the sensitivity to recent data; otherwise, to retain more historical information.

[0041] Smoothing parameters refer to the calculation of single exponential smoothing formula and double exponential smoothing formula: ; in, Indicates time The original observation value of Indicates time The single exponential smoothing value of represents the smoothing parameter, Indicates time Smoothing value at -1; ; in, Indicates time The double exponential smoothing value of Indicates time The original observation value of represents the smoothing parameter of the trend component, represents the horizontal component at the current moment, Indicates the trend component at the current moment, represents the front horizontal component at the current moment, Represents the previous trend component at the current moment.

[0042] Capturing the changing patterns of the distribution network's operating status involves trend feature extraction, anomaly detection, and prediction and early warning. First, trend feature extraction involves obtaining the time derivative of the smoothed value to reflect the rate of change of the trend, analyzing the periodic characteristics of the smoothed value, and identifying potential periodic change patterns. Secondly, a normal range threshold is defined (such as the mean plus or minus three standard deviations), and smoothed values ​​outside this range are marked as potential anomalies. In combination with an incremental learning mechanism, the threshold range is dynamically adjusted to adapt to changes in the operating status. Finally, based on the extracted low-frequency trend components and combined with a time series prediction model (such as ARIMA or LSTM), the operating status for a period of time in the future is predicted. If the prediction result approaches or exceeds the warning threshold, an alarm mechanism is triggered, prompting operation and maintenance personnel to take action. Joint thresholds are set based on the multidimensional feature matrix and the correlation between various physical quantities. When the voltage drops by more than 10% and the current rises by more than 20%, an alarm is triggered; When the temperature exceeds 80% of the rated value and lasts for more than 5 minutes, an alarm is triggered.

[0043] When capturing the changing laws of the distribution network's operating status based on the medium-frequency fluctuation layer, we first analyze the multi-source heterogeneous data, align the time stamps of the time series data of physical quantities such as voltage, current, and temperature to ensure the consistency of the time dimension of each data source, apply wavelet threshold denoising or low-pass filtering methods to remove high-frequency noise interference, retain the medium-frequency fluctuation characteristics, and normalize the data of each physical quantity to the same dimension range [0, 1]. Secondly, decompose the original time series data into Intrinsic mode functions and a residual term: ; in, Indicates the The intrinsic mode functions, represents the residual term, Represents the original time series data; Finally, the intermediate frequency fluctuation components are extracted and captured, the frequency characteristics of the IMF components are extracted, and the components reflecting the intermediate frequency fluctuations are screened out. Spectral analysis: Fast Fourier transform (FFT) is performed on each IMF component to calculate its spectral distribution. IMF components with a frequency range in the intermediate frequency band (such as 0.1 Hz to 10 Hz) are selected. Energy distribution is evaluated, the energy proportion of each IMF component is calculated, and components with significantly concentrated energy are retained. The screened IMF components are superimposed to generate an intermediate frequency fluctuation signal: ; in, represents a medium frequency fluctuation signal, represents the first intrinsic mode function, represents a set of IMF components in the medium frequency range, represents an index of the intrinsic mode function.

[0044] The last capture refers to identifying potential periodic patterns by autocorrelation analysis of the medium frequency fluctuation signal, applying the Hilbert transform to calculate the instantaneous frequency and amplitude, further characterizing the dynamic characteristics of the fluctuation, defining a normal range threshold (such as mean plus or minus 3 times the standard deviation), marking fluctuations outside the range as potential abnormal points, combining an incremental learning mechanism to dynamically adjust the threshold range to adapt to changes in operating state, based on the extracted medium frequency fluctuation signal, combining a time series prediction model to predict the operating state in the future period.

[0045] In the embodiment of the present application, S3: decompose the operating state change law of the power distribution network in the power distribution network dynamic characteristic modeling stage, extract the power distribution network fault characteristics to complete the diagnosis.

[0046] Among them, the decomposition of the operating state change law of the power distribution network in the power distribution network dynamic characteristic modeling stage includes empirical mode decomposition and variational mode decomposition, through the process of empirical mode decomposition and variational mode decomposition, the low-frequency trend layer of the power distribution network operating state change law component characteristics, the medium frequency fluctuation layer of the power distribution network operating state change law component characteristics and the high frequency disturbance layer of the power distribution network operating state change law component characteristics are extracted, and the power distribution network fault characteristic extraction is completed.

[0047] In the embodiment of the present application, the comparison between the present application and the prior art is shown in Table 1 as follows: Table 1 Comparison between the present application and the prior art Technical aspects Prior art means Invention means Data acquisition preprocessing architecture Traditional sensor networks, single data acquisition; lack of adaptive learning mechanism, low efficiency of anomaly detection, unclear time dimension information. Distributed sensor network, comprehensive coverage of voltage, current, temperature and other physical quantities, adaptive learning mechanism improves anomaly detection accuracy, introduces timestamp marking mechanism, enhances time dimension information Dynamic characteristic modeling stage Single frequency domain or time domain analysis method, cannot fully capture dynamic characteristics, prediction model is simple, difficult to deal with complex changes. Multi-scale decomposition, low-frequency trend layer, medium-frequency fluctuation layer, high-frequency disturbance layer, fully capture dynamic characteristics, combine empirical mode decomposition EMD and variational mode decomposition VMD, improve flexibility, sliding window exponential smoothing algorithm accurately extracts trend component Fault characteristic extraction and diagnosis Simple data processing, lack of multi-dimensional feature fusion, limited temperature monitoring means, weak hotspot identification ability Construct a multi-dimensional feature matrix for comprehensive analysis, use infrared thermal imaging technology to accurately monitor temperature hotspots, dynamically adjust the threshold range based on historical data to improve anomaly detection accuracy Prediction and early warning mechanism Fixed threshold setting, prone to false positives or false negatives, single prediction model, slow response speed Joint threshold setting, improve the effectiveness of the alarm mechanism, combine ARIMA or LSTM time series prediction model, discover potential risks in advance, dynamically adjust the sliding window length, optimize response speed and monitoring accuracy. System integration and implementation Low degree of system modularization, poor scalability, low monitoring frequency supported, insufficient real-time performance Modular design, easy to expand and maintain, support millisecond level timestamp accuracy, ensure data real-time performance and accuracy under high frequency monitoring requirements

[0048] Table 1 describes the data acquisition preprocessing architecture, the prior art relies on traditional sensor networks, while the present application adopts a distributed sensor network and introduces an adaptive learning mechanism and a timestamp marking mechanism. In the dynamic characteristic modeling stage, the prior art usually uses a single analysis method, while the present application can more comprehensively capture the dynamic characteristics of the power distribution network through multi-scale decomposition and the combination of multiple analysis methods. Fault characteristic extraction and diagnosis, the prior art is relatively simple in data processing and temperature monitoring, the present application improves the accuracy and comprehensiveness of diagnosis by constructing a multi-dimensional feature matrix and infrared thermal imaging technology. Prediction and early warning mechanism, the fixed threshold setting of the prior art is prone to false positives or false negatives, the present application improves the accuracy and response speed of prediction and early warning by combining threshold setting and dynamically adjusting the sliding window length. System integration and implementation, the modularity of the existing system is low and the supported monitoring frequency is low, the present application enhances the scalability and real-time performance of the system through modular design and high-frequency monitoring support.

[0049] In another embodiment, the technical means according to the present application solve the problems in the background art, in the specific embodiments of the present application, first, the distributed sensor network in the background technology is applied to collect voltage, current and temperature and other physical quantities in the power distribution network. These sensors are distributed on the key nodes of the power distribution network, ensuring the comprehensiveness and real-time performance of the data. Through this network, high-frequency and accurate data segments can be obtained, laying a solid foundation for subsequent analysis. In order to process these raw data, an adaptive learning mechanism is adopted for anomaly detection method, which filters the undiagnosed raw data. Specifically, a sliding window (such as data in the past 5 minutes or 30 minutes) is defined, and statistical indicators (such as mean and standard deviation) in the current time period are calculated. Based on these statistical indicators, the threshold range for anomaly detection is dynamically generated, i.e. the upper threshold and the lower threshold, which is used to identify potential abnormal data points. This method not only improves the accuracy of anomaly detection, but also reduces the possibility of false positives.

[0050] Next, in the dynamic characteristic modeling stage, the multi-source heterogeneous data is decomposed into multiple intrinsic mode functions (IMFs) using empirical mode decomposition (EMD) and variational mode decomposition (VMD) techniques, thereby capturing dynamic characteristics in different frequency bands. The trend component of voltage, current, and temperature is extracted by the exponential smoothing algorithm of the sliding window for the low-frequency trend layer, revealing the change law of the long-term operation state of the distribution network. The EMD technique is used for the medium-frequency fluctuation layer to decompose the data into multiple IMF components, and the components reflecting the medium-frequency fluctuations are selected through frequency spectrum analysis and energy distribution evaluation, further describing the periodic and quasi-periodic dynamic characteristics. The time resolution analysis method is combined to determine the sudden disturbance component for the high-frequency disturbance layer, especially the severe changes related to equipment failure or external interference. Through these multi-level analysis methods, we can comprehensively capture the change law of the operation state of the distribution network, providing strong support for subsequent fault diagnosis.

[0051] In the fault characteristic extraction and diagnosis link, the filtered original data is converted into a unified format, and a timestamp marking mechanism is introduced to establish time dimension information. Through normalization processing of physical quantities such as voltage, current, and temperature, a multi-dimensional feature matrix is formed. On this basis, infrared thermal imaging technology is used to monitor the surface temperature of the distribution network, especially the non-contact temperature monitoring of embedded sensors and environmental temperature sensors, to help capture temperature data in local hot areas. For these temperature data, a normalization formula is defined, and a self-adaptive learning dynamic normalization mapping interval adjustment mechanism is established to ensure the accuracy and consistency of temperature data. In addition, the temperature data distribution in the current time period is analyzed based on the sliding time window technique, and different types of abnormal values are identified, including smoothness, periodicity, and suddenness.

[0052] To achieve effective prediction and early warning, time series prediction models such as ARIMA and LSTM are combined to predict the operation state in the future period. Based on the extracted low-frequency trend component and medium-frequency fluctuation signal, a joint threshold is set. When the prediction result approaches or exceeds these thresholds, the alarm mechanism is triggered to remind the operation and maintenance personnel to take measures. At the same time, a sliding time window adjustment strategy based on feedback mechanism is designed to dynamically adjust the window length according to the load change of the distribution network, optimizing the response speed and monitoring accuracy.

[0053] Finally, through the modular design, the system is easy to expand and maintain, supports millisecond level timestamp precision, meets the real-time and accuracy requirements under high frequency monitoring requirements, the whole technical scheme not only makes full use of the advantages of the background technology, but also realizes innovation and improvement in multiple key technical links, significantly improves the comprehensiveness, accuracy and response speed of power distribution network fault diagnosis, provides strong technical support for the safe and stable operation of power distribution network, and this whole process solution from data acquisition to preprocessing, dynamic characteristic modeling and fault diagnosis shows the great potential and value of the application in practical application.

[0054] In summary, the application significantly improves the accuracy and efficiency of power distribution network fault diagnosis through innovative technical means. Firstly, in the data acquisition and preprocessing stage, the hierarchical architecture combined with distributed sensor network and adaptive learning mechanism is adopted, which not only realizes the comprehensive acquisition and efficient screening of multi-source heterogeneous data such as voltage, current and temperature, but also enhances the identification ability of abnormal data, and ensures the integrity of time dimension information of data through timestamp marking mechanism. Secondly, in the dynamic characteristic modeling stage, the multi-scale decomposition method (including low-frequency trend layer, medium-frequency fluctuation layer and high-frequency disturbance layer) is introduced, and advanced technologies such as sliding window exponential smoothing algorithm, empirical mode decomposition (EMD) and variational mode decomposition (VMD) are used to accurately extract and analyze the dynamic characteristics in different frequency ranges, so as to fully capture the change law of power distribution network operation state. In addition, by constructing multi-dimensional feature matrix and applying infrared thermal imaging technology, the accuracy and reliability of fault characteristic extraction and diagnosis are further improved. In terms of prediction and early warning mechanism, the application adopts joint threshold setting and time series prediction model based on ARIMA or LSTM, which can discover potential risks in advance and trigger alarm in time. At the same time, by dynamically adjusting the sliding window length, the response speed and monitoring accuracy are optimized. Finally, the modular design supports the high expansibility of the system, and the millisecond level timestamp precision ensures the real-time and accuracy under high frequency monitoring requirements. In summary, the application has significant advantages in improving the comprehensiveness, accuracy and response speed of power distribution network fault diagnosis, and provides strong technical support for the safe and stable operation of power distribution network.

Claims

1. A method for fault diagnosis of a distribution network for distributed energy, characterized in that: include: Collect multi-source heterogeneous data of distributed energy, and use a layered data collection and pre-processing architecture to layer the collection process and complete the collection pre-processing process; After completing the pre-processing process of the acquisition, the dynamic characteristics modeling stage of the distribution network is established; The changing rules of the distribution network's operating status in the established distribution network dynamic characteristic modeling phase are decomposed, and the distribution network fault characteristics are extracted to complete the diagnosis.

2. The method for fault diagnosis of a distributed energy distribution network according to claim 1, characterized in that: The data acquisition and preprocessing architecture includes a bottom-level architecture, a middle-level architecture, and a top-level architecture. The bottom-level architecture utilizes a distributed sensor network to collect physical quantities of voltage, current, and temperature in the power distribution network. The middle-level architecture includes an anomaly detection mechanism that uses an adaptive learning mechanism to filter undiagnosed raw data from the power grid. The top-level architecture converts the filtered raw data into a unified format through a feature extraction operation and introduces a timestamp marking mechanism to establish time dimension information for the unified format raw data. The screening of undiagnosed raw data of the power grid data includes using sliding window technology to perform real-time analysis on the raw data, calculating statistical indicators within the current time period, dynamically generating a threshold range for anomaly detection based on the calculated statistical indicators, and defining upper and lower threshold ranges for anomaly detection based on the calculated statistical indicators.

3. The method for fault diagnosis of a distributed energy distribution network according to claim 2, characterized in that: The converting of the filtered raw data into a unified format includes normalizing the values ​​of the physical quantities of voltage, current, and temperature to the same dimensional range, and concatenating the characteristic vectors of voltage, current, and temperature in chronological order to form a multi-dimensional characteristic matrix; The temperature includes introducing sliding time window technology, analyzing the temperature data distribution in the current time period, deploying a distributed temperature sensor network at the nodes of the distribution network, and the distributed temperature sensor network includes using infrared temperature measurement technology to obtain the surface temperature of the distribution network and monitor the surface temperature of the distribution network.

4. The method for fault diagnosis of a distributed energy distribution network according to claim 2, characterized in that: The surface temperature of the distribution network obtained by using infrared temperature measurement technology includes internal temperature and external temperature. The internal temperature includes the transformer winding temperature of the distributed temperature sensor measured by an embedded sensor; the external temperature includes the temperature change of the environment around the distributed temperature sensor collected by an outdoor ambient temperature sensor; Monitoring the surface temperature of the distribution network includes non-contact temperature monitoring of the operating status of embedded sensors and ambient temperature sensors using infrared thermal imaging technology to capture temperature data of local hot spots; The characteristic vectors of voltage, current and temperature are spliced ​​in time sequence, including analyzing the temperature data distribution in the current time period based on the sliding time window technology, and dynamically updating and The value of Indicates the maximum value of temperature data, Indicates the minimum value of temperature data; The temperature data distribution includes smoothness, periodicity and mutation, and outliers of different temperature data distributions are identified according to the time series characteristics of the temperature data.

5. The method for diagnosing faults in a distribution network for distributed energy according to claim 4, characterized in that: The establishment of time dimension information for the unified format raw data includes capturing abnormal values ​​of different temperature data distributions, adding the abnormal values ​​to the organized time series format, and designing a sliding time window adjustment strategy based on a feedback mechanism according to the fluctuation characteristics of the distribution network operation state; When the load of the distribution network is detected to increase, the time length of the sliding window is shortened to improve the time response speed; Otherwise, the time length of the sliding window is extended to reduce the time response speed.

6. The method for fault diagnosis of a distributed energy distribution network according to claim 5, characterized in that: The stage of establishing the dynamic characteristics modeling of the distribution network includes using the data layered analysis method of time dimension information and data acquisition preprocessing architecture to construct a distribution network dynamic characteristics framework that reflects the changing laws of the distribution network operation status; The data layered analysis method includes layering and parsing the pre-processed multi-source heterogeneous data according to the time scale to extract dynamic features at different levels, including a low-frequency trend layer, a medium-frequency fluctuation layer, and a high-frequency disturbance layer; The low-frequency trend layer uses an exponential smoothing algorithm with a sliding window to extract the trend components of the voltage, current and temperature of the low-frequency trend layer of the underlying architecture, and capture the changing laws of the operating status of the distribution network; The medium frequency fluctuation layer includes the use of empirical mode decomposition technology to decompose multi-source heterogeneous data into N inherent mode functions to capture the changing laws of the operating status of the distribution network; The high-frequency disturbance layer includes using a time resolution analysis method to determine the sudden disturbance component of the operation state of the distribution network and capture the change law of the operation state of the distribution network.

7. The method for diagnosing faults in a distribution network for distributed energy according to claim 6, characterized in that: The decomposition of the distribution network's operating state change law in the dynamic characteristic modeling stage of the distribution network is performed, including empirical mode decomposition and variational mode decomposition. Through the empirical mode decomposition and variational mode decomposition processes, the distribution network operating state change law component characteristics of the low-frequency trend layer, the distribution network operating state change law component characteristics of the medium-frequency fluctuation layer, and the distribution network operating state change law component characteristics of the high-frequency disturbance layer are extracted to complete the distribution network fault characteristic extraction.

8. A distribution network fault diagnosis system for distributed energy, characterized in that: include: The acquisition layering module uses a layered data acquisition preprocessing architecture to layer the acquisition process and complete the acquisition preprocessing process; The dynamic characteristics modeling phase module establishes the dynamic characteristics modeling phase of the distribution network after completing the pre-processing process of the acquisition; The distribution network fault characteristic diagnosis module decomposes the change law of the distribution network's operating state in the established distribution network dynamic characteristic modeling stage, extracts the distribution network fault characteristics and completes the diagnosis.

9. A computer-readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the distributed energy distribution network fault diagnosis method according to any one of claims 1 to 7 is implemented.

Images