A power system safety monitoring method and electronic device

Abstract

The application discloses a power system safety monitoring method and electronic equipment, and relates to the technical field of power system safety monitoring. Time sequence data obtained by the application is mapped into image data, and then the image data is preprocessed. Next, the preprocessed image data is input into a safety monitoring model constructed based on a convolutional neural network to obtain a safety monitoring result. Moreover, the application performs visual processing on the image data and the monitoring result, and can quickly and accurately realize visual monitoring on the state of power system equipment.

Description

Technical Field

[0001] This invention relates to the field of power system safety monitoring technology, and in particular to a power system safety monitoring method and electronic equipment. Background Technology

[0002] A power system is a complex, large-scale system comprised of power generation, transmission, transformation, distribution, and consumption. Its purpose is to deliver electrical energy from power plants to users and ensure the quality and reliability of power supply. The safe operation of the power system is of great significance to national economic and social development; therefore, real-time monitoring and analysis of the operating status of various devices within the power system are essential.

[0003] Power system safety monitoring refers to the observation and detection of the operating status of critical equipment in a power system to prevent malfunctions or performance degradation from affecting normal system operation and to avoid accidents that endanger safety. Examples include monitoring parameters such as temperature, voltage, current, frequency, and impedance of equipment like transformers, circuit breakers, busbars, and transmission lines, as well as monitoring abnormal sounds, vibrations, and gas signals from the equipment. Power system safety monitoring can promptly detect equipment aging, damage, or malfunctions, allowing for appropriate repair or replacement measures.

[0004] Traditional power system safety monitoring methods mainly rely on manual inspections or data collected by sensors, which have the following problems:

[0005] 1) Manual inspection is inefficient and prone to omissions or misjudgments. Due to the large number and wide distribution of equipment in the power system, manual inspection requires a lot of manpower and time, and is easily affected by personnel quality, experience and environmental factors, which can easily lead to missed inspections or misjudgments.

[0006] 2) Limited number of sensors, unable to cover all devices. Due to the cost and installation difficulty of sensors, as well as the limitations of wireless communication, sensors cannot cover all devices and locations in the power system, resulting in some devices being unable to be effectively monitored.

[0007] 3) Sensor data contains noise and outliers, requiring complex preprocessing. Due to the inherent accuracy and stability of the sensor, as well as external interference and signal transmission losses, the data acquired by the sensor often contains noise and outliers, requiring preprocessing operations such as filtering, smoothing, and interpolation, which increases the complexity of data processing.

[0008] 4) Sensor data has high dimensionality, making it difficult to extract effective features. Due to the wide variety of equipment and parameters in power systems, the data collected by sensors has high dimensionality, and there are correlations and redundancies among the data. Dimensionality reduction, feature selection, or feature extraction are required to facilitate subsequent analysis and modeling.

[0009] 5) Uneven distribution of sensor data makes it difficult to establish a unified model. Due to the differences and heterogeneity among equipment in the power system, and the influence of operating conditions and environmental factors, the data collected by sensors is unevenly distributed and may have multiple modes or state transitions, making it difficult to describe with a unified model.

[0010] Due to the aforementioned problems, traditional power system safety monitoring methods are unable to quickly and accurately achieve visualized monitoring of the status of power system equipment. Summary of the Invention

[0011] To address the aforementioned problems in the existing technology, this invention provides a power system safety monitoring method and electronic device.

[0012] To achieve the above objectives, the present invention provides the following solution:

[0013] A power system safety monitoring method, comprising:

[0014] Acquire time-series data; the time-series data is the operating data of the power system collected by sensors;

[0015] Map the time series data to image data;

[0016] The image data is preprocessed;

[0017] Obtain a security monitoring model; the security monitoring model is a trained neural network model.

[0018] The preprocessed image data is input into the security monitoring model to obtain the security monitoring results;

[0019] The image data and monitoring results are then visualized.

[0020] Optionally, mapping the time-series data to image data specifically includes:

[0021] The time series data is mapped to first image data according to the first dimension;

[0022] The time series data is mapped into second image data according to the second dimension;

[0023] The time series data is mapped to third image data according to the third dimension;

[0024] The image data is obtained by merging the first image data, the second image data, and the third image data.

[0025] Optionally, the sensor type is used as the first dimension; the location of each device in the power system is used as the second dimension; and a set time period is used as the third dimension.

[0026] Optionally, mapping the time series data to first image data according to a first dimension specifically includes:

[0027] The time series data is divided into multiple sub-tensors of shape M×T according to the first dimension; where M is the number of power system operation data collected by the sensor, and T is the time point;

[0028] Expand the M×T subtensor into a vector of length M×T along the first dimension;

[0029] Adjust the vector of length M×T to have a side length of... A square matrix rounded up;

[0030] The side length is The square matrix rounded up is equivalent to an image, thus obtaining the first image data.

[0031] Optionally, mapping the time series data to second image data according to the second dimension specifically includes:

[0032] The time series data is divided into multiple sub-tensors of shape N×T according to the second dimension; N is the number of sensors, and T is the time point;

[0033] Expand multiple subtensors of shape N×T into vectors of length N×T along the second dimension;

[0034] Adjust the vector of length N×T to have a side length of... A square matrix rounded up;

[0035] The side length is The square matrix rounded up is equivalent to an image, thus obtaining the second image data.

[0036] Optionally, the time series data is mapped to third image data according to a third dimension, specifically including:

[0037] The time series data is divided into multiple sub-tensors of shape N×M×t according to the third dimension; N is the number of sensors, and T is the time point.

[0038] Multiple subtensors of shape N×M×t are expanded into a matrix of shape N×Mt along the third dimension;

[0039] The matrix of shape N×Mt is equivalent to an image to obtain the third image data.

[0040] Optionally, the security monitoring model includes, in sequence according to the data processing procedure, an input layer, a first convolutional layer, a second convolutional layer, a first max pooling layer, a third convolutional layer, a fourth convolutional layer, a second max pooling layer, a fifth convolutional layer, a sixth convolutional layer, a global average pooling layer, a fully connected layer, and an output layer.

[0041] Optionally, the preprocessing includes data normalization, data denoising, and data augmentation.

[0042] Optionally, before inputting the preprocessed image data into the security monitoring model to obtain the security monitoring result, the method further includes:

[0043] The security monitoring model is compressed using model compression techniques, including weight quantization, model pruning, and knowledge distillation.

[0044] After the compressed security monitoring model is converted to ONNX format, it is transferred to the target platform.

[0045] An electronic device, comprising:

[0046] Memory, used to store computer programs;

[0047] A processor, connected to the memory, is used to retrieve and execute the computer program to implement the power system security monitoring method provided above.

[0048] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0049] 1) This invention utilizes convolutional neural networks to construct a security monitoring model, which can automatically learn features, greatly simplifying the selection of features and improving the representativeness and robustness of features.

[0050] 2) This invention converts time series data into image data, which is convenient for direct use in neural network models, providing a brand-new perspective on data processing and utilization for power systems and helping to reveal hidden patterns in the data.

[0051] 3) This invention utilizes a safety monitoring model constructed using a convolutional neural network for safety monitoring, which can quickly process large amounts of image data, thereby achieving real-time monitoring of the power system.

[0052] 4) This invention applies neural networks to power system security monitoring, which can improve the accuracy of equipment status identification and prediction.

[0053] 5) This invention visualizes image data and monitoring results, which can intuitively display the status and problems of the power system through images. This not only helps technicians to quickly understand and locate problems, but also provides decision-makers with intuitive decision-making basis. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart of the power system safety monitoring method provided by the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] The purpose of this invention is to provide a power system safety monitoring method and electronic device that can quickly and accurately achieve visual monitoring of the status of power system equipment.

[0058] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] A power system is a complex network composed of multiple devices (such as transformers, generators, and transmission lines). These devices are affected by various internal and external factors (such as load changes, ambient temperature, and fault events) during operation. In order to detect and address these impacts in a timely manner and ensure the safe and stable operation of the power system, it is necessary to monitor the operating status of the equipment in real time.

[0060] The primary method for monitoring equipment operating status is to use sensors to collect various physical quantities of the equipment, such as temperature, voltage, and current. Different types of physical quantities reflect different aspects of the equipment; for example, temperature reflects the equipment's thermal balance, voltage reflects the potential difference, and current reflects the equipment's power output. Therefore, it is necessary to install various types of sensors in the power system to obtain comprehensive and accurate data from the equipment.

[0061] Sensor data records the equipment's operation in time-series format, meaning each point in time corresponds to the value of one or more physical quantities. Time-series data can reflect the equipment's trends and patterns over time, as well as potential anomalies or malfunctions. The sensor data acquisition frequency should be determined based on the equipment's characteristics and monitoring needs, generally once per second or per minute. Too low a acquisition frequency can lead to data loss or delay, affecting monitoring effectiveness; too high a frequency can result in data redundancy or noise, increasing the difficulty of data processing.

[0062] Based on this, the present invention provides a method for power system safety monitoring. For example... Figure 1 As shown, the power system safety monitoring method includes:

[0063] Step 100: Acquire time series data. The time series data is the operating data of the power system collected using sensors.

[0064] Step 101: Map the time series data to image data.

[0065] In practical applications, to convert time-series data acquired by sensors into image data, the time-series data needs to be organized into a two-dimensional matrix. This is because an image is a two-dimensional array of pixels, with each pixel corresponding to a grayscale or color value. Therefore, to map the time-series data to the pixels of the image, the three dimensions of the time-series data (sensor, physical quantity, and time) need to be mapped to the two dimensions of the image (rows and columns). Assuming there are N sensors, each sensor acquires M physical quantities, and each physical quantity has T time points, the sensor data can be represented as an N×M×T three-dimensional tensor X. Then, the three-dimensional tensor X is divided into several two-dimensional matrices (i.e., sub-tensors) X according to different methods (i.e., different dimensions). i Each two-dimensional matrix X i Each corresponds to one image. Based on this, the present invention can be segmented in the following three ways:

[0066] Method 1: Segmentation based on sensor type (i.e., the first dimension). During segmentation, each image contains only data from the same type of sensor, such as temperature or voltage data. This highlights the temporal variation patterns of different types of physical quantities. The specific segmentation method for this approach is as follows:

[0067] 1) Divide the three-dimensional tensor X into N sub-tensors X1, X2, ..., XN according to the first dimension. N Each subtensor has a shape of M×T.

[0068] 2) Convert each subtensor X n Expanded along the first dimension, it forms a vector of length M×T. n=1,2,...,N.

[0069] 3) Reshape each vector into a square matrix with a side length of . Round up to the nearest integer. If the length of this vector is not a perfect square, pad the end of the vector with zeros.

[0070] 4) By treating each square matrix as an image, the first image data can be obtained.

[0071] Method 2: Segmentation by Equipment Location. During segmentation, each image contains only equipment data from the same location, such as a transformer or generator. This highlights the differences in operational status between equipment at different locations. The segmentation method for this method is as follows:

[0072] 1) Divide the three-dimensional tensor X' into M sub-tensors X1, X2, ..., X' according to the second dimension. M Each subtensor has a shape of N×T.

[0073] 2) Convert each subtensor X m Expanding along the second dimension, this vector has a length of N×T. m = 1, 2, ..., M.

[0074] 3) Reshape each vector into a square matrix with a side length of . Round up to the nearest integer. If the length of the vector is not a perfect square, pad the end of the vector with zeros.

[0075] 4) By treating each square as an image, the second image data can be obtained.

[0076] Method 3: Segmentation by Time Period. During segmentation, each image contains only data from all devices within a specific time period, such as one day or one hour. This highlights the overall operation of the power system across different time periods. The segmentation method for this approach is as follows:

[0077] 1) Divide the three-dimensional tensor X into T / t subtensors X1, X2, ..., Xt according to the third dimension. T / t Each subtensor has a shape of N×M×t, where t is the time length of the segmentation, such as 24 or 60.

[0078] 2) Convert each subtensor X i Expanding along the third dimension, the matrix has the shape N×Mt, where i = 1, 2, ..., T / t.

[0079] 3) By treating each matrix as an image, the third image data can be obtained.

[0080] Finally, by merging the datasets obtained from the three methods (i.e., the first image data, the second image data, and the third image data), the required image data can be obtained.

[0081] Step 102: Preprocess the image data. The purpose of preprocessing is to improve image quality and the learning effect of the convolutional neural network.

[0082] In practical applications, the preprocessing process can include the following three steps:

[0083] (1) Normalization: Converting the numerical values ​​in image data into decimals in the range [0, 1] to eliminate the differences in dimensions and numerical ranges between different physical quantities. Normalization can make each pixel in the image have the same importance, and avoid some physical quantities affecting the parameter updates of the convolutional neural network due to excessively large or small values.

[0084] In this invention, mean-variance normalization is used, which involves subtracting the mean from each value in the image matrix and then dividing by the standard deviation, i.e.:

[0085] Among them, X 原 It is the original image matrix, X norm It is the normalized image matrix, X mean and X std These are the mean and standard deviation of the original image matrix, respectively.

[0086] (2) Noise Reduction: Removing noise and outliers from the normalized image to improve image clarity and information accuracy. Noise and outliers may be caused by errors or interference during sensor data acquisition, or by data conversion or segmentation. Noise and outliers affect the grayscale or color distribution in the image, leading to image blurring or distortion.

[0087] In this invention, median filtering is used to replace each pixel in the image with the median value of its neighborhood, that is:

[0088] X denoise [i, j] = median({X norm [i+k, j+l]} k，l∈N ).

[0089] Among them, X denoise [i, j] is the denoised image matrix, N is the neighborhood size (3×3), and i and j are the pixel indices of the image. That is, X[i, j] represents the pixel in the i-th row and j-th column of image X. norm [i+k, j+l] is the normalized image matrix X. normThe pixel in the (i+k)th row and (j+l)th column. k and l are the length and width of the convolution kernel (neighborhood). Median filtering can effectively remove random noise such as salt-and-pepper noise.

[0090] (3) Enhancement: Enhance the contrast and brightness of the image to improve image visualization and feature extraction capabilities. Contrast and brightness are two important attributes in an image, reflecting the degree of grayscale or color variation and the overall brightness level, respectively. Too low or too high contrast and brightness will lead to poor deep learning performance.

[0091] In this invention, a histogram equalization method is used to redistribute the grayscale or color values ​​in an image to make them conform to a uniform distribution, that is:

[0092]

[0093] Among them, X enhance This is the enhanced image matrix, where C(*) is the number of values ​​less than or equal to * in the denoised image matrix. max This is the maximum number of values ​​in the original image matrix. Histogram equalization can effectively improve image contrast and dynamic range.

[0094] For example, in practical applications, X enhance [i, j] is a 128×128×1 matrix.

[0095] Step 103: Obtain the security monitoring model. The security monitoring model is a pre-trained neural network model.

[0096] In practical applications, the security monitoring model, according to the data processing procedure, includes the following layers in sequence: input layer, first convolutional layer, second convolutional layer, first max pooling layer, third convolutional layer, fourth convolutional layer, second max pooling layer, fifth convolutional layer, sixth convolutional layer, global average pooling layer, fully connected layer, and output layer.

[0097] The input layer receives the input image, namely X. enhance [i, j].

[0098] The first convolutional layer convolves the input image using multiple filters. This layer can capture basic features of the image, such as edges and textures, and is activated using the ReLU function.

[0099] The second convolutional layer continues to use convolution for feature extraction. Here, it still uses 32 3x3 filters with a stride of 1, Same padding, and the ReLU function.

[0100] The first max pooling layer reduces the dimensionality of the feature map by taking the maximum value of a 2x2 region for spatial downsampling.

[0101] To capture more complex features, the third convolutional layer uses 64 3x3 filters with a stride of 1, Same padding, and the ReLU activation function.

[0102] The fourth convolutional layer also uses 64 3x3 filters with a stride of 1, Same padding, and applies the ReLU function.

[0103] The second max-pooling layer again uses 2x2 max-pooling to reduce the dimension of the feature map.

[0104] In the fifth convolutional layer, 128 3x3 filters with a stride of 1 are used, with Same padding and ReLU activation.

[0105] The sixth convolutional layer uses 128 3x3 filters with a stride of 1, Same padding, and applies the ReLU activation function again.

[0106] The Global Average Pooling (GAP) layer performs average pooling on each feature map to obtain a single value, thereby reducing the number of parameters and preparing for the next layer.

[0107] A densely connected layer with 256 nodes is used as a fully connected layer to combine features from the previous layer and prepare for the classification task. Following this layer, a Dropout layer (with a dropout rate of 0.5) is used to reduce the risk of overfitting.

[0108] Based on the characteristics of this task, normal is defined as 0 and abnormal as 1. The output layer provides an output value for each category and uses the Softmax function to convert it into a probability distribution. When the output value is less than 0.5, the current state is considered normal. Conversely, when it is greater than 0.5, the power system is considered to have experienced an anomaly.

[0109] Furthermore, to improve monitoring accuracy, a pre-trained and tuned network training strategy was used to train the convolutional neural network to obtain the security monitoring model. Pre-training and tuning are common strategies in deep learning, especially when dealing with a relatively small dataset or a complex model.

[0110] Among them, (1) pre-training stage: In this stage, a large amount of data is used for training. This data is relatively coarse, and its annotation has not undergone secondary review. In this process, the following steps are completed:

[0111] Step 1: Initialize the model and train the first to fourth convolutional layers of the above network.

[0112] Step 2: Use a larger learning rate for training, because you are learning new features rather than optimizing.

[0113] Step 3: Train the network until it converges or reaches a preset number of epochs.

[0114] (2) Fine-tuning stage: In this stage, a carefully selected dataset is used, and data augmentation techniques, such as random rotation, cropping, and horizontal flipping, are employed to increase the diversity of the small dataset. The training method is as follows:

[0115] Step 1: Initialize the model and use the parameters obtained from the solidified pre-training environment for the first to fourth convolutional layers.

[0116] Step 2: Train with a small learning rate and use an early stopping strategy to avoid overfitting.

[0117] Step 104: Input the preprocessed image data into the security monitoring model to obtain the security monitoring results.

[0118] Step 105: Visualize the image data and monitoring results.

[0119] Furthermore, prior to step 104, i.e., before the formal adoption of the constructed model, the main focus is on deploying the convolutional neural network, i.e., transferring the trained model to the target platform so that it can perform inference in a real-world environment. The deployment method employed in this invention can be:

[0120] (1) Model compression: To make the model more suitable for mobile or edge devices, model compression techniques are used, namely weight quantization, model pruning and knowledge distillation.

[0121] (2) Convert to ONNX format: ONNX (Open Neural Network Exchange) is an open format for representing deep learning models. By converting the model to ONNX format, it can be deployed across multiple deep learning frameworks.

[0122] Furthermore, the present invention also provides an electronic device comprising: a memory and a processor.

[0123] Memory is used to store computer programs.

[0124] The processor is connected to the memory to retrieve and execute computer programs to implement the power system security monitoring method provided above.

[0125] Furthermore, when the computer program in the aforementioned memory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0126] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0127] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for power system safety monitoring, characterized in that, include: Obtain time series data; The time series data is the operating data of the power system collected by sensors; Map the time series data to image data; The image data is preprocessed; Obtain a security monitoring model; the security monitoring model is a trained neural network model. The preprocessed image data is input into the security monitoring model to obtain the security monitoring results; The image data and the security monitoring results are visualized. The process of mapping the time series data to image data includes: The time-series data is mapped to first image data according to a first dimension; the sensor type is used as the first dimension. The time series data is mapped into second image data according to the second dimension; the location of each device in the power system is used as the second dimension. The time series data is mapped to third image data according to the third dimension; the set time period is used as the third dimension; The image data is obtained by merging the first image data, the second image data, and the third image data. Specifically, mapping the time series data to first image data according to the first dimension includes: Dividing the time series data into shapes according to the first dimension is... Multiple subtensors; among which, M The number of power system operation data collected by sensors. T For a point in time; The shape is The subtensor expanded along the first dimension has a length of ; The length is The vector is adjusted to the side length is A square matrix rounded up; The side length is The square matrix rounded up is equivalent to an image, thus obtaining the first image data; Mapping the time series data to second image data according to the second dimension specifically includes: Dividing the time series data into shapes according to the second dimension is... Multiple subtensors; N is the number of sensors, T For a point in time; The shape is Multiple subtensors expanded along the second dimension have a length of ; The length is The vector is adjusted to the side length is A square matrix rounded up; The side length is The square matrix rounded up is equivalent to an image, thus obtaining the second image data; Mapping the time series data to third image data according to the third dimension specifically includes: Dividing the time series data into shapes according to the third dimension is... Multiple subtensors; N is the number of sensors, T For a point in time; The shape is The multiple subtensors unfolded according to the third dimension are the shape Matrix; The shape is The matrix is ​​equivalent to an image, thus obtaining the third image data.

2. The power system safety monitoring method according to claim 1, characterized in that, The security monitoring model, in accordance with the data processing procedure, includes, in sequence, an input layer, a first convolutional layer, a second convolutional layer, a first max pooling layer, a third convolutional layer, a fourth convolutional layer, a second max pooling layer, a fifth convolutional layer, a sixth convolutional layer, a global average pooling layer, a fully connected layer, and an output layer.

3. The power system safety monitoring method according to claim 1, characterized in that, The preprocessing includes data normalization, data denoising, and data augmentation.

4. The power system safety monitoring method according to claim 1, characterized in that, Before inputting the preprocessed image data into the security monitoring model to obtain the security monitoring result, the method further includes: The security monitoring model is compressed using model compression techniques, including weight quantization, model pruning, and knowledge distillation. After the compressed security monitoring model is converted to ONNX format, it is transferred to the target platform.

5. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, connected to the memory, is used to retrieve and execute the computer program to implement the power system security monitoring method as described in any one of claims 1-4.

Images