Transformer substation monitoring method and system based on Internet of Things

Abstract

The invention provides a transformer substation monitoring method and system based on the Internet of Things, and the method comprises the steps: obtaining an original electromagnetic disturbance signal of a transformer substation, and constructing an electromagnetic disturbance feature matrix; carrying out the anomaly detection of the electromagnetic disturbance signal through employing a self-adaptive attention mechanism in combination with asymmetric anomaly detection loss; inputting the time sequence input vector into a fault classification model to perform fault type identification, outputting a fault type label and severity, and performing fault propagation path calculation on the fault type label and severity to obtain a fault propagation path; and simulating fault evolution through a digital twin model, predicting a fault development trend, a secondary fault influence range and a future state vector, and finally generating a maintenance priority, a time window and a resource scheduling scheme. The system breaks through the limitation of a traditional monitoring technology, improves the intelligent level of a transformer substation, reduces the maintenance cost, and improves the operation safety and reliability of a power grid.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to a substation monitoring method and system based on the Internet of Things. Background Art

[0002] Substations are critical infrastructure for power grids, responsible for power conversion and distribution. As power grids expand, traditional monitoring systems, relying on SCADA systems, status sensors (current, voltage, temperature), and video surveillance, remain limited in terms of fault warning, accurate diagnosis, and intelligent operations and maintenance.

[0003] Early fault warning cannot be achieved:

[0004] Existing monitoring methods mainly rely on parameters such as current, voltage, and temperature. These physical quantities respond slowly to early faults such as partial discharge, short-term arcing, and insulation aging, resulting in delayed early warning and easily missing the best maintenance opportunity.

[0005] Insufficient data analysis capabilities make it difficult to accurately identify failure modes:

[0006] Traditional monitoring methods use fixed thresholds or statistical analysis, which can easily lead to false alarms or missed alarms when loads fluctuate significantly or when abnormal patterns are complex. Furthermore, AI-based anomaly detection relies on large amounts of fault data, while actual substation fault data is scarce, limiting the model's generalization capabilities.

[0007] Digital twin technology has difficulty accurately predicting fault evolution:

[0008] Existing digital twin technology is mainly based on static physical models, which makes it difficult to update the health status of equipment in real time, resulting in large fault prediction errors and making it difficult to effectively simulate the impact of factors such as equipment aging and load changes on faults.

[0009] Rigid O&M strategies and high maintenance costs:

[0010] Traditional substation operation and maintenance relies on fixed-cycle inspections or post-maintenance, and lacks intelligent decision-making based on real-time status. This may lead to unnecessary maintenance or missed optimal maintenance windows, increasing maintenance costs and affecting grid stability.

[0011] Existing technologies still have large technical gaps in early fault warning, accurate fault analysis, and intelligent operation and maintenance optimization, and a new intelligent monitoring solution is urgently needed. Summary of the Invention

[0012] The purpose of the present invention is to design a substation monitoring method and system based on the Internet of Things. Through electromagnetic disturbance signal analysis + digital twin modeling + intelligent operation and maintenance optimization, it breaks through the limitations of existing monitoring solutions and realizes accurate fault warning, simulation and maintenance optimization.

[0013] In order to achieve the above-mentioned object, in a first aspect of the present invention, a substation monitoring method based on the Internet of Things is provided, the method comprising:

[0014] S1. Obtaining an original electromagnetic disturbance signal from a substation, preprocessing the original electromagnetic disturbance signal, performing variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and constructing an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column represents a different disturbance feature;

[0015] S2. Construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with an asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, obtaining the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point;

[0016] S3. Construct an abnormal feature matrix based on the corresponding disturbance features, aggregate the abnormal feature matrix within the time window into a time series input vector, input the time series input vector into a fault classification model to identify the fault type, output the fault type label and severity, and perform fault propagation path calculation on the fault type label and severity to obtain the fault propagation path;

[0017] S4. Simulating a digital twin model based on the fault type label, severity, and fault propagation path, simulating fault evolution through the digital twin model, and predicting fault development trends, secondary fault impact range, and future state vectors;

[0018] S5. Generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

[0019] Preferably, the preprocessing includes removing high-frequency interference components by using a time-frequency filtering method based on adaptive noise estimation, and then performing a short-time Fourier transform on the filtered signal to obtain a time-frequency distribution matrix;

[0020] The variational modal decomposition includes: adaptively extracting multimodal signal features through variational modal decomposition, and optimizing decomposition parameters in combination with a device-specific frequency band selection mechanism;

[0021] The smoothing process includes: using exponentially weighted moving average to smooth the time-frequency distribution matrix to obtain a final electromagnetic disturbance characteristic matrix.

[0022] Preferably, the S2 specifically includes:

[0023] Adopting the adaptive attention mechanism to construct the temporal input matrix within the time window, the attention weight of time step i within the window is obtained, the contribution of the outliers in the time dimension is adaptively adjusted, and the feature importance score of the attention weight is learned through the Transformer encoder;

[0024] The reconstruction error of time step i is determined based on the adaptive attention mechanism, and the weighted asymmetric anomaly scoring formula is introduced to calculate the total anomaly score at the current time t;

[0025] Adopting time-decay adaptive threshold to dynamically adjust abnormality judgment criteria and obtain adaptive abnormality threshold;

[0026] If the total anomaly score is greater than the adaptive anomaly threshold, the electromagnetic disturbance at time t is determined to be abnormal, and the abnormal point information is output.

[0027] Preferably, the total anomaly score is calculated as follows:

[0028]

[0029] Among them, A s (t) is the total anomaly score, i is the time step i, T is the total number of time steps, β i (t) is the attention weight of time step i in the window, ε i (t) is the reconstruction error at time step i, γ is the abnormality sensitivity adjustment factor, and γ>1 can enhance the penalty for severe abnormalities and improve the fault recognition rate.

[0030] Preferably, the fault classification model is constructed as follows:

[0031] The fault classification model uses a multi-layer classification network with optimized graph structure for identification;

[0032] Construct a temporal topological graph G of electromagnetic disturbance features, and use graph convolutional networks to learn topological features, defining node v i and edge ij The relationship between:

[0033]

[0034] in, The updated feature vector of node i in layer l, is the feature vector of node i in layer l; is the set of neighbor nodes of node i; W (l) is the weight matrix of the graph convolution layer; Z ij is the normalization factor of the adjacency matrix; σ is the nonlinear activation function, j is the node j, is the feature vector of node j in layer l; b (l) is the bias term;

[0035] The time series input vector is input into the fault classification model to calculate the fault type label and fault severity.

[0036] Preferably, a graph convolutional network is used to calculate the fault impact range based on the fault type label and severity to obtain the fault propagation path.

[0037] Preferably, the S4 specifically includes:

[0038] Establish the physical topology of the equipment in the digital twin environment and construct a dynamic feature vector of the substation operating status; the dynamic feature vector includes the voltage state vector, current state vector, equipment temperature distribution, electromagnetic disturbance feature matrix, and fault severity label;

[0039] A fault evolution prediction model is constructed using a physics-enhanced neural network to analyze the dynamic evolution of substation faults. The loss function of the fault evolution prediction model consists of a physical constraint term and a data error term.

[0040] Based on the dynamic feature vector and fault propagation path, a physics-enhanced neural network is used to predict the fault development trend and estimate the possible secondary fault points.

[0041] Based on the timing topology graph, the fault propagation probability is determined, and the predicted device state is updated based on the fault propagation probability, and the future state vector is obtained and the simulation data is stored.

[0042] Preferably, the S5 specifically includes:

[0043] Determine a comprehensive risk score based on anomaly scores, fault propagation paths, and historical fault data;

[0044] The device's operational stability score is calculated by weighted summation based on the device's current voltage, current, and temperature fluctuations.

[0045] The final maintenance priority is calculated based on the maintenance cost, combined with the comprehensive risk score and the operation stability score.

[0046] Preferably, the S5 further includes:

[0047] Calculate the optimal maintenance time T based on the final maintenance priority and the predicted failure development trend m (i), calculated as:

[0048]

[0049] Among them, T f (i) is the expected failure time of equipment i; T m(i) is the optimal maintenance time window. The smaller the value, the earlier the maintenance should be carried out.

[0050] The optimal maintenance task scheduling plan is designed by combining the final maintenance priority with the best maintenance time.

[0051] In a second aspect of the present invention, a substation monitoring system based on the Internet of Things comprises:

[0052] An electromagnetic disturbance signal acquisition unit is used to acquire the original electromagnetic disturbance signal of the substation, preprocess the original electromagnetic disturbance signal, perform variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and construct an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column represents a different disturbance feature;

[0053] The anomaly detection unit is used to construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with an asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, and obtain the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point;

[0054] a fault monitoring unit, configured to construct an abnormal feature matrix based on the corresponding disturbance feature, aggregate the abnormal feature matrix within the time window into a time series input vector, input the time series input vector into a fault classification model to identify the fault type, output a fault type label and severity, and perform fault propagation path calculation on the fault type label and severity to obtain the fault propagation path;

[0055] A fault trend prediction unit is used to simulate a digital twin model based on the fault type label, severity, and fault propagation path, simulate fault evolution through the digital twin model, and predict fault development trends, secondary fault impact range, and future state vectors;

[0056] The maintenance optimization unit is used to generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

[0057] The beneficial technical effects of the present invention are at least as follows:

[0058] (1) This invention uses electromagnetic disturbance signals as a new monitoring method to detect early faults such as partial discharge, short-term arcing, and insulation aging. By extracting abnormal features through variational mode decomposition and time series deep learning, it improves detection accuracy and reduces false positives and missed negatives.

[0059] (2) The present invention adopts a hybrid physics + AI modeling method to dynamically adjust the twin model parameters, so that it can map electromagnetic disturbance data in real time and predict the fault propagation path, thereby improving the simulation accuracy of complex fault evolution.

[0060] (3) Based on electromagnetic disturbance detection and digital twin prediction, an adaptive fault warning mechanism is established to dynamically adjust the warning level (low / medium / high risk). Combining historical data, economic factors and load conditions, the optimal maintenance strategy (immediate repair / delayed maintenance / observation) is calculated to reduce unnecessary maintenance and optimize substation operation and maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0062] Figure 1 This is a flow chart of the substation monitoring method based on the Internet of Things of the present invention.

[0063] Figure 2 This is a framework diagram of the substation monitoring system based on the Internet of Things of the present invention. DETAILED DESCRIPTION

[0064] The following describes embodiments of the present invention in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended only to explain the present invention and are not to be construed as limiting the present invention.

[0065] In one or more embodiments, Figure 1 As shown, a substation monitoring method based on the Internet of Things is disclosed, and the method includes the following steps S1-S5:

[0066] S1. Obtain the original electromagnetic disturbance signal of the substation, preprocess the original electromagnetic disturbance signal, perform variational modal decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and construct an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column is a different disturbance feature.

[0067] Specifically, highly sensitive electromagnetic sensors are deployed around key substation equipment to collect electromagnetic disturbance signals in real time. These sensors' measurement range covers locations prone to partial discharge or short-duration arcing, such as transformer windings, switch contacts, and cable connections. Assume the original electromagnetic signal collected at time t is:

[0068] X raw(t)=[x1(t),x2(t),...,x N (t)];

[0069] Among them, x i (t) represents the original electromagnetic signal collected by the i-th sensor at time t, and N is the number of sensors.

[0070] Furthermore, considering the strong interference in the substation environment (such as high-frequency noise and electromagnetic coupling), we denoise the original signal. First, a time-frequency filtering method based on adaptive noise estimation is used to remove the high-frequency interference component. Then, a short-time Fourier transform (STFT) is performed on the filtered signal to obtain the time-frequency distribution matrix:

[0071]

[0072] Among them, X stft (t,f) is the time-frequency signal matrix at time t, w(tk) is the windowing function, f is the frequency component, and k is the time index. This matrix is used for subsequent signal decomposition and feature extraction.

[0073] Furthermore, in order to extract fault information of different modes from the complex electromagnetic disturbance signal, we use variational mode decomposition (VMD) to stft (t,f) is adaptively decomposed to obtain K intrinsic mode components (IMFs), namely:

[0074] F emd (t)=[IMF1(t),IMF2(t),...,IMF K (t)];

[0075] Among them, F emd (t) is the electromagnetic disturbance characteristic matrix, IMF k (t) represents the kth modal decomposition signal, and K is the number of decomposed signal modes. The advantage of VMD over traditional wavelet transform methods is that its decomposition mode can be adaptively adjusted to more accurately match electromagnetic disturbance signals of different fault types.

[0076] The characteristic frequency ranges of electromagnetic disturbance signals may vary for different substation equipment. Therefore, we further introduce a device-specific feature selection mechanism to dynamically adjust the number of frequency bands, K, used in the VMD decomposition. Through adaptive optimization, we select the optimal modal decomposition parameters so that the extracted modal components can best reflect the health status of the equipment.

[0077] Furthermore, from F emd(t) extracts multiple time-frequency features and constructs a high-dimensional feature vector as the input for subsequent anomaly detection. It is assumed that M most representative features are selected, including energy ratio, root mean square value, kurtosis, etc., to form the final input matrix:

[0078] Φ(t)=[φ1(t),φ2(t),...,φ M (t)];

[0079] Among them, φ m (t) represents the mth electromagnetic disturbance feature, and M is the number of features.

[0080] Furthermore, a time series smoothing mechanism is introduced to eliminate the impact of instantaneous errors on anomaly detection. Exponentially weighted moving average (EWMA) is used to smooth Φ(t) to obtain the final feature input matrix F emd :

[0081] F emd (t)=αΦ(t)+(1-α)F emd (t-1);

[0082] Among them, α is a smoothing factor (0<α<1), which is used to control the impact of short-term fluctuations.

[0083] Finally, we will F emd As data input, it is passed to the anomaly detection model. During this process, we ensure the following:

[0084] Time alignment of data ensures that the signals from all sensors are calculated within the same time window.

[0085] Normalization is used to eliminate the impact of amplitude differences between different devices on model learning.

[0086] Record and store abnormal features for subsequent digital twin simulation analysis.

[0087] Final output F emd , used for anomaly detection in the next step.

[0088] S2. Construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, and obtain the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point.

[0089] Specifically, the goal of this step is to generate the electromagnetic disturbance characteristic matrix F emdAnomaly detection is performed to identify possible abnormal conditions in the equipment and provide high-confidence input for subsequent fault classification and severity assessment (step 3). In substation environments, electromagnetic disturbance signals have complex time-frequency structures and contain a large number of noise components. Traditional fixed threshold methods (such as mean plus standard deviation) are susceptible to background noise and load fluctuations, resulting in high false alarm rates or serious omissions. Therefore, this step uses a deep learning method based on Dynamic Adaptive Attention Network (DAN), combined with an innovative asymmetric anomaly scoring mechanism, to ensure the accuracy and stability of anomaly detection.

[0090] Furthermore, the electromagnetic disturbance characteristic matrix F generated in step 1 is used emd , where each row corresponds to a time step t and each column is a different perturbation feature.

[0091] Construct a time window W and take the data of T time steps before the current time t as the time series input F seq (t):

[0092] F seq (t)=[F emd (t-T+1),...,F emd (t)];

[0093] Where T is the time window size, F seq (t) is a (T×M)-dimensional matrix, where M is the feature dimension.

[0094] Furthermore, since electromagnetic disturbance signals exhibit different abnormal patterns under different equipment and load conditions, this step adopts a dynamic attention mechanism (DAN) to model the timing features and combines it with asymmetric anomaly detection loss to ensure robustness to different types of faults.

[0095] DAN calculates the attention weight β within the window at time step i i (t), so that the contribution of the outlier in the time dimension is adaptively adjusted:

[0096]

[0097] Among them, g i (t) is the feature importance score at time step i, learned by the Transformer encoder.

[0098] Furthermore, DAN is used to calculate the reconstruction error ε at time step i i (t), and then calculate the total anomaly score A at the current time t using the weighted asymmetric anomaly scoring formula s (t):

[0099]

[0100] Among them, γ is the abnormality sensitivity adjustment factor. When γ>1, it can increase the penalty for severe abnormalities and improve the fault recognition rate.

[0101] Furthermore, since the load fluctuation of the substation may cause the abnormal threshold of electromagnetic disturbance to change over time, the time-attenuated adaptive abnormal threshold (TAAT) is used for dynamic adjustment:

[0102] T A (t) = μ A (t)+λ·σ A (t);

[0103] Where: μ A (t) is the sliding mean of the past H time steps:

[0104]

[0105] Among them, σ A (t) is the sliding standard deviation; λ is the dynamic adjustment coefficient, which is optimized by historical fault data.

[0106] Abnormal determination:

[0107] When A s (t)>T A At (t), the electromagnetic disturbance at time t is determined to be abnormal, and the abnormal point information is output:

[0108] Anomaly score A s (t), the corresponding disturbance feature F emd (t), the weight distribution β(t) of the outlier point.

[0109] S3. Construct an abnormal feature matrix based on the corresponding disturbance, aggregate the abnormal feature matrix within the time window into a time series input vector, input the time series input vector into the fault classification model to identify the fault type, output the fault type label and severity, and calculate the fault propagation path of the fault type label and severity to obtain the fault propagation path.

[0110] Specifically, receive the anomaly detection results generated in step 2, including the anomaly score A s (t) and abnormal feature matrix F emd (t), and construct the classification input dataset.

[0111] Furthermore, the abnormal features within the time window W are aggregated into a time series input vector F cls (t), that is:

[0112] F cls (t)=[F emd(t-T+1),...,F emd (t)];

[0113] Among them, T is the time window size, F cls (t) is a (T×M)-dimensional matrix, where M is the feature dimension.

[0114] Build a fault classification model:

[0115] A multi-layer classification network based on graph structure optimization (GOCN, Graph Optimized Classification Network) is used to identify different types of failure modes.

[0116] First, we construct a temporal topological graph G of electromagnetic disturbance features and use a graph convolutional network (GCN) to learn the topological features. We define the node v i and edge ij The relationship between:

[0117]

[0118] in, is the feature vector of node i in layer l; is the set of neighbor nodes of node i; W (l) is the weight matrix of the graph convolution layer; Z ij is the normalization factor of the adjacency matrix; σ is the nonlinear activation function.

[0119] Based on the graph structure optimization classification network GOCN, the classification output C is calculated f (fault category) and fault severity S f :

[0120] C f ,S f =GOCN(F cls (t));

[0121] Among them, C f is the fault type label, such as partial discharge (PD), short circuit (SC), insulation aging (IA), transient arc (TA), etc.; S f is the severity of the fault (low / medium / high).

[0122] Fault propagation path calculation:

[0123] In order to predict the impact range of the current fault on other equipment in the substation, we establish a propagation graph P based on the power topology structure. f , and use GCN to calculate the fault impact range:

[0124] P f =GCN propagation (Cf ,S f );

[0125] Among them, P f Indicates a list of other devices that may be affected by the fault.

[0126] Generate classification results and pass them to step S4:

[0127] Output fault classification result C f , used for fault modeling and simulation analysis of digital twin systems;

[0128] Output fault severity S f , providing initial conditions for fault development trend prediction;

[0129] Output fault propagation path P f , used to predict the impact range of fault propagation.

[0130] S4. Simulate the digital twin model based on the fault type label, severity, and fault propagation path, simulate the fault evolution through the digital twin model, and predict the fault development trend, secondary fault impact range, and future state vector.

[0131] Specifically, receive the fault classification result C generated in step S3 f , fault severity S f and fault propagation path P f , and build simulation input for digital twins.

[0132] Furthermore, the physical topology of the equipment is established in the digital twin environment, and the dynamic feature vector D of the substation operation status is constructed. state (t), including:

[0133] D state (t)=[V(t),I(t),T(t),F emd (t),S f ];

[0134] Where: V(t) is the voltage state vector; I(t) is the current state vector; T(t) is the device temperature distribution; F emd (t) is the electromagnetic disturbance characteristic matrix; S f This is the fault severity label.

[0135] Build a fault evolution prediction model based on a hybrid physics-data driven modeling approach:

[0136] A physics-enhanced neural network (PE-NN) is used to model the dynamic evolution of substation faults. The loss function of the model consists of a physical constraint term and a data error term:

[0137]

[0138] in, is the data error term, i.e., the error between the twin simulation output and the true measurement value; is a physical constraint term, which ensures that the simulation model satisfies the basic physical laws of the power system (such as power conservation, heat conduction equation, etc.); p is the weight of the physical constraint;

[0139] Using PE-NN to predict fault development trend f , and estimate the possible secondary fault point R f ,Right now:

[0140] T f ,R f =PE-NN(D state (t),P f );

[0141] Among them, T f Represents the fault impact time series and predicts the evolution of the fault in the next k time steps; R f The secondary fault impact range indicates other devices affected by the current fault.

[0142] Furthermore, based on the power network topology, the fault propagation probability matrix M is calculated. p , defined as follows:

[0143]

[0144] Among them, M p (i, j) represents the probability of a fault propagating from device i to device j; w ij is the connection weight from device i to device j; S f is the fault severity factor; ∑ k w ik is the normalization factor.

[0145] Update the device state and calculate the future state vector D state (t+Δt) and store the simulation data.

[0146] Furthermore, simulation results are generated and passed to step 5:

[0147] Output failure development trend prediction T f, used for operation and maintenance optimization decision-making;

[0148] Output secondary fault impact range R f ,provides input for fault propagation path analysis;

[0149] Output device operating status data D state (t+Δt), which is used to optimize the substation maintenance strategy.

[0150] S5. Generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

[0151] Specifically, the goal of this step is to calculate the optimal maintenance strategy based on the fault evolution results generated in step 4, combined with factors such as equipment health status, fault propagation impact, maintenance cost, and resource allocation, ultimately ensuring substation operational stability, reducing unnecessary maintenance costs, and minimizing fault downtime.

[0152] Furthermore, the failure risk of the equipment is comprehensively evaluated, considering the current failure degree S of the equipment f (i) Fault propagation effect P f (i) and historical failure data H f (i) Define the comprehensive risk score R s (i)

[0153]

[0154] Among them, R s (i): comprehensive failure risk score of device i; S f (i): The current fault severity of device i (calculated in step 3); M p (i, j): fault propagation probability (calculated in step 4); R f (i): other devices that may be affected by device i; H f (i): historical fault record score of device i; ω h : Weight coefficient of historical fault data.

[0155] Calculate the operational stability score U of the equipment s (i), taking into account the current voltage, current, and temperature fluctuations, is defined as follows:

[0156]

[0157] Among them, U s (i): Operation stability score of device i. A larger value indicates a more unstable device state. V i ,I i ,Ti : Current voltage, current, and temperature of device i; V nom ,I nom ,T nom : Rated voltage, current, and temperature of the equipment; α v ,α i ,α t : The weights of different physical quantities.

[0158] Calculate the final maintenance priority P m (i), comprehensive consideration of R s (i) and U s (i), combined with the maintenance cost C m (i) Perform optimization calculations:

[0159]

[0160] Among them, C m (i): Maintenance cost of equipment i (including labor cost, spare parts cost, etc.). P m (i): Maintenance priority of device i. A larger value indicates that maintenance is required earlier.

[0161] Furthermore, according to P m (i) Calculate the optimal maintenance time T m (i) Ensure that maintenance is timely, as defined below:

[0162]

[0163] Among them, T f (i): estimated failure time of device i (calculated in step 4); T m (i): Optimal maintenance time window. The smaller the value, the earlier the maintenance should be carried out. m (i) For smaller equipment (i.e., non-critical equipment), adjust the maintenance plan to reduce the impact on substation operations.

[0164] Furthermore, the maintenance resource scheduling matrix S is defined m , which is used to optimize the allocation of maintenance tasks and ensure the reasonable scheduling of personnel and spare parts. The optimization objective function is constructed to minimize the total maintenance cost while maximizing the system stability, which is defined as follows:

[0165]

[0166] Among them, S m (i): If equipment i is repaired during the current maintenance cycle, then S m (i)=1, otherwise S m (i) = 0; λ s : System stability weight coefficient.

[0167] Solve the optimization problem and obtain the optimal maintenance task scheduling plan.

[0168] Generate the final maintenance strategy and output:

[0169] Output maintenance priority P m (i) Provide decision-making basis for equipment maintenance plan.

[0170] Output the optimal maintenance time window T m (i) Ensure that maintenance is performed at the optimal time.

[0171] Output maintenance resource scheduling matrix S m , ensure the rational allocation of maintenance resources and improve the operating efficiency of substations.

[0172] In one or more embodiments, Figure 2 As shown, a substation monitoring system based on the Internet of Things is disclosed, and the system includes:

[0173] The electromagnetic disturbance signal acquisition unit 1 is used to acquire the original electromagnetic disturbance signal of the substation, preprocess the original electromagnetic disturbance signal, perform variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and construct an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column represents a different disturbance feature;

[0174] Anomaly detection unit 2 is used to construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with an asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, and obtain the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point;

[0175] Fault monitoring unit 3, configured to construct an abnormal feature matrix based on the corresponding disturbance features, aggregate the abnormal feature matrix within the time window into a time series input vector, input the time series input vector into a fault classification model to identify the fault type, output a fault type label and severity, and perform fault propagation path calculation on the fault type label and severity to obtain the fault propagation path;

[0176] A fault trend prediction unit 4 is configured to simulate a digital twin model based on the fault type label, severity, and fault propagation path, simulate fault evolution through the digital twin model, and predict fault development trends, secondary fault impact range, and future state vectors;

[0177] The maintenance optimization unit 5 is used to generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

[0178] It is worth noting that the specific workflow of the IoT-based substation monitoring system provided in the embodiment of the present invention is the same as the workflow of the IoT-based substation monitoring method described in the above embodiment, and will not be repeated here.

[0179] Compared with the prior art, the Internet of Things-based substation monitoring system provided by the embodiment of the present invention obtains the original electromagnetic disturbance signal of the substation, preprocesses the original electromagnetic disturbance signal, performs variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and constructs an electromagnetic disturbance feature matrix; wherein, each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column is a different disturbance feature; according to the electromagnetic disturbance feature matrix, a time series input based on a time window is constructed, and an adaptive attention mechanism is used in combination with an asymmetric anomaly detection loss to perform anomaly detection of the electromagnetic disturbance signal, and obtains the anomaly score corresponding to its electromagnetic disturbance signal, the corresponding disturbance feature and the weight distribution of the anomaly point; the corresponding disturbance feature is converted into the weight distribution of the anomaly point. An abnormal feature matrix is constructed, and the abnormal feature matrix within the time window is aggregated into a time series input vector. The time series input vector is input into a fault classification model to identify the fault type, output the fault type label and severity, and calculate the fault propagation path for the fault type label and severity to obtain the fault propagation path; based on the fault type label, severity and fault propagation path, a digital twin model is simulated, and the fault evolution is simulated through the digital twin model to predict the fault development trend, the impact range of secondary faults and the future state vector; based on the abnormal score, fault propagation path and historical fault data, maintenance priority, time window and resource scheduling plan are generated to minimize the total maintenance cost while maximizing system stability.

[0180] The embodiment of the present invention also provides a substation monitoring device based on the Internet of Things, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned embodiment of the substation monitoring method based on the Internet of Things are implemented, for example Figure 1 or, the processor implements the functions of the modules in the above-mentioned system embodiments when executing the computer program.

[0181] Exemplarily, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to implement the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program in the IoT-based substation monitoring device.

[0182] The IoT-based substation monitoring equipment can be a computing device such as a desktop computer, laptop, PDA, or cloud server. The IoT-based substation monitoring equipment can include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the IoT-based substation monitoring equipment can also include input / output devices, network access devices, buses, and the like.

[0183] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor serves as the control center of the IoT-based substation monitoring device, connecting various parts of the entire IoT-based substation monitoring device using various interfaces and lines.

[0184] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the IoT-based substation monitoring device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created based on the operation of the air conditioning controller, etc. In addition, the memory can include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0185] Wherein, if the module integrated in the substation monitoring equipment based on the Internet of Things is implemented in the form of 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 present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0186] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0187] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A substation monitoring method based on the Internet of Things, characterized in that: The method comprises: S1. Obtaining an original electromagnetic disturbance signal from a substation, preprocessing the original electromagnetic disturbance signal, performing variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and constructing an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column represents a different disturbance feature; S2. Construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with an asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, obtaining the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point; S3. Construct an abnormal feature matrix based on the corresponding disturbance features, aggregate the abnormal feature matrix within the time window into a time series input vector, input the time series input vector into a fault classification model to identify the fault type, output the fault type label and severity, and perform fault propagation path calculation on the fault type label and severity to obtain the fault propagation path; S4. Simulating a digital twin model based on the fault type label, severity, and fault propagation path, simulating fault evolution through the digital twin model, and predicting fault development trends, secondary fault impact range, and future state vectors; S5. Generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

2. The substation monitoring method based on the Internet of Things according to claim 1 is characterized in that: The preprocessing includes using a time-frequency filtering method based on adaptive noise estimation to remove high-frequency interference components, and then performing a short-time Fourier transform on the filtered signal to obtain a time-frequency distribution matrix; The variational modal decomposition includes: adaptively extracting multimodal signal features through variational modal decomposition, and optimizing decomposition parameters in combination with a device-specific frequency band selection mechanism; The smoothing process includes: using exponentially weighted moving average to smooth the time-frequency distribution matrix to obtain a final electromagnetic disturbance characteristic matrix.

3. The substation monitoring method based on the Internet of Things according to claim 1, characterized in that: Said S2 specifically includes: Adopting the adaptive attention mechanism to construct the temporal input matrix within the time window, the attention weight of time step i within the window is obtained, the contribution of the outliers in the time dimension is adaptively adjusted, and the feature importance score of the attention weight is learned through the Transformer encoder; The reconstruction error of time step i is determined based on the adaptive attention mechanism, and the weighted asymmetric anomaly scoring formula is introduced to calculate the total anomaly score at the current time t; Adopting time-decay adaptive threshold to dynamically adjust abnormality judgment criteria and obtain adaptive abnormality threshold; If the total anomaly score is greater than the adaptive anomaly threshold, the electromagnetic disturbance at time t is determined to be abnormal, and the abnormal point information is output.

4. The substation monitoring method based on the Internet of Things according to claim 3 is characterized in that: The total anomaly score is calculated as follows: Among them, A s (t) is the total anomaly score, i is the time step i, T is the total number of time steps, β i (t) is the attention weight of time step i in the window, ε i (t) is the reconstruction error at time step i, γ is the abnormality sensitivity adjustment factor, and γ>1 can enhance the penalty for severe abnormalities and improve the fault recognition rate.

5. The substation monitoring method based on the Internet of Things according to claim 3 is characterized in that: The fault classification model is constructed as follows: The fault classification model uses a multi-layer classification network with optimized graph structure for identification; Construct a temporal topological graph G of electromagnetic disturbance features, and use graph convolutional networks to learn topological features, defining node v i and edge ij The relationship between: in, The updated feature vector of node i in layer l, is the feature vector of node i in layer l; is the set of neighbor nodes of node i; W (l) is the weight matrix of the graph convolution layer; Z ij is the normalization factor of the adjacency matrix; σ is the nonlinear activation function, j is the node j, is the feature vector of node j in layer l; b (l) is the bias term; The time series input vector is input into the fault classification model to calculate the fault type label and fault severity.

6. The substation monitoring method based on the Internet of Things according to claim 5, characterized in that: Based on the fault type label and severity, a graph convolutional network is used to calculate the fault impact range and obtain the fault propagation path.

7. The substation monitoring method based on the Internet of Things according to claim 1, characterized in that: Said S4 specifically includes: Establish the physical topology of the equipment in the digital twin environment and construct a dynamic feature vector of the substation operating status; the dynamic feature vector includes the voltage state vector, current state vector, equipment temperature distribution, electromagnetic disturbance feature matrix, and fault severity label; A fault evolution prediction model is constructed using a physics-enhanced neural network to analyze the dynamic evolution of substation faults. The loss function of the fault evolution prediction model consists of a physical constraint term and a data error term. Based on the dynamic feature vector and fault propagation path, a physics-enhanced neural network is used to predict the fault development trend and estimate the possible secondary fault points. Based on the timing topology graph, the fault propagation probability is determined, and the predicted device state is updated based on the fault propagation probability, and the future state vector is obtained and the simulation data is stored.

8. The substation monitoring method based on the Internet of Things according to claim 1, characterized in that: Said S5 specifically includes: Determine a comprehensive risk score based on anomaly scores, fault propagation paths, and historical fault data; The device's operational stability score is calculated by weighted summation based on the device's current voltage, current, and temperature fluctuations. The final maintenance priority is calculated based on the maintenance cost, combined with the comprehensive risk score and the operation stability score.

9. The substation monitoring method based on the Internet of Things according to claim 8, characterized in that: Said S5 further includes: Calculate the optimal maintenance time T based on the final maintenance priority and the predicted failure development trend m (i), calculated as: Among them, T f (i) is the expected failure time of equipment i; T m (i) is the optimal maintenance time window. The smaller the value, the earlier the maintenance should be carried out. The optimal maintenance task scheduling plan is designed by combining the final maintenance priority with the best maintenance time.

10. The substation monitoring system based on the Internet of Things is characterized by: The system comprises: An electromagnetic disturbance signal acquisition unit is used to acquire the original electromagnetic disturbance signal of the substation, preprocess the original electromagnetic disturbance signal, perform variational mode decomposition and smoothing on the preprocessed electromagnetic disturbance signal, and construct an electromagnetic disturbance feature matrix; wherein each row of the electromagnetic disturbance feature matrix corresponds to a time step, and each column represents a different disturbance feature; The anomaly detection unit is used to construct a time-window-based time series input based on the electromagnetic disturbance feature matrix, and use an adaptive attention mechanism combined with an asymmetric anomaly detection loss to perform anomaly detection on the electromagnetic disturbance signal, and obtain the anomaly score corresponding to the electromagnetic disturbance signal, the corresponding disturbance feature, and the weight distribution of the anomaly point; a fault monitoring unit, configured to construct an abnormal feature matrix based on the corresponding disturbance features, aggregate the abnormal feature matrix within a time window into a time series input vector, input the time series input vector into a fault classification model to identify the fault type, output a fault type label and severity, and perform fault propagation path calculation on the fault type label and severity to obtain the fault propagation path; A fault trend prediction unit is used to simulate a digital twin model based on the fault type label, severity, and fault propagation path, simulate fault evolution through the digital twin model, and predict fault development trends, secondary fault impact range, and future state vectors; The maintenance optimization unit is used to generate maintenance priorities, time windows, and resource scheduling plans based on anomaly scores, fault propagation paths, and historical fault data to minimize total maintenance costs while maximizing system stability.

Images