An electrical power distribution cabinet fault diagnosis method based on big data analysis

Abstract

The application discloses a kind of electrical distribution cabinet fault diagnosis methods based on big data analysis, comprising the following steps: step one: acquisition and preprocessing from the multi-source sensor data of electrical distribution cabinet;Step two: extract time domain features from sensor data set by time domain analysis method;Step three: the time domain signal in time domain feature set is converted into frequency domain signal by fast fourier transform, and each frequency domain feature is calculated;Step four: correlation analysis is carried out to time domain feature set and frequency domain feature set by pearson correlation coefficient;Step five: based on time domain frequency domain feature set, the improved HDP-HMM algorithm is used to identify the hidden state set of electrical distribution cabinet, and the optimal hidden state sequence is obtained by hidden state sequence inference;Step six: according to the optimal hidden state sequence, generate fault report and maintenance suggestion.The application combines time domain frequency domain features with improved HDP-HMM algorithm, realizes electrical distribution cabinet fault intelligent diagnosis and prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, and in particular to an electrical power distribution cabinet fault diagnosis method based on big data analysis. BACKGROUND

[0002] The electrical power distribution cabinet is an important part of the power system, used for receiving and distributing power. The reliability and stability of the power distribution cabinet are crucial to the safety of the power system. However, due to the influence of factors such as temperature, humidity, current, voltage, etc. during long-term operation of the power distribution cabinet, equipment may fail, causing power supply interruption or equipment damage, and even larger-scale power accidents. Therefore, the fault diagnosis and early warning of the electrical power distribution cabinet becomes particularly important.

[0003] Traditional power distribution cabinet fault diagnosis methods mainly rely on manual inspection and regular maintenance. Although this method can find abnormal conditions of equipment to some extent, it lacks real-time and systematicity and cannot timely warn potential faults. Traditional methods often rely on the experience of operators, which may lead to missed faults or inaccurate diagnosis, especially in complex equipment or large environmental changes. With the continuous expansion of the power distribution system, traditional manual inspection and regular inspection are increasingly unable to meet the efficient operation needs of modern power systems, and an intelligent and real-time fault diagnosis solution is urgently needed.

[0004] In recent years, with the advancement of sensor technology, various sensors have been applied to the state monitoring of electrical power distribution cabinets. These sensors can collect real-time data such as current, voltage, temperature, humidity, vibration, etc. in multiple dimensions, providing rich monitoring information. However, how to effectively extract effective features from massive sensor data and conduct intelligent diagnosis is still a technical problem. Although existing fault diagnosis methods can use sensor data for fault detection, they mostly rely on traditional statistical analysis or rule-based models. These methods often face difficulties in feature selection, poor model adaptability, etc. when dealing with complex and variable data, resulting in insufficient accuracy and reliability of fault detection. Current machine learning-based fault diagnosis methods have made some progress, but most existing technologies only rely on single time-domain or frequency-domain features for analysis, lacking comprehensive analysis of time-domain and frequency-domain features, resulting in poor diagnosis effect under complex fault patterns. Many methods have not been able to effectively combine time-domain and frequency-domain features, thus failing to fully exploit the potential of big data.

[0005] Therefore, how to provide an electrical power distribution cabinet fault diagnosis method based on big data analysis is a problem that needs to be solved by those skilled in the art. SUMMARY

[0006] One purpose of the present application is to provide an electrical power distribution cabinet fault diagnosis method based on big data analysis, which realizes intelligent diagnosis and prediction of electrical power distribution cabinet faults by combining time domain and frequency domain features with an improved HDP-HMM algorithm. This method automatically extracts effective features, removes redundant information, and improves diagnostic accuracy. The improved HDP-HMM algorithm dynamically adjusts the number of hidden states, improves the adaptability of the model, can accurately identify the current fault mode and predict future faults, provides detailed fault reports and maintenance recommendations, effectively reduces the probability of sudden failures, optimizes the maintenance and management strategies of the power system, and improves the stability and safety of the power distribution system.

[0007] According to an electrical power distribution cabinet fault diagnosis method based on big data analysis, the method comprises the following steps:

[0008] Step one: collect and preprocess multi-source sensor data from the electrical power distribution cabinet to obtain a unified structure sensor data set;

[0009] Step two: extract time domain features from the sensor data set by a time domain analysis method to obtain a time domain feature set;

[0010] Step three: convert the time domain signals in the time domain feature set into frequency domain signals by fast Fourier transform, and calculate each frequency domain feature to obtain a frequency domain feature set;

[0011] Step four: perform correlation analysis on the time domain feature set and the frequency domain feature set by Pearson correlation coefficient to obtain a time domain and frequency domain feature set;

[0012] Step five: identify the hidden state set of the electrical power distribution cabinet based on the time domain and frequency domain feature set by an improved HDP-HMM algorithm, calculate the hidden state transition probability, and obtain the optimal hidden state sequence through hidden state sequence inference;

[0013] Step six: generate a fault report and maintenance recommendations according to the optimal hidden state sequence.

[0014] Optionally, the collection is performed at a predetermined time interval; the multi-source sensor data includes current data, voltage data, temperature data, humidity data and vibration data; the preprocessing includes outlier rejection, missing value filling, timestamp alignment and normalization processing for different types of data respectively to obtain a unified structure sensor data set.

[0015] Optionally, the time domain analysis method specifically comprises:

[0016] The original signal in the sensor data set is subjected to a difference processing to obtain a first difference signal, and the difference processing is to calculate the difference between each data point and its previous data point in the original signal;

[0017] checking a difference between a maximum value and a minimum value of the signal in the once-differential signal, and if the difference is greater than or equal to a preset difference threshold, considering that a trend of the current once-differential signal is not eliminated;

[0018] performing a twice-differential processing on the once-differential signal to obtain a twice-differential signal, the twice-differential processing being a differential processing on the once-differential signal again;

[0019] repeating the iteration until the difference between the maximum value and the minimum value of the signal in the differential signal is less than the preset difference threshold, to obtain the smoothed signal data;

[0020] segmenting the smoothed signal data according to a preset size of a time window;

[0021] calculating a time domain feature in each time window to obtain a time domain feature set, the time domain feature including a time domain signal mean value, a standard deviation, a maximum value, a minimum value, a peak factor and a fluctuation rate;

[0022] the time domain signal mean value is used to represent a center position of the signal;

[0023] the time domain signal standard deviation is used to represent a fluctuation amplitude of the signal;

[0024] the time domain signal maximum value and the minimum value are used to represent extreme fluctuations of the signal;

[0025] the time domain signal peak factor is a ratio of the signal maximum value to the signal mean value, and is used to represent a sharp degree of the signal;

[0026] the time domain signal fluctuation rate is a ratio of the signal standard deviation to the signal mean value, and is used to represent a fluctuation intensity of the signal.

[0027] Optionally, the step three specifically includes:

[0028] performing a fast Fourier transform on each time domain signal in the time domain feature set to convert the time domain signal from a time domain representation to a frequency domain representation;

[0029] the fast Fourier transform converts the time domain signal into amplitude and phase information of each frequency component to obtain a frequency domain signal;

[0030] calculating a spectral amplitude of each frequency component in the frequency domain signal to obtain a spectral amplitude map, the spectral amplitude representing an energy distribution of the frequency domain signal on each frequency component;

[0031] finding a frequency component with a maximum spectral amplitude from the spectral amplitude map, and taking the frequency component as a main frequency component, the main frequency component representing a main vibration mode or an operating frequency of the electrical power distribution cabinet;

[0032] a frequency bandwidth of each frequency component, the frequency bandwidth being a range occupied by the frequency component on the spectrum amplitude map;

[0033] The frequency domain features include a spectrum amplitude, a main frequency component, and a frequency bandwidth.

[0034] Optionally, the correlation analysis specifically includes:

[0035] Each time domain feature in the time domain feature set and each frequency domain feature in the frequency domain feature set are paired with each other to calculate a Pearson correlation coefficient;

[0036] The Pearson correlation coefficient is an index of a linear relationship between the time domain feature and the frequency domain feature, and the Pearson correlation coefficient value ranges from -1 to 1, wherein 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents no linear correlation;

[0037] If the Pearson correlation coefficient between the time domain feature and the frequency domain feature is greater than or equal to a set linear correlation threshold, it is considered that the two are highly correlated, and one redundant feature is removed;

[0038] If the Pearson correlation coefficient is less than the set linear correlation threshold, the corresponding time domain feature and the frequency domain feature are retained;

[0039] The iteration comparison is repeated until all time domain features and frequency domain features are compared, and a time domain and frequency domain feature set is obtained.

[0040] Optionally, the improved HDP-HMM algorithm specifically includes:

[0041] A time sequence data of a preset size is randomly selected from the time domain and frequency domain feature set as a first observation sequence;

[0042] Corresponding time domain features and frequency domain features in the first observation sequence are extracted, and a first observation vector is obtained by arranging the time domain features and the frequency domain features in time sequence;

[0043] The first observation vector is taken as a first hidden state, and time sequence data of the same size is selected from the time domain and frequency domain feature set in time sequence to obtain a second observation vector;

[0044] An Euclidean distance between the first observation vector and the second observation vector is calculated;

[0045] If the Euclidean distance is greater than or equal to a preset distance threshold, the second observation vector is taken as a second hidden state;

[0046] If the Euclidean distance is less than the preset distance threshold, the second observation vector is classified into the first hidden state;

[0047] The iteration is repeated until the time domain and frequency domain feature set is traversed completely, and a hidden state set is obtained, each hidden state in the hidden state set corresponding to a failure mode of the electrical power distribution cabinet;

[0048] A hidden state space is constructed according to the hidden state set, and the hidden state space is a failure mode space;

[0049] An observation probability distribution of each hidden state is obtained based on a Gaussian distribution of an observation vector in the hidden state space;

[0050] A hidden state transition probability is obtained by calculating a corresponding conditional probability according to the observation probability distribution of each hidden state, and the hidden state transition probability is a transition probability between each pair of hidden states;

[0051] A best hidden state sequence is obtained by hidden state sequence inference according to the transition probability of each pair of hidden states;

[0052] The best hidden state sequence is a most possible failure mode sequence of the electrical power distribution cabinet in an entire observation time period.

[0053] Optionally, the hidden state sequence inference according to the transition probability of each pair of hidden states to obtain the best hidden state sequence is specifically:

[0054] A hidden state transition probability matrix is constructed according to the transition probability of each pair of hidden states;

[0055] The hidden state transition probability from an initial hidden state to each hidden state is traversed, and a hidden state with a maximum hidden state transition probability is selected as a target hidden state for next transition;

[0056] The initial hidden state is a hidden state in a left upper corner of the hidden state transition probability matrix;

[0057] For each time point, a maximum hidden state transition probability of each time point is selected step by step according to an observation probability of a current hidden state and the hidden state transition probability, until a last time point;

[0058] The best hidden state sequence is obtained by backtracking from the last time point to the initial time point in the hidden state transition probability matrix.

[0059] Optionally, the step six is specifically:

[0060] The failure report includes a failure mode, an occurrence time range, an influence range and a device damage degree, and is specifically:

[0061] A failure mode of the electrical power distribution cabinet corresponding to a hidden state at each time point is determined according to the best hidden state sequence;

[0062] A possible occurrence time range of the failure mode is obtained by analyzing a start time and an end time of the hidden state in the hidden state sequence;

[0063] in combination with the failure mode and the device historical data, speculate the damage degree of the failure mode to the device;

[0064] generate a maintenance suggestion based on the failure report, the maintenance suggestion including performing load adjustment or load balancing and performing device repair or replacing aged components.

[0065] The beneficial effects of the present application are:

[0066] The electrical power distribution cabinet fault diagnosis method based on big data analysis provided by the present application overcomes multiple problems existing in the prior art by combining time domain features and frequency domain features and introducing an improved HDP-HMM algorithm, realizing the intelligentization, automation and high efficiency of electrical power distribution cabinet fault diagnosis. Compared with traditional rule-based or statistical analysis-based fault diagnosis methods, the present method has significant advantages. By automatically extracting time domain and frequency domain features and using Pearson correlation analysis to remove redundant features, the present method maximizes the retention of effective information and avoids the possibility of information loss or misjudgment, thereby improving the accuracy of diagnosis. Especially, the use of fast Fourier transform (FFT) to convert time domain features into frequency domain features further enriches the dimension of feature information, enabling more comprehensive analysis and identification of power distribution cabinet failure modes. In addition, the use of the improved HDP-HMM algorithm dynamically and adaptively optimizes the number of hidden states, avoiding the problem of fixed number of hidden states in traditional HMM algorithms and improving the adaptability and flexibility of the model when facing complex failure modes. The improved algorithm not only accurately identifies the current failure state, but also effectively predicts future failure modes and their occurrence times, thereby providing reliable data support for preventive maintenance of electrical power distribution cabinets. Most importantly, the present method can generate detailed failure reports based on failure modes and device historical operation data and provide maintenance suggestions based on failure types, helping maintenance personnel optimize equipment maintenance plans, reducing the probability of sudden failures and improving the operational stability and safety of power distribution systems. Overall, the present application not only realizes accurate diagnosis and effective prediction of electrical power distribution cabinet failures, but also greatly improves the intelligent management level of power systems through data-driven methods, having wide application prospects and practical significance. BRIEF DESCRIPTION OF DRAWINGS

[0067] The accompanying drawings are included to provide a further understanding of the present application, and constitute a part of the specification, illustrate embodiments of the present application and are used, together with the description, to explain the present application, and do not limit the present application. In the drawings:

[0068] Figure 1 The overall flowchart of the electrical power distribution cabinet fault diagnosis method based on big data analysis provided by the present application is shown in the figure;

[0069] Figure 2A fast Fourier transform process schematic diagram of an electrical power distribution cabinet fault diagnosis method based on big data analysis is provided for the present application.

[0070] Figure 3 An improved HDP-HMM algorithm step flow chart of an electrical power distribution cabinet fault diagnosis method based on big data analysis is provided for the present application. DETAILED DESCRIPTION

[0071] The present application will now be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, and only illustrate the basic structure of the present application in a schematic manner, and therefore only show the components related to the present application.

[0072] REFERENCE Figures 1-3 An electrical power distribution cabinet fault diagnosis method based on big data analysis, comprising the following steps:

[0073] Step one: collect and preprocess multi-source sensor data from the electrical power distribution cabinet to obtain a uniform structure sensor data set;

[0074] Step two: extract time domain features from the sensor data set through time domain analysis method to obtain a time domain feature set;

[0075] Step three: convert the time domain signal in the time domain feature set into a frequency domain signal through fast Fourier transform, and calculate each frequency domain feature to obtain a frequency domain feature set;

[0076] Step four: perform correlation analysis on the time domain feature set and the frequency domain feature set through Pearson correlation coefficient to obtain a time domain frequency domain feature set;

[0077] Step five: identify the hidden state set of the electrical power distribution cabinet based on the time domain frequency domain feature set through the improved HDP-HMM algorithm, calculate the hidden state transition probability, and obtain the optimal hidden state sequence through hidden state sequence inference;

[0078] Step six: generate a fault report and maintenance suggestion according to the optimal hidden state sequence.

[0079] This step effectively improves the accuracy and robustness of fault detection by combining time domain and frequency domain features. The method ensures the consistency and accuracy of the data through the collection and preprocessing of multi-source sensor data; adopts time domain analysis and fast Fourier transform to comprehensively extract signal features, enhancing the representativeness of feature information. Redundant features are removed through Pearson correlation analysis, effectively improving the effectiveness of the data. The improved HDP-HMM algorithm can dynamically identify hidden states and accurately infer fault patterns, further optimizing fault diagnosis. The fault report and maintenance suggestion generated based on the optimal hidden state sequence provide strong support for preventive maintenance and management decisions of the equipment, improving the operation efficiency and safety of the power distribution system.

[0080] In this embodiment, the collection is performed at predetermined time intervals; the multi-source sensor data includes current data, voltage data, temperature data, humidity data, and vibration data; the preprocessing includes outlier rejection, missing value filling, timestamp alignment, and normalization processing for different types of data, to obtain a unified structure of the sensor data set.

[0081] This step improves the reliability and consistency of the data by collecting and preprocessing the multi-source sensor data of the electrical distribution cabinet. The collected data covers multiple dimensions such as current, voltage, temperature, humidity, and vibration, comprehensively reflecting the running state of the equipment. By performing outlier rejection, missing value filling, timestamp alignment, and normalization processing on different types of data, the accuracy and standardization of the data are ensured, thus eliminating possible data bias and inconsistency. This preprocessing process provides a high-quality data basis for subsequent feature extraction and fault diagnosis, further enhancing the accuracy and operability of the diagnosis results, effectively improving the real-time and precision of the distribution cabinet fault monitoring.

[0082] In this embodiment, the time domain analysis method is specifically:

[0083] The original signal in the sensor data set is subjected to a first difference processing to obtain a first difference signal, and the difference processing is to calculate the difference between each data point and its previous data point in the original signal.

[0084] The difference between the maximum value and the minimum value of the first difference signal is checked, and if the difference is greater than or equal to a preset difference threshold, it is considered that the current first difference signal trend has not been eliminated.

[0085] The first difference signal is subjected to a second difference processing to obtain a second difference signal, and the second difference processing is to perform a first difference processing on the first difference signal.

[0086] The iteration is repeated until the difference between the maximum value and the minimum value of the difference signal is less than the preset difference threshold, to obtain a smoothed signal data.

[0087] The smoothed signal data is segmented by a time window of a preset size;

[0088] The time domain features in each time window are calculated to obtain a time domain feature set, and the time domain features include time domain signal mean, standard deviation, maximum value, minimum value, peak factor, and volatility.

[0089] The time domain signal mean is used to represent the center position of the signal.

[0090] The time domain signal standard deviation is used to represent the fluctuation amplitude of the signal.

[0091] The time domain signal maximum value and minimum value are used to represent the extreme fluctuation of the signal;

[0092] The time domain signal peak factor is the ratio of the signal maximum value to the signal mean value, and is used to represent the sharpness of the signal;

[0093] The time domain signal fluctuation rate is the ratio of the signal standard deviation to the signal mean value, and is used to represent the fluctuation intensity of the signal.

[0094] This step solves the trend component and noise influence in the signal through time domain analysis of the electrical power distribution cabinet sensor data, improves the accuracy and reliability of feature extraction. By performing difference processing on the original signal, the trend part in the signal is removed, making the signal more stable, thereby reducing the interference of the trend change on subsequent analysis. Further trend elimination is performed using quadratic difference processing, making the signal more stable and facilitating subsequent processing. By setting a preset difference threshold, it is ensured that the signal after each difference processing meets the stabilization standard, enhancing the quality of the signal data. On the basis of the stabilized signal, time window segmentation processing is performed to extract time domain features such as mean value, standard deviation, peak factor and fluctuation rate, providing accurate and reliable feature data for fault diagnosis. These time domain features effectively describe the fluctuation, fluctuation amplitude and sharpness of the signal, providing accurate basic information for electrical power distribution cabinet fault diagnosis, which helps to improve the diagnosis accuracy and support the accurate prediction of the early warning system.

[0095] In this embodiment, step three is specifically:

[0096] Performing fast Fourier transform on each time domain signal in the time domain feature set to convert the time domain signal from time domain representation to frequency domain representation;

[0097] The fast Fourier transform converts the time domain signal into amplitude and phase information of each frequency component to obtain a frequency domain signal;

[0098] Calculating the spectral amplitude of each frequency component in the frequency domain signal to obtain a spectral amplitude map, where the spectral amplitude represents the energy distribution of the frequency domain signal at each frequency component;

[0099] Finding the frequency component with the maximum spectral amplitude from the spectral amplitude map, and taking it as the main frequency component, which represents the main vibration mode or operating frequency of the electrical power distribution cabinet;

[0100] Calculating the frequency bandwidth of each frequency component according to the spectral amplitude map, where the frequency bandwidth is the range occupied by the frequency component on the spectral amplitude map;

[0101] The frequency domain features include spectral amplitude, main frequency component and frequency bandwidth.

[0102] The step converts the time-domain signal into a frequency-domain signal through fast Fourier transform (FFT), effectively revealing the frequency components in the signal and providing a deeper understanding of the electrical power distribution cabinet failure mode. By calculating the spectral amplitude, the energy distribution of the signal at each frequency component can be intuitively displayed, effectively identifying the main frequency components in the signal. The main frequency components reflect the main vibration mode or operating frequency of the electrical power distribution cabinet, which is of great significance for identifying periodic problems such as mechanical failure and power anomalies. The calculation of the frequency bandwidth helps to reveal the width of the frequency distribution of the signal, further analyzing the complexity or abnormality of the device operation. The extraction of frequency domain features provides key information for subsequent failure mode identification, making fault diagnosis more accurate and effectively improving the fault detection capability of the electrical power distribution cabinet, enhancing the stability and safety of the system. Through the analysis of frequency domain features, the fault source can be more accurately located, potential fault risks can be warned in advance, and the efficient operation of the power distribution system can be ensured.

[0103] In the embodiment, the correlation analysis is specifically:

[0104] Each time-domain feature in the time-domain feature set and each frequency-domain feature in the frequency-domain feature set are paired respectively, and the Pearson correlation coefficient is calculated;

[0105] The Pearson correlation coefficient is an index of the linear relationship between the time-domain features and the frequency-domain features, and the value of the Pearson correlation coefficient ranges from -1 to 1, where 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents no linear correlation;

[0106] If the Pearson correlation coefficient between the time-domain features and the frequency-domain features is greater than or equal to the set linear correlation threshold, it is considered that the two are highly correlated, and one redundant feature is removed;

[0107] If the Pearson correlation coefficient is less than the set linear correlation threshold, the corresponding time-domain feature and frequency-domain feature are retained;

[0108] Repeat the iteration comparison until all time-domain features and frequency-domain features are compared, and obtain the time-domain frequency-domain feature set.

[0109] The step effectively screens and optimizes the feature data by performing Pearson correlation analysis on the time domain feature set and the frequency domain feature set, and improves the accuracy of fault diagnosis. By calculating the Pearson correlation coefficient between the time domain features and the frequency domain features, the linear relationship between them can be quantified, so as to identify which features have redundant information. In the case of high correlation, redundant features are removed to avoid unnecessary feature interference and ensure the simplification and efficiency of the data set. If the correlation between the features is low, the relevant features are retained to ensure that as much useful information as possible is retained. This process compares all time domain and frequency domain features through iteration to finally obtain an optimized time domain and frequency domain feature set, which can better reflect the working state of the electrical power distribution cabinet. The optimization process effectively reduces feature redundancy and improves the accuracy of subsequent fault pattern recognition, and provides accurate basis for real-time monitoring and early warning of the electrical power distribution cabinet.

[0110] In the embodiment, the improved HDP-HMM algorithm is specifically:

[0111] Randomly select a time series data of a preset size from the time domain and frequency domain feature set as a first observation sequence;

[0112] Extract the corresponding time domain features and frequency domain features in the first observation sequence, and arrange them in time sequence to obtain a first observation vector;

[0113] Take the first observation vector as a first hidden state, and select time series data of the same size from the time domain and frequency domain feature set in time sequence to obtain a second observation vector;

[0114] Calculate the Euclidean distance between the first observation vector and the second observation vector;

[0115] If the Euclidean distance is greater than or equal to a preset distance threshold, the second observation vector is taken as a second hidden state;

[0116] If the Euclidean distance is less than the preset distance threshold, the second observation vector is classified as the first hidden state;

[0117] Repeat the iteration until the time domain and frequency domain feature set is completely traversed to obtain a hidden state set, each hidden state in the hidden state set corresponding to a fault mode of the electrical power distribution cabinet;

[0118] Construct a hidden state space according to the hidden state set, the hidden state space being a fault mode space;

[0119] Based on the observation vectors in the hidden state space, the observation probability distribution of each hidden state is obtained through Gaussian distribution;

[0120] According to the observation probability distribution of each hidden state, the hidden state transition probability is obtained by calculating the corresponding conditional probability, the hidden state transition probability being the transition probability between each pair of hidden states.

[0121] According to the transition probability of each pair of hidden states, the hidden state sequence is inferred to obtain an optimal hidden state sequence.

[0122] The optimal hidden state sequence is the most likely fault mode sequence of the electrical power distribution cabinet within the entire observation time period.

[0123] This step uses an improved HDP-HMM algorithm to adaptively identify the hidden states of the electrical power distribution cabinet, significantly improving the diagnostic accuracy of the fault mode. By selecting time series data from the time and frequency domain feature set and calculating the Euclidean distance, this method can accurately divide the hidden states, ensuring that each hidden state corresponds to the actual fault mode of the electrical power distribution cabinet. This method can automatically eliminate redundant data and optimize the construction of the hidden state set, thereby improving the efficiency and accuracy of the diagnostic system. The observation probability distribution is calculated using a Gaussian distribution, and the optimal hidden state sequence is derived from the hidden state transition probability, ensuring accurate inference of the fault mode. The final optimal hidden state sequence provides a dynamic evolution process of the fault mode of the power distribution cabinet over the entire time period, making fault prediction and maintenance planning more scientific and accurate. This improved HDP-HMM algorithm has higher flexibility and adaptability, can handle complex fault modes and provide reliable early warning, significantly improving the fault diagnosis capability and system stability of the electrical power distribution cabinet.

[0124] In this embodiment, the hidden state sequence is inferred according to the transition probability of each pair of hidden states to obtain an optimal hidden state sequence, which is specifically:

[0125] A hidden state transition probability matrix is constructed according to the transition probability of each pair of hidden states.

[0126] The hidden state transition probability from the initial hidden state to each hidden state is traversed, and the hidden state with the maximum hidden state transition probability is selected as the target hidden state for the next transition.

[0127] The initial hidden state is the top-left corner hidden state of the hidden state transition probability matrix.

[0128] For each time, the maximum hidden state transition probability of each time is gradually selected according to the observation probability and hidden state transition probability of the current hidden state until the last time.

[0129] Starting from the last time, the hidden state transition probability matrix is traced back to the initial time to obtain the optimal hidden state sequence.

[0130] The step significantly improves the accuracy and reliability of electrical power distribution cabinet fault diagnosis through the hidden state sequence inference method based on hidden state transition probability. By constructing a hidden state transition probability matrix, the transition rules between hidden states can be clearly described, providing a solid foundation for subsequent hidden state inference. By traversing each pair of hidden state transition probability and selecting the maximum probability path, the most likely hidden state at each time is ensured, gradually forming the optimal hidden state sequence. This method can accurately identify the fault mode of the power distribution cabinet, especially in complex equipment fault modes, with strong adaptability and flexibility. When backtracking the optimal hidden state path, the fault evolution process of the equipment is accurately restored, thereby improving the accuracy of fault prediction and maintenance. This method effectively avoids the dependence on a fixed number of hidden states in traditional fault diagnosis, can handle more diverse fault modes, and provides accurate data support for equipment maintenance decisions.

[0131] In this embodiment, step six is specifically:

[0132] The fault report includes fault mode, time range of occurrence, impact range, and degree of damage to the equipment, specifically:

[0133] According to the optimal hidden state sequence, determine the fault mode of the electrical power distribution cabinet corresponding to the hidden state at each time;

[0134] By analyzing the start and end times of the hidden states in the hidden state sequence, obtain the possible time range of occurrence of the fault mode;

[0135] Combine the fault mode and historical data of the equipment to infer the damage degree of the fault mode to the equipment;

[0136] Generate maintenance recommendations based on the fault report, including load adjustment or load balancing, and equipment maintenance or replacement of aging components.

[0137] This step greatly improves the accuracy and predictability of fault diagnosis by generating detailed fault reports based on the optimal hidden state sequence. By analyzing the start and end times of each hidden state in the hidden state sequence, the time range of occurrence of the fault mode can be accurately determined, providing an important basis for preventive maintenance of the equipment. At the same time, combined with the historical data of the equipment, the invention can infer the damage degree of the fault mode to the equipment, thereby providing accurate recommendations for maintenance. The generated maintenance recommendations include load adjustment, load balancing, equipment maintenance, or replacement of aging components, etc., helping maintenance personnel to reasonably arrange maintenance plans, optimize the maintenance and management strategy of the power distribution cabinet, and reduce the risk of equipment failure and system interruption. This method not only improves the stability of the equipment, but also effectively prolongs the service life of the equipment, and improves the safety and reliability of the power system.

[0138] Example 1:

[0139] To verify the feasibility of the application in practice, the application is applied to the power distribution system of a certain city, which is equipped with various sensors to monitor the operating status of the distribution cabinet in real time. The distribution cabinet is installed with current, voltage, temperature, humidity and vibration sensors, and through these sensor data, the system can monitor the working environment and operation of the distribution cabinet, and timely capture potential failure risks. In actual application, the multi-source sensor data of the distribution cabinet is collected, and the method of the application is used to process and analyze these data.

[0140] In the operation and maintenance of the distribution cabinet, current, voltage, temperature, humidity and vibration data are first collected, with a data collection frequency of once per minute. After data collection, the collected raw data is preprocessed, including outlier rejection, missing value filling, timestamp alignment and normalization processing. Outlier rejection uses a statistical method based on Z-score to remove data points exceeding 3 standard deviations; missing values are filled using the mean interpolation method to ensure data continuity and consistency. After these preprocessing, all sensor data is formatted uniformly and aligned by timestamp, and finally a unified structure of sensor data set is obtained.

[0141] The data in the time domain feature set is analyzed in the time domain, and a first difference and second difference processing method is adopted to remove the trend component in the original signal and obtain the smoothed signal data. These smoothed signal data are further segmented according to the preset time window, and time domain features such as mean, standard deviation, maximum, minimum, peak factor and volatility are calculated. These time domain features effectively reflect the different change patterns of the distribution cabinet under normal operation and failure state.

[0142] In the frequency domain analysis stage, each signal in the time domain feature set is subjected to fast Fourier transform (FFT) to convert the time domain signal to the frequency domain signal. By calculating the spectral amplitude and main frequency component, the main vibration mode and operating frequency of the electrical distribution cabinet are identified. In the spectral amplitude diagram, the frequency component with the largest spectral amplitude is found, and it is confirmed that this frequency component represents the normal operating frequency of the electrical distribution cabinet. Through the calculation of the frequency bandwidth, the signal frequency distribution of the electrical distribution cabinet is further understood. The extraction of frequency domain features provides an important basis for subsequent fault diagnosis.

[0143] Based on the time domain and frequency domain feature set, the correlation analysis is carried out through the Pearson correlation coefficient. By calculating the correlation between the time domain features and the frequency domain features, the redundant features are eliminated, and the optimized time domain and frequency domain feature set is obtained. Through these feature sets, the improved HDP-HMM algorithm is used for fault mode recognition and hidden state inference. Specifically, a segment of time series data with a predetermined size is randomly selected from the time domain and frequency domain feature set as the first observation sequence, the Euclidean distance is calculated, and through iterative calculation, the hidden state set is finally obtained, and the hidden state transition probability is calculated through the hidden state transition probability matrix.

[0144] During the application of the method of the application, the power distribution cabinet experiences an overload fault. Before the fault occurs, the system detects that the sensor data of current, temperature and vibration etc. fluctuate obviously, and through the combination of time domain and frequency domain features, the system predicts in advance that the equipment may have an overload fault. After the fault occurs, the system infers the fault mode of the power distribution cabinet through the hidden state transition probability, and generates a fault report which lists the fault type, the time range of occurrence, the influence range and the equipment damage in detail. According to the report, the system proposes maintenance suggestions of load adjustment and equipment repair, which effectively avoids further damage to the equipment.

[0145] To verify the effectiveness of the method of the application, the following is the comparison data of the method of the application and the traditional method in fault diagnosis.

[0146] Table 1 Comparison data table of fault diagnosis methods

[0147] Diagnostic method Diagnostic accuracy Diagnostic response time Fault prediction lead time Maintenance recommendation effectiveness Threshold-based fault diagnostic method 65% 72 hours None Low Support vector machine (SVM)-based classification method 85% 24 hours 12 hours Medium Rule-based method based on expert experience 75% 48 hours None Medium The method of the invention 95% 12 hours 36 hours High

[0148] As can be seen from Table 1, the method of the present application has significant advantages over the traditional three fault diagnosis methods in many aspects. First, in terms of diagnostic accuracy, the fault diagnosis method based on big data analysis reaches 95%, which is much higher than the other three methods. The traditional threshold-based diagnosis method often cannot cope with complex equipment failure modes due to its over-reliance on fixed thresholds, and the accuracy is only 65%; while the support vector machine (SVM) and rule-based reasoning method based on expert experience can better identify some known failure modes, but cannot achieve the high accuracy of the method of the present application. Secondly, in terms of diagnostic response time, the method of the present application can quickly respond to fault diagnosis through real-time acquisition and efficient feature processing, with a response time of 12 hours, while the response time of other methods is generally longer, especially the threshold-based diagnosis method, which takes as long as 72 hours, and cannot reflect the actual fault condition of the power distribution cabinet in time. The fault prediction lead time is another important indicator, which shows the advantage of the method of the present application in early warning capability. Through dynamic identification of failure modes and hidden state transitions, the method of the present application can predict faults 36 hours in advance, while traditional methods usually cannot achieve early warning of faults. Finally, the effectiveness of the maintenance suggestion is also a prominent advantage of the present application. By combining historical data and current state, the method of the present application can provide more accurate maintenance suggestions, significantly improving the maintenance efficiency of the power distribution cabinet and the service life of the equipment, and avoiding unnecessary maintenance and failure. Overall, the method of the present application significantly improves the efficiency of fault diagnosis, early warning and maintenance management through big data analysis and intelligent algorithms, and has stronger practicality and adaptability.

[0149] The above description is only a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, and any skilled person in the art can make equivalent replacements or changes to the technical range disclosed in the present application according to the technical solution and inventive concept of the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for diagnosing faults in an electrical distribution cabinet based on big data analysis, characterized by, The method comprises the following steps: Step 1: collecting and preprocessing multi-source sensor data from an electrical power distribution cabinet to obtain a uniform structure sensor data set; Step 2: extracting time domain features from the sensor data set by a time domain analysis method to obtain a time domain feature set; Step 3: converting time domain signals in the time domain feature set into frequency domain signals by a fast Fourier transform, and calculating each frequency domain feature to obtain a frequency domain feature set; Step 4: performing correlation analysis on the time domain feature set and the frequency domain feature set by a Pearson correlation coefficient to obtain a time domain frequency domain feature set; Step 5: identifying a hidden state set of the electrical power distribution cabinet based on the time domain frequency domain feature set by an improved HDP-HMM algorithm, calculating a hidden state transition probability, and obtaining an optimal hidden state sequence by hidden state sequence inference; The improved HDP-HMM algorithm specifically comprises: randomly selecting a time sequence data of a preset size from the time domain frequency domain feature set as a first observation sequence; extracting corresponding time domain features and frequency domain features in the first observation sequence and arranging the time domain features and the frequency domain features in a time sequence to obtain a first observation vector; taking the first observation vector as a first hidden state, selecting time sequence data of the same size from the time domain frequency domain feature set in a time sequence to obtain a second observation vector; calculating the Euclidean distance between the first observation vector and the second observation vector; if the Euclidean distance is greater than or equal to a preset distance threshold, taking the second observation vector as a second hidden state; if the Euclidean distance is less than the preset distance threshold, taking the second observation vector as the first hidden state; iterating repeatedly until the time domain frequency domain feature set is traversed completely to obtain a hidden state set, each hidden state in the hidden state set corresponding to a fault mode of the electrical power distribution cabinet; constructing a hidden state space according to the hidden state set, the hidden state space being a fault mode space; obtaining an observation probability distribution of each hidden state by a Gaussian distribution based on observation vectors in the hidden state space; obtaining a hidden state transition probability by calculating a corresponding conditional probability according to the observation probability distribution of each hidden state, the hidden state transition probability being a transition probability between each pair of hidden states; inference of a hidden state sequence according to each pair of hidden state transition probabilities to obtain an optimal hidden state sequence; the optimal hidden state sequence being a most possible fault mode sequence of the electrical power distribution cabinet in an entire observation time period; the inference of the hidden state sequence according to each pair of hidden state transition probabilities to obtain the optimal hidden state sequence specifically comprising: constructing a hidden state transition probability matrix according to each pair of hidden state transition probabilities; traversing hidden state transition probabilities from an initial hidden state to each hidden state, and selecting a hidden state transition probability maximum as a target hidden state for next transition; the initial hidden state being a hidden state in the left upper corner of the hidden state transition probability matrix; for each time, gradually selecting a hidden state transition probability maximum of each time according to an observation probability of a current hidden state and the hidden state transition probability until a last time; starting from the last time, backtracking from the hidden state transition probability matrix to the initial time to obtain the optimal hidden state sequence; Step 6: generating a fault report and a maintenance suggestion according to the optimal hidden state sequence.

2. The method for diagnosing faults in an electrical switchgear cabinet based on big data analysis according to claim 1, characterized in that, The collection is performed at predetermined time intervals; the multi-source sensor data includes current data, voltage data, temperature data, humidity data, and vibration data; the preprocessing step includes outlier removal, missing value filling, timestamp alignment, and normalization processing for different types of data, respectively, to obtain a unified structure of the sensor data set.

3. The method for diagnosing faults in an electrical distribution cabinet based on big data analysis according to claim 1, characterized in that, The time domain analysis method specifically includes: First difference processing is performed on the original signal in the sensor data set to obtain a first difference signal, and the difference processing is to calculate the difference between each data point in the original signal and the previous data point; Check the difference between the maximum and minimum values of the first difference signal, if the difference is greater than or equal to a preset difference threshold, it is considered that the trend of the current first difference signal has not been eliminated; Second difference processing is performed on the first difference signal to obtain a second difference signal, and the second difference processing is to perform first difference processing on the first difference signal again; Iterate repeatedly until the difference between the maximum and minimum values of the difference signal is less than the preset difference threshold to obtain a smoothed signal data; The smoothed signal data is segmented by a time window of a preset size; Calculate the time domain features in each time window to obtain a time domain feature set, and the time domain features include time domain signal mean, standard deviation, maximum value, minimum value, peak factor, and volatility; The time domain signal mean is used to represent the center position of the signal; The time domain signal standard deviation is used to represent the fluctuation amplitude of the signal; The time domain signal maximum value and minimum value are used to represent the extreme fluctuation of the signal; The time domain signal peak factor is the ratio of the signal maximum value to the signal mean, which is used to represent the sharpness of the signal; The time domain signal volatility is the ratio of the signal standard deviation to the signal mean, which is used to represent the fluctuation intensity of the signal.

4. The method for diagnosing faults in an electrical distribution cabinet based on big data analysis according to claim 1, characterized in that, Step three specifically includes: Fast Fourier transform is performed on each time domain signal in the time domain feature set to convert the time domain signal from time domain representation to frequency domain representation; The fast Fourier transform converts the time domain signal into amplitude and phase information of each frequency component to obtain a frequency domain signal; Calculate the frequency spectrum amplitude of each frequency component in the frequency domain signal to obtain a frequency spectrum amplitude graph, and the frequency spectrum amplitude represents the energy distribution of the frequency domain signal at each frequency component; Find the frequency component with the maximum frequency spectrum amplitude from the frequency spectrum amplitude graph, and take it as the main frequency component, which represents the main vibration mode or operating frequency of the electrical distribution cabinet; Calculate the frequency bandwidth of each frequency component according to the frequency spectrum amplitude graph, and the frequency bandwidth is the range occupied by the frequency component on the frequency spectrum amplitude graph; The frequency domain features include frequency spectrum amplitude, main frequency component, and frequency bandwidth.

5. The method for diagnosing faults in an electrical distribution cabinet based on big data analysis as claimed in claim 1, wherein, The correlation analysis specifically includes: Each time domain feature in the time domain feature set is paired with each frequency domain feature in the frequency domain feature set to calculate the Pearson correlation coefficient; The Pearson correlation coefficient is an indicator of the linear relationship between the time domain feature and the frequency domain feature, and the value of the Pearson correlation coefficient ranges from -1 to 1, where 1 represents complete positive correlation, -1 represents complete negative correlation, and 0 represents no linear correlation. If the Pearson correlation coefficient between the time domain feature and the frequency domain feature is greater than or equal to a set linear correlation threshold, it is considered that the two are highly correlated, and one redundant feature is removed; If the Pearson correlation coefficient is less than the set linear correlation threshold, the corresponding time domain feature and the frequency domain feature are retained; Repeat the iteration comparison until all time domain features and frequency domain features are compared, and obtain the time domain frequency domain feature set.

6. The method for diagnosing faults in an electrical distribution cabinet based on big data analysis as claimed in claim 1, wherein, The step six is specifically: The fault report includes fault mode, occurrence time range, influence range and equipment damage degree, specifically: According to the optimal hidden state sequence, determine the fault mode of the electrical power distribution cabinet corresponding to the hidden state at each time; By analyzing the start and end time of the hidden state in the hidden state sequence, the possible occurrence time range of the fault mode is obtained; Combined with the fault mode and the historical data of the equipment, the damage degree of the fault mode to the equipment is inferred; Based on the fault report, maintenance suggestions are generated, including load adjustment or load balancing and equipment maintenance or replacement of aging components.

Images