Power distribution room fault detection method and system combining edge calculation and machine learning

Abstract

The invention provides a power distribution room fault detection method and system combining edge calculation and machine learning. Voltage, current, temperature, humidity, equipment vibration signals and sound signals in the power distribution room are collected in real time through the sensor module; analyzing the collected real-time data through a pre-trained differential autoregressive moving average model to carry out future data prediction; performing denoising and normalization preprocessing on the real-time data and the prediction data; performing time-frequency domain analysis on the preprocessed data, and extracting time-frequency features for fault diagnosis and prediction after dimension reduction through principal component analysis; a pre-trained fault diagnosis model based on a support vector machine analyzes the dimensionality-reduced time-frequency characteristics to obtain the fault condition of the power distribution room; wherein the differential autoregressive moving average model and the fault diagnosis model are operated by a plurality of edge computing controllers built according to a parallel computing architecture.

Description

Technical Field

[0001] The present invention relates to the field of edge computing technology, and in particular to a distribution room fault detection method and system combining edge computing with machine learning. Background Art

[0002] With the continuous advancement of power distribution room monitoring technology, the functions and applications of existing equipment have become more diversified. However, there are still problems such as insufficient data processing capabilities, difficulty in processing high-dimensional data, poor equipment compatibility and high maintenance costs.

[0003] The following are some typical monitoring equipment products for power distribution rooms: Smart meter: a meter with multiple measurement functions, often used to monitor and record power usage in real time. Monitoring module: a module that integrates multiple sensors (such as temperature, humidity, and pressure sensors) to facilitate integration with existing systems. Cloud monitoring platform: provides data display, analysis, and alarm functions through the Internet, which is convenient for users to monitor remotely. There are problems with these products. They collect voltage, current, temperature, and humidity data and preset thresholds to alarm abnormalities. Some transmit data to the cloud and process complex data through cloud computing. False alarms often occur by setting thresholds. Data is concentrated and processed in the cloud. The existing communication facilities in the power distribution room cannot guarantee the real-time nature of the data, and cloud services are overwhelmed as the power distribution equipment increases year by year. To solve this problem, it is necessary to improve edge computing capabilities, reduce dependence on the cloud by sinking some data processing to local devices (edge ​​devices), and achieve local real-time analysis and rapid response. However, the computing power of edge devices limits the application of gap methods such as machine learning. Summary of the invention

[0004] In order to solve the above technical problems or at least partially solve the above technical problems, the present invention provides a distribution room fault detection method and system combining edge computing and machine learning.

[0005] In a first aspect, the present invention provides a method for detecting faults in a power distribution room by combining edge computing with machine learning, comprising: collecting voltage, current, temperature, humidity, equipment vibration signals and sound signals in the power distribution room in real time as real-time data through a sensor module;

[0006] Predict future data based on real-time data using a pre-trained differential autoregressive moving average model;

[0007] De-noising and normalizing the real-time data and forecast data; performing time-frequency domain analysis on the pre-processed data, and extracting time-frequency features for fault diagnosis and prediction after dimensionality reduction through principal component analysis;

[0008] The pre-trained fault diagnosis model based on support vector machine analyzes the reduced-dimensional time-frequency features to obtain the fault conditions of the power distribution room;

[0009] Among them, the differential autoregressive moving average model and the fault diagnosis model are run by multiple edge computing controllers built according to the parallel computing architecture.

[0010] Furthermore, the real-time data and the forecast data are preprocessed by denoising and normalizing, including: using the moving average method to smooth the data to reduce the impact of noise:

[0011]

[0012] Among them, x[ni] is the original signal of the data, y[n] is the signal after data denoising, and N is the window size;

[0013] The signal data is normalized to the interval [0,1] according to the following formula to reduce the dimension effect:

[0014]

[0015] Where y' is the normalized signal, y is the signal before normalization, and Y is the set of signals before normalization; max(Y) is the maximum value in the set Y, and min(Y) is the minimum value in the set Y.

[0016] Furthermore, the preprocessed data is analyzed in the time-frequency domain to extract the time-frequency features including:

[0017] Extract time domain features based on time domain signals, including mean, standard deviation, peak value, and waveform factor;

[0018] Use fast Fourier transform to convert time domain data to frequency domain and get frequency domain signal:

[0019]

[0020] Among them, ω is the frequency value in the frequency domain;

[0021] Extract features from frequency domain signals, including main frequency, harmonic distortion rate, and harmonic amount.

[0022] Furthermore, the steps of establishing the difference autoregressive moving average model include:

[0023] Perform a sequence stationarity test on the collected real-time data, and use the unit root test to check whether the real-time data is stationary; if the real-time data is not stationary, the difference autoregressive moving average model needs to be differentiated:

[0024] Where B is the lag operator, defined as Bx[n]=x[n-1], d is the difference order, x is the original signal of the real-time data, is the differential result of the original signal x of real-time data, [n] is the time sequence number of the signal;

[0025] Autoregressive part: Where p is the order of the autoregressive term, φ0 is a constant, φ1, φ2…φ p is the autoregressive coefficient;

[0026] Moving average part: Θ(B)ε[n]=θ0ε[n]+θ1Bε[n]+θ2B 2 ε[n]+…+θ p B q ε[n], where q is the order of the translation term, ε[n] is a white noise process, φ0, φ1…φ p is the shift coefficient, and the autocorrelation function and partial autocorrelation function are used to identify the parameters of the difference autoregressive moving average model;

[0027]

[0028] Estimate the parameters of the differenced autoregressive moving average model using the maximum likelihood estimation algorithm.

[0029] Furthermore, the L1 regularized fault diagnosis model is trained, and the steps include:

[0030] Select a kernel function, which can be any one of a linear kernel, a polynomial kernel, or a radial basis function kernel; set the L1 regularization coefficient λ; use the minimum angle regression method to solve the optimization problem with L1 regularization. In the L1 regularized fault diagnosis model, the optimization problem is to minimize the following objective function:

[0031]

[0032] in, is the loss function of the fault diagnosis model, Wf i +b is the time-frequency feature f i The prediction result W is the weight, b is the bias; m i is the time-frequency feature f i The corresponding true label, K is the sample size, W j is the jth item in the weight, and the dimension of the weight is U;

[0033] During the training process, the parameters of the fault diagnosis model are updated by stochastic gradient descent:

[0034] Among them, η1 is the learning rate of the training process, is the gradient of the objective function.

[0035] Furthermore, a Bagging or Boosting ensemble learning method is used to train multiple base classifiers for fault diagnosis based on support vector machines, and the prediction results of the base classifiers are weighted and summed to obtain the final fault diagnosis result decision.

[0036] Furthermore, for the newly added fault types, the weights are fine-tuned on the basis of the original fault diagnosis model for incremental learning. The model weight parameter update formula is:

[0037]

[0038] Among them, W new is the model weight parameter after incremental learning, W old is the model weight parameter before incremental learning, m is the true label of the new data, is the predicted value of the model, f is the new input feature vector, and η2 is the learning rate of incremental learning.

[0039] Furthermore, the fault diagnosis model is compressed and quantized, wherein model compression is achieved by removing weights that have little impact on the model's prediction effect through parameter pruning; and weight quantization is achieved by converting the model's floating-point parameters into low-precision integers to reduce memory usage and computational complexity during model runtime.

[0040] In a second aspect, the present invention provides a power distribution room fault detection device combining edge computing with machine learning, including: a sensor module responsible for real-time collection of various parameters in the power distribution room, including voltage, current, temperature, humidity, equipment vibration signals and sound signals, the sensor module including: a voltage sensor for measuring the voltage of the power distribution system, a current sensor for measuring the current of the power distribution system, a temperature sensor for monitoring the temperature inside the power distribution room, a humidity sensor for monitoring the humidity inside the power distribution room, a vibration sensor for monitoring the vibration signal of the equipment, and a voiceprint sensor for monitoring the sound signal of the equipment;

[0041] A data processing module, which is responsible for data reception, preprocessing, feature extraction and fault detection to implement the power distribution room fault detection method combining edge computing with machine learning. The data processing module includes: a microcontroller / microprocessor and a memory that supports parallel computing;

[0042] Connection between sensors and data processing modules;

[0043] A communication module, which realizes data transmission between the device and the external system; a connection between the data processing module and the communication module;

[0044] A power supply module, which provides a stable power supply for the device;

[0045] Display and control module,The display and control module provides a visual display of the device status and supports user interaction.

[0046] In a third aspect, the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the distribution room fault detection method combining edge computing and machine learning.

[0047] The above technical solution provided by the embodiment of the present invention has the following advantages compared with the prior art:

[0048] This application predicts the operating status and fault diagnosis of the distribution system through the trained differential autoregressive moving average model and fault diagnosis model. The differential autoregressive moving average model outputs the prediction of current, voltage, temperature, vibration and sound, and the fault diagnosis model analyzes the real-time and predicted signals to obtain the fault type and possible causes. The differential autoregressive moving average model and the fault diagnosis model are run by multiple edge computing controllers built according to the parallel computing architecture. It supports edge computing without data upload and has better timeliness. This application reduces the dimension of features through principal component analysis, reduces the amount of analysis calculations, and better supports edge computing. When the data noise is large, the system's anti-interference ability is improved through data preprocessing. The introduction of support vector machines combined with incremental learning and ensemble learning algorithms can better capture the complex changes of parameters and improve the accuracy and robustness of fault detection.

[0049] Through model pruning, quantization and L1 regularization, the model size and computational complexity are reduced, allowing the system to run efficiently on a low-power MCU, reducing memory and computing requirements and better supporting edge computing.

[0050] Through incremental learning, the algorithm can flexibly adapt to different input characteristics and fault types, easily add or adjust monitoring parameters, quickly adapt to new fault modes, and improve system scalability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0053] Figure 1 A flow chart of a method for detecting faults in a power distribution room that combines edge computing with machine learning;

[0054] Figure 2 Schematic diagram of a distribution room fault detection device that combines edge computing and machine learning. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0057] Example 1

[0058] See also Figure 1 As shown in the figure, in order to realize intelligent fault diagnosis and predictive monitoring, a distribution room fault detection method combining edge computing and machine learning is designed. This algorithm will be based on the combination of data collection, feature extraction, fault diagnosis and prediction model, and reduce the computing requirements of the microcontroller and reduce costs, including:

[0059] Multiple microcontrollers / microprocessors built in parallel computing architecture are used to run the differential autoregressive moving average model and the fault diagnosis model.

[0060] The sensor module collects voltage, current, temperature, humidity, equipment vibration signals and sound signals in the distribution room in real time as implementation data.

[0061] The difference autoregressive moving average model analyzes real-time data to predict future data;

[0062] In the specific implementation process, the steps of establishing the difference autoregressive moving average model include:

[0063] Perform a sequence stationarity test on the collected real-time data; in the specific implementation process, use the unit root test to check whether the real-time data is stable. If the real-time data is not stable, the difference autoregressive moving average model needs to be differentiated. Where B is the lag operator defined as Bx[n]=x[n-1], d is the difference order, x is the original signal of the real-time data, is the differential result of the original signal x of real-time data, [n] is the time sequence number of the signal;

[0064] Autoregressive part: Where p is the order of the autoregressive term, φ0 is a constant that can be adjusted for training, φ1, φ2…φ p is the autoregressive coefficient.

[0065] Moving average part: Θ(B)ε[n]=θ0ε[n]+θ1Bε[n]+θ2B 2 ε[n]+…+θ p B q ε[n], where q is the order of the translation term, ε[n] is a white noise process, φ0, φ1…φ p is the shift coefficient. Use the autocorrelation function and partial autocorrelation function to identify the parameters of the difference autoregressive moving average model.

[0066]

[0067] Estimate the parameters of the differenced autoregressive moving average model using the maximum likelihood estimation algorithm.

[0068] The real-time data and forecast data are preprocessed by denoising and normalization; the preprocessing includes: denoising and normalization. Denoising: Use the moving average method to smooth the data to reduce the impact of noise.

[0069]

[0070] Among them, x[ni] is the original signal of the data, y[n] is the denoised signal, and N is the window size.

[0071] Normalization: Normalize the signal data to the [0,1] interval to reduce the dimension effect:

[0072]

[0073] Among them, y' is the normalized signal, y is the signal before normalization, and Y is the set of signals before normalization; max(Y) is the maximum value in the set Y, and min(Y) is the minimum value in the set Y. In order to improve the accuracy and reliability of the data, the data is denoised. The data is normalized to convert features of different dimensions into a unified standard scale, so that different features have a balanced impact on the loss function during the model training process.

[0074] Perform time-frequency domain analysis on the preprocessed data and extract time-frequency features for fault diagnosis and prediction;

[0075] Extract time domain features based on time domain signals, including mean, standard deviation, peak value, and waveform factor;

[0076] Use fast Fourier transform to convert the time domain signal to the frequency domain and get the frequency domain signal:

[0077]

[0078] Among them, ω is the frequency value in the frequency domain;

[0079] Frequency domain features are extracted from the frequency domain signal. The frequency domain features include main frequency, harmonic distortion rate, and harmonic amount.

[0080] Feature dimensionality reduction: Use principal component analysis to reduce the dimensionality of the extracted time-frequency features to reduce computational complexity and memory usage.

[0081] Through the pre-trained support vector machine-based fault diagnosis model, the reduced-dimensional time-frequency features are analyzed to obtain the fault conditions of the power distribution room;

[0082] Support vector machine is used as the fault diagnosis model. Support vector machine performs well in small samples and high-dimensional space and is suitable for equipment status classification.

[0083] Training the L1 regularized fault diagnosis model. The steps to reduce the dimension of the support vector machine by L1 regularization include:

[0084] Select a kernel function, which can be any one of a linear kernel, a polynomial kernel, or a radial basis function kernel; set the L1 regularization coefficient λ; use the minimum angle regression method to solve the optimization problem with L1 regularization. In the L1 regularized fault diagnosis model, the optimization problem is to minimize the following objective function:

[0085]

[0086] in, is the loss function of the fault diagnosis model, Wf i +b is the time-frequency feature f i The prediction result W is the weight, b is the bias; m i is the time-frequency feature f i The corresponding true label, K is the sample size, W j is the jth item in the weight, and the dimension of the weight is U;

[0087] During the training process, the parameters of the fault diagnosis model are updated by stochastic gradient descent:

[0088] Among them, η1 is the learning rate of the training process, is the gradient of the objective function.

[0089] In order to improve the real-time and adaptability of the model, for the newly added fault types, the weights are fine-tuned on the basis of the original model for incremental learning. The model weight parameter update formula is:

[0090]

[0091] Among them, W new is the model weight parameter after incremental learning, W old is the model weight parameter before incremental learning, m is the true label of the new data, is the predicted value of the model, f is the new input feature vector, and η2 is the learning rate of incremental learning.

[0092] Model Compression and Quantization,In order to run efficiently on resource-constrained microcontrollers, the model needs to be compressed and quantized.,Model compression is mainly achieved through parameter pruning, that is, removing weights that have little impact on the model's prediction effect.,Weight quantization is to convert the model's floating-point parameters into low-precision integers to reduce memory usage and computation.,The specific quantization process can be achieved through the following formula:,W, q = round(W scale), where scale is the quantization scale factor, and round() is used to map the floating point numbers in the weight W processed by the quantization scale factor to the integer range. This can effectively reduce the storage requirements and inference time of the model.

[0093] Parallel computing and distributed processing, in the process of data processing and model training, efficiency can be improved through parallel computing and distributed processing. Divide the data set into multiple small batches and use multi-core microcontrollers or multiple MCUs for parallel processing. For example, current, temperature, and voltage data can be processed in parallel on different processing units, or multi-threaded data preprocessing can be performed to speed up the calculation speed of the entire system.

[0094] Ensemble learning,In order to further improve the robustness and accuracy of the model, multiple base classifiers based on support vector machines are combined for fault diagnosis.,For example, Bagging or Boosting ensemble learning methods are used to train multiple base classifiers based on support vector machines for fault diagnosis, and the prediction results of the base classifiers are weighted and summed to obtain the final fault diagnosis result decision.

[0095] Finally, fault diagnosis and predictive monitoring predicts the operating status of the distribution system and diagnoses faults through the trained differential autoregressive moving average model and fault diagnosis model. The differential autoregressive moving average model outputs the prediction of current, voltage, temperature, vibration and sound, and the fault diagnosis model analyzes the real-time and predicted signals to obtain the fault type and possible cause. At the same time, the predictive diagnosis results can be visualized to help users monitor the operating status of the equipment more intuitively and perform maintenance in a timely manner.

[0096] Example 2

[0097] like Figure 2 As shown, the present invention provides a distribution room fault detection device combining edge computing with machine learning, including: a sensor module is responsible for real-time collection of various parameters in the distribution room, including voltage, current, temperature, humidity, equipment vibration signals and sound signals. The sensor module includes: a voltage sensor: measures the voltage of the distribution system. A current sensor: measures the current of the distribution system, using a Hall effect sensor. A temperature sensor: monitors the internal temperature of the distribution room to ensure that the equipment operates within a safe range. A humidity sensor: monitors the internal humidity of the distribution room to prevent the equipment from getting damp. A vibration sensor: monitors the vibration signal of the equipment. A voiceprint sensor: monitors the sound signal of the equipment.

[0098] Data processing module, which is responsible for data reception, preprocessing, feature extraction and fault detection. The data processing module includes: a microcontroller / microprocessor, such as ESP 32 or ARM Cortex-A series, responsible for data processing and model reasoning; a memory: used to store historical data and model parameters (such as Flash or SD card).

[0099] Connection between the sensor and the data processing module. Analog signal connection: The analog signal output by the sensor is input to the ADC pin of the microcontroller / microprocessor through the analog-to-digital converter. Digital signal connection: The digital sensor is connected to the microcontroller / microprocessor through the I2C or SPI interface.

[0100] Communication module, which realizes data transmission between the device and external systems, such as the cloud or monitoring platform. The communication module includes: wireless communication module: such as Wi-Fi, LoRa or NB-IoT module, to realize remote transmission of data. Wired communication interface: such as RS485 or Ethernet interface, for communication between devices.

[0101] The connection between the data processing module and the communication module: Specifically, the data processing module is connected to the communication module through a UART, I2C or SPI interface to achieve data upload and download.

[0102] The power supply module provides a stable power supply for the device to ensure continuous operation of the device. The power supply module includes: a power adapter: converts AC power to DC power to power the device. A DC-to-DC level conversion unit is used to convert the voltage provided by the battery backup or power adapter to the level required for operation; battery backup: such as lithium battery, to ensure that the device can still operate in the event of a power outage. Connection between the power supply module and other modules.

[0103] Display and control module: The display and control module provides visual display of equipment status and supports user interaction. The display and control module includes: LCD or OLED display: displays real-time data, alarm information and equipment status. Buttons or touch screen: allows users to configure and set parameters.

[0104] Example 3

[0105] An embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed, the power distribution room fault detection method combining edge computing with machine learning is implemented, including:

[0106] Predict future data based on real-time data using a pre-trained differential autoregressive moving average model;

[0107] Perform denoising and normalization preprocessing on real-time data and forecast data; perform time-frequency domain analysis on preprocessed data, and extract time-frequency features after dimensionality reduction through principal component analysis for fault diagnosis and prediction;

[0108] Through the pre-trained support vector machine-based fault diagnosis model, the reduced-dimensional time-frequency features are analyzed to obtain the fault conditions of the power distribution room;

[0109] Among them, the differential autoregressive moving average model and the fault diagnosis model are run by multiple edge computing controllers built according to the parallel computing architecture.

[0110] A computer-readable storage medium provided in an embodiment of the present invention stores a computer program that is not limited to the method operations described above, but can also execute related operations in a distribution room fault detection method combining edge computing and machine learning provided in any embodiment of the present invention.

[0111] In the embodiments provided by the present invention, it should be understood that the disclosed methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the description division of the circuit is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling shown or discussed can be indirect coupling through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0112] The foregoing is merely a specific embodiment of the present invention, which enables those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A distribution room fault detection method combining edge computing and machine learning, characterized in that: include: The sensor module collects voltage, current, temperature, humidity, equipment vibration signals and sound signals in the power distribution room as real-time data; The pre-trained differential autoregressive moving average model analyzes real-time data to predict future data; De-noising and normalizing the real-time data and forecast data; performing time-frequency domain analysis on the pre-processed data, and extracting time-frequency features for fault diagnosis and prediction after dimensionality reduction through principal component analysis; The pre-trained fault diagnosis model based on support vector machine analyzes the reduced-dimensional time-frequency features to obtain the fault conditions of the power distribution room; Among them, the differential autoregressive moving average model and the fault diagnosis model are run by multiple edge computing controllers built according to the parallel computing architecture.

2. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: The real-time data and forecast data are preprocessed by denoising and normalizing, including: using the moving average method to smooth the data to reduce the impact of noise: Among them, x[ni] is the original signal of the data, y[n] is the signal after data denoising, and N is the window size; The signal data is normalized to the interval [0,1] according to the following formula to reduce the dimension effect: Where y' is the normalized signal, y is the signal before normalization, and Y is the set of signals before normalization; max(Y) is the maximum value in the set Y, and min(Y) is the minimum value in the set Y.

3. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: Perform time-frequency domain analysis on the preprocessed data and extract time-frequency features including: Extract time domain features based on time domain signals, including mean, standard deviation, peak value, and waveform factor; Use fast Fourier transform to convert time domain data to frequency domain and get frequency domain signal: Among them, ω is the frequency value in the frequency domain; Extract features from frequency domain signals, including main frequency, harmonic distortion rate, and harmonic amount.

4. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: The steps to build a difference autoregressive moving average model include: Perform a sequence stationarity test on the collected real-time data, and use the unit root test to check whether the real-time data is stationary; if the real-time data is not stationary, the difference autoregressive moving average model needs to be differentiated: Where B is the lag operator, defined as Bx[n]=x[n-1], d is the difference order, x is the original signal of the real-time data, is the differential result of the original signal x of real-time data, [n] is the time sequence number of the signal; Autoregressive part: Where p is the order of the autoregressive term, φ0 is a constant, φ1, φ2…φ p is the autoregressive coefficient; Moving average part: Θ(B)ε[n]=θ0ε[n]+θ1Bε[n]+θ2B 2 ε[n]+…+θ p B q ε[n], where q is the order of the translation term, ε[n] is a white noise process, φ0, φ1…φ p is the shift coefficient, and the autocorrelation function and partial autocorrelation function are used to identify the parameters of the difference autoregressive moving average model; Estimate the parameters of the differenced autoregressive moving average model using the maximum likelihood estimation algorithm.

5. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: Training the L1 regularized fault diagnosis model includes the following steps: Select a kernel function, which can be any one of a linear kernel, a polynomial kernel, or a radial basis function kernel; set the L1 regularization coefficient λ; use the minimum angle regression method to solve the optimization problem with L1 regularization. In the L1 regularized fault diagnosis model, the optimization problem is to minimize the following objective function: in, is the loss function of the fault diagnosis model, Wf i +b is the time-frequency feature f i The prediction result W is the weight, b is the bias; m i is the time-frequency feature f i The corresponding true label, K is the sample size, W j is the jth item in the weight, and the dimension of the weight is U; During the training process, the parameters of the fault diagnosis model are updated by stochastic gradient descent: Among them, η1 is the learning rate of the training process, is the gradient of the objective function.

6. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: The Bagging or Boosting ensemble learning method is used to train multiple base classifiers based on support vector machines for fault diagnosis, and the prediction results of the base classifiers are weighted and summed to obtain the final fault diagnosis result decision.

7. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: For the newly added fault types, the weights are fine-tuned on the basis of the original fault diagnosis model for incremental learning. The model weight parameter update formula is: Among them, W new is the model weight parameter after incremental learning, W old is the model weight parameter before incremental learning, m is the true label of the new data, is the predicted value of the model, f is the new input feature vector, and η2 is the learning rate of incremental learning.

8. The power distribution room fault detection method combining edge computing and machine learning according to claim 1 is characterized in that: The fault diagnosis model is compressed and quantized. Model compression is achieved by removing weights that have little impact on the model's prediction effect through parameter pruning. Weight quantization is achieved by converting the model's floating-point parameters into low-precision integers to reduce memory usage and computational complexity when the model is running.

9. A power distribution room fault detection device combining edge computing and machine learning, characterized in that: include: The sensor module is responsible for real-time collection of various parameters in the power distribution room, including voltage, current, temperature, humidity, equipment vibration signals and sound signals. The sensor module includes: voltage sensor for measuring the voltage of the power distribution system, current sensor for measuring the current of the power distribution system, temperature sensor for monitoring the temperature inside the power distribution room, humidity sensor for monitoring the humidity inside the power distribution room, vibration sensor for monitoring the vibration signal of the equipment, and voiceprint sensor for monitoring the sound signal of the equipment; A data processing module, which is responsible for data reception, preprocessing, feature extraction and fault detection to implement the power distribution room fault detection method combining edge computing and machine learning as described in any one of claims 1 to 8, and the data processing module includes: a microcontroller / microprocessor and a memory that supports parallel computing; Connection between sensors and data processing modules; A communication module, which realizes data transmission between the device and the external system; a connection between the data processing module and the communication module; A power supply module, which provides a stable power supply for the device; Display and control module,The display and control module provides a visual display of the device status and supports user interaction.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by the processor, the distribution room fault detection method combining edge computing and machine learning as described in any one of claims 1 to 8 is implemented.

Images