An Internet of Things-based circuit short-circuit fault monitoring system and method

Abstract

The present invention relates to the field of Internet of Things technology, and particularly to a circuit short - circuit fault monitoring system and method based on the Internet of Things; the present invention identifies fault modes based on current characteristic waveforms, makes predictions based on historical data, uses a current sensor with high - speed sampling to capture current waveform data in real - time, identifies and marks abnormal changes in the current waveform; constructs an RNN deep - learning model, extracts features from the waveform, and compares them with known fault modes to identify potential short - circuit faults. The constructed deep - learning model makes inferences based on real - time data, quickly identifies abnormal waveform features, thereby achieving a fast response, greatly shortening the fault detection time, and effectively dealing with short - circuit faults in high - speed circuits. It uses distributed data aggregation and analysis technology to process power flow data of multiple nodes in real - time, constructs a Bayesian network model, identifies potential short - circuit faults when the power flow is abnormal, and locates the specific nodes and paths where the faults occur.

Description

Technical Field

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

[0002] Current circuit short - circuit fault monitoring systems mainly rely on current sensors, fuses or circuit breakers. Such devices identify short - circuit faults by detecting a sudden increase in current and then cut off the circuit for line protection; for the monitoring of power systems, it mainly relies on single - point current or voltage monitoring. Generally, sensors are arranged at multiple nodes of the circuit to monitor the power flow in real - time, and trigger a protection mechanism when an anomaly is detected, so as to realize the monitoring of circuit short - circuit faults.

[0003] Although the above - mentioned traditional circuit protection methods are simple and effective, they still have disadvantages. It can be seen that the above - mentioned several fault monitoring methods mainly focus on the change in current amplitude and ignore the characteristics of the current waveform. However, in complex circuits, potential short - circuit faults often do not cause an instant large - scale change in current, but are manifested through subtle abnormalities in the waveform. If short - circuit faults are identified only by a sudden increase in current, it will undoubtedly lead to missed reports or inability to give early warnings; at the same time, the circuit short - circuit fault monitoring system relies on the action of physical components (such as fuses). These components require a certain response time when detecting current anomalies. In some high - speed circuits or dynamic load environments, short - circuit faults will develop into serious problems in an extremely short time, and the response speed of the above - mentioned monitoring methods cannot meet the requirements; the power - off of current sensors, fuses or circuit breakers all responds after the fault occurs and cannot identify potential fault risks in advance;

[0004] For the monitoring method, traditional monitoring methods cannot comprehensively cover all nodes in the entire power system, it is difficult to reflect the overall situation, and it is easy to cause false alarms or missed reports; also due to limited monitoring points, when an anomaly is detected, it is necessary to conduct a step - by - step investigation to determine the fault location, which is likely to delay the fault handling time. Therefore, there is an urgent need for a circuit short - circuit fault monitoring system and method based on the Internet of Things to solve such problems. Summary of the Invention

[0005] Therefore, the present invention provides a circuit short - circuit fault monitoring system and method based on the Internet of Things to overcome the problems in the prior art that it is impossible to effectively monitor the characteristics of the current waveform, overly rely on the action of physical components (such as fuses), and has a certain delay in response; the single - point current and voltage monitoring methods of the power system cannot comprehensively cover all nodes in the entire power system, are prone to false alarms or missed reports, and are likely to delay the fault handling time.

[0006] To achieve the above object, on the one hand, the present invention provides a circuit short - circuit fault monitoring system based on the Internet of Things, including:

[0007] Real-time waveform capture module, which captures the current waveform data in the circuit in real time and quickly detects waveform anomalies;

[0008] Fault identification module, which is based on the real-time waveform data, uses deep learning algorithms to identify faults, and predicts potential fault trends according to historical data;

[0009] Multi-node power flow monitoring module, which conducts real-time monitoring of power flow at multiple key nodes in the power system and supports fault location;

[0010] Distributed data aggregation and anomaly detection module, which aggregates the power flow data from multiple nodes, uses Bayesian networks for global anomaly detection, and accurately locates the fault position;

[0011] Management module, which executes fault response according to the detection results of the distributed data aggregation and anomaly detection module.

[0012] Furthermore, the real-time waveform capture module includes:

[0013] High-speed sampling current sensor, which is deployed at key positions in the circuit and continuously captures the instantaneous waveform changes of the current;

[0014] Edge processing unit, which conducts preliminary analysis and anomaly marking on the collected waveform data, and transmits the key data to the fault identification module for further processing;

[0015] The fault identification module includes:

[0016] RNN deep learning model, which extracts features from the data transmitted by the real-time waveform capture module, identifies possible short-circuit faults, the model identifies anomalies through real-time inference, and feeds back the results to the monitoring center;

[0017] Fault prediction engine, which uses historical waveform data to predict possible fault trends, and issues an alarm in advance when waveforms similar to historical fault characteristics are detected;

[0018] The multi-node power flow monitoring module includes:

[0019] Multi-node monitoring devices, which are distributed at different positions in the power system, collect the power flow data of each node in real time, and transmit the data to the circuit short-circuit fault monitoring system through the Internet of Things;

[0020] Data transmission and synchronization, which conducts data synchronization between nodes;

[0021] The distributed data aggregation and anomaly detection module includes:

[0022] A data aggregation platform integrates the data transmitted by multiple nodes it receives to form a global power flow map.

[0023] A Bayesian network model analyzes abnormal phenomena in power flow based on the global data, quickly identifies short - circuit faults, and locates the fault nodes and paths.

[0024] The management module includes:

[0025] A fault response system automatically triggers preset response mechanisms, including power cut and alarm, according to the analysis results of the fault identification module and the Bayesian network.

[0026] On the other hand, the present invention also provides an Internet - of - Things - based method for monitoring circuit short - circuit faults, including:

[0027] Step S1, data acquisition: The real - time waveform capture module continuously collects current waveform data, and the edge processing unit performs preliminary processing on it and sends the abnormal data to the fault identification module.

[0028] Step S2, data analysis: The fault identification module deeply analyzes the real - time data, identifies short - circuit faults, and simultaneously predicts future possible fault trends, and outputs the results to the management module.

[0029] Step S3, global monitoring: The multi - node power flow monitoring module continuously tracks the states of all nodes in the power system.

[0030] Step S4, fault detection and location: The distributed data aggregation and anomaly detection module summarizes the data of multiple nodes, conducts global analysis through the Bayesian network model, identifies anomalies, and locates the fault points.

[0031] Step S5, response and management: The management module executes emergency response measures according to the detected fault information.

[0032] Further, in step S1, the data acquisition method is as follows:

[0033] Introduce an adaptive sampling rate f s (t), which dynamically adjusts the sampling rate according to the changes in the current waveform, where f s (t) represents the sampling rate at time t, f base represents the basic sampling rate, α represents the sampling rate gain coefficient, which controls the amplitude of the adaptive sampling rate, β represents the current change sensitivity coefficient, which controls the response degree of the sampling rate to the current change rate, represents the instantaneous change rate of the current I(t), which reflects the mutation of the current waveform, and γ represents the non - linear enhancement index of the change rate; when the current changes violently, the sampling rate is increased to capture more details; when the current changes smoothly, the sampling rate is decreased to reduce redundant data.

[0034] The current signal I(t) is passed through an A / D converter and sampled at the dynamic sampling rate of f s (t) to generate a discrete current signal sequence I[n], where n is the discrete time index. I[n] is the discrete sampled data point, representing the current value collected at the nth moment, and η[n] represents the sampling noise. A filter is introduced to process it.

[0035] High-dimensional feature extraction is performed on the discrete sampling points I[n] to construct a feature vector F(n) for preliminary anomaly detection. where F(n) represents the feature vector, which contains various signal features. and respectively represent the gradient and second-order gradient of the current, Η{I[n]} represents the signal envelope after Hilbert transform, Ρ(n) represents the average power feature within a specific window, indicating the power change of the current, and Ε(n) represents the energy feature, indicating the energy accumulation of the signal within a specific time window.

[0036] Furthermore, in step S1, the data acquisition method also includes:

[0037] The dynamic threshold detection algorithm is used to perform preliminary anomaly marking on the feature vector F(n). Θ(n) = μ + κ·σ, where Θ(n) represents the dynamic threshold, which is dynamically calculated based on the mean μ and standard deviation σ of the feature vector, and κ represents the sensitivity coefficient.

[0038] When any component of F(n) exceeds the threshold Θ(n), it is considered that an anomaly may exist, and this data is marked as a potential fault signal.

[0039] The waveform data marked as abnormal is preferentially transmitted to the fault identification module.

[0040] The normal waveform data not marked as abnormal is cached and compressed and stored using a compression algorithm based on wavelet transform:

[0041] where W(n) represents the compressed signal representation, ψ i (n) represents the wavelet basis function, a i represents the wavelet coefficient, indicating the features of the signal at each scale, and N represents the number of levels of signal decomposition in the wavelet transform. The data acquisition module not only realizes high-precision current waveform capture but also can perform preliminary processing and anomaly marking during acquisition, significantly improving the real-time performance, response speed, and detection accuracy of the system.

[0042] Furthermore, the data analysis method in step S2 is as follows:

[0043] Perform adaptive multi-scale wavelet transform on the real-time collected current signal I[n] to extract the time-frequency features of the signal:

[0044] where W I (a, b) represents the wavelet transform coefficient of the current signal I[n], a represents the scale parameter, which controls the frequency resolution of the signal, b represents the translation parameter, which controls the time resolution of the signal, ψ represents the mother wavelet function, which is used to extract the local features of the signal, and ψ * represents the conjugate function of the mother wavelet and is used for the inverse wavelet transform;

[0045] Flexibly adapt to the characteristics of different current waveforms by adaptively adjusting the wavelet scale a and the translation parameter b. Especially in multi-band signals, it can effectively separate the fault signal from the normal signal;

[0046] On the basis of wavelet transform, further extract the multi-dimensional feature matrix F(a, b), which includes time-frequency features, energy features, and instantaneous phase information. where W I (a, b) represents the time-frequency feature, ε I (a) represents the energy feature, which represents the energy distribution at each scale, and P I (a) represents the average power feature, which represents the power distribution within a specific time window, and φ I (a, b) = arg(W I (a, b)), which represents the instantaneous phase feature and captures the phase change of the signal;

[0047] Use the deep residual network ResNet to analyze the multi-dimensional feature matrix F(a, b). The residual structure of ResNet can effectively solve the problem of gradient disappearance in deep networks. The network structure is:

[0048] Ο = ResNet θ (F(a, b)), where Ο represents the output fault classification result, that is, whether there is a short-circuit fault in the current signal, and ResNet θ represents the deep residual network with parameters θ; maintain high-precision recognition through the structure of the residual network when the network depth increases;

[0049] Use the weighted cross-entropy loss function for network training:

[0050] where L represents the value of the loss function, C represents the number of classifications of faults and non-faults, w i represents the class weight, which is adjusted for the problem of sample imbalance, and y i and represent the true label and the predicted probability; balance the impact brought by sample imbalance through the weighted loss function.

[0051] Furthermore, the data analysis method in step S2 also includes:

[0052] Using a long short-term memory network (LSTM) to perform time series modeling on historical fault data:

[0053] where represents the predicted fault state at the future time t + k, and LSTM φ represents the LSTM network with parameters φ, and F t to F t-m represent the multi-dimensional feature matrices from the current time t to the past m-th time;

[0054] Combining the historical multi-dimensional feature matrices, predicting short-term faults, giving early warnings based on long-term trends, and helping to prevent potential system risks;

[0055] Using a Bayesian neural network (BNN) to evaluate the uncertainty of the prediction results, quantifying the confidence interval of the prediction, and the output of the BNN includes the predicted value and the variance of the prediction: where μ t+k , μ t+k and are the predicted mean and variance respectively.

[0056] Furthermore, the global monitoring method in step S3 is:

[0057] Installing monitoring devices at key nodes of the power system, including transformers, switches, and buses, to measure current, voltage, and power parameters in real time;

[0058] The monitoring device at each node continuously collects real-time data on the power flow;

[0059] The collected real-time data is transmitted to the central processing unit through the Internet of Things network, and the data from different nodes is synchronously processed in the central processing unit;

[0060] The central processing unit aggregates the data from different nodes to form a real-time global view of the entire power system.

[0061] Furthermore, the fault detection and location method in step S4 is:

[0062] Locally process the data from different nodes on their respective acquisition devices, and then transmit it to the data aggregation platform through a distributed computing architecture;

[0063] After data aggregation, preprocess the aggregated data to eliminate noise and irrelevant fluctuations, and use an adaptive noise filter for data cleaning;

[0064] Construct a Bayesian network based on the nodes of the power system, where the nodes represent various state variables in the power system, including voltage, current, and power, and the edges represent the causal relationships between the state variables, and the dependence relationships between the nodes in the power system are shown through the Bayesian network;

[0065] Use the Bayesian network to reason about the real-time data and calculate the fault probability of the system in the current state;

[0066] Bayesian update:

[0067] Among them, P(V i |D clean (t)) represents the posterior probability of node V clean after the given cleaned data D i . P(D clean (t)|V i ) represents the likelihood function of observing the data D i (t) in the state of node V clean . P(V i ) represents the prior probability of node V i . P(D clean (t)) represents the marginal probability of the data;

[0068] By updating the Bayesian network in real time, re-evaluate the fault probability of each node in the power system at each moment and identify abnormal situations in advance;

[0069] For each node V i , when the posterior probability P(V i |D clean (t)) exceeds the predetermined threshold, the system marks the node as abnormal.

[0070] Furthermore, the fault detection and location method in step S4 also includes:

[0071] When multiple node anomalies are detected, use the causal chain analysis of the Bayesian network to identify possible fault sources. The fault source node is usually the first node to become abnormal or the node that has the greatest impact on downstream nodes. Fault source location algorithm:

[0072] When the fault source is confirmed, immediately trigger the alarm mechanism to notify the maintenance personnel to take emergency measures.

[0073] Compared with the prior art, the present invention has the following beneficial effects:

[0074] In the present invention, fault mode recognition is performed based on the current characteristic waveform, prediction is performed based on historical data, the current waveform data is captured in real time by a high-speed sampling current sensor, and abnormal changes in the current waveform are identified and marked.

[0075] In the present invention, an RNN deep learning model is constructed to extract features from waveforms and compare them with known fault patterns to identify potential short - circuit faults. The constructed deep learning model makes inferences based on real - time data, quickly identifies abnormal waveform features, thereby achieving a fast response, significantly shortening the fault detection time, and effectively dealing with short - circuit faults in high - speed circuits.

[0076] In the present invention, historical waveform data is analyzed and learned to predict potential fault trends in the circuit; when the waveform features are similar to those of short - circuit faults in history, an alarm is issued in advance; the probability of faults is reduced, and the service life of circuit equipment is extended.

[0077] In the present invention, monitoring devices are arranged at multiple nodes of the power system, and the monitoring devices are interconnected through the Internet of Things to monitor the real - time flow of electricity between each node. Any abnormal changes in any node are captured from a global perspective, reducing the monitoring blind area.

[0078] In the present invention, distributed data aggregation and analysis technology is used to process power flow data of multiple nodes in real - time, a Bayesian network model is constructed, potential short - circuit faults are identified when the power flow is abnormal, and the specific nodes and paths where the faults occur are located.

[0079] In the present invention, data from different nodes are analyzed simultaneously in a multi - node monitoring method. When an abnormality occurs in a certain node, the relevant data is immediately compared with the power flow data of other nodes to quickly locate the fault source; compared with the step - by - step troubleshooting of traditional systems, this method significantly improves the fault response speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic flow chart of the method for monitoring circuit short - circuit faults based on the Internet of Things according to the present invention;

[0081] Figure 2 It is a schematic structural diagram of the system for monitoring circuit short - circuit faults based on the Internet of Things according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0082] In order to enable those skilled in the art of the present technology to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0083] It should be noted that the terms "first", "second", etc. in the specification, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0084] The present invention will be further described in detail below with reference to the accompanying drawings:

[0085] Embodiment 1

[0086] Please refer to Figure 1 - Figure 2 , the present invention provides an Internet of Things-based circuit short-circuit fault monitoring system, including:

[0087] A real-time waveform capture module that captures the current waveform data in the circuit in real time and quickly detects waveform anomalies;

[0088] The real-time waveform capture module includes:

[0089] A high-speed sampling current sensor deployed at key positions in the circuit to continuously capture the instantaneous waveform changes of the current;

[0090] An edge processing unit that performs preliminary analysis and anomaly marking on the collected waveform data, and transmits key data to the fault identification module for further processing;

[0091] A fault identification module that, based on the real-time waveform data, uses deep learning algorithms for fault identification and predicts potential fault trends according to historical data;

[0092] The fault identification module includes:

[0093] An RNN deep learning model that extracts features from the data transmitted by the real-time waveform capture module, identifies possible short-circuit faults, and the model identifies anomalies through real-time inference and feeds the results back to the monitoring center;

[0094] A fault prediction engine that uses historical waveform data to predict possible fault trends and issues an alarm in advance when waveforms similar to historical fault characteristics are detected;

[0095] A multi-node power flow monitoring module that performs real-time monitoring of power flow at multiple key nodes in the power system to support fault location;

[0096] The multi-node power flow monitoring module includes:

[0097] Multi-node monitoring devices are distributed at different positions in the power system, collect power flow data of each node in real time, and transmit the data to the circuit short-circuit fault monitoring system through the Internet of Things;

[0098] Data transmission and synchronization to synchronize data between nodes;

[0099] Distributed data aggregation and anomaly detection module, which aggregates power flow data from multiple nodes, uses Bayesian network for global anomaly detection, and accurately locates the fault position;

[0100] The distributed data aggregation and anomaly detection module includes:

[0101] Data aggregation platform, which integrates the data transmitted from multiple nodes based on the received data to form a global power flow map;

[0102] Bayesian network model, which analyzes abnormal phenomena in power flow based on global data, quickly identifies short-circuit faults, and locates fault nodes and paths;

[0103] Management module, which executes fault response according to the detection results of the distributed data aggregation and anomaly detection module;

[0104] The management module includes:

[0105] Fault response system, which automatically triggers a preset response mechanism according to the analysis results of the fault identification module and the Bayesian network, including power-off and alarm.

[0106] The present invention also provides an Internet of Things-based circuit short-circuit fault monitoring method, including the following steps:

[0107] Step S1, data acquisition, the real-time waveform capture module continuously collects current waveform data, the edge processing unit performs preliminary processing on it, and sends the abnormal data to the fault identification module;

[0108] Introduce an adaptive sampling rate f s (t), dynamically adjust the sampling rate according to the change of the current waveform, where f s (t) represents the sampling rate at time t, f base represents the basic sampling rate, α represents the sampling rate gain coefficient to control the amplitude of the adaptive sampling rate, β represents the current change sensitivity coefficient to control the response degree of the sampling rate to the current change rate, represents the instantaneous change rate of the current I(t), reflecting the abrupt change of the current waveform, and γ represents the non-linear enhancement index of the change rate; when the current changes violently, the sampling rate is increased to capture more details; when the current changes smoothly, the sampling rate is decreased to reduce redundant data;

[0109] The current signal I(t) passes through an A / D converter and is sampled at the dynamic sampling rate of f s (t) to generate a discrete current signal sequence I[n], where n is the discrete time index, I[n] is the discrete sampling data point, representing the current value collected at the nth moment, and η[n] represents the sampling noise, and a filter is introduced to process it;

[0110] High-dimensional feature extraction is performed on the discrete sampling points I[n], and a feature vector F(n) is constructed for preliminary anomaly detection, where F(n) represents the feature vector, which contains various signal features, and respectively represent the gradient and second-order gradient of the current, Η{I[n]} represents the signal envelope after Hilbert transform, Ρ(n) represents the average power feature within a specific window, indicating the power change of the current, and Ε(n) represents the energy feature, indicating the energy accumulation of the signal within a specific time window;

[0111] The dynamic threshold detection algorithm is used to perform preliminary anomaly marking on the feature vector F(n), Θ(n) = μ + κ·σ, where Θ(n) represents the dynamic threshold, which is dynamically calculated based on the mean μ and standard deviation σ of the feature vector, and κ represents the sensitivity coefficient;

[0112] When any component of F(n) exceeds the threshold Θ(n), it is considered that an anomaly may exist, and this data is marked as a potential fault signal;

[0113] The waveform data marked as abnormal is preferentially transmitted to the fault identification module;

[0114] The normal waveform data not marked as abnormal is cached, and a compression algorithm based on wavelet transform is used to compress and store it:

[0115] where W(n) represents the compressed signal representation, ψ i (n) represents the wavelet basis function, a i represents the wavelet coefficient, indicating the features of the signal at each scale, and N represents the number of levels of signal decomposition in the wavelet transform; the data acquisition module not only realizes high-precision current waveform capture, but also can perform preliminary processing and anomaly marking during acquisition, significantly improving the real-time performance, response speed and detection accuracy of the system;

[0116] Specifically, an adaptive sampling rate is adopted in data acquisition. The sampling frequency is adjusted according to the dynamic changes of the current waveform, increasing the sampling density during severe fluctuations and reducing redundant data during stable periods. The dynamic adjustment mechanism enables the system to accurately capture subtle anomalies in the current waveform. In addition, through real-time feature extraction and anomaly marking, potential faults are initially identified during the acquisition stage, shortening the response time for fault detection.

[0117] Step S2, data analysis. The fault identification module deeply analyzes the real-time data to identify short-circuit faults and simultaneously predicts future possible fault trends, outputting the results to the management module.

[0118] Perform adaptive multi-scale wavelet transform on the real-time acquired current signal I[n] to extract the time-frequency features of the signal:

[0119] Where W I (a, b) represents the wavelet transform coefficients of the current signal I[n]. a represents the scale parameter, which controls the frequency resolution of the signal, b represents the translation parameter, which controls the time resolution of the signal, ψ represents the mother wavelet function, used to extract the local features of the signal, ψ * represents the conjugate function of the mother wavelet, used for the inverse wavelet transform;

[0120] Flexibly adapt to the characteristics of different current waveforms by adaptively adjusting the wavelet scale a and the translation parameter b. Especially in multi-band signals, it can effectively separate fault signals from normal signals;

[0121] On the basis of wavelet transform, further extract the multi-dimensional feature matrix F(a, b), including time-frequency features, energy features, and instantaneous phase information. Where W I (a, b) represents the time-frequency features, ε I (a) represents the energy features, indicating the energy distribution at each scale, P I (a) represents the average power features, indicating the power distribution within a specific time window, φ I (a, b) = arg(W I (a, b)), represents the instantaneous phase features, capturing the phase changes of the signal;

[0122] Use the deep residual network ResNet to analyze the multi-dimensional feature matrix F(a, b). The residual structure of ResNet can effectively solve the problem of gradient disappearance in deep networks. The network structure:

[0123] Ο = ResNet θ (F(a, b)), where Ο represents the output fault classification result, that is, whether there is a short-circuit fault in the current signal, ResNet θDenote a deep residual network with parameter θ; maintain high-precision recognition through the structure of the residual network even when the network depth increases;

[0124] Use a weighted cross-entropy loss function for network training:

[0125] where L represents the value of the loss function, C represents the number of classifications of faults and non-faults, w i represents the class weight, which is adjusted for the problem of sample imbalance, y i and represent the true label and the predicted probability; balance the influence brought by sample imbalance through the weighted loss function;

[0126] Use a long short-term memory network LSTM to perform time series modeling on historical fault data:

[0127] where represents the fault state at the future time t + k predicted, and LSTM φ represents an LSTM network with parameter φ, F t to F t-m represent the multi-dimensional feature matrix from the current time t to the past mth time;

[0128] Combine the historical multi-dimensional feature matrix, predict faults in the short term, give early warnings based on the long-term trend, and help prevent potential system risks;

[0129] Use a Bayesian neural network BNN to evaluate the uncertainty of the prediction results, quantify the confidence interval of the prediction, and the output of the BNN includes the predicted value and the variance of the prediction: where μ t+k , μt + k and are the mean and variance of the prediction respectively;

[0130] Specifically, in the data analysis stage, use adaptive multi-scale wavelet transform to extract the time-frequency features of the signal, and combine with a deep residual network (ResNet) for fault mode recognition; the multi-level analysis method enables the system to maintain high-precision fault recognition ability even when facing complex current waveforms; predict future possible fault trends in advance through combining historical data for long short-term memory network LSTM modeling.

[0131] Step S3, global monitoring, the multi-node power flow monitoring module tracks the states of all nodes in the power system in real time;

[0132] Install monitoring devices at key nodes of the power system, including transformers, switches, and busbars, to measure current, voltage, and power parameters in real time;

[0133] The monitoring devices at each node continuously collect real-time data on power flow;

[0134] The collected real-time data is transmitted to the central processing unit through the Internet of Things network, where the data from different nodes is synchronously processed;

[0135] The central processing unit aggregates the data from different nodes to form a real-time global view of the entire power system;

[0136] Specifically, in terms of global monitoring, monitoring devices are deployed at key nodes to achieve comprehensive real-time monitoring of the power system. The data is synchronized and transmitted through the Internet of Things and aggregated at the central processing unit to form a global view of the power system.

[0137] Step S4, Fault Detection and Location. The distributed data aggregation and anomaly detection module summarizes the data from multiple nodes and conducts global analysis through the Bayesian network model to identify anomalies and locate the fault points;

[0138] The data from different nodes is locally processed on their respective collection devices and then transmitted to the data aggregation platform through the distributed computing architecture;

[0139] After data aggregation, preprocess the aggregated data to eliminate noise and irrelevant fluctuations, and use an adaptive noise filter for data cleaning;

[0140] Construct a Bayesian network based on the nodes of the power system, where the nodes represent various state variables in the power system, including voltage, current, and power, and the edges represent the causal relationships between state variables. The dependence relationships between nodes in the power system are shown through the Bayesian network;

[0141] Use the Bayesian network to infer the real-time data and calculate the fault probability of the system in the current state;

[0142] Bayesian Update:

[0143] where P(V i |D clean (t)) represents the posterior probability of node V clean after the given cleaned data D i , P(D clean (t)|V i ) represents the likelihood function of observing the data D i (t) in the state of node V clean , P(V i ) represents the prior probability of node V i , and P(D clean (t)) represents the marginal probability of the data;

[0144] By updating the Bayesian network in real time, re-evaluate the fault probabilities of each node in the power system at each moment to identify abnormal situations in advance;

[0145] For each node V i , when the posterior probability P(V i |D clean (t)) exceeds a predetermined threshold, the system marks this node as abnormal;

[0146] When multiple node anomalies are detected, use the causal chain analysis of the Bayesian network to identify possible fault sources. The fault source node is usually the first node to become abnormal or the node that has the greatest impact on downstream nodes. Fault source location algorithm:

[0147] When the fault source is confirmed, immediately trigger the alarm mechanism to notify the maintenance personnel to take emergency measures;

[0148] Specifically, during the fault detection and location process, use distributed data aggregation and the Bayesian network model to perform global analysis on the aggregated data of multiple nodes. Calculate the fault probabilities of each node through the inference ability of the Bayesian network, identify abnormal situations in advance, and determine the location of the fault source using causal chain analysis; through real-time updates, the system can dynamically adjust and re-evaluate the status of each node, and can quickly locate and respond when a fault occurs.

[0149] Step S5, Response and Management. The management module executes emergency response measures according to the detected fault information.

[0150] So far, the technical solution of the present invention has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to relevant technical features, and the technical solutions after these changes or substitutions will all fall within the protection scope of the present invention.

Claims

1. A circuit short circuit fault monitoring system based on the Internet of Things, characterized in that: include: Real-time waveform capture module, real-time capture of current waveform data in the circuit; The fault identification module uses deep learning algorithms to identify faults based on real-time waveform data and predict potential fault trends based on historical data; Multi-node power flow monitoring module, which performs real-time monitoring of power flow at multiple key nodes in the power system and supports fault location; Distributed data aggregation and anomaly detection module aggregates power flow data from multiple nodes, uses Bayesian networks to perform global anomaly detection, and accurately locates fault locations; The management module executes fault response according to the detection results of the distributed data aggregation and anomaly detection module.

2. According to the circuit short circuit fault monitoring system based on the Internet of Things according to claim 1, it is characterized in that: The real-time waveform capture module includes: High-speed sampling current sensors are deployed at key locations in the circuit to continuously capture instantaneous waveform changes of the current; The edge processing unit performs preliminary analysis and anomaly marking on the collected waveform data, and transmits key data to the fault identification module for further processing; The fault identification module includes: The RNN deep learning model extracts features from the data transmitted by the real-time waveform capture module to identify possible short-circuit faults. The model identifies anomalies through real-time reasoning and feeds the results back to the monitoring center. The fault prediction engine uses historical waveform data to predict possible fault trends and issues an early alarm when a waveform similar to historical fault characteristics is detected; The multi-node power flow monitoring module includes: Multi-node monitoring equipment, distributed at different locations in the power system, collects power flow data at each node in real time and transmits the data to the circuit short-circuit fault monitoring system through the Internet of Things; Data transmission and synchronization: data synchronization between nodes; The distributed data aggregation and anomaly detection module includes: The data aggregation platform integrates the data received from multiple nodes to form a global power flow map; The Bayesian network model analyzes abnormal phenomena in power flow based on global data, quickly identifies short-circuit faults, and locates faulty nodes and paths; The management modules include: The fault response system automatically triggers the preset response mechanism, including power off and alarm, based on the analysis results of the fault identification module and Bayesian network.

3. A circuit short circuit fault monitoring method based on the Internet of Things, characterized in that: A circuit short circuit fault monitoring system based on the Internet of Things according to any one of claims 1 to 2 is used, comprising the following steps: Step S1, data acquisition, the real-time waveform capture module continuously collects current waveform data, the edge processing unit performs preliminary processing on it, and sends abnormal data to the fault identification module; Step S2, data analysis, the fault identification module conducts in-depth analysis on the real-time data, identifies short-circuit faults, and predicts possible future fault trends, and outputs the results to the management module; Step S3, global monitoring, a multi-node power flow monitoring module tracks the status of each node in the power system in real time; Step S4, fault detection and location, the distributed data aggregation and anomaly detection module aggregates the data of multiple nodes, performs global analysis through the Bayesian network model, identifies anomalies and locates the fault point; Step S5, response and management, the management module executes emergency response measures according to the detected fault information.

4. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 3 is characterized in that: In step S1, the data collection method is: Introducing adaptive sampling rate f s (t), dynamically adjust the sampling rate according to the change of current waveform, where f s (t) represents the sampling rate at time t, f base represents the basic sampling rate, α represents the sampling rate gain coefficient, which controls the amplitude of the adaptive sampling rate, β represents the current change sensitivity coefficient, which controls the degree of response of the sampling rate to the current change rate, It represents the instantaneous rate of change of current I(t), reflecting the sudden change of current waveform, and γ represents the nonlinear enhancement index of the rate of change; The current signal I(t) is converted into f by an A / D converter. s (t) is sampled at a dynamic sampling rate to generate a discrete current signal sequence I[n], where n is a discrete time index, I[n] is a discrete sampling data point, indicating the current value collected at the nth moment, and η[n] represents the sampling noise; Perform high-dimensional feature extraction on discrete sampling points I[n] and construct feature vector F(n) for preliminary anomaly detection. Where F(n) represents the feature vector, which contains a variety of signal features. and They represent the gradient and second-order gradient of the current respectively, H{I[n]} represents the signal envelope after Hilbert transform, P(n) represents the average power characteristic within the sampling time window, indicating the power change of the current, E(n) represents the energy characteristic, indicating the energy accumulation of the signal within the sampling time window.

5. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 4 is characterized in that: In step S1, the data collection method further includes: Use the dynamic threshold detection algorithm to perform preliminary abnormal marking on the feature vector F(n), Θ(n) = μ + κ·σ, where Θ(n) represents the dynamic threshold, which is dynamically calculated based on the mean μ and standard deviation σ of the feature vector, and κ represents the sensitivity coefficient; When any component of F(n) exceeds the threshold Θ(n), it is considered that there may be an abnormality and the data is marked as a potential fault signal; The waveform data marked as abnormal is transmitted to the fault identification module in priority; The regular waveform data that is not marked as abnormal is cached and compressed and stored using a compression algorithm based on wavelet transform: Where W(n) represents the compressed signal representation, ψ i (n) represents the wavelet basis function, a i It represents the wavelet coefficient, which indicates the characteristics of the signal at each scale, and N represents the number of levels of signal decomposition in the wavelet transform.

6. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 5 is characterized in that: The data analysis method in step S2 is: Perform adaptive multi-scale wavelet transform on the current signal I[n] collected in real time to extract the time-frequency characteristics of the signal: Where W I (a, b) represents the wavelet transform coefficient of the current signal I[n], a represents the scale parameter, which controls the frequency resolution of the signal, b represents the translation parameter, which controls the time resolution of the signal, ψ represents the mother wavelet function, which is used to extract the local features of the signal, and ψ * Represents the conjugate function of the mother wavelet, used for the inverse transform of the wavelet transform; On the basis of wavelet transform, the multidimensional feature matrix F(a,b) is further extracted, which contains time-frequency features, energy features and instantaneous phase information. Where W I (a, b) represents the time-frequency characteristics, ε I (a) represents the energy characteristics, indicating the energy distribution at each scale, P I (a) represents the average power characteristic, which represents the power distribution within the sampling time window, φ I (a,b)=arg(W I (a, b)) represents the instantaneous phase feature, capturing the phase change of the signal; Use the deep residual network ResNet to analyze the multi-dimensional feature matrix F(a,b). The network structure is: Ο=ResNet θ (F(a,b)), where Ο represents the output fault classification result, that is, whether the current signal has a short circuit fault, ResNet θ represents a deep residual network with parameters θ; The network is trained using the weighted cross entropy loss function: Where L represents the loss function value, C represents the number of fault and non-fault classifications, and w i Represents the category weight, which is adjusted for the sample imbalance problem, y i and Represents the true label and predicted probability.

7. A circuit short circuit fault monitoring method based on the Internet of Things according to claim 6, characterized in that: The data analysis method in step S2 also includes: Use long short-term memory network LSTM to perform time series modeling on historical fault data: in Indicates the predicted fault status at the future time t+k, LSTM φ represents an LSTM network with parameter φ, F t To F t-m Represents the multidimensional feature matrix from the current time t to the mth time in the past; The Bayesian neural network BNN is used to evaluate the uncertainty of the prediction results and quantify the confidence interval of the prediction. The output of BNN includes the predicted value and the predicted variance: in and are the predicted mean and variance, respectively.

8. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 7 is characterized in that: The global monitoring method in step S3 is: Install monitoring equipment at key nodes of the power system, including transformers, switches, and busbars, to measure current, voltage, and power parameters in real time; Monitoring equipment at each node continuously collects real-time data on power flow; The collected real-time data is transmitted to the central processing unit through the IoT network, where the data from different nodes are processed synchronously; The central processing unit aggregates data from different nodes to form a real-time global view of the entire power system.

9. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 8, characterized in that: The fault detection and location method in step S4 is: Data from different nodes are processed locally on their respective collection devices and then transmitted to the data aggregation platform through a distributed computing architecture; After data aggregation, the aggregated data is preprocessed to eliminate noise and irrelevant fluctuations, and an adaptive noise filter is used for data cleaning; Construct a Bayesian network based on power system nodes, where the nodes represent the various state variables in the power system, including voltage, current, and power, and the edges represent the causal relationship between the state variables. The Bayesian network is used to display the dependency relationship between the nodes in the power system. Use Bayesian networks to reason about real-time data and calculate the probability of failure of the system in its current state; Bayesian Update: Where P(V i |D clean (t)) represents the given cleaned data D clean (t) later, node V i The posterior probability, P(D clean (t)|V i ) indicates that at node V i Data D is observed in the state clean The likelihood function of (t), P(V i ) represents node V i The prior probability, P(D clean (t)) represents the marginal probability of the data; For each node V i , when the posterior probability P(V i |D clean (t)) exceeds a predetermined threshold, the system marks the node as abnormal.

10. The circuit short circuit fault monitoring method based on the Internet of Things according to claim 9, characterized in that: The fault detection and location method in step S4 also includes: When multiple node anomalies are detected, the causal chain analysis of the Bayesian network is used to identify possible fault sources. The fault source location algorithm: When the fault source is confirmed, the alarm mechanism is immediately triggered to notify maintenance personnel to take emergency measures.

Images