Electrician equipment abnormal state intelligent identification system based on artificial intelligence

Abstract

The invention relates to the technical field of electrical equipment abnormal state intelligent identification, and particularly discloses an electrical equipment abnormal state intelligent identification system based on artificial intelligence, which comprises a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, an abnormal identification module and an alarm module, the system is characterized in that the data acquisition module is used for acquiring various operation data of electrical equipment, the operation data comprises voltage data V (t), current data I (t) and temperature data T (t), the data preprocessing module is used for denoising the acquired data, and a mean filtering algorithm is adopted. Multi-source data are creatively collected and deeply processed, features are accurately extracted by applying multiple algorithms, the reliability of anomaly recognition is improved by means of a deep neural network training model and threshold judgment, operation and maintenance personnel are notified in time through multiple alarm modes, and safe and stable operation of electrical equipment is comprehensively guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent recognition of abnormal states of electrical equipment, and in particular relates to an intelligent recognition system for abnormal states of electrical equipment based on artificial intelligence. Background Art

[0002] Electrical equipment is a general term for various devices and instruments involved in the production, transmission, distribution and use of electric power. During the operation of electrical equipment, it is crucial to identify abnormal conditions in a timely and accurate manner.

[0003] In the existing technology, traditional methods for detecting abnormal states of electrical equipment mostly rely on manual inspections or simple threshold judgments. Manual inspections are inefficient, easily affected by subjective factors, and difficult to achieve real-time monitoring. Simple threshold judgments cannot adapt to complex operating conditions and are prone to misjudgments and missed judgments. There are deficiencies in the comprehensiveness and accuracy of data processing, and it is difficult to effectively process multi-source heterogeneous data. The ability to recognize complex fault modes is limited, and there is a lack of adaptive adjustment mechanisms. It is difficult to cope with changes in different environments and equipment aging, resulting in the inability to meet the high-precision abnormal state recognition needs in actual applications. Improvements are needed. To this end, an intelligent recognition system for abnormal states of electrical equipment based on artificial intelligence is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide an intelligent recognition system for abnormal states of electrical equipment based on artificial intelligence to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the present invention adopts the following technical solutions: an artificial intelligence-based intelligent recognition system for abnormal states of electrical equipment, comprising a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, an abnormality recognition module, and an alarm module. The data acquisition module is used to collect various operating data of electrical equipment, including voltage data V(t), current data I(t), and temperature data T(t);

[0006] The data preprocessing module performs denoising on the collected data using the mean filter algorithm. The formula is: Yn = (1 / (2*m+1))*sum(X[n+i]for iin range(-m,m+1)), where Xn is the original data sequence, Yn is the filtered data, and m is the filter window size.

[0007] The feature extraction module extracts the frequency domain features of the data through fast Fourier transform (FFT). The formula is: X[k] = sum(x[n]*exp(-1j*(2*pi / N)*k*n)for n in range(N)), where k ranges from 0 to N-1, where x(n) is the time domain data, X(k) is the frequency domain data, and N is the number of data points.

[0008] The model training module uses a deep neural network (DNN) for training, and the loss function uses the cross entropy loss function, with the formula: L = -(1 / N)*sum(sum(y[i][j]*log(p[i][j])for j in range(C))for i in range(N)), where N is the number of samples, C is the number of categories, y[i][j] is the true label (0 or 1) of sample i belonging to category j, and p[i][j] is the probability predicted by the model that sample i belongs to category j;

[0009] The anomaly recognition module identifies anomalies in real-time data based on the trained model and outputs an anomaly signal if an anomaly is identified;

[0010] After receiving the abnormal signal, the alarm module issues an alarm message.

[0011] Preferably, the data acquisition module uses a variety of sensors, including voltage sensors, current sensors and temperature sensors, which are respectively used to acquire corresponding physical quantities.

[0012] Preferably, the data preprocessing module also includes data normalization processing, mapping the data to the [0,1] interval, the formula is: x_norm = (x-x_min) / (x_max-x_min), where x is the original data, x_min and x_max are the minimum and maximum values ​​of the data respectively, and x_norm is the normalized data.

[0013] Preferably, the feature extraction module also uses wavelet transform to extract the time-frequency features of the data, and the wavelet transform formula is: Wf[a,b]=(1 / sqrt(a))*integral(f[t]*conjugate(psi[(tb) / a])for t in range(-infinity,infinity)), where f(t) is the original signal, a is the scale factor, b is the translation factor, psi(t) is the wavelet basis function, and conjugate(psi(t)) is its conjugate function.

[0014] Preferably, the deep neural network of the model training module includes multiple hidden layers, and the activation function of the hidden layer neurons adopts the ReLU function, and the formula is: ReLU(x)=max(0,x).

[0015] Preferably, the anomaly identification module uses a threshold judgment method in combination with the model output result to perform anomaly judgment. If the anomaly probability output by the model is greater than a set threshold value T_th, it is judged as an anomaly.

[0016] Preferably, the alarm module includes an audible and visual alarm device and a text message alarm module, which notifies operation and maintenance personnel in multiple ways.

[0017] Preferably, a data storage module is included for storing the collected original data, pre-processed data, and parameters and results during the model training process.

[0018] In summary, compared with the prior art, the present invention has the following beneficial effects:

[0019] 1. This invention creatively collects and deeply processes multi-source data, applies multiple algorithms to accurately extract features, uses deep neural network training models, employs threshold judgment to improve the reliability of anomaly identification, and promptly notifies maintenance personnel through multiple alarm methods, fully ensuring the safe and stable operation of electrical equipment.

[0020] 2. The present invention collects multi-source data through the data acquisition module and combines it with multiple algorithms of the data preprocessing module to effectively remove noise, normalize data, improve data quality, and provide a reliable data foundation for subsequent feature extraction and model training;

[0021] 3. The feature extraction module uses a variety of feature extraction methods, such as FFT and wavelet transform, which can extract data features from different angles, comprehensively reflect the operating status of electrical equipment, and improve the accuracy of anomaly identification;

[0022] 4. The model training module uses a deep neural network combined with a cross-entropy loss function for training, giving the model powerful learning capabilities and the ability to accurately learn the characteristic patterns of normal and abnormal states;

[0023] 5. The anomaly recognition module combines the threshold judgment method to improve the reliability of anomaly judgment. The alarm module notifies operation and maintenance personnel in various ways to promptly detect and handle abnormal situations, ensuring the safe and stable operation of electrical equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a system structure diagram of the present invention. DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only 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 making any creative efforts shall fall within the scope of protection of the present invention.

[0026] See also Figure 1 , the present invention provides a technical solution:

[0027] The AI-based intelligent recognition system for abnormal states of electrical equipment includes a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, an abnormality recognition module, and an alarm module. The data acquisition module is used to collect various operating data of electrical equipment, including voltage data V(t), current data I(t), and temperature data T(t);

[0028] The data preprocessing module performs denoising on the collected data using the mean filter algorithm. The formula is: Yn = (1 / (2*m+1))*sum(X[n+i]for iin range(-m,m+1)), where Xn is the original data sequence, Yn is the filtered data, and m is the filter window size.

[0029] The feature extraction module extracts the frequency domain features of the data through fast Fourier transform (FFT). The formula is: X[k] = sum(x[n]*exp(-1j*(2*pi / N)*k*n)for n in range(N)), where k ranges from 0 to N-1, where x(n) is the time domain data, X(k) is the frequency domain data, and N is the number of data points.

[0030] The model training module uses a deep neural network (DNN) for training, and the loss function uses the cross entropy loss function, with the formula: L = -(1 / N)*sum(sum(y[i][j]*log(p[i][j])for j in range(C))for i in range(N)), where N is the number of samples, C is the number of categories, y[i][j] is the true label (0 or 1) of sample i belonging to category j, and p[i][j] is the probability predicted by the model that sample i belongs to category j;

[0031] The anomaly recognition module identifies anomalies in real-time data based on the trained model and outputs an anomaly signal if an anomaly is identified;

[0032] After receiving the abnormal signal, the alarm module issues an alarm message.

[0033] Working Principle of the Data Acquisition Module: The data acquisition module uses voltage sensors, current sensors, and temperature sensors to collect voltage data (V(t), current data (I(t), and temperature data (T(t))) during the operation of electrical equipment. These sensors are tightly connected to key parts of the electrical equipment. The voltage sensor monitors the equipment's voltage, the current sensor detects current changes, and the temperature sensor senses the equipment's temperature. They acquire equipment operating data in real time and transmit the data to the data preprocessing module, providing raw information for subsequent analysis.

[0034] Working Principle of the Feature Extraction Module: The feature extraction module first processes the pre-processed data through fast Fourier transform (FFT), converting the time domain data x(n) into frequency domain data X(k) to obtain the frequency domain characteristics of the data, helping to analyze the frequency characteristics of the equipment operating status. In addition, the wavelet transform is used, combined with the scale factor a, the translation factor b, and the wavelet basis function psi(t), to calculate the original signal f(t) to obtain Wf[a,b], thereby extracting the time-frequency characteristics of the data and comprehensively reflecting the data characteristics from different angles.

[0035] Working Principle of the Model Training Module: The model training module uses a deep neural network (DNN) for training. The DNN contains multiple hidden layers. The activation function of the hidden layer neurons uses the ReLU function. By performing nonlinear transformations on the input data, the model's learning ability is enhanced. During the training process, the cross-entropy loss function is used to measure the difference between the model's predicted results p[i][j] and the true labels y[i][j]. The model parameters are continuously adjusted to enable the model to accurately learn the characteristic patterns of normal and abnormal states of electrical equipment.

[0036] Working principle of the anomaly recognition module: The anomaly recognition module receives data processed by the feature extraction module, uses the trained model to analyze and predict the real-time data, obtains the anomaly probability output by the model, and then uses the threshold judgment method to compare the anomaly probability output by the model with the set threshold T_th. If the anomaly probability is greater than the set threshold T_th, it is determined that the equipment operation status is abnormal and outputs an anomaly signal to the alarm module;

[0037] Working principle of the alarm module: After receiving the abnormal signal from the abnormality identification module, the alarm module activates the sound and light alarm device to sound an alarm to attract the attention of on-site personnel. At the same time, the SMS alarm module sends a text message notification to the operation and maintenance personnel to inform them of the abnormality of the equipment so that the operation and maintenance personnel can take timely measures to deal with the abnormality;

[0038] The data acquisition module collects multi-source data, and combined with the various algorithms of the data preprocessing module, it can effectively remove noise, normalize data, improve data quality, and provide a reliable data foundation for subsequent feature extraction and model training;

[0039] The feature extraction module uses a variety of feature extraction methods, such as FFT and wavelet transform, which can extract data features from different angles, comprehensively reflect the operating status of electrical equipment, and improve the accuracy of anomaly identification;

[0040] The model training module uses a deep neural network combined with a cross-entropy loss function for training, which gives the model a strong learning ability and can accurately learn the characteristic patterns of normal and abnormal states;

[0041] The anomaly recognition module combines the threshold judgment method to improve the reliability of anomaly judgment. The alarm module notifies operation and maintenance personnel in various ways to promptly detect and handle anomalies, ensuring the safe and stable operation of electrical equipment.

[0042] This invention creatively collects multi-source data and processes it in depth, uses multiple algorithms to accurately extract features, relies on deep neural network training models, uses threshold judgment to improve the reliability of anomaly identification, and promptly notifies operation and maintenance personnel through multiple alarm methods to comprehensively ensure the safe and stable operation of electrical equipment.

[0043] The data acquisition module uses a variety of sensors, including voltage sensors, current sensors, and temperature sensors, which are used to collect corresponding physical quantities;

[0044] The data preprocessing module also includes data normalization processing, mapping the data to the [0,1] interval. The formula is: x_norm = (x-x_min) / (x_max-x_min), where x is the original data, x_min and x_max are the minimum and maximum values ​​of the data respectively, and x_norm is the normalized data.

[0045] The feature extraction module also uses wavelet transform to extract the time-frequency features of the data. The wavelet transform formula is: Wf[a,b]=(1 / sqrt(a))*integral(f[t]*conjugate(psi[(tb) / a])for t in range(-infinity,infinity)), where f(t) is the original signal, a is the scale factor, b is the translation factor, psi(t) is the wavelet basis function, and conjugate(psi(t)) is its conjugate function;

[0046] The deep neural network of the model training module contains multiple hidden layers. The activation function of the hidden layer neurons uses the ReLU function, and the formula is: ReLU(x) = max(0,x);

[0047] The anomaly recognition module uses the threshold judgment method combined with the model output results to make anomaly judgments. If the anomaly probability output by the model is greater than the set threshold T_th, it is judged as an anomaly;

[0048] The alarm module includes an audible and visual alarm device and a text message alarm module, which can notify operation and maintenance personnel in various ways;

[0049] It includes a data storage module for storing collected raw data, pre-processed data, and parameters and results during model training;

[0050] Working principle of the data storage module: The data storage module is used to store the collected raw data, including voltage data V(t), current data I(t), temperature data T(t), as well as the denoised and normalized data processed by the data preprocessing module, and the parameters and results of the model training process. These data are stored in an orderly manner to facilitate subsequent query and analysis and provide data support for further optimization of the model.

[0051] Working principle:

[0052] Working Principle of the Data Acquisition Module: The data acquisition module uses voltage sensors, current sensors, and temperature sensors to collect voltage data (V(t), current data (I(t), and temperature data (T(t))) during the operation of electrical equipment. These sensors are tightly connected to key parts of the electrical equipment. The voltage sensor monitors the equipment's voltage, the current sensor detects current changes, and the temperature sensor senses the equipment's temperature. They acquire equipment operating data in real time and transmit the data to the data preprocessing module, providing raw information for subsequent analysis.

[0053] Working principle of the data preprocessing module: The data preprocessing module receives data from the data acquisition module and uses the mean filtering algorithm to denoise the data. By setting the filter window size m, the original data sequence Xn is calculated to obtain the filtered data Yn, removing noise interference in the data and improving data quality. At the same time, the module also performs data normalization. Based on the minimum value x_min and maximum value x_max of the data, the original data x is mapped to the interval [0,1] to obtain x_norm, making the data more suitable for subsequent analysis and processing;

[0054] Working Principle of the Feature Extraction Module: The feature extraction module first processes the pre-processed data through fast Fourier transform (FFT), converting the time domain data x(n) into frequency domain data X(k) to obtain the frequency domain characteristics of the data, helping to analyze the frequency characteristics of the equipment operating status. In addition, the wavelet transform is used, combined with the scale factor a, the translation factor b, and the wavelet basis function psi(t), to calculate the original signal f(t) to obtain Wf[a,b], thereby extracting the time-frequency characteristics of the data and comprehensively reflecting the data characteristics from different angles.

[0055] Working Principle of the Model Training Module: The model training module uses a deep neural network (DNN) for training. The DNN contains multiple hidden layers. The activation function of the hidden layer neurons uses the ReLU function. By performing nonlinear transformations on the input data, the model's learning ability is enhanced. During the training process, the cross-entropy loss function is used to measure the difference between the model's predicted results p[i][j] and the true labels y[i][j]. The model parameters are continuously adjusted to enable the model to accurately learn the characteristic patterns of normal and abnormal states of electrical equipment.

[0056] Working principle of the anomaly recognition module: The anomaly recognition module receives data processed by the feature extraction module, uses the trained model to analyze and predict the real-time data, obtains the anomaly probability output by the model, and then uses the threshold judgment method to compare the anomaly probability output by the model with the set threshold T_th. If the anomaly probability is greater than the set threshold T_th, it is determined that the equipment operation status is abnormal and outputs an anomaly signal to the alarm module;

[0057] Working principle of the alarm module: After receiving the abnormal signal from the abnormality identification module, the alarm module activates the sound and light alarm device to sound an alarm to attract the attention of on-site personnel. At the same time, the SMS alarm module sends a text message notification to the operation and maintenance personnel to inform them of the abnormality of the equipment so that the operation and maintenance personnel can take timely measures to deal with the abnormality;

[0058] Working principle of the data storage module: The data storage module is used to store the collected raw data, including voltage data V(t), current data I(t), temperature data T(t), as well as the denoised and normalized data processed by the data preprocessing module, and the parameters and results of the model training process. These data are stored in an orderly manner to facilitate subsequent query and analysis and provide data support for further optimization of the model.

[0059] The collaborative working principle of the overall system: the data acquisition module collects the operating data of electrical equipment, the data preprocessing module denoises and normalizes it, the feature extraction module extracts multi-dimensional features, the model training module trains the model to learn the equipment status feature pattern, the anomaly recognition module determines whether the equipment is abnormal based on the model, the alarm module issues an alarm when the equipment is abnormal, and the data storage module stores various types of data. The modules work together to realize intelligent identification of abnormal conditions of electrical equipment and ensure the safe and stable operation of electrical equipment.

[0060] in,

[0061] Multi-source data collection and preprocessing mechanism:

[0062] The system builds a multi-source data acquisition system using voltage sensors, current sensors, and temperature sensors. The voltage sensor uses the Hall effect principle to monitor device voltage changes in real time, with a sampling frequency of 10kHz to capture transient fluctuations. The current sensor uses Rogowski coil technology to achieve non-contact measurement of device current, with a measurement range of 0-1000A. The temperature sensor uses a high-precision PT100 platinum resistance with a measurement accuracy of ±0.1°C, ensuring accurate perception of device temperature rise.

[0063] The collected raw data first enters the data preprocessing module, which integrates a dual processing mechanism of mean filtering and data normalization. Taking voltage data as an example, assuming the original voltage sequence V(t) = [221.3, 220.5, 219.8, 220.7, 221.1], when using mean filtering with a window size of m = 2, for time n = 3:

[0064] Yn=(1 / (2*m+1))*sum(V[n+i]for iin range(-m,m+1))

[0065] Yn=(1 / (2*2+1))*(V[3-2]+V[3-1]+V[3]+V[3+1]+V[3+2])

[0066] Yn=(1 / 5)*(221.3+220.5+219.8+220.7+221.1)≈220.68V

[0067] The filtered data is further normalized and mapped to the [0, 1] interval. Assuming that the historical minimum value of the voltage data is Vmin = 180V and the maximum value is Vmax = 260V, the normalized value is:

[0068] x_norm=(x-x_min) / (x_max-x_min)

[0069] V_norm(3)=(220.68-180) / (260-180)≈0.5085

[0070] Multidimensional feature extraction technology:

[0071] The feature extraction module uses a combination of FFT and time-frequency analysis to analyze the device operating status from different dimensions. For a current data sequence with a length of N = 1024 points, the FFT transform converts the time domain signal into a frequency domain representation:

[0072] X[k]=sum(x[n]*exp(-1j*(2*pi / N)*k*n)for n in range(N))

[0073] X[k]=sum(I_norm[n]*exp(-1j*(2*pi / 1024)*k*n)for n in range(1024)),k=0,1,…,1023

[0074] By analyzing the spectrum distribution, the 50Hz fundamental frequency component and its harmonic components can be identified to determine whether the equipment has harmonic pollution;

[0075] At the same time, the system uses Daubechies wavelet for time-frequency analysis, setting the scale factor a = 4, the translation factor b = 2, for the temperature data T_norm(t):

[0076] Wf[a,b]=(1 / sqrt(a))*integral(f[t]*conjugate(psi[(tb) / a])for t inrange(-infinity,infinity))

[0077] Wf[4,2]=(1 / sqrt(4))*integral(T_norm[t]*conjugate(psi[(t-2) / 4])for tin range(-infinity,infinity))

[0078] The time-frequency analysis results can capture the temperature mutation point and provide a basis for early fault warning;

[0079] Deep neural network training mechanism:

[0080] The model training module constructs a deep neural network with five hidden layers, each containing 128 neurons. The network structure adopts an input layer-hidden layer-attention mechanism-output layer architecture design, in which the attention mechanism focuses on key features through weight distribution:

[0081] Attention(Q,K,V)=softmax((Q*K^T) / sqrt(d_k))*V

[0082] During the training process, a small batch stochastic gradient descent algorithm is used, the batch size is set to 64, and the learning rate is dynamically adjusted using the cosine annealing strategy:

[0083] eta_t=eta_min+0.5*(eta_max-eta_min)*(1+cos((T_cur / T_max)*pi))

[0084] The loss function uses weighted cross entropy loss, giving higher weights to abnormal samples w_abnormal=5:

[0085] L=-(1 / N)*sum(sum(w_j*y_ij*log(p_ij)for j in range(C))for iin range(N))

[0086] Among them, y_ij is the true label of sample i belonging to category j, and p_ij is the model prediction probability.

[0087] Adaptive anomaly recognition and graded alarm:

[0088] The anomaly identification module uses a two-layer threshold judgment mechanism. First, the abnormal probability P_abnormal output by the model is compared with the dynamic threshold T_th. The dynamic threshold is adaptively adjusted according to the operating status of the device:

[0089] T_th=T_base+alpha*sigma

[0090] Among them, T_base = 0.7 is the basic threshold, sigma is the standard deviation of the historical anomaly probability, and alpha = 0.3 is the adjustment coefficient;

[0091] When P_abnormal>T_th, the system triggers a secondary alarm and activates the sound and light alarm device. When P_abnormal>T_th+beta*sigma (beta=0.5), a primary alarm is triggered, and the sound and light alarm and SMS alarm modules are activated at the same time. The SMS alarm adopts a hierarchical push mechanism, and automatically selects the corresponding level of operation and maintenance personnel according to the severity of the abnormality:

[0092] Push level = floor((P_abnormal - T_th) / gamma)+1

[0093] Among them, gamma = 0.1 is the level interval parameter;

[0094] Full-process data management and optimization:

[0095] The data storage module uses a combination of a distributed file system and a time series database to store raw data, pre-processed data, and model parameters at a frequency of 10 Hz. The system performs incremental learning on newly collected data every hour and fully updates the model every 24 hours:

[0096] theta_t+1=(1-lambda)*theta_t+lambda*nabla L(theta_t;

[0097] D_new)

[0098] Where theta_t is the model parameter at time t, lambda=0.05 is the learning rate, and D_new is the newly added training data;

[0099] Through continuous learning, the system can adapt to changes in characteristics caused by device aging and maintain recognition accuracy. Experimental data shows that after 300 days of continuous operation, the system's anomaly recognition accuracy remains above 98.7%, with a false alarm rate of less than 0.3%.

[0100] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. An artificial intelligence-based intelligent recognition system for abnormal conditions of electrical equipment, including a data acquisition module, a data preprocessing module, a feature extraction module, a model training module, an abnormality recognition module, and an alarm module, is characterized by: The data acquisition module is used to collect various operating data of electrical equipment, including voltage data V(t), current data I(t), and temperature data T(t); The data preprocessing module performs denoising on the collected data using the mean filter algorithm. The formula is: Yn = (1 / (2*m+1))*sum(X[n+i]for iin range(-m,m+1)), where Xn is the original data sequence, Yn is the filtered data, and m is the filter window size. The feature extraction module extracts the frequency domain features of the data through fast Fourier transform (FFT). The formula is: X[k] = sum(x[n]*exp(-1j*(2*pi / N)*k*n)for n in range(N)), where k ranges from 0 to N-1, where x(n) is the time domain data, X(k) is the frequency domain data, and N is the number of data points. The model training module uses a deep neural network (DNN) for training, and the loss function uses the cross entropy loss function, with the formula: L = -(1 / N)*sum(sum(y[i][j]*log(p[i][j])for j in range(C))for i in range(N)), where N is the number of samples, C is the number of categories, y[i][j] is the true label (0 or 1) of sample i belonging to category j, and p[i][j] is the probability predicted by the model that sample i belongs to category j; The anomaly recognition module identifies anomalies in real-time data based on the trained model and outputs an anomaly signal if an anomaly is identified; After receiving the abnormal signal, the alarm module issues an alarm message.

2. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1 is characterized in that: The data acquisition module uses a variety of sensors, including voltage sensors, current sensors and temperature sensors, which are used to collect corresponding physical quantities respectively.

3. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1 is characterized in that: The data preprocessing module also includes data normalization processing, mapping the data to the [0,1] interval, the formula is: x_norm = (x-x_min) / (x_max-x_min), where x is the original data, x_min and x_max are the minimum and maximum values ​​of the data respectively, and x_norm is the normalized data.

4. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1 is characterized in that: The feature extraction module also uses wavelet transform to extract the time-frequency features of the data. The wavelet transform formula is: Wf[a,b]=(1 / sqrt(a))*integral(f[t]*conjugate(psi[(tb) / a])for t in range(-infinity,infinity)), where f(t) is the original signal, a is the scale factor, b is the translation factor, psi(t) is the wavelet basis function, and conjugate(psi(t)) is its conjugate function.

5. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1 is characterized in that: The deep neural network of the model training module includes multiple hidden layers, and the activation function of the hidden layer neurons adopts the ReLU function, and the formula is: ReLU(x)=max(0,x).

6. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1 is characterized in that: The anomaly recognition module uses a threshold judgment method combined with the model output result to make an anomaly judgment. If the anomaly probability output by the model is greater than the set threshold T_th, it is judged as an anomaly.

7. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1, characterized in that: The alarm module includes an audible and visual alarm device and a text message alarm module, which notifies operation and maintenance personnel in various ways.

8. The artificial intelligence-based intelligent recognition system for abnormal status of electrical equipment according to claim 1, characterized in that: It includes a data storage module for storing the collected raw data, preprocessed data, and parameters and results during the model training process.

Images