Breaker intelligent control system and method based on Internet of Things
By using an IoT-based smart circuit breaker system, combined with wavelet transform and multi-objective particle swarm optimization algorithm, efficient fault detection and dynamic protection of the power system are achieved. This solves the shortcomings of traditional circuit breakers in fault detection and protection strategies, and improves the safety, stability and equipment lifespan of the power system.
Patent Information
- Application Number
- CN202511381110.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Priority Date
- 2025-07-24
- Filing Date
- 2025-09-25
- Publication Date
- 2026-01-09
AI Technical Summary
Traditional circuit breakers lack in-depth analysis capabilities in fault detection, making it difficult to capture microsecond-level high-frequency transient fault signals in real time. Furthermore, their protection strategies cannot balance power supply reliability with equipment lifespan, leading to misjudgments or missed judgments, which affect the safe and stable operation of the power system.
An IoT-based intelligent control system is adopted, which utilizes a signal acquisition module, a fault determination module, an edge computing module, and a dynamic protection module. It combines wavelet transform, time series isolated forest algorithm, and multi-objective particle swarm optimization algorithm to achieve real-time acquisition and fault determination of current, voltage, temperature, and vibration signals, and dynamically adjust the protection strategy.
It enables accurate detection of high-frequency transient faults, reduces false alarms and missed alarms, extends the service life of circuit breakers, and improves the operating efficiency and economy of power systems.
Smart Images
Figure CN121308367A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the field of intelligent control of circuit breakers, in particular to an intelligent control system and method of circuit breakers based on the Internet of Things. BACKGROUND
[0002] In the process of the continuous advancement of the intelligentization and high efficiency of the power system today, the control technology of the circuit breaker, as a key device to ensure the stability and safety of power supply, is crucial.
[0003] The traditional circuit breaker has a single means in terms of fault detection, lacks deep analysis capability of operation data, and is difficult to capture real-time and accurate high-frequency transient fault signals such as arc fault and weak pulse before insulation breakdown; its fault judgment mostly depends on preset fixed threshold, which cannot be flexibly adapted when the power grid has load fluctuations and equipment aging, and is prone to misjudgment or omission, so that the fault cannot be discovered and handled in time, seriously affecting the safe and stable operation of the power system; the traditional protection strategy often focuses on a single target, such as only paying attention to cutting off the fault current, ignoring the loss caused to users by power failure and the wear and tear of the circuit breaker itself due to frequent operation; such one-sided protection strategy cannot find a balance between ensuring the reliability of power supply and prolonging the service life of the equipment, resulting in low efficiency and economy of the power system.
[0004] Therefore, it is necessary to provide an intelligent control system and method of circuit breakers based on the Internet of Things to solve the above technical problems. SUMMARY
[0005] In order to solve the technical problems proposed in the background art, the present application provides an intelligent control system and method of circuit breakers based on the Internet of Things.
[0006] The object of the present application can be achieved by the following technical solutions:
[0007] The present application provides an intelligent control system of circuit breakers based on the Internet of Things, which specifically comprises: an intelligent circuit breaker terminal, a cloud platform and a database, the intelligent circuit breaker terminal comprising a signal acquisition module and a fault judgment module, and the cloud platform comprising an edge computing module and a dynamic protection module.
[0008] The signal acquisition module acquires the operation data of the intelligent circuit breaker in real time according to the integrated current, voltage, temperature and vibration sensors, and adds a wavelet transform coprocessor to support parallel computing of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal, which has the following process:
[0009] The intelligent circuit breaker is provided with a current sensor, a voltage sensor, a temperature sensor and a vibration sensor. Real-time current, real-time voltage, real-time temperature and real-time vibration data are collected by the above-mentioned sensors, and are marked as α, β, υ and η respectively. The signals of α, β, υ and η are amplified to improve the amplitude of the signals. The amplified signals are filtered by a Butterworth filter to remove high-frequency noise in the signals. The transfer function of the filter is set as The filtered signals are as follows: filtered current signal Filtered voltage signal Filtered temperature signal Filtered vibration signal Wherein, F represents Fourier transform, F -1 represents inverse Fourier transform; the filtered current signal is decomposed in time-frequency domain by a wavelet transform coprocessor. The wavelet transform coprocessor is provided with a Morlet wavelet basis function ψ and supports parallel calculation of multi-scale wavelet coefficients. The wavelet transform calculation logic is as follows: α W r,e (r,e) = α·ψ m , r represents a scale parameter, and e represents a translation parameter; a plurality of different scales {r1, r2,..., rm} are selected in sequence for parallel calculation, and m represents the total number of scale parameters. In this way, wavelet coefficients {W α (r1,e), W α (r2,e),..., W α (r m ,e)} under different scales are obtained; different scales correspond to different frequency resolutions, larger scales correspond to lower frequencies, and smaller scales correspond to higher frequencies; through multi-scale decomposition, the current signal can be decomposed into different frequency subbands, so as to extract the energy features of the high frequency band; the filtered current signal, voltage signal, temperature signal and vibration signal are sent to a cloud platform and a fault determination module;
[0010] The square sum of the wavelet coefficients under the corresponding scale is calculated to obtain the energy value E i , W α (r i ,e) is set as the wavelet coefficient of the corresponding high frequency band scale r i , and the energy value calculation logic under this scale is as follows: E i = |W α (r i ,e)| 2 , the energy values under all high frequency band scales are added to obtain the total energy E high of the high frequency band;
[0011] The proportion P of the high frequency energy to the total energy is calculated, and the calculation logic is as follows: Wherein The total energy across all scales is calculated; a dynamic energy threshold is extracted from the database. If P exceeds the dynamic energy threshold, the fault determination module is triggered.
[0012] The fault determination module uses the time series isolated forest algorithm to identify abnormal subsequences in the running data and outputs a binary classification result to determine whether an anomaly exists. The specific process is as follows:
[0013] The current, voltage, temperature, and vibration signals are divided into subsequences using a sliding window of fixed size with a step size of 1. The statistical vectors of each subsequence are then obtained, including the mean. The variance value σ and the kurtosis value θ are used to obtain the statistical vector d = {ζ, σ, θ} corresponding to each subsequence;
[0014] Constructing an isolated forest for time series data involves randomly selecting a subset of the statistical vectors from all subsequences as training data, D = {d1, d2, ..., dn}. z}, where Z represents the number of training data. For each isolated tree T, a statistical vector and a split point are randomly selected. The training data D is split according to the statistical vector and the split point. Data smaller than the split point is placed in the left subtree, and data greater than or equal to the split point is placed in the right subtree. Then, the left and right subtrees are recursively split until each subset contains only one data point. This step is repeated until multiple isolated trees are constructed, forming an isolated forest.
[0015] For a new subsequence's statistics vector, input it into each isolated tree in the constructed isolated forest, and compute the path length XT in each tree. k Let k represent the number of the isolated tree. Then, the average path length of all trees is taken to obtain the anomaly score XE of this data subsequence. The calculation logic for the average path length is as follows: Where K represents the total number of isolated trees. The anomaly score is calculated based on the average path length as follows: Where c(n) represents the harmonic number;
[0016] An abnormal score threshold μ, an abnormal counter, and an abnormal sequence threshold are set in the intelligent circuit breaker terminal. If the abnormal score XE > μ, the subsequence is determined to be abnormal; otherwise, it is determined to be normal. When the abnormal counter detects an abnormal subsequence, the abnormal counter value is incremented by 1. When the abnormal counter value is greater than the abnormal sequence threshold, an abnormal signal is generated and sent to the cloud platform.
[0017] The edge computing module, based on the features extracted from the operational data, inputs a pre-defined binary neural network (BNN) fault classification model and outputs specific fault classification results. The specific process is as follows:
[0018] Upon receiving an abnormal signal, time-domain and frequency-domain features are extracted from the current, voltage, temperature, and vibration signals. Time-domain features include peak values and RMS values, while frequency-domain features include the spectrum and power spectrum. The extracted feature vector is labeled as G = {g1, g2, ..., g...} n}, where n represents the total number of features; the extracted feature vectors are binarized, converted to +1 or -1 form to meet the input requirements of the binary neural network BNN. The binarization calculation logic is as follows: Where g j It is the j-th element in the eigenvector. These are the binarized elements, where 'sign' represents the sign function, and the binarized feature vector.
[0019] A binary neural network fault classification model is trained using the binarized feature vectors. This model consists of an input layer, hidden layers, and an output layer. The hidden layers include multiple binary convolutional layers, pooling layers, and fully connected layers. The cross-entropy loss function L is chosen to measure the difference between the model's predictions and the true labels. Its calculation logic is as follows: Where C represents the number of fault types, yv represents the v-th component of the true label, sv represents the probability value of the v-th fault type predicted by the model, and log represents the natural logarithm;
[0020] The real-time collected feature vectors are input into the preset binary neural network fault classification model. The model is propagated through the input layer and hidden layer to the output layer, where the predicted probability value of each fault type is stored. The fault type with the highest predicted probability value is selected as the fault classification result according to the argmax function. The calculation logic of the argmax function is: fault type = argmax v(pv).
[0021] The dynamic protection module uses a multi-objective particle swarm optimization algorithm to search for the optimal solution set on the Pareto front with the objective function as the output, generating an adaptive protection strategy to send opening and closing commands and fault retry commands to the smart circuit breaker terminal. The specific process is as follows:
[0022] Retrieve the overload protection current threshold, tripping action delay, and fault retry interval from the database, and label them as I. th t trip and T delay This is integrated into a protection strategy parameter vector γ = [I th , t trip T delay Minimizing the power outage duration is denoted as the first objective function f1, and minimizing the circuit breaker mechanical wear is denoted as the second objective function f2. The dual objective functions are modeled accordingly, and their calculation logic is as follows: Where td a Let φ represent the duration of the a-th power outage event, φ be the total number of power outage events within a week, a be the number of the power outage event within a week, and N be the number of the power outage event within a week. op The circuit breaker is represented by the number of opening and closing cycles, KF represents the wear coefficient for a single operation, KL represents the average operating current, and I represents the circuit breaker's operating current. avg Represented as rated current, I rated It is represented as the current overload influence coefficient; it should be noted that the power outage duration and the total number of power outage events are collected by the cloud platform and sent to the database for storage, and the cloud platform can adjust the data;
[0023] Initialize the target particle swarm: the position vector of each particle is... b represents the position vector of any particle. It should be noted that the particle's position vector directly corresponds to the parameter vector of the protection strategy; each particle has a velocity vector used to update its position. ο represents the total number of velocity vectors; initialize the empty set P. elite It is used to store historical Pareto front solutions; the position and velocity of the particles are randomly initialized, and then the initial individual optimal position (phest) and global optimal position (gbest) of each particle are set.
[0024] Based on the protection strategy vector of each particle's position vector, the comprehensive objective function under the current power grid operation state is calculated and used as the fitness value of the particle. The calculation logic is as follows: YQ = f1(γ) × 0.7 + f2(γ) × 0.3 to obtain the comprehensive objective function YQ, which is used as the fitness value. Here, 0.7 is the fixed weight of the first objective function and 0.3 is the fixed weight of the second objective function.
[0025] The fitness value of each particle is compared with the fitness value of its individual best position. If the current fitness value is greater than the fitness value of its individual best position, the individual best position of the particle is updated. Among the individual best positions of all particles, the position with the best fitness value is found and the global best position is updated.
[0026] The velocity and position of the optimized particles are updated according to the multi-objective particle swarm optimization algorithm. The calculation logic is as follows: Where θ is the inertia weight, ε1 and ε2 are cognitive and social learning factors, λ1 and λ2 are random numbers in the interval [0,1], and pbest b This is represented as the individual optimal solution for particle b;
[0027] Based on the updated particle velocity and position, a set of optimal solutions is obtained through the Pareto front, which is then mapped to the corresponding I... th new、t trip new and T delayThe system generates opening / closing commands and fault retry commands based on this information and sends them to the intelligent circuit breaker terminal. The logic for generating opening / closing commands is as follows: Fault retry instruction generation logic: If it is the first fault, retry will be performed after 5 minutes; if it is a series of faults, retry will be performed after 30 minutes and manual intervention will be requested.
[0028] The second aspect of this invention provides a smart control method for circuit breakers based on the Internet of Things, the specific steps of which are as follows:
[0029] Step 1, Signal Acquisition: Based on integrated current, voltage, temperature and vibration sensors, the operating data of the smart circuit breaker is acquired in real time, and a wavelet transform coprocessor is added to support parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal;
[0030] Step 2, Fault Determination: Based on the time series isolated forest algorithm, identify abnormal subsequences in the running data and output binary classification results to determine whether an anomaly exists;
[0031] Step 3, Edge Computing: Based on the features extracted from the operational data, input the preset binary neural network fault classification model and output the fault classification results;
[0032] Step 4, Dynamic Protection: Using a multi-objective particle swarm optimization algorithm, the optimal solution set is searched on the Pareto front with the objective function as input, and an adaptive protection strategy is generated to send opening and closing commands and fault retry commands to the intelligent circuit breaker terminal.
[0033] Compared with the prior art, the beneficial effects of the present invention are:
[0034] The dynamic protection module utilizes a multi-objective particle swarm optimization algorithm, comprehensively considering two objectives: minimizing power outage duration and minimizing circuit breaker mechanical wear. It searches for the optimal solution set on the Pareto front, enabling the protection strategy to be dynamically adjusted according to actual conditions, balancing power outage losses and equipment lifespan losses. Through optimized opening and closing commands and fault retry commands, it can not only effectively protect the power system but also extend the service life of circuit breakers and reduce equipment replacement.
[0035] The edge computing module extracts time-domain and frequency-domain features from various signals, which are then binarized and input into a binary neural network fault classification model. This model, trained using a cross-entropy loss function, can accurately distinguish different fault types, providing strong support for subsequent targeted maintenance and handling. Compared with traditional fault classification methods, this reduces fault diagnosis time and improves maintenance efficiency. Attached Figure Description
[0036] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. The following drawings are not drawn to scale according to the actual size, but are intended to illustrate the main idea of the present invention.
[0037] Figure 1 This is a block diagram showing the module connection of the present invention.
[0038] Figure 2 This is a diagram illustrating the method steps of the present invention. Detailed Implementation
[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are also within the scope of protection of the present invention.
[0040] Please refer to Figure 1 As shown, the present invention provides an Internet of Things-based intelligent control system for circuit breakers, specifically including: an intelligent circuit breaker terminal, a cloud platform, and a database. The intelligent circuit breaker terminal includes a signal acquisition module and a fault determination module, and the cloud platform includes an edge computing module and a dynamic protection module.
[0041] The signal acquisition module collects real-time operating data of the smart circuit breaker based on integrated current, voltage, temperature, and vibration sensors. It also includes a wavelet transform coprocessor to support parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal. The process is as follows:
[0042] The intelligent circuit breaker is equipped with a current sensor, a voltage sensor, a temperature sensor, and a vibration sensor. These sensors collect real-time current, voltage, temperature, and vibration data, which are labeled α, β, υ, and η, respectively. The signals of α, β, υ, and η are amplified to increase their amplitude. The amplified signals are then filtered by a Butterworth filter to remove high-frequency noise. The transfer function of the filter is set as follows: The filtered signals are: the filtered current signal. Filtered voltage signal Filtered temperature signal Filtered vibration signal Where F represents the Fourier transform, F -1 This represents the inverse Fourier transform. The filtered current signal is decomposed in the time-frequency domain using a wavelet transform coprocessor. The wavelet transform coprocessor includes Morlet wavelet basis functions ψ and supports parallel computation of multi-scale wavelet coefficients. Its wavelet transform calculation logic is as follows: Wα (r,e)=α·ψ r,e r represents the scale parameter, and e represents the translation parameter; multiple different scales {r1, r2, ..., r} are selected sequentially. m Parallel computation is performed, where m represents the total number of scale parameters, to obtain wavelet coefficients {W} at different scales. α (r1,e), W α (r2,e).....W α (r m Different scales correspond to different frequency resolutions; larger scales correspond to lower frequencies, and smaller scales correspond to higher frequencies. Through multi-scale decomposition, the current signal can be decomposed into different frequency sub-bands, thereby extracting the energy characteristics of the high-frequency band. The filtered current signal, voltage signal, temperature signal, and vibration signal are sent to the cloud platform and fault judgment module.
[0043] The energy value E is obtained by calculating the sum of squares of the wavelet coefficients at the corresponding scale. i Set W α (r i e) represents the wavelet coefficients corresponding to the high-frequency scale ri. The energy value calculation logic at this scale is: E i =|W α (r i ,e)| 2 The total energy E of the high-frequency band is obtained by summing the energies at all high-frequency scales. high ;
[0044] The calculation logic for the proportion P of high-frequency energy to total energy is as follows: in The total energy across all scales is calculated; a dynamic energy threshold is extracted from the database. If P exceeds the dynamic energy threshold, the fault determination module is triggered.
[0045] The fault determination module uses the time series isolated forest algorithm to identify abnormal subsequences in the running data and outputs a binary classification result to determine whether an anomaly exists. The specific process is as follows:
[0046] The current, voltage, temperature, and vibration signals are segmented into subsequences using a sliding window of fixed size with a step size of 1. The statistical vectors of each subsequence are then obtained, including the mean ζ, variance σ, and kurtosis θ. This yields the corresponding statistical vector for each subsequence.
[0047] Constructing an isolated forest for time series data involves randomly selecting a subset of the statistical vectors from all subsequences as training data, D = {d1, d2, ..., dn}. z}, where Z represents the number of training data. For each isolated tree T, a statistical vector and a split point are randomly selected. The training data D is split according to the statistical vector and the split point. Data smaller than the split point is placed in the left subtree, and data greater than or equal to the split point is placed in the right subtree. Then, the left and right subtrees are recursively split until each subset contains only one data point. This step is repeated until multiple isolated trees are constructed, forming an isolated forest.
[0048] For a new subsequence's statistics vector, input it into each isolated tree in the constructed isolated forest, and compute the path length XT in each tree. k Let k represent the number of the isolated tree. Then, the average path length of all trees is taken to obtain the anomaly score XE of this data subsequence. The calculation logic for the average path length is as follows: Where K represents the total number of isolated trees. The anomaly score is calculated based on the average path length as follows: Where c(n) represents the harmonic number;
[0049] An abnormal score threshold μ, an abnormal counter, and an abnormal sequence threshold are set in the intelligent circuit breaker terminal. If the abnormal score XE > μ, the subsequence is determined to be abnormal; otherwise, it is determined to be normal. When the abnormal counter detects an abnormal subsequence, the abnormal counter value is incremented by 1. When the abnormal counter value is greater than the abnormal sequence threshold, an abnormal signal is generated and sent to the cloud platform.
[0050] The edge computing module, based on the features extracted from the operational data, inputs a pre-defined binary neural network (BNN) fault classification model and outputs specific fault classification results. The specific process is as follows:
[0051] Upon receiving an abnormal signal, time-domain and frequency-domain features are extracted from the current, voltage, temperature, and vibration signals. Time-domain features include peak values and RMS values, while frequency-domain features include the spectrum and power spectrum. The extracted feature vector is labeled as G = {g1, g2, ..., g...} n}, where n represents the total number of features; the extracted feature vectors are binarized, converted to +1 or -1 form to meet the input requirements of the binary neural network BNN. The binarization calculation logic is as follows: Where g j It is the j-th element in the eigenvector. These are the binarized elements, where 'sign' represents the sign function, and the binarized feature vector.
[0052] A binary neural network fault classification model is trained using the binarized feature vectors. This model consists of an input layer, hidden layers, and an output layer. The hidden layers include multiple binary convolutional layers, pooling layers, and fully connected layers. The cross-entropy loss function L is chosen to measure the difference between the model's predictions and the true labels. Its calculation logic is as follows: Where C represents the number of fault types, yv represents the v-th component of the true label, sv represents the probability value of the v-th fault type predicted by the model, and log represents the natural logarithm;
[0053] The real-time collected feature vectors are input into the preset binary neural network fault classification model. The model is propagated through the input layer and hidden layer to the output layer, where the predicted probability value of each fault type is stored. The fault type with the highest predicted probability value is selected as the fault classification result according to the argmax function. The calculation logic of the argmax function is: fault type = argmax v(pv).
[0054] The dynamic protection module uses a multi-objective particle swarm optimization algorithm to search for the optimal solution set on the Pareto front with the objective function as the output, generating an adaptive protection strategy to send opening and closing commands and fault retry commands to the smart circuit breaker terminal. The specific process is as follows:
[0055] Retrieve the overload protection current threshold, tripping action delay, and fault retry interval from the database, and label them as I. th t trip and T delay This is integrated into a protection strategy parameter vector γ = [I th , t trip , t delay Minimizing the power outage duration is denoted as the first objective function f1, and minimizing the circuit breaker mechanical wear is denoted as the second objective function f2. The dual objective functions are modeled accordingly, and their calculation logic is as follows: Where td a Let φ represent the duration of the a-th power outage event, φ be the total number of power outage events within a week, a be the number of the power outage event within a week, and N be the number of the power outage event within a week. op The circuit breaker is represented by the number of opening and closing cycles, KF represents the wear coefficient for a single operation, KL represents the average operating current, and I represents the circuit breaker's operating current. avg Represented as rated current, I rated It is represented as the current overload influence coefficient; it should be noted that the power outage duration and the total number of power outage events are collected by the cloud platform and sent to the database for storage, and the cloud platform can adjust the data;
[0056] Initialize the target particle swarm: the position vector of each particle is...
[0057] b represents the position vector of any particle. It should be noted that the particle's position vector directly corresponds to the parameter vector of the protection strategy; each particle has a velocity vector used to update its position. ο represents the total number of velocity vectors; initialize the empty set P. elite It is used to store historical Pareto front solutions; the position and velocity of the particles are randomly initialized, and then the initial individual optimal position (phest) and global optimal position (gbest) of each particle are set.
[0058] Based on the protection strategy vector of each particle's position vector, the comprehensive objective function under the current power grid operation state is calculated and used as the fitness value of the particle. The calculation logic is as follows: YQ = f1(γ) × 0.7 + f2(γ) × 0.3 to obtain the comprehensive objective function YQ, which is used as the fitness value. Here, 0.7 is the fixed weight of the first objective function and 0.3 is the fixed weight of the second objective function.
[0059] The fitness value of each particle is compared with the fitness value of its individual best position. If the current fitness value is greater than the fitness value of its individual best position, the individual best position of the particle is updated. Among the individual best positions of all particles, the position with the best fitness value is found and the global best position is updated.
[0060] The velocity and position of the optimized particles are updated according to the multi-objective particle swarm optimization algorithm. The calculation logic is as follows: Where θ is the inertia weight, ε1 and ε2 are cognitive and social learning factors, λ1 and λ2 are random numbers in the interval [0,1], and pbest b This is represented as the individual optimal solution for particle b;
[0061] Based on the updated particle velocity and position, a set of optimal solutions is obtained through the Pareto front, which is then mapped to the corresponding I... th new、t trip new and T delay The system generates opening / closing commands and fault retry commands based on this information and sends them to the intelligent circuit breaker terminal. The logic for generating opening / closing commands is as follows: Fault retry instruction generation logic: If it is the first fault, retry will be performed after 5 minutes; if it is a series of faults, retry will be performed after 30 minutes and manual intervention will be requested.
[0062] Please refer to Figure 2 As shown, the second aspect of the present invention provides a smart control method for circuit breakers based on the Internet of Things, the specific steps of which are as follows:
[0063] Step 1, Signal Acquisition: Based on integrated current, voltage, temperature and vibration sensors, the operating data of the smart circuit breaker is acquired in real time, and a wavelet transform coprocessor is added to support parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal;
[0064] Step 2, Fault Determination: Based on the time series isolated forest algorithm, identify abnormal subsequences in the running data and output binary classification results to determine whether an anomaly exists;
[0065] Step 3, Edge Computing: Based on the features extracted from the operational data, input the preset binary neural network fault classification model and output the fault classification results;
[0066] Step 4, Dynamic Protection: Using a multi-objective particle swarm optimization algorithm, the optimal solution set is searched on the Pareto front with the objective function as input, and an adaptive protection strategy is generated to send opening and closing commands and fault retry commands to the intelligent circuit breaker terminal.
[0067] The foregoing description is illustrative of the invention and should not be construed as limiting it. Although several exemplary embodiments of the invention have been described, those skilled in the art will readily understand that many modifications can be made to the exemplary embodiments without departing from the novel teachings and advantages of the invention. Therefore, all such modifications are intended to be included within the scope of the invention as defined in the claims. It should be understood that the foregoing description is illustrative of the invention and should not be construed as limiting it to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The invention is defined by the claims and their equivalents.
Claims
1. An IoT-based intelligent control system for circuit breakers, comprising an intelligent circuit breaker terminal, a cloud platform, and a database, wherein the intelligent circuit breaker terminal includes a signal acquisition module and a fault determination module, and the cloud platform includes an edge computing module and a dynamic protection module, characterized in that: The signal acquisition module performs time-frequency domain decomposition on the filtered current signal through a wavelet transform coprocessor, and supports parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal; the fault determination module identifies abnormal subsequences in the running data based on the time series isolated forest algorithm, and outputs binary classification results to determine whether there is an anomaly. The edge computing module, based on the features extracted from the operational data, inputs a preset binary neural network fault classification model and outputs the fault classification result. The dynamic protection module generates an adaptive protection strategy based on the optimal solution set obtained by the multi-objective particle swarm optimization algorithm, and sends opening and closing commands and fault retry commands to the intelligent circuit breaker terminal. The specific process is as follows: based on the updated particle velocity and position, a set of optimal solutions is obtained through the Pareto front, and these solutions are mapped to the corresponding I... th new、t trip new and T delay The system generates opening / closing commands and fault retry commands based on this information and sends them to the intelligent circuit breaker terminal. The logic for generating opening / closing commands is as follows: Fault retry instruction generation logic: If it is the first fault, retry will be performed after 5 minutes; if it is a series of faults, retry will be performed after 30 minutes and manual intervention will be requested.
2. The IoT-based intelligent control system for circuit breakers according to claim 1, characterized in that, The edge computing module performs binarization processing based on the features of the extracted runtime data, specifically as follows: Upon receiving an abnormal signal, time-domain and frequency-domain features are extracted from the current, voltage, temperature, and vibration signals. Time-domain features include peak values and RMS values, while frequency-domain features include the spectrum and power spectrum. The extracted feature vector is labeled as G = {g1, g2, ..., g...} n }, where n represents the total number of features; the extracted feature vectors are binarized, converted to +1 or -1 form to meet the input requirements of the binary neural network BNN. The binarization calculation logic is as follows: Where g j It is the j-th element in the eigenvector. These are the binarized elements, where 'sign' represents the sign function, and the binarized feature vector.
3. The IoT-based intelligent control system for circuit breakers according to claim 2, characterized in that, The edge computing module takes a preset binary neural network (BNN) fault classification model as input and outputs specific fault classification results. The specific process is as follows: A binary neural network fault classification model is trained using the binarized feature vectors. This model consists of an input layer, hidden layers, and an output layer. The hidden layers include multiple binary convolutional layers, pooling layers, and fully connected layers. The cross-entropy loss function L is chosen to measure the difference between the model's predictions and the true labels. Its calculation logic is as follows: Where C represents the number of fault types, yv represents the v-th component of the true label, sv represents the probability value of the v-th fault type predicted by the model, and log represents the natural logarithm; The real-time collected feature vectors are input into a pre-defined binary neural network fault classification model. The model propagates through the input layer and hidden layer to the output layer, where the predicted probability value of each fault type is stored. The fault type with the highest predicted probability value is selected as the fault classification result according to the argmax function. The calculation logic of the argmax function is: fault type = argmax v(pv).
4. The IoT-based intelligent control system for circuit breakers according to claim 1, characterized in that, The dynamic protection module models the objective function, and the specific process is as follows: Retrieve the overload protection current threshold, tripping action delay, and fault retry interval from the database, and label them as I. th t trip and T delay This is integrated into a protection strategy parameter vector γ = [I th ,t trip ,T delay Minimizing the power outage duration is denoted as the first objective function f1, and minimizing the circuit breaker mechanical wear is denoted as the second objective function f2. The dual objective functions are modeled accordingly, and their calculation logic is as follows: Where td a Let φ represent the duration of the a-th power outage event, φ be the total number of power outage events within a week, a be the number of the power outage event within a week, and N be the number of the power outage event within a week. op The circuit breaker is represented by the number of opening and closing cycles, KF represents the wear coefficient for a single operation, KL represents the average operating current, and I represents the circuit breaker's operating current. avg Represented as rated current, I rated It is expressed as the current overload influence coefficient.
5. The IoT-based intelligent control system for circuit breakers according to claim 4, characterized in that, The dynamic protection module initializes the target particles and uses a multi-target particle swarm optimization algorithm to update and optimize the velocity and position of the particles. The specific process is as follows: Initialize the target particle swarm: the position vector of each particle is... b represents the position vector of any particle. It should be noted that the particle's position vector directly corresponds to the parameter vector of the protection strategy; each particle has a velocity vector used to update its position. ο represents the total number of velocity vectors; initialize the empty set P. elite It is used to store historical Pareto front solutions; the position and velocity of the particles are randomly initialized, and then the initial individual optimal position (phest) and global optimal position (gbest) of each particle are set. Based on the protection strategy vector of each particle's position vector, the comprehensive objective function under the current power grid operation state is calculated and used as the fitness value of the particle. The calculation logic is as follows: YQ = f1(γ) × 0.7 + f2(γ) × 0.3 to obtain the comprehensive objective function YQ, which is used as the fitness value. Here, 0.7 is the fixed weight of the first objective function and 0.3 is the fixed weight of the second objective function. The fitness value of each particle is compared with the fitness value of its individual best position. If the current fitness value is greater than the fitness value of its individual best position, the individual best position of the particle is updated. Among the individual best positions of all particles, the position with the best fitness value is found and the global best position is updated. The velocity and position of the optimized particles are updated according to the multi-objective particle swarm optimization algorithm. The calculation logic is as follows: Where θ is the inertia weight, ε1 and ε2 are cognitive and social learning factors, λ1 and λ2 are random numbers in the interval [0,1], and pbest b Let be the individual optimal solution for particle b.
6. The IoT-based intelligent control system for circuit breakers according to claim 1, characterized in that, The signal acquisition module collects real-time operating data of the intelligent circuit breaker based on integrated current, voltage, temperature, and vibration sensors, and amplifies and filters the operating data. The specific process is as follows: The intelligent circuit breaker is equipped with a current sensor, a voltage sensor, a temperature sensor, and a vibration sensor. These sensors collect real-time current, voltage, temperature, and vibration data, which are labeled α, β, υ, and η, respectively. The signals of α, β, υ, and η are amplified, and the amplified signals are then filtered by a Butterworth filter to remove high-frequency noise. The transfer function of the filter is set as follows: The filtered signals are: the filtered current signal. Filtered voltage signal Filtered temperature signal Filtered vibration signal Where F represents the Fourier transform, F -1 This represents the inverse Fourier transform.
7. The IoT-based intelligent control system for circuit breakers according to claim 6, characterized in that, The signal acquisition module performs time-frequency domain decomposition on the filtered current signal through a wavelet transform coprocessor, supporting parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal. The process is as follows: The wavelet transform coprocessor includes Morlet wavelet basis functions ψ and supports parallel computation of multi-scale wavelet coefficients. Its wavelet transform computation logic is as follows: W α (r,e)=α·ψ r,e r represents the scale parameter, and e represents the translation parameter; multiple different scales {r1, r2, ..., r} are selected sequentially. m Parallel computation is performed, where m represents the total number of scale parameters, to obtain wavelet coefficients {W} at different scales. α (r1,e), W α (r2,e).....W α (r m Different scales correspond to different frequency resolutions; larger scales correspond to lower frequencies, and smaller scales correspond to higher frequencies. Through multi-scale decomposition, current signals can be decomposed into different frequency sub-bands, thereby extracting the energy characteristics of high-frequency bands. The filtered current signal, voltage signal, temperature signal, and vibration signal are sent to the cloud platform and fault diagnosis module. The energy value E is obtained by calculating the sum of squares of the wavelet coefficients at the corresponding scale. i Set W α (r i e) represents the corresponding high-frequency band scale r i The wavelet coefficients, and the energy value calculation logic at this scale, are: E i =|W α (r i ,e)| 2 The total energy E of the high-frequency band is obtained by summing the energies at all high-frequency scales. high ; The calculation logic for the proportion P of high-frequency energy to total energy is as follows: in The total energy across all scales is calculated; a dynamic energy threshold is extracted from the database. If P exceeds the dynamic energy threshold, the fault determination module is triggered.
8. The IoT-based intelligent control system for circuit breakers according to claim 1, characterized in that, The fault determination module constructs a time-series isolated forest, and the specific process is as follows: The current, voltage, temperature, and vibration signals are divided into subsequences using a sliding window of fixed size with a step size of 1. The statistical vectors of each subsequence are then obtained, including the mean. variance σ and kurtosis Obtain the statistical vectors corresponding to each subsequence Constructing an isolated forest for time series data involves randomly selecting a subset of the statistical vectors from all subsequences as training data, D = {d1, d2, ..., dn}. z Z represents the number of training data. For each isolated tree T, a statistical vector and a split point are randomly selected. The training data D is split according to the statistical vector and the split point. Data smaller than the split point is placed in the left subtree, and data greater than or equal to the split point is placed in the right subtree. Then, the left and right subtrees are recursively split until each subset contains only one data point. This step is repeated until multiple isolated trees are built, forming an isolated forest.
9. The IoT-based intelligent control system for circuit breakers according to claim 8, characterized in that, The fault determination module identifies abnormal subsequences in the operating data and outputs a binary classification result to determine whether an anomaly exists. The specific process is as follows: For a new subsequence's statistics vector, input it into each isolated tree in the constructed isolated forest, and compute the path length XT in each tree. k Let k represent the number of the isolated tree. Then, the average path length of all trees is taken to obtain the anomaly score XE of this data subsequence. The calculation logic for the average path length is as follows: Where K represents the total number of isolated trees. The anomaly score is calculated based on the average path length as follows: Where c(n) represents the harmonic number; An abnormal score threshold μ, an abnormal counter, and an abnormal sequence threshold are set in the intelligent circuit breaker terminal. If the abnormal score XE > μ, the subsequence is determined to be abnormal; otherwise, it is determined to be normal. When the anomaly counter detects an anomaly in a subsequence, the anomaly counter value is incremented by 1; When the value of the anomaly counter exceeds the anomaly sequence threshold, an anomaly signal is generated and sent to the cloud platform.
10. A circuit breaker intelligent control method based on the Internet of Things (IoT), used to implement the circuit breaker intelligent control system based on the IoT as described in any one of claims 1-9, characterized in that, include: Step 1, Signal Acquisition: Based on integrated current, voltage, temperature and vibration sensors, the operating data of the smart circuit breaker is acquired in real time, and a wavelet transform coprocessor is added to support parallel calculation of multi-scale wavelet coefficients for time-frequency domain decomposition of the current signal; Step 2, Fault Determination: Based on the time series isolated forest algorithm, identify abnormal subsequences in the running data and output binary classification results to determine whether an anomaly exists; Step 3, Edge Computing: Based on the features extracted from the operational data, input the preset binary neural network fault classification model and output the fault classification results; Step 4, Dynamic Protection: Using a multi-objective particle swarm optimization algorithm, the optimal solution set is searched on the Pareto front with the objective function as input, and an adaptive protection strategy is generated to send opening and closing commands and fault retry commands to the intelligent circuit breaker terminal.