An intelligent state recognition and regulation method and system for an electrical device

Abstract

The application relates to the technical field of electrical equipment monitoring, and particularly discloses an intelligent state recognition and regulation method and system for electrical equipment, which realizes real-time collection of equipment operation data through a plurality of sensors such as a current sensor and a voltage sensor, combines historical and power grid load information, carries out filtering and normalization pretreatment, extracts features by using a convolutional neural network, inputs a support vector machine model to recognize normal, slight abnormal and other states of the equipment, generates a regulation strategy in combination with the power grid load and operation parameters according to the state result, adjusts parameters when the state is slightly abnormal, alarms and cuts off power when the state is seriously abnormal or faulty, and realizes real-time monitoring and feedback optimization. The application realizes accurate recognition and effective regulation of the operation state of the electrical equipment through multi-sensor data collection, an advanced data processing algorithm and an intelligent regulation strategy, improves the operation reliability and stability of the electrical equipment, and guarantees the safe and efficient operation of the power system.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment monitoring technology, specifically to an intelligent status identification and control method and system for electrical equipment. Background Technology

[0002] With the continuous development of power systems and the increasing demands for intelligence, the stable operation of electrical equipment is crucial for ensuring the reliability and security of power supply. Traditional methods for monitoring the condition of electrical equipment mainly rely on manual inspections and simple instrument measurements, which suffer from low monitoring efficiency, poor real-time performance, and difficulty in detecting potential faults. Although some existing automated monitoring technologies can collect some equipment operating data, there is still room for improvement in data processing, accuracy of condition identification, and intelligence of control strategies. For example, the feature extraction for complex operating states of electrical equipment is not comprehensive enough, resulting in low accuracy of condition identification; when equipment malfunctions, control strategies lack flexibility and adaptability, and cannot be effectively adjusted according to the actual situation of the equipment and the grid load.

[0003] Therefore, a method and system for intelligent status identification and control of electrical equipment has become an urgent problem to be solved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for intelligent state identification and control of electrical equipment. Through multi-sensor data acquisition, advanced data processing algorithms and intelligent control strategies, the method and system can accurately identify and effectively control the operating status of electrical equipment, improve the reliability and stability of electrical equipment operation, and ensure the safe and efficient operation of the power system.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for intelligent state identification and control of electrical equipment, comprising the following steps:

[0006] S1. Using current sensors, voltage sensors, temperature sensors, vibration sensors, and humidity sensors, the system collects real-time data on the current, voltage, temperature, vibration frequency, and ambient humidity of the electrical equipment. Simultaneously, it acquires historical operating data, maintenance records, and power grid load information of the equipment through a network communication module.

[0007] S2. Filter the collected raw data to remove noise, and use the normalization method to standardize the data of different types and magnitudes.

[0008] S3. Use a convolutional neural network to extract features from the preprocessed data to obtain key feature vectors that reflect the operating status of electrical equipment.

[0009] S4. Input the extracted feature vectors into the state recognition model built on support vector machine and trained with a large amount of historical data to determine whether the electrical equipment is in normal operation, minor abnormality, serious abnormality or fault state.

[0010] S5. Based on the equipment status identification results, combined with the power grid load information and equipment operating parameters, generate control strategies: when the equipment is in a slight abnormality, adjust the operating parameters; when it is in a serious abnormality or fault, issue an alarm and cut off the power supply according to the preset protection logic.

[0011] S6. The control module sends the control strategy to the electrical equipment actuator, monitors the control effect in real time, and provides feedback for optimization.

[0012] Furthermore, the preprocessed multidimensional time-series data is divided into three-dimensional tensor inputs according to time windows, and the tensor dimension is [time step × number of sensor types × feature dimension];

[0013] The construction of a multi-layer convolutional neural network architecture includes, in sequence:

[0014] The first layer is a one-dimensional convolutional layer with 16 convolutional kernels of size 3, a stride of 1, and the activation function ReLU, used to extract short-term temporal features;

[0015] The max pooling layer, with a pooling window size of 2 and a stride of 2, is used for feature dimensionality reduction and noise suppression.

[0016] The second one-dimensional convolutional layer uses 32 convolutional kernels of size 5 with a stride of 1 and the activation function LeakyReLU to extract mid-term temporal dependencies.

[0017] The global average pooling layer compresses the temporal dimension into a feature vector;

[0018] Batch normalization layers and Dropout layers are inserted between convolutional layers, with a Dropout rate of 0.2, to prevent overfitting and accelerate training;

[0019] The features are mapped to a low-dimensional space through a fully connected layer, generating a key feature vector with a dimension of 64. This key feature vector contains time-frequency domain features of the electrical equipment's operating status, multi-sensor correlation features, and environmental impact features.

[0020] The network is trained using a triplet loss function to enhance the clustering of features of similar states and the separation of features of dissimilar states. The training data includes normal, abnormal, and fault state samples.

[0021] Furthermore, the specific methods for determining the operating status of electrical equipment are as follows:

[0022] The 64-dimensional key feature vector extracted by the convolutional neural network is used as input, and the four states of the device—normal, minor abnormality, severe abnormality, and fault—are used as output labels.

[0023] A multi-class state recognition model is constructed using the SVM-RBF kernel function. The four-class classification problem is transformed into six binary classification sub-problems using a One-vs-One strategy. Each sub-problem is responsible for distinguishing two classes of states.

[0024] The key feature vectors are Z-score standardized, and their Mahalanobis distances to the four state centers are calculated. The final state category is determined by a voting mechanism: six binary classifiers are used for voting. If the number of votes for a certain state exceeds the threshold and the Mahalanobis distance from the sample to that state is the smallest, the device is determined to be in that state. When the difference between the number of votes for any two states is less than 5%, a fuzzy state determination is triggered.

[0025] The method for determining the fuzzy state is as follows: calculate the distance from the key feature vector to each classification hyperplane. If all distances are greater than the threshold, the state is determined to be unknown. Otherwise, select the two closest categories and make a secondary determination based on the device's historical state transition probability.

[0026] When five consecutive samples are classified as unknown, the incremental learning process is initiated. The 20 most representative samples are selected through active learning for manual annotation, and the training set is updated. A sliding window mechanism is adopted, where the earliest 50 samples are discarded for every 100 newly added labeled samples to maintain a stable training set size.

[0027] Furthermore, in step S5, when the equipment is in a slight abnormality, the specific method for adjusting the operating parameters is as follows: based on the parameter distribution under normal conditions in the equipment's historical operating data, combined with the current power grid load information, the voltage, current, and power parameters of the electrical equipment are adjusted through preset parameter adjustment rules, and the adjusted equipment operating data is recorded for subsequent model optimization.

[0028] Furthermore, in step S6, when monitoring the control effect in real time, the effectiveness of the control strategy is judged by comparing the changes in the operating parameters of the electrical equipment before and after the control and whether the status identification results return to normal; if the control effect does not meet expectations, the control strategy is adjusted and re-executed according to the preset feedback optimization rules.

[0029] The present invention also provides an intelligent state identification and control system for electrical equipment, used to implement the above method, comprising:

[0030] Data acquisition module: includes current sensor, voltage sensor, temperature sensor, vibration sensor, humidity sensor and network communication module, used to collect electrical equipment operation data, historical data, maintenance records and power grid load information in real time;

[0031] Data preprocessing module: Filter and normalize the collected raw data;

[0032] Feature extraction module: Based on a convolutional neural network, extract features from the preprocessed data to generate key feature vectors;

[0033] Status recognition module: Adopt a support vector machine model to judge the operating status of electrical equipment according to the extracted key feature vectors;

[0034] Regulation strategy generation module: Generate corresponding regulation strategies according to the equipment status, grid load and operating parameters;

[0035] Regulation execution and feedback module: Send the regulation strategy to the actuator, and monitor the regulation effect in real time, and optimize according to the feedback.

[0036] The advantages of the present invention compared with the prior art are as follows:

[0037] The present invention collects rich data through multiple types of sensors, combines historical and grid load information, and adopts advanced data preprocessing and feature extraction algorithms, which can comprehensively and accurately mine the operating state characteristics of electrical equipment and improve the accuracy of status recognition.

[0038] The status recognition model of the present invention based on a support vector machine, combined with multiple judgment strategies and an incremental learning mechanism, can effectively handle complex status judgment situations and improve the reliability and adaptability of status recognition.

[0039] The present invention generates reasonable regulation strategies according to different equipment states and grid conditions, and continuously improves the regulation effect through a feedback optimization mechanism, which can effectively ensure the safe and stable operation of electrical equipment and improve the reliability and efficiency of the power system.

[0040] The intelligent status recognition and regulation system supporting the present invention, with each module working together, realizes the full-process automation and intelligence from data collection to regulation optimization, and has good practicability and popularization value. Brief description of the drawings

[0041] Figure 1 is a flow chart of an intelligent status recognition and regulation method for an electrical equipment of the present invention.

[0042] Figure 2 is a system block diagram of an intelligent status recognition and regulation system for an electrical equipment of the present invention. Detailed implementation manners

[0043] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present invention.

[0044] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0045] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0046] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0047] The following detailed description, in conjunction with the accompanying drawings, provides a method and system for intelligent state identification and control of electrical equipment according to the present invention.

[0048] Combined with appendix Figure 1-2 This invention will be described in detail below.

[0049] A method for intelligent state identification and control of electrical equipment specifically includes the following steps:

[0050] Data Acquisition: Utilizing current, voltage, temperature, vibration, and humidity sensors, real-time data on current, voltage, temperature, vibration frequency, and ambient humidity of the electrical equipment are collected. Simultaneously, historical operating data, maintenance records, and power grid load information are acquired via a network communication module. The collaborative work of multiple sensors comprehensively collects equipment operating data, which, combined with historical and power grid load information, provides a rich data foundation for subsequent analysis.

[0051] Data preprocessing: The collected raw data is filtered to remove noise, and normalization methods are used to standardize data of different types and magnitudes. Filtering and normalization improve data quality, making different types of data comparable and facilitating subsequent algorithm processing.

[0052] Feature Extraction: Convolutional neural networks are used to extract features from the preprocessed data, obtaining key feature vectors reflecting the operating status of electrical equipment. Specifically, the preprocessed multidimensional time-series data is divided into three-dimensional tensor inputs according to time windows, with tensor dimensions of [time step × number of sensor types × feature dimension]. A multi-layer convolutional neural network architecture is constructed, including: a first one-dimensional convolutional layer with 16 kernels of size 3, a stride of 1, and ReLU activation function, used to extract short-term time-series features; a max-pooling layer with a pooling window size of 2 and a stride of 2, used for feature dimensionality reduction and noise suppression; and a second one-dimensional convolutional layer with 32 kernels of size 5, a stride of 1, and LeakyReLU activation function, used to extract mid-term time-series dependencies. The architecture employs a global average pooling layer to compress the temporal dimension into feature vectors. Batch normalization and Dropout layers are inserted between convolutional layers, with a Dropout rate of 0.2 to prevent overfitting and accelerate training. Fully connected layers map features to a low-dimensional space, generating a 64-dimensional key feature vector. This key feature vector contains time-frequency domain features of electrical equipment operating states, multi-sensor correlation features, and environmental impact features. A triplet loss function is used to train the network to enhance the clustering of similar state features and the separation of dissimilar state features. The training data includes normal, abnormal, and fault state samples. This convolutional neural network architecture can fully mine the feature information in the data, improving the accuracy and effectiveness of feature extraction.

[0053] State Recognition: The extracted feature vectors are input into a state recognition model built on support vector machines and trained on a large amount of historical data to determine whether electrical equipment is in normal operation, minor abnormality, severe abnormality, or fault state. Specifically, the method is as follows: 64-dimensional key feature vectors extracted by a convolutional neural network are used as input, with the four states of normal, minor abnormality, severe abnormality, and fault as output labels; a multi-class state recognition model is built using the SVM-RBF kernel function, and a one-vs-one strategy is used to transform the four-class problem into six binary sub-problems, each responsible for distinguishing two states; the key feature vectors are Z-score standardized, and their Mahalanobis distances to the centers of the four states are calculated; the final state category is determined through a voting mechanism: six binary classifiers vote, and if the number of votes for a certain state exceeds a threshold, and the sample's distance to that state is... The device is determined to be in the state if the Mahalanobis distance is minimized. When the difference in votes between any two states is less than 5%, a fuzzy state determination is triggered. The method for fuzzy state determination is as follows: calculate the distance from the key feature vector to each classification hyperplane; if all distances are greater than a threshold, the state is determined to be unknown; otherwise, select the two closest categories and perform a secondary determination based on the device's historical state transition probabilities. When five consecutive samples are determined to be in an unknown state, an incremental learning process is initiated. The 20 most representative samples are selected through active learning for manual annotation, updating the training set. A sliding window mechanism is used; for every 100 newly annotated samples, the oldest 50 samples are discarded to maintain a stable training set size. This state recognition method integrates multiple strategies, improving the accuracy and reliability of state recognition and effectively handling complex state determination situations.

[0054] Control Strategy Generation: Based on equipment status identification results, combined with grid load information and equipment operating parameters, a control strategy is generated. When the equipment experiences minor anomalies, operating parameters are adjusted; when it experiences severe anomalies or faults, an alarm is issued and power is cut off according to preset protection logic. Specifically, when the equipment experiences minor anomalies, the operating parameters are adjusted according to the parameter distribution under normal conditions in historical operating data, combined with current grid load information, using preset parameter adjustment rules. The adjusted equipment operating data is recorded for subsequent model optimization. This control strategy can make reasonable adjustments based on different equipment states and grid conditions, ensuring safe equipment operation.

[0055] Control Execution and Feedback Optimization: The control module sends the control strategy to the electrical equipment actuators, monitors the control effect in real time, and provides feedback optimization. During real-time monitoring, the effectiveness of the control strategy is determined by comparing the changes in the operating parameters of the electrical equipment before and after control, and whether the status identification results have returned to normal. If the control effect does not meet expectations, the control strategy is adjusted and re-executed according to preset feedback optimization rules. Through this feedback optimization mechanism, the control strategy is continuously improved, enhancing the control effect.

[0056] Based on the above method, the present invention also provides an intelligent state identification and control system for electrical equipment, comprising:

[0057] Data acquisition module: Includes current sensor, voltage sensor, temperature sensor, vibration sensor, humidity sensor and network communication module, used to collect electrical equipment operation data, historical data, maintenance records and power grid load information in real time.

[0058] Data preprocessing module: Filters and normalizes the collected raw data to improve data quality.

[0059] Feature extraction module: Based on convolutional neural networks, it extracts features from the preprocessed data and generates key feature vectors.

[0060] Status recognition module: It adopts a support vector machine model to determine the operating status of electrical equipment based on the extracted feature vectors.

[0061] Control strategy generation module: Generates corresponding control strategies based on equipment status, grid load, and operating parameters.

[0062] Control Execution and Feedback Module: Sends control strategies to the implementing agencies, monitors the control effects in real time, and optimizes based on feedback.

[0063] The specific implementation process of the intelligent state identification and control method and system for electrical equipment of the present invention is as follows:

[0064] Data acquisition: Current sensors, voltage sensors, temperature sensors, vibration sensors, and humidity sensors are installed on the transformer to collect real-time data on the transformer's current, voltage, temperature, vibration frequency, and ambient humidity. At the same time, historical operating data, maintenance records, and load information of the power grid are obtained through a network communication module.

[0065] Data preprocessing: The collected raw data is filtered to remove noise interference; normalization methods are used to standardize data of different types and magnitudes to make the data comparable.

[0066] Feature Extraction: The preprocessed multidimensional temporal data is segmented into three-dimensional tensors according to time windows and input into a convolutional neural network. The first one-dimensional convolutional layer of this network uses 16 convolutional kernels of size 3 with a stride of 1 and the ReLU activation function to extract short-term temporal features; after max pooling, feature dimensionality reduction and noise suppression are performed. The second one-dimensional convolutional layer uses 32 convolutional kernels of size 5 with a stride of 1 and the LeakyReLU activation function to extract mid-term temporal dependencies; then, a global average pooling layer compresses the temporal dimension into a feature vector. Batch normalization layers and Dropout layers with a dropout rate of 0.2 are inserted between the convolutional layers to prevent overfitting and accelerate training. Finally, a fully connected layer maps the features to a low-dimensional space, generating a 64-dimensional key feature vector. Training data containing normal, abnormal, and fault state samples is used, and the network is trained using a triplet loss function.

[0067] State Recognition: The extracted 64-dimensional key feature vectors are input into a multi-class state recognition model built based on the SVM-RBF kernel function. A one-vs-one strategy is used to transform the four-class classification problem into six binary classification sub-problems. The key feature vectors are Z-score standardized, and their Mahalanobis distances to the state centers of the four classes are calculated. A voting mechanism is used to determine the transformer's operating state. When ambiguous states occur, they are handled according to the corresponding judgment method; when five consecutive samples are judged as unknown states, an incremental learning process is initiated to update the training set.

[0068] Control strategy generation: If the transformer is in a slightly abnormal state, the voltage, current and power parameters of the transformer will be adjusted according to the parameter distribution under normal conditions in its historical operating data and combined with the current grid load information, and the adjusted operating data will be recorded; if it is in a seriously abnormal or faulty state, an alarm will be issued and the power supply will be cut off according to the preset protection logic.

[0069] Control execution and feedback optimization: The control module sends the control strategy to the transformer's actuators, and monitors the changes in transformer operating parameters and status identification results in real time before and after control. If the control effect does not meet expectations, the control strategy is adjusted and re-executed according to preset feedback optimization rules.

[0070] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for intelligent state identification and control of electrical equipment, characterized in that: Includes the following steps: S1. Using current sensors, voltage sensors, temperature sensors, vibration sensors, and humidity sensors, the system collects real-time data on the current, voltage, temperature, vibration frequency, and ambient humidity of the electrical equipment. Simultaneously, it acquires historical operating data, maintenance records, and power grid load information of the equipment through a network communication module. S2. Filter the collected raw data to remove noise, and use the normalization method to standardize the data of different types and magnitudes. S3. Use a convolutional neural network to extract features from the preprocessed data to obtain key feature vectors that reflect the operating status of electrical equipment. S4. Input the extracted feature vectors into the state recognition model built on support vector machine and trained with a large amount of historical data to determine whether the electrical equipment is in normal operation, minor abnormality, serious abnormality or fault state. The specific methods for determining the operating status of electrical equipment are as follows: The 64-dimensional key feature vector extracted by the convolutional neural network is used as input, and the four states of the device—normal, minor abnormality, severe abnormality, and fault—are used as output labels. A multi-class state recognition model is constructed using the SVM-RBF kernel function. The four-class classification problem is transformed into six binary classification sub-problems using a One-vs-One strategy. Each sub-problem is responsible for distinguishing two classes of states. The key feature vectors are Z-score standardized, and their Mahalanobis distances to the four state centers are calculated. The final state category is determined by a voting mechanism: six binary classifiers are used for voting. If the number of votes for a certain state exceeds the threshold and the Mahalanobis distance from the sample to that state is the smallest, the device is determined to be in that state. When the difference between the number of votes for any two states is less than 5%, a fuzzy state determination is triggered. The method for determining the fuzzy state is as follows: calculate the distance from the key feature vector to each classification hyperplane. If all distances are greater than the threshold, the state is determined to be unknown. Otherwise, select the two closest categories and make a secondary determination based on the device's historical state transition probability. When 5 consecutive samples are determined to be in an unknown state, the incremental learning process is initiated. The training set is updated by actively learning to select the 20 most representative samples for manual annotation; a sliding window mechanism is used to discard the earliest 50 samples for every 100 new labeled samples to keep the size of the training set stable. S5. Based on the equipment status identification results, combined with the power grid load information and equipment operating parameters, generate control strategies: when the equipment is in a slight abnormality, adjust the operating parameters; when it is in a serious abnormality or fault, issue an alarm and cut off the power supply according to the preset protection logic. S6. The control module sends the control strategy to the electrical equipment actuator, monitors the control effect in real time, and provides feedback for optimization.

2. The intelligent state recognition and regulation method for electrical equipment according to claim 1, characterized in that: The preprocessed multidimensional time series data is divided into three-dimensional tensor inputs according to time windows, and the tensor dimension is [time step × number of sensor types × feature dimension]; The construction of a multi-layer convolutional neural network architecture includes, in sequence: The first layer is a one-dimensional convolutional layer with 16 convolutional kernels of size 3, a stride of 1, and the activation function ReLU, used to extract short-term temporal features; The max pooling layer, with a pooling window size of 2 and a stride of 2, is used for feature dimensionality reduction and noise suppression. The second one-dimensional convolutional layer uses 32 convolutional kernels of size 5 with a stride of 1 and the activation function LeakyReLU to extract mid-term temporal dependencies. The global average pooling layer compresses the temporal dimension into a feature vector; Batch normalization layers and Dropout layers are inserted between convolutional layers, with a Dropout rate of 0.2, to prevent overfitting and accelerate training; The features are mapped to a low-dimensional space through a fully connected layer, generating a key feature vector with a dimension of 64. This key feature vector contains time-frequency domain features of the electrical equipment's operating status, multi-sensor correlation features, and environmental impact features. The network is trained using a triplet loss function to enhance the clustering of features of similar states and the separation of features of dissimilar states. The training data includes normal, abnormal, and fault state samples. 3.The method of claim 2, wherein: In step S5, when the equipment is in a slight abnormality, the specific method for adjusting the operating parameters is as follows: based on the parameter distribution under normal conditions in the equipment's historical operating data, combined with the current power grid load information, the voltage, current, and power parameters of the electrical equipment are adjusted according to the preset parameter adjustment rules, and the adjusted equipment operating data is recorded for subsequent model optimization.

4. The intelligent state recognition and regulation method of an electrical device according to claim 3, characterized in that: In step S6, when monitoring the control effect in real time, the effectiveness of the control strategy is judged by comparing the changes in the operating parameters of the electrical equipment before and after the control and whether the status identification results return to normal. If the control effect does not meet expectations, the control strategy will be adjusted and re-executed according to the preset feedback optimization rules.

5. An intelligent state identification and control system for electrical equipment, used to implement the intelligent state identification and control method for electrical equipment according to any one of claims 1-4, comprising: Data acquisition module: includes current sensor, voltage sensor, temperature sensor, vibration sensor, humidity sensor and network communication module, used to collect electrical equipment operation data, historical data, maintenance records and power grid load information in real time; Data preprocessing module: performs filtering and normalization on the collected raw data; Feature extraction module: Based on convolutional neural network, it extracts features from preprocessed data and generates key feature vectors; Status recognition module: It adopts a support vector machine model to determine the operating status of electrical equipment based on the extracted key feature vectors; Control strategy generation module: Generates corresponding control strategies based on equipment status, grid load, and operating parameters; Control Execution and Feedback Module: Sends control strategies to the implementing agencies, monitors the control effects in real time, and optimizes based on feedback.

Images