A knowledge distillation-based voltage metering reliability evaluation method in a complex environment

Abstract

This invention discloses a method for assessing the reliability of voltage metering in complex environments based on knowledge distillation, comprising the following steps: collecting relevant voltage metering data; data preprocessing; constructing a deep neural network model for voltage reliability assessment; compressing the model through knowledge distillation; training the deep neural network using training set data and optimizing network weights; testing the model using test set data; and determining the reliability of the target voltage through threshold comparison. This invention incorporates multiple units such as voltage transformers, energy meters, and secondary circuits into the assessment model, providing a more comprehensive assessment of voltage metering reliability; employing deep neural network modeling and compressing the model through knowledge distillation effectively improves the system's computational efficiency and accuracy, enabling the assessment model to adapt to resource-constrained industrial control computer platforms; and adapting to various environmental changes to provide more accurate and stable voltage metering assessments, significantly improving the reliability of the voltage metering system.

Description

Technical Field

[0001] This application relates to the field of power data acquisition technology, and in particular to a method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation. Background Technology

[0002] Voltage metering reliability assessment is the process of evaluating the ability of voltage measurement equipment or systems to maintain accuracy and stability during long-term use. As a core component of modern power systems, the operation of smart grids relies on accurate and real-time power data acquisition and analysis. Voltage metering is not only the foundation for monitoring grid operation status but also a key basis for multiple functions such as fault diagnosis, load management, and power quality optimization. With the gradual construction and modernization of smart grids, the reliability of voltage metering systems directly affects the safety and stability of the power system and the efficient utilization of electrical energy. To address potential voltage fluctuations and voltage dips during grid operation, assessing the accuracy, timeliness, and reliability of voltage metering becomes particularly important. By introducing high-precision metering equipment, advanced monitoring technologies, and data analysis methods, the reliability of voltage metering can be improved, potential problems in the power system can be detected in a timely manner, thereby ensuring the efficient operation of the grid and the stability of power supply.

[0003] Existing voltage measurement methods mainly include traditional analog measurement methods and digital voltage measurement methods. Traditional analog measurement methods rely primarily on mechanical instruments (such as voltmeters and ammeters), directly measuring voltage values ​​through electrical principles. Their accuracy and response speed are relatively poor, and they are easily affected by external environmental factors, leading to low reliability of measurement results. Digital voltage measurement methods utilize digital signal processing technology to convert voltage signals into digital signals, which are then processed and stored by a computer system. This method offers higher measurement accuracy, faster response, and stronger anti-interference capabilities, but it places higher demands on hardware and may be affected by system latency, data transmission errors, and other factors. Therefore, most existing methods neglect the impact of environmental factors (such as atmospheric humidity, altitude, and temperature) on voltage accuracy and cannot comprehensively handle the interactions between multiple components such as voltage transformers, energy meters, and secondary circuits. Furthermore, existing technologies often lack effective modeling and optimization methods when dealing with complex, high-dimensional, multi-factor data, especially on resource-constrained industrial control computer platforms where real-time, efficient computation is difficult. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a voltage measurement reliability assessment method based on knowledge distillation in complex environments. This method comprehensively considers the influence of multiple environmental factors through a deep neural network model and combines knowledge distillation technology to compress the model. While ensuring high-precision assessment, it adapts to complex working conditions and improves computational efficiency, thereby solving the reliability assessment problem of existing technologies under environmental changes and computational resource constraints.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following means:

[0006] A method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation includes the following steps:

[0007] S1. Collect relevant data for voltage measurement;

[0008] S2, Data Preprocessing;

[0009] S3. Construct a deep neural network model for voltage reliability assessment;

[0010] S4. Model compression is achieved through knowledge distillation;

[0011] S5. Train a deep neural network using the training set data and optimize the network weights;

[0012] S6. Test the model using the test set data;

[0013] S7. Determine whether the target voltage is reliable by comparing threshold values.

[0014] Furthermore, the specific method for step S1 is as follows:

[0015] Raw data is collected from the three core units of the voltage metering system, namely a voltage transformer, an energy meter, and a secondary circuit. The raw data is then organized into a dataset X, the expression of which is as follows:

[0016] X=[D1,D2,D3,D4,D5,D6,D7,D8,D9,D 10 ]

[0017] Where D1, D2, D3, D4, and D5 represent the atmospheric humidity, altitude, air pollution level, operating time, and maximum operating current of the voltage transformer, respectively; D6 and D7 represent the operating time and maximum operating current of the energy meter, respectively; and D8, D9, and D5 represent the operating time and maximum operating current of the energy meter, respectively. 10 These represent the temperature, operating time, and maximum operating current of the secondary circuit, respectively.

[0018] The i-th sample in dataset X is X i The j-th feature is D j .

[0019] Furthermore, the specific method for step S2 is as follows:

[0020] If dataset X contains missing values, imputation is used to handle these missing values, ensuring that each feature is used during training. The features in dataset X are then standardized so that their values ​​fall within the same range. This standardization process is implemented using the following expression:

[0021]

[0022] Where, μ j and σ j Let represent the mean and standard deviation of the j-th feature, respectively. This represents the standardized value;

[0023] The standardized dataset is divided into a training set and a test set, with the training set accounting for 70% and the test set accounting for 30%.

[0024] Furthermore, the specific method for step S3 is as follows:

[0025] By constructing a deep neural network (DNN) model, a multi-layer neural network is used to perform non-linear mapping on the input feature data. The neural network includes an input layer, multiple hidden layers, and an output layer.

[0026] The input layer consists of 10 standardized features;

[0027] The hidden layer contains k fully connected layers, and the output expression of each layer is as follows:

[0028] h l =σ(W l ·h l-1 +b l )

[0029] Among them, W l Let b represent the weight matrix of the l-th layer. l σ represents the bias term, and σ represents the activation function.

[0030] The output layer calculates the reliability score for each sample using the softmax activation function, as shown in the following expression:

[0031]

[0032] Where p(X) i ) represents the reliability score of the sample, h k (X i ) represents sample X i In network output, h k (Xj ) represents sample X j Output in the network.

[0033] Furthermore, the specific method for step S4 is as follows:

[0034] During training, knowledge learned by the teacher model is transferred to the student model through knowledge distillation. The teacher model is obtained through a deep neural network using the dataset and training method described above. Soft labels are provided by the probability distribution output by the softmax activation function. The student model learns these soft labels, thereby mimicking the decision-making process of the teacher model.

[0035] The loss function of the student model includes the knowledge loss from the teacher model and the cross-entropy loss of the standard labels. The specific expression of its loss function is as follows:

[0036]

[0037] Where λ represents the weight parameter, D KL P represents the Kullback-Leibler divergence, which measures the difference between two probability distributions. teacher (X) and P student (X) represent the output probability distributions of the teacher and student models, respectively, and L... CE Let be the cross-entropy loss function, representing the student model's ability to predict the true labels, and y and These represent the true labels of the samples and the predicted labels of the student model, respectively.

[0038] Furthermore, the specific method for step S5 is as follows:

[0039] The network weight parameters are updated and optimized using gradient descent. The specific update and optimization expression is as follows:

[0040]

[0041] Where θ represents the parameters to be optimized in the network, and η represents the learning rate. This represents the gradient of the loss function with respect to the parameters.

[0042] Furthermore, the specific method for step S6 is as follows:

[0043] After training, the test set is input into the updated and optimized model, the model's output probability p is calculated, and it is compared with the true label y. The accuracy of the model is calculated as follows:

[0044]

[0045] in, Indicates the indicator function, y i and Let represent the true label of the i-th sample and the prediction result of the model, respectively.

[0046] Furthermore, the specific method for step S7 is as follows:

[0047] For the model output a probability value p(X) i By setting a threshold, the expression for converting it into a category label is as follows:

[0048]

[0049] Final output This indicates whether the voltage is within the standard range, directly assesses whether the voltage of the model system meets the design requirements, and reflects the operating status of the power grid. If the voltage deviates from the standard range, it means that the metering equipment has made an error or malfunction, or that the power grid has experienced abnormal fluctuations.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] This invention comprehensively considers multiple units, including voltage transformers, energy meters, secondary circuits, and their environmental factors, in the evaluation model, enabling a more comprehensive assessment of voltage metering reliability. By employing deep neural network (Network-T) modeling and compressing the model using knowledge distillation (Network-S), the system's computational efficiency and accuracy are effectively improved. This allows the evaluation model to adapt to resource-constrained industrial control computer platforms, resulting in stronger application prospects. Furthermore, by training the deep neural network and combining it with distillation technology, the system can adapt to various environmental changes and provide more accurate and stable voltage metering assessments. It operates stably and continuously optimizes in complex power environments, significantly improving the reliability of the voltage metering system. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0053] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and specific examples.

[0054] This invention aims to evaluate the reliability of voltage metering systems in complex environments. It comprehensively considers the impact of multiple influencing factors (such as atmospheric humidity, altitude, meter operating time, and temperature) on voltage metering accuracy, ensuring that voltage metering continuously and accurately reflects the actual situation of the power grid and providing a basis for system maintenance and optimization. This invention employs deep neural network (DNN) modeling and optimizes the model through knowledge distillation to ensure efficient operation even on resource-constrained industrial control computers.

[0055] like Figure 1As shown, a method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation includes the following steps:

[0056] S1. Collect relevant voltage measurement data, using the following methods:

[0057] Raw data is collected from the three core units of the voltage metering system, namely a voltage transformer, an energy meter, and a secondary circuit. The raw data is then organized into a dataset X, the expression of which is as follows:

[0058] X=[D1,D2,D3,D4,D5,D6,D7,D8,D9,D 10 ]

[0059] Where D1, D2, D3, D4, and D5 represent the atmospheric humidity, altitude, air pollution level, operating time, and maximum operating current of the voltage transformer, respectively; D6 and D7 represent the operating time and maximum operating current of the energy meter, respectively; and D8, D9, and D5 represent the operating time and maximum operating current of the energy meter, respectively. 10 These represent the temperature, operating time, and maximum operating current of the secondary circuit, respectively.

[0060] The i-th sample in dataset X is X i The j-th feature is D j .

[0061] S2. Data preprocessing, the specific methods are as follows:

[0062] If dataset X contains missing values, imputation methods are used to handle these missing values, ensuring that each feature is used during training. Since the dimensions of various influencing factors differ, each feature needs to be standardized to ensure its values ​​are within the same range (e.g., zero mean, unit variance). The standardization process for each feature in dataset X, ensuring its values ​​are within the same range, is implemented using the following expression:

[0063]

[0064] Where, μ j and σ j Let represent the mean and standard deviation of the j-th feature, respectively. This represents the standardized value;

[0065] The standardized dataset is divided into a training set and a test set, with the training set accounting for 70% and the test set accounting for 30%.

[0066] S3. Construct a deep neural network model for voltage reliability assessment. The specific method is as follows:

[0067] By constructing a deep neural network (DNN) model, a multi-layer neural network is used to perform non-linear mapping on the input feature data. The neural network includes an input layer, multiple hidden layers, and an output layer.

[0068] The input layer consists of 10 standardized features.

[0069] The hidden layer contains k fully connected layers, and the output expression of each layer is as follows:

[0070] h l =σ(W l ·h l-1 +b l )

[0071] Among them, W l Let b represent the weight matrix of the l-th layer. l σ represents the bias term, and σ represents the activation function.

[0072] The output layer calculates the reliability score for each sample using the softmax activation function, as shown in the following expression:

[0073]

[0074] Where p(X) i ) represents the reliability score of the sample, h k (X i ) represents sample X i In network output, h k (X j ) represents sample X j Output in the network.

[0075] S4. Model compression is performed through knowledge distillation, as follows:

[0076] During training, knowledge distillation is used to transfer the knowledge learned by the teacher model (Network-T) to the student model (Network-S). The teacher model is obtained through a deep neural network using the aforementioned dataset and training method, while the student model is a smaller neural network that uses the same input features but with fewer layers and neurons. Knowledge distillation allows the student model to operate efficiently in computationally limited environments. Soft labels are provided by the probability distribution output by the softmax activation function, and the student model learns these soft labels to mimic the decision-making process of the teacher model.

[0077] The loss function of the student model includes the knowledge loss from the teacher model and the cross-entropy loss of the standard labels. The specific expression of its loss function is as follows:

[0078]

[0079] Where λ represents the weight parameter, D KL P represents the Kullback-Leibler divergence, which measures the difference between two probability distributions. teacher (X) and P student (X) represent the output probability distributions of the teacher and student models, respectively, and L... CE Let be the cross-entropy loss function, representing the student model's ability to predict the true labels, and y and These represent the true labels of the samples and the predicted labels of the student model, respectively. Through the distillation process, the student model not only learns the standard labels but also absorbs the rich inter-class relationship information provided by the teacher model, thereby improving its performance.

[0080] S5. Train a deep neural network using the training set data and optimize the network weights. The specific method is as follows:

[0081] The network weight parameters are updated and optimized using gradient descent. The specific update and optimization expression is as follows:

[0082]

[0083] Where θ represents the parameters to be optimized in the network, and η represents the learning rate. This represents the gradient of the loss function with respect to the parameters.

[0084] S6. Test the model using the test set data, as follows:

[0085] After training, the test set is input into the updated and optimized model, the model's output probability p is calculated, and it is compared with the true label y. The accuracy of the model is calculated as follows:

[0086]

[0087] in, Indicates the indicator function, y i and Let represent the true label of the i-th sample and the prediction result of the model, respectively.

[0088] S7. Determine whether the target voltage is reliable by comparing threshold values. The specific method is as follows:

[0089] For the model output a probability value p(X) i By setting a threshold, the expression for converting it into a category label is as follows:

[0090]

[0091] Final output This indicates whether the voltage is within the standard range, directly assesses whether the voltage of the model system meets the design requirements, and reflects the operating status of the power grid. If the voltage deviates from the standard range, it means that the metering equipment has made an error or malfunction, or that the power grid has experienced abnormal fluctuations.

[0092] The above description is merely an embodiment of the present invention. It should be reiterated that those skilled in the art can make several improvements to the present invention without departing from the principle of the present invention, and these improvements are also included within the scope of protection of the claims of the present invention.

Claims

1. A method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation, characterized in that: The steps include the following: S1. Collect relevant data for voltage measurement; S2, Data Preprocessing; S3. Construct a deep neural network model for voltage reliability assessment; S4. Model compression is achieved through knowledge distillation; S5. Train a deep neural network using the training set data and optimize the network weights; S6. Test the model using the test set data; S7. Determine whether the target voltage is reliable by comparing threshold values; The specific method for step S1 is as follows: Raw data is collected from the three core units of the voltage metering system, namely a voltage transformer, an energy meter, and a secondary circuit. The raw data is then organized into a dataset X, the expression of which is as follows: X=[D1,D2,D3,D4,D5,D6,D7,D8,D9,D 10 ] Where D1, D2, D3, D4, and D5 represent the atmospheric humidity, altitude, air pollution level, operating time, and maximum operating current of the voltage transformer, respectively; D6 and D7 represent the operating time and maximum operating current of the energy meter, respectively; and D8, D9, and D5 represent the operating time and maximum operating current of the energy meter, respectively. 10 These represent the temperature, operating time, and maximum operating current of the secondary circuit, respectively. The i-th sample in dataset X is X i The j-th feature is D j ; The specific method for step S2 is as follows: If dataset X contains missing values, imputation is used to handle these missing values, ensuring that each feature is used during training. The features in dataset X are then standardized so that their values ​​fall within the same range. This standardization process is implemented using the following expression: Where, μ j and σ j Let represent the mean and standard deviation of the j-th feature, respectively. This represents the standardized value; The standardized dataset is divided into a training set and a test set, with the training set accounting for 70% and the test set accounting for 30%. The specific method for step S3 is as follows: By constructing a deep neural network model, a multi-layer neural network is used to perform non-linear mapping on the input feature data. The neural network includes an input layer, multiple hidden layers, and an output layer. The input layer consists of 10 standardized features; The hidden layer contains k fully connected layers, and the output expression of each layer is as follows: h l =σ(W l ·h l-1 +b l ) Among them, W l Let b represent the weight matrix of the l-th layer. l σ represents the bias term, and σ represents the activation function. The output layer calculates the reliability score for each sample using the softmax activation function, as shown in the following expression: Where p(X) i ) represents the reliability score of the sample, h k (X i ) represents sample X i In network output, h k (X j ) represents sample X j Output in the network.

2. The method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation as described in claim 1, characterized in that: The specific method for step S4 is as follows: During training, knowledge learned by the teacher model is transferred to the student model through knowledge distillation. The teacher model is obtained through a deep neural network using the dataset and training method described above. Soft labels are provided by the probability distribution output by the softmax activation function. The student model learns these soft labels, thereby mimicking the decision-making process of the teacher model. The loss function of the student model includes the knowledge loss from the teacher model and the cross-entropy loss of the standard labels. The specific expression of its loss function is as follows: Where λ represents the weight parameter, D KL P represents the Kullback-Leibler divergence, which measures the difference between two probability distributions. teacher (X) and P student (X) represent the output probability distributions of the teacher and student models, respectively, and L... CE Let be the cross-entropy loss function, representing the student model's ability to predict the true labels, and y and These represent the true labels of the samples and the predicted labels of the student model, respectively.

3. The method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation as described in claim 1, characterized in that: The specific method for step S5 is as follows: The network weight parameters are updated and optimized using gradient descent. The specific update and optimization expression is as follows: Where θ represents the parameters to be optimized in the network, and η represents the learning rate. This represents the gradient of the loss function with respect to the parameters.

4. The method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation as described in claim 1, characterized in that: The specific method for step S6 is as follows: After training, the test set is input into the updated and optimized model, the model's output probability p is calculated, and it is compared with the true label y. The accuracy of the model is calculated as follows: in, Indicates the indicator function, y i and Let represent the true label of the i-th sample and the prediction result of the model, respectively.

5. The method for assessing the reliability of voltage measurement in complex environments based on knowledge distillation as described in claim 1, characterized in that: The specific method for step S7 is as follows: For the model output a probability value p(X) i By setting a threshold, the expression for converting it into a category label is as follows: Final output It indicates whether the voltage is within the standard range, directly assesses whether the voltage of the model system meets the design requirements, and reflects the operating status of the power grid.

Images