Intelligent DC Circuit Breaker Arc Energy Suppression System and Its Method

Abstract

The present invention relates to the technical field of DC power distribution networks, and more specifically to an intelligent DC circuit breaker arc energy suppression system and method. The system comprises a high-frequency signal injection module that generates and injects a high-frequency carrier signal into the arc path; an arc feature recognition module that collects arc voltage and current signals, identifies the arc type using the VMD-HHT algorithm, and generates feature information; a magnetic flux disturbance observation module that detects current and magnetic flux, calculates the magnetic flux disturbance value and phase difference; a deep learning control module that receives the arc feature information and phase difference and generates an IGBT control timing signal using a deep reinforcement learning algorithm; and a multi-port commutation control module that receives this signal and controls the switching frequency of multiple parallel IGBT submodules, efficiently suppressing arc energy based on a PWM control method. A dual-heuristic algorithm control module combines a genetic algorithm and a particle swarm algorithm to adaptively distribute the drive current to each port, achieving parallel balancing of IGBT modules.

Description

Technical Field

[0001] The present invention relates to the technical field of DC distribution networks, and particularly to an intelligent DC circuit breaker arc energy suppression system and method, which are applicable to arc fault handling and energy suppression in DC distribution networks. Background Art

[0002] With the wide application of DC distribution networks, as a key protection device, the performance of DC circuit breakers directly affects the safety and stability of the system. During the opening process of DC circuit breakers, since there is no natural zero-crossing of DC current, continuous arc discharge phenomena are extremely likely to occur. These arcs will not only damage the circuit breaker itself, but also cause system voltage fluctuations, energy losses, and even lead to safety accidents.

[0003] In the prior art, the arc suppression of DC circuit breakers mainly adopts passive methods, such as increasing the contact distance, using arc extinguishing devices, adding absorption circuits, etc. Although these methods can suppress arcs to a certain extent, there are problems such as slow response speed, low energy conversion efficiency, and poor adaptability. Especially in high-voltage and high-current environments, the traditional arc extinguishing methods have limited effects and cannot meet the requirements of modern DC distribution networks for fast breaking and energy suppression.

[0004] In recent years, active arc suppression technologies based on power electronic devices have gradually received attention, but the existing technologies still have the following deficiencies: First, the arc characteristics are not accurately identified, and it is impossible to adopt differential suppression strategies for different types of arcs; second, the control algorithm is relatively simple and cannot achieve the optimal suppression of arc energy; third, the balanced control problem during parallel connection of multiple modules has not been effectively solved, affecting the stability and reliability of the system.

[0005] Therefore, how to achieve fast and efficient suppression of DC circuit breaker arcs and improve the safety and stability of DC distribution networks has become a technical problem to be solved urgently. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent DC circuit breaker arc energy suppression system and method, aiming to overcome the disadvantages of low arc suppression efficiency, poor adaptability, and inability to achieve intelligent control in the prior art, and to achieve efficient suppression of DC circuit breaker arc energy.

[0007] The present invention proposes an intelligent DC circuit breaker arc energy suppression system, including:

[0008] A high-frequency signal injection module, configured to:

[0009] Generate a high-frequency carrier signal;

[0010] Inject the high-frequency carrier signal into the DC circuit breaker arc path;

[0011] The arc feature recognition module, electrically connected to the high-frequency signal injection module, is used for:

[0012] Collecting the arc voltage signal and the arc current signal;

[0013] Based on the arc voltage signal and the arc current signal, identifying the arc type by using the VMD-HHT algorithm;

[0014] Generating arc feature information;

[0015] The magnetic flux disturbance observation module, electrically connected to the arc feature recognition module, is used for:

[0016] Real-time detecting the current magnetic flux in the circuit;

[0017] Converting the current magnetic flux into a voltage signal;

[0018] Calculating the magnetic flux disturbance value based on the voltage signal;

[0019] Generating the phase difference between the observation signal and the desired signal;

[0020] The deep learning control module, electrically connected to the magnetic flux disturbance observation module and the arc feature recognition module, is used for:

[0021] Receiving the arc feature information and the phase difference;

[0022] Based on the arc feature information and the phase difference, generating a control strategy through the deep reinforcement learning algorithm;

[0023] Outputting the IGBT control timing signal;

[0024] The multi-port commutation control module, electrically connected to the deep learning control module, is used for:

[0025] Receiving the IGBT control timing signal;

[0026] Controlling the switching frequency of the multi-level IGBT parallel sub-module;

[0027] Based on the PWM control method, achieving efficient suppression of the arc energy;

[0028] The dual heuristic algorithm control module, electrically connected to the multi-port commutation control module, is used for:

[0029] Based on the combination of the improved genetic algorithm and the particle swarm algorithm, constructing a dual heuristic intelligent algorithm;

[0030] Adaptive allocation of the driving current for each port;

[0031] Realizing the parallel balance of the IGBT module;

[0032] A communication interface module, electrically connected to the deep learning control module and the dual heuristic algorithm control module, is configured to:

[0033] Receive the arc feature information and the control strategy;

[0034] Transmit the arc feature information and the control strategy to the dispatching center;

[0035] Receive the control instructions returned by the dispatching center and transfer them to each functional module.

[0036] Preferably, the high-frequency signal injection module includes:

[0037] A high-frequency signal generation unit, configured to generate a high-frequency carrier signal with a specific frequency;

[0038] A signal modulation unit, electrically connected to the high-frequency signal generation unit, is configured to modulate the high-frequency carrier signal with the arc signal of the DC circuit breaker;

[0039] A high-frequency transformer unit, electrically connected to the signal modulation unit, is configured to couple the modulated high-frequency carrier signal to the arc path of the DC circuit breaker;

[0040] A signal injection switch unit, electrically connected to the high-frequency transformer unit, is configured to control the injection timing of the high-frequency carrier signal.

[0041] Preferably, the arc feature recognition module includes:

[0042] A sensing and acquisition unit, configured to collect the arc voltage signal and the arc current signal in real time;

[0043] A signal processing unit, electrically connected to the sensing and acquisition unit, is configured to filter and amplify the arc voltage signal and the arc current signal;

[0044] An arc feature extraction unit, electrically connected to the signal processing unit, is configured to:

[0045] Perform time-frequency analysis on the processed arc voltage signal and arc current signal by using the fast Fourier algorithm;

[0046] Extract the arc spectrum features;

[0047] An arc type recognition unit, electrically connected to the arc feature extraction unit, is configured to analyze the arc spectrum features based on the VMD-HHT algorithm and identify the arc as one of a continuous DC short arc, a DC small-gap arc, a DC long-gap arc, and a pure resistive or inductive arc.

[0048] Preferably, the flux disturbance observation module includes:

[0049] A flux observer for sampling magnetic flux by detecting current in a circuit;

[0050] An integrator electrically connected to the flux observer for converting the magnetic flux into a voltage;

[0051] An inverter electrically connected to the integrator for inverting the voltage to obtain a flux perturbation value;

[0052] A phase comparator electrically connected to the inverter for calculating the phase difference between the observed signal and the desired signal.

[0053] Preferably, the deep learning control module includes:

[0054] A feature input unit for receiving the arc feature information and the phase difference;

[0055] A neural network processing unit electrically connected to the feature input unit for:

[0056] Constructing a training sample based on the arc feature information and the phase difference;

[0057] Training a deep neural network to generate a frequency estimation model;

[0058] A reinforcement learning unit electrically connected to the neural network processing unit for:

[0059] Training a deep reinforcement learning algorithm based on the frequency estimation model;

[0060] Optimizing a control strategy through a reward function;

[0061] Wherein, the reward function is: , is the multi-path composite branch current, is the output voltage of the high-frequency transformer, is the loss of the high-frequency transformer.

[0062] Preferably, the multi-port commutation control module includes:

[0063] A multi-stage commutation structure unit for:

[0064] Constructing multiple commutation channels;

[0065] Each stage of the commutation branch includes 2 parallel IGBTs and a series-connected thyristor;

[0066] A fast recovery diode is connected in parallel between the IGBT and the thyristor;

[0067] A PWM signal control unit electrically connected to the multi-stage commutation structure unit for:

[0068] Receive the IGBT control timing signal;

[0069] Generate a PWM control signal;

[0070] Control the switching frequency of the IGBT;

[0071] An intelligent drive unit, electrically connected to the PWM signal control unit, for:

[0072] Realize intelligent drive with variable conduction angles;

[0073] Dynamically adjust the conduction angle of the IGBT according to system requirements;

[0074] Change the commutation timing.

[0075] Preferably, the dual heuristic algorithm control module includes:

[0076] A genetic algorithm unit for globally optimizing the port current distribution scheme;

[0077] A particle swarm algorithm unit for local precise optimization;

[0078] An algorithm fusion unit, electrically connected to the genetic algorithm unit and the particle swarm algorithm unit, for:

[0079] Construct a dual heuristic intelligent algorithm by combining the genetic algorithm and the particle swarm algorithm;

[0080] Dynamically adjust the optimization parameters according to the arc type and system state;

[0081] A current distribution unit, electrically connected to the algorithm fusion unit, for:

[0082] Allocate the drive current of each port based on the optimization result of the dual heuristic intelligent algorithm;

[0083] Realize the parallel balance of the IGBT module.

[0084] Preferably, the communication interface module includes:

[0085] A 5G communication unit for realizing high-speed data transmission;

[0086] A data processing unit, electrically connected to the 5G communication unit, for:

[0087] Encode the arc feature information and the control strategy;

[0088] Decode the control instruction returned by the dispatching center;

[0089] A security encryption unit, electrically connected to the data processing unit, for encrypting and protecting the transmitted data;

[0090] A cloud interface unit, electrically connected to the security encryption unit, is used for data interaction with the dispatching center.

[0091] Preferably, the system further includes:

[0092] A multi-stage variable FPI filter module, electrically connected to the high-frequency signal injection module, is used for:

[0093] Suppressing high-order harmonics in the voltage signal;

[0094] Separating the high-frequency signal from the power frequency signal during the occurrence of an arc;

[0095] A variable high-frequency transformer module, electrically connected to the high-frequency signal injection module, is used for:

[0096] Absorbing the impact of arc energy on the system;

[0097] Appearing inductive in the case of an arc and capacitive in the case of an AC signal;

[0098] Among them, the variable high-frequency transformer module includes a primary coil and multiple secondary coils, and the secondary coils are respectively connected to variable resistive switches at different levels.

[0099] An intelligent DC circuit breaker arc energy suppression method includes the following steps:

[0100] Arc discharge stage:

[0101] Detecting DC circuit breaker fault information through a sensor;

[0102] Starting the inverter of the energy storage system to provide power frequency current;

[0103] Based on the high-frequency signal injection technology, performing closed-loop control on the DC circuit breaker;

[0104] Arc feature recognition stage:

[0105] Collecting the arc voltage signal and the arc current signal;

[0106] Performing time-frequency analysis on the arc voltage signal and the arc current signal by using the fast Fourier algorithm;

[0107] Identifying the arc type based on the VMD-HHT algorithm;

[0108] Magnetic flux disturbance observation stage:

[0109] Real-time detecting the current in the circuit and sampling the magnetic flux;

[0110] Converting the magnetic flux into a voltage signal;

[0111] Invert the said voltage signal to obtain a magnetic flux disturbance value;

[0112] Calculate the phase difference between the observed signal and the desired signal;

[0113] Deep learning control stage:

[0114] Construct training samples to train a deep neural network;

[0115] Generate a frequency estimation model;

[0116] Train a deep reinforcement learning algorithm based on the said frequency estimation model;

[0117] Optimize the control strategy through a reward function;

[0118] Arc suppression stage:

[0119] Generate an IGBT control timing signal based on the said control strategy;

[0120] Control the switching frequency of the multi-level IGBT parallel sub-module;

[0121] Suppress the arc energy based on the PWM control method;

[0122] Current balancing stage:

[0123] Construct a dual heuristic intelligent algorithm based on an improved genetic algorithm and a particle swarm algorithm;

[0124] Adaptive allocate the drive current of each port;

[0125] Achieve IGBT module parallel balance;

[0126] Energy restoration stage:

[0127] When the arc is extinguished, control the coupling circuit to upload the data waveform to the dispatching center;

[0128] Receive the control instructions issued by the dispatching center;

[0129] Start the energy restoration process.

[0130] The present invention realizes active arc control through high-frequency signal injection, combines deep learning and a dual heuristic intelligent algorithm to accurately identify and suppress arc energy, and at the same time adopts a multi-port commutation topology and a magnetic flux disturbance observation technology to improve the system response speed and control accuracy.

[0131] The present invention has the following beneficial effects:

[0132] 1. Adopt the active arc extinguishing control technology of high-frequency signal injection, significantly improve the arc suppression efficiency, and shorten the arc extinguishing time from the traditional dozens of milliseconds to the sub-millisecond level;

[0133] 2. Arc feature recognition technology based on the VMD-HHT algorithm can accurately identify different types of arcs, adopt targeted differentiated suppression strategies, and significantly improve system adaptability;

[0134] 3. Integrating deep reinforcement learning and dual-heuristic intelligent algorithms to achieve adaptive optimization of arc suppression control and parallel balancing of IGBT modules, effectively improving system stability;

[0135] 4. The magnetic flux disturbance observation technology achieves millisecond-level response to arc transient characteristics, significantly improving system control accuracy and dynamic performance;

[0136] 5. The multi-port commutation topology provides multi-path current diversion channels. Combined with a multi-stage variable FPI filter and a variable high-frequency transformer, it effectively reduces the impact of arc energy on the system.

[0137] 6. Integrate 5G communication technology to achieve high-speed data interaction between the system and the dispatching center, and improve overall operational efficiency and intelligence level.

[0138] In summary, the present invention significantly improves the efficiency and reliability of arc suppression in DC circuit breakers, and provides strong technical support for the safe operation of DC distribution networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0139] Figure 1 It is a schematic diagram of the overall structure of the intelligent DC circuit breaker arc energy suppression system of the present invention;

[0140] Figure 2 It is a structural diagram of the high-frequency signal injection module of the present invention;

[0141] Figure 3 It is a structural diagram of the arc feature recognition module of the present invention;

[0142] Figure 4 It is a structural diagram of the magnetic flux disturbance observation module of the present invention;

[0143] Figure 5 It is a structural diagram of the deep learning control module of the present invention;

[0144] Figure 6 1 is a schematic structural diagram of a multi-port commutation control module of the present invention;

[0145] Figure 7 It is a structural diagram of the dual heuristic algorithm control module of the present invention;

[0146] Figure 8 It is a structural diagram of the communication interface module of the present invention;

[0147] Figure 9It is the frequency response characteristic diagram of the multi-stage variable FPI filter module of the present invention;

[0148] Figure 10 It is the impedance characteristic diagram of the variable high-frequency transformer module of the present invention;

[0149] Figure 11 It is the flow chart of the method for suppressing the arc energy of the intelligent DC circuit breaker of the present invention;

[0150] Figure 12 It is the processing flow chart of the arc discharge stage of the present invention;

[0151] Figure 13 It is the processing flow chart of the arc feature recognition stage of the present invention;

[0152] Figure 14 It is the processing flow chart of the deep learning control stage of the present invention;

[0153] Figure 15 It is the processing flow chart of the arc suppression stage of the present invention;

[0154] Figure 16 It is the processing flow chart of the current balancing stage of the present invention;

[0155] Figure 17 It is the processing flow chart of the power restoration stage of the present invention; Detailed implementation manners

[0156] Please refer to the appendix Figure 1-17 , the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0157] Referring to Figure 1 , the intelligent DC circuit breaker arc energy suppression system provided by the present invention includes a high-frequency signal injection module 1, an arc feature recognition module 2, a magnetic flux disturbance observation module 3, a deep learning control module 4, a multi-port commutation control module 5, a dual heuristic algorithm control module 6, and a communication interface module 7.

[0158] The high-frequency signal injection module 1 is used to generate a high-frequency carrier signal and inject it into the arc path of the DC circuit breaker. Preferably, the frequency range of the high-frequency carrier signal is from 10 kHz to 100 kHz, and this frequency range can effectively penetrate the arc channel and provide an appropriate energy carrier. The arc feature recognition module 2 is electrically connected to the high-frequency signal injection module 1, and is used to collect the arc voltage signal and the arc current signal, and identify the arc type based on these signals using the VMD-HHT algorithm to generate arc feature information. The magnetic flux disturbance observation module 3 is electrically connected to the arc feature recognition module 2, and is used to detect the current magnetic flux in the circuit in real time, convert it into a voltage signal, and calculate the magnetic flux disturbance value to generate the phase difference between the observation signal and the desired signal.

[0159] The deep learning control module 4 is electrically connected to the magnetic flux disturbance observation module 3 and the arc feature recognition module 2, and is used to receive the arc feature information and the phase difference, generate a control strategy through the deep reinforcement learning algorithm, and output the IGBT control timing signal. The multi-port commutation control module 5 is electrically connected to the deep learning control module 4, and is used to receive the IGBT control timing signal, control the switching frequency of the multi-stage IGBT parallel sub-module, and realize the efficient suppression of the arc energy based on the PWM control method.

[0160] The dual heuristic algorithm control module 6 is electrically connected to the multi-port commutation control module 5, and is used to construct a dual heuristic intelligent algorithm based on the combination of the improved genetic algorithm and the particle swarm algorithm, adaptively allocate the drive current of each port, and realize the parallel balance of the IGBT module. The communication interface module 7 is electrically connected to the deep learning control module 4 and the dual heuristic algorithm control module 6, and is used to receive the arc feature information and the control strategy, transmit them to the dispatching center, and receive the control instructions returned by the dispatching center and transfer them to each functional module.

[0161] In an embodiment of the present invention, as Figure 2 shown, the high-frequency signal injection module 1 includes a high-frequency signal generation unit 11, a signal modulation unit 12, a high-frequency transformer unit 13, and a signal injection switch unit 14. The high-frequency signal generation unit 11 is used to generate a high-frequency carrier signal with a specific frequency. Preferably, this unit adopts digital synthesis technology and can accurately adjust the frequency in the range of 10 kHz to 100 kHz. The signal modulation unit 12 is electrically connected to the high-frequency signal generation unit 11, and is used to modulate the high-frequency carrier signal with the arc signal of the DC circuit breaker to form a composite signal with a specific modulation depth. The high-frequency transformer unit 13 is electrically connected to the signal modulation unit 12, and is used to couple the modulated high-frequency carrier signal to the arc path of the DC circuit breaker. This transformer uses nanocrystalline core material and has low loss and high saturation magnetic flux density. The signal injection switch unit 14 is electrically connected to the high-frequency transformer unit 13, and is used to control the injection timing of the high-frequency carrier signal to ensure signal injection at an appropriate time.

[0162] As Figure 3 shown, the arc feature recognition module 2 includes a sensing and acquisition unit 21, a signal processing unit 22, an arc feature extraction unit 23, and an arc type recognition unit 24. The sensing and acquisition unit 21 is used to collect arc voltage signals and arc current signals in real time, and the sampling frequency is preferably 1 MHz to ensure capturing the high-frequency characteristics of the arc. The signal processing unit 22 is electrically connected to the sensing and acquisition unit 21 and is used to filter and amplify the arc voltage signals and arc current signals to eliminate background noise and improve the signal quality. The arc feature extraction unit 23 is electrically connected to the signal processing unit 22 and is used to perform time-frequency analysis on the processed arc voltage signals and arc current signals by using the fast Fourier algorithm to extract arc spectrum features. The arc type recognition unit 24 is electrically connected to the arc feature extraction unit 23 and is used to analyze the arc spectrum features based on the VMD-HHT algorithm to identify the arc as one of a continuous DC short arc, a DC small-gap arc, a DC long-gap arc, and a pure resistive or inductive arc.

[0163] The VMD-HHT algorithm (Variational Mode Decomposition-Hilbert Huang Transform algorithm) is an advanced signal processing method, especially suitable for arc feature analysis. This algorithm first decomposes the complex signal into a finite number of Intrinsic Mode Functions (IMFs) through variational mode decomposition, and then applies the Hilbert transform to each IMF to obtain instantaneous frequency and amplitude characteristics. The specific implementation steps are as follows:

[0164] The variational mode decomposition is based on the following variational problem:

[0165] ,

[0166] where: is the k-th mode, is the central frequency of the k-th mode, is the Dirac function, * represents the convolution operation, represents the partial derivative with respect to time, is the total number of modes, represents the square of the 2-norm, is the imaginary unit, is the time variable.

[0167] The algorithm solves this variational problem through the alternating direction method of multipliers to obtain K IMF components.

[0168] Apply the Hilbert transform to each IMF component:

[0169] ,

[0170] where: is The Hilbert transform, where P.V. represents the Cauchy principal value integral, is the integration variable.

[0171] Construct the analytic signal:

[0172] ,

[0173] where: is the instantaneous amplitude, is the instantaneous phase, is the imaginary unit.

[0174] Instantaneous frequency calculation:

[0175] ,

[0176] where: is the instantaneous frequency, is the derivative of the phase with respect to time.

[0177] Based on the instantaneous frequency and amplitude characteristics of each IMF component, construct the feature vector:

[0178] ,

[0179] where: is the average amplitude of the k-th IMF, is the main frequency of the k-th IMF, is the energy of the k-th IMF.

[0180] Basis for arc type judgment:

[0181] Continuous DC short arc: The high-frequency energy component in the feature vector accounts for a large proportion and decays rapidly;

[0182] DC small-gap arc: The intermediate-frequency energy component dominates;

[0183] DC long-gap arc: The low-frequency energy component dominates and decays slowly;

[0184] Pure resistive arc: The spectrum is evenly distributed and the high-frequency component has a long duration;

[0185] Inductive arc: The spectrum is concentrated and the amplitude decays slowly;

[0186] Such as Figure 4As shown in the figure, the magnetic flux disturbance observation module 3 includes a magnetic flux observer 31, an integrator 32, an inverter 33, and a phase comparator 34. The magnetic flux observer 31 is used to sample the magnetic flux by detecting the current in the circuit, preferably using a Hall effect sensor with a sensitivity of 0.1 mV / A. The integrator 32 is electrically connected to the magnetic flux observer 31 and is used to convert the magnetic flux into a voltage. A high-precision operational amplifier circuit is adopted, and the integration time constant can be adjusted in the range of 1 μs to 100 μs. The inverter 33 is electrically connected to the integrator 32 and is used to invert the voltage to obtain the magnetic flux disturbance value, with a gain accuracy better than ±0.1%. The phase comparator 34 is electrically connected to the inverter 33 and is used to calculate the phase difference between the observed signal and the desired signal, adopting digital phase detection technology with a phase resolution of 0.1°.

[0187] The magnetic flux disturbance observation system can be represented by the following mathematical model:

[0188] ,

[0189] Where: is the magnetic flux observation value, is the initial magnetic flux value, is the induced electromotive force, is the time-varying disturbance, is the environmental noise, is the time variable.

[0190] Convert the above formula into state space form:

[0191] ,

[0192] Where: is the combined effect of the time-varying disturbance and the environmental noise, is the system damping coefficient, is the natural frequency of the system.

[0193] Through Laplace transform, we can get:

[0194] ,

[0195] Where: is the Laplace transform of the system state variable, is the Laplace transform of the time-varying disturbance, is the Laplace transform of the environmental noise, is the complex frequency variable.

[0196] Phase difference calculation:

[0197] ,

[0198] Where: is the phase difference (unit: degree). is the frequency-domain representation of the observed signal. is the frequency-domain representation of the desired signal. denotes taking the imaginary part of a complex number. denotes taking the real part of a complex number. is the angular frequency (unit: rad / s).

[0199] As Figure 5 shown, the deep learning control module 4 includes a feature input unit 41, a neural network processing unit 42, and a reinforcement learning unit 43. The feature input unit 41 is used to receive arc feature information and phase difference, preprocess and normalize the data. The data sampling rate is 1 MHz, and the processed data accuracy is 16 bits. The neural network processing unit 42 is electrically connected to the feature input unit 41 and is used to construct training samples based on the arc feature information and phase difference, and train a deep neural network to generate a frequency estimation model. Preferably, the neural network adopts a five-layer structure, including an input layer, three hidden layers, and an output layer. The number of neurons in the hidden layers is 64, 128, and 64 respectively. The reinforcement learning unit 43 is electrically connected to the neural network processing unit 42 and is used to train a deep reinforcement learning algorithm based on the frequency estimation model and optimize the control strategy through a reward function.

[0200] The deep reinforcement learning control algorithm is implemented based on the Deep Q-Network (DQN). Its core is to approximate the Q-value function through a neural network. The algorithm process includes:

[0201] 1. State space definition:

[0202] ,

[0203] Wherein: is the state space, is the system state vector (including features such as arc voltage, arc current, phase difference, etc.), is the dimension of the state space.

[0204] 2. Action space definition:

[0205] ,

[0206] Wherein: is the action space, is the control action (including IGBT switching timing, PWM modulation depth, etc.), is the dimension of the action space.

[0207] 3. Reward function:

[0208] ,

[0209] The specific implementation is:

[0210] ,

[0211] Wherein: is the reward value, is the multi-path composite branch current (unit: A), is the output voltage of the high-frequency transformer (unit: V), is the loss of the high-frequency transformer (unit: W), , , are the weight coefficients (dimensionless), is the reference current value (unit: A), is the optimal output voltage value (unit: V). Preferably, , , .

[0212] 4. Q-value function update:

[0213] ,

[0214] Wherein: is the Q-value of taking action in state , is the immediate reward, is the discount factor (set to 0.95), is the learning rate, represents the maximum Q-value in the next state .

[0215] 5. Deep network training:

[0216] Use the mean squared error (MSE) as the loss function:

[0217] ,

[0218] Wherein: is the loss function value, represents the expectation operation, is the immediate reward, is the discount factor, is the Q-value of taking action in the state of the target network , is the Q-value of taking action in the state of the main network , are the parameters of the main network, are the parameters of the target network, which are updated every 100 iterations.

[0219] Such asFigure 6 As shown in the figure, the multi-port commutation control module 5 includes a multi-stage commutation structure unit 51, a PWM signal control unit 52, and an intelligent drive unit 53. The multi-stage commutation structure unit 51 is used to construct multiple commutation channels. Each stage of the commutation branch includes 2 parallel IGBTs and a series-connected thyristor. A fast recovery diode is connected in parallel between the IGBT and the thyristor. In this embodiment, a 4-stage commutation structure is adopted, and a total of 8 IGBTs and 4 thyristors are configured. The PWM signal control unit 52 is electrically connected to the multi-stage commutation structure unit 51, and is used to receive the IGBT control timing signal, generate a PWM control signal, and control the switching frequency of the IGBT. The digital PWM technology is adopted, and the frequency resolution reaches 0.1 kHz, and the modulation depth can be adjusted in the range of 0 to 100%. The intelligent drive unit 53 is electrically connected to the PWM signal control unit 52, and is used to realize the intelligent drive with variable conduction angle, dynamically adjust the conduction angle of the IGBT according to the system requirements, and change the commutation timing. Preferably, the conduction angle can be adjusted in the range of 30° to 150°, and the step accuracy is 0.5°.

[0220] As Figure 7 shown, the double heuristic algorithm control module 6 includes a genetic algorithm unit 61, a particle swarm algorithm unit 62, an algorithm fusion unit 63, and a current distribution unit 64. The genetic algorithm unit 61 is used to globally optimize the port current distribution scheme. The population size is set to 100, the crossover probability is 0.8, and the mutation probability is 0.05. The particle swarm algorithm unit 62 is used for local precise optimization. The number of particles is set to 50, the inertia weight is 0.7, the global learning factor is 1.5, and the local learning factor is 1.2. The algorithm fusion unit 63 is electrically connected to the genetic algorithm unit 61 and the particle swarm algorithm unit 62, and is used to combine the genetic algorithm and the particle swarm algorithm to construct a double heuristic intelligent algorithm, and dynamically adjust the optimization parameters according to the arc type and system state. The current distribution unit 64 is electrically connected to the algorithm fusion unit 63, and is used to distribute the drive current of each port based on the optimization result of the double heuristic intelligent algorithm to achieve the parallel balance of the IGBT modules.

[0221] The implementation process of the double heuristic intelligent algorithm is as follows:

[0222] 1. Initialize parameters:

[0223] Genetic algorithm parameters: population size , crossover probability , mutation probability ;

[0224] Particle swarm algorithm parameters: number of particles , inertia weight , learning factor ;

[0225] Algorithm fusion parameters: dynamic weight factor ;

[0226] 2. Genetic algorithm global search:

[0227] Coding: Real number coding is adopted, and each gene represents the driving current ratio of a port

[0228] Fitness function:

[0229] ,

[0230] Where: is the fitness function value of the genetic algorithm, is the current value of the i-th port (unit: A), is the average current value (unit: A), is the temperature of the i-th IGBT (unit: °C), is the average temperature (unit: °C), is the system loss (unit: W), 、 z 、 are weight coefficients (dimensionless), is the total number of ports.

[0231] Selection operation: Roulette wheel selection method is adopted;

[0232] Crossover operation: Arithmetic crossover is adopted;

[0233] Mutation operation: Non-uniform mutation is adopted;

[0234] Iterative optimization: Execute = 200 generations of evolution;

[0235] 3. Particle swarm algorithm local fine search:

[0236] The particle position represents the driving current ratio of each port

[0237] Particle velocity update:

[0238] ,

[0239] Where: is the velocity of the -th particle at the -th iteration, is the position of the -th particle at the -th iteration, is the individual optimal position of the -th particle, gbest is the global optimal position, is the inertia weight (dimensionless), 、 is the learning factor (dimensionless), , is a random number in the interval [0, 1].

[0240] Particle position update:

[0241] ,

[0242] where: is the position of the th particle at the th iteration, is the position of the th particle at the th iteration, is the velocity of the th particle at the th iteration.

[0243] Fitness function:

[0244] ,

[0245] where: is the fitness function value of the particle swarm algorithm, is the actual current value (unit: A) of the th port, is the target current value (unit: A), is the voltage value (unit: V) of the th port, is the target voltage value (unit: V), is the weight coefficient (dimensionless), is the total number of ports.

[0246] Iterative optimization: Execute = 100 iterations

[0247] 4. Algorithm fusion strategy:

[0248] Dynamic weight adjustment:

[0249] ,

[0250] where: is the dynamic weight factor of the current iteration, is the minimum value of the weight factor, is the maximum value of the weight factor, is the current iteration number, is the maximum iteration number.

[0251] Solution fusion:

[0252] ,

[0253] Wherein: is the fusion solution (representing the driving current ratio of each port), is the optimal solution of the genetic algorithm, is the optimal solution of the particle swarm optimization algorithm, is the dynamic weight factor.

[0254] 5. Current distribution implementation:

[0255] Calculate the driving current ratio of each port according to the fusion solution;

[0256] Map it to the duty cycle of the IGBT drive signal;

[0257] Achieve precise current distribution through PWM control.

[0258] As Figure 8 shown, the communication interface module 7 includes a 5G communication unit 71, a data processing unit 72, a security encryption unit 73, and a cloud interface unit 74. The 5G communication unit 71 is used to achieve high-speed data transmission, supporting communication delays in the millisecond level and transmission rates in the Gbps level. The data processing unit 72 is electrically connected to the 5G communication unit 71 and is used to encode the arc feature information and control strategy and decode the control instructions returned by the dispatching center. An efficient data compression algorithm is adopted, and the compression ratio can reach 10:1 while maintaining data integrity. The security encryption unit 73 is electrically connected to the data processing unit 72 and is used to encrypt and protect the transmitted data. The AES-256 encryption standard is adopted to ensure the security of data transmission. The cloud interface unit 74 is electrically connected to the security encryption unit 73 and is used to interact with the dispatching center, supporting multiple protocols such as REST API and MQTT, and flexibly adapting to the requirements of different dispatching systems.

[0259] In this embodiment, the system further includes a multi-stage variable FPI filter module 8 and a variable high-frequency transformer module 9. The multi-stage variable FPI filter module 8 is electrically connected to the high-frequency signal injection module 1 and is used to suppress the high-order harmonics in the voltage signal and separate the high-frequency signal from the power frequency signal during the occurrence of the arc. The filter adopts a Foster-π network structure, the variable inductance range is from 10 μH to 1 mH, the variable capacitance range is from 0.1 μF to 10 μF, and the resonance frequency can be adjusted in the range of 1 kHz to 50 kHz. The variable high-frequency transformer module 9 is electrically connected to the high-frequency signal injection module 1 and is used to absorb the impact of the arc energy on the system, showing inductive in the case of an arc and capacitive in the case of an AC signal. The transformer includes a primary coil and multiple secondary coils. The number of turns of the primary coil winding is 200, the total number of turns of the secondary coils is 600, which are distributed on 4 independent magnetic cores, and each secondary coil is respectively connected to variable resistive switches at different levels.

[0260] Reference Figure 11 The intelligent DC circuit breaker arc energy suppression method provided by the present invention includes the following steps:

[0261] Step S1: Arc discharge stage. In this stage, the system first detects the DC circuit breaker fault information through sensors. Preferably, high-sensitivity voltage sensors and current sensors are used, with a sampling frequency of 1 MHz and a response time of less than 1 μs to ensure that transient arc characteristics can be captured.

[0262] After detecting the fault, the system starts the inverter of the energy storage system to provide power frequency current. This step aims to maintain a stable power frequency current and reduce the voltage fluctuation on both sides of the circuit breaker. In this embodiment, the inverter output frequency is 50 Hz, and the power range is from 1 kW to 50 kW, which can be dynamically adjusted according to system requirements.

[0263] Subsequently, the system performs closed-loop control on the DC circuit breaker based on high-frequency signal injection technology. The frequency of the high-frequency signal is preferably in the range of 30 kHz to 50 kHz, and the signal amplitude is 5% to 10% of the DC voltage. This parameter setting can ensure that the signal penetrates the arc channel without causing excessive fluctuations in the system.

[0264] Step S2: Arc feature recognition stage. In this stage, the system first collects the arc voltage signal and the arc current signal. To ensure data quality, a 16-bit high-precision ADC is used, with a sampling rate of 1 MHz. The signal conditioning circuit includes a band-pass filter to eliminate interference.

[0265] Then, the system performs time-frequency analysis on the arc voltage signal and the arc current signal using the fast Fourier algorithm. The FFT algorithm uses a radix-2 butterfly operation, and the window function selects a Hanning window. The analysis frequency range is from 0 Hz to 500 kHz, and the frequency resolution is 10 Hz.

[0266] Finally, the arc type is identified based on the VMD-HHT algorithm. The number of modes K of the VMD algorithm is preferably set to 5, the penalty factor α is set to 2000, and the convergence tolerance ε is set to 1e-7. According to the characteristic parameters of different arc types, the system classifies the arcs into the following types:

[0267] Continuous DC short arc: High-frequency energy ratio > 60%, decay time constant < 5 ms

[0268] DC small-gap arc: Intermediate-frequency (5 kHz - 20 kHz) energy ratio > 50%

[0269] DC long-gap arc: Low-frequency (< 5 kHz) energy ratio > 70%, decay time constant > 20 ms

[0270] Resistive arc: Spectrum evenly distributed, high-frequency duration > 15 ms

[0271] Sensible arc: Spectrum concentrated, amplitude attenuation rate < 0.2 / ms

[0272] Step S3: Flux perturbation observation stage. In this stage, the system first samples the magnetic flux by detecting the current in the circuit in real time. A Hall effect sensor is used, with a range of 0 to 1000 A, linearity better than ±0.5%, and a bandwidth of 100 kHz.

[0273] Next, the system converts the magnetic flux into a voltage signal. A high-precision integration circuit is used, with an integration time constant of 10 μs and a temperature drift less than 50 ppm / °C to ensure stability at different operating temperatures.

[0274] Then, the system takes the inverse of the voltage signal to obtain the flux perturbation value. The gain of the inverting amplifier is precisely set to -1.000 ± 0.001, and the common-mode rejection ratio is greater than 80 dB to ensure signal integrity.

[0275] Finally, the system calculates the phase difference between the observed signal and the desired signal. The phase resolution of the digital phase detection circuit is 0.1°, the measurement range is ±180°, and the response time is less than 10 μs, meeting the requirements of real-time control.

[0276] Step S4: Deep learning control stage. In the deep learning control stage, the system first constructs training samples to train the deep neural network. The training samples include arc feature vectors under different working conditions and the corresponding control effects, with the number of samples not less than 10,000 groups, covering various common and extreme working conditions.

[0277] Next, the system generates a frequency estimation model. A five-layer neural network structure is used, the activation function is selected as ReLU, the optimization algorithm is Adam, the initial learning rate is 0.001, the attenuation rate is 0.95, and it decays once every 1000 steps. The number of training iterations is 50,000 times, the batch size is 128, and the proportion of the validation set is 20%.

[0278] Based on the frequency estimation model, the system trains a deep reinforcement learning algorithm. A double DQN structure is used, the size of the experience replay buffer is 10,000, the target network update frequency is 100 steps, the exploration strategy uses ε-greedy, the initial value of ε is 0.9, the minimum value is 0.1, and the attenuation rate is 0.995.

[0279] Finally, the system optimizes the control strategy through the reward function. The weight parameters of the reward function are determined through a large number of experiments: α = 0.6, β = 0.3, γ = 0.1, and this set of parameters achieves the best balance in terms of arc suppression efficiency and system stability.

[0280] Step S5: Arc suppression stage. In the arc suppression stage, the system generates IGBT control timing signals based on a control strategy. The time accuracy of the control timing is 0.1 μs, the pulse width can be adjusted in the range of 1 μs to 100 μs, and the rise time and fall time are both less than 0.5 μs.

[0281] Then, the system controls the switching frequency of the multi-stage IGBT parallel sub-module. The switching frequency of each stage of IGBT is in the range of 10 kHz to 50 kHz, and the frequency difference between different stages is maintained above 5 kHz to avoid resonance phenomena.

[0282] Finally, the system realizes the suppression of arc energy based on the PWM control method. The modulation depth of the PWM signal is dynamically adjusted in the range of 30% to 90%, the carrier frequency is 50 kHz, and the dead time is set to 2 μs to ensure that the upper and lower bridge arm IGBTs do not conduct simultaneously.

[0283] Step S6: Current balancing stage. In the current balancing stage, the system constructs a dual heuristic intelligent algorithm based on an improved genetic algorithm and a particle swarm algorithm. The population size of the genetic algorithm is 100, the number of iterations is 200, the crossover probability is 0.8, and the mutation probability is 0.05; the number of particles of the particle swarm algorithm is 50, the number of iterations is 100, the inertia weight is 0.7, and the learning factors c1 = 1.5 and c2 = 1.2.

[0284] Next, the system adaptively distributes the drive current of each port. The distribution ratio of the drive current is optimized according to the temperature, current-carrying capacity, and dynamic characteristics of each IGBT module to ensure that the current imbalance is less than 5%.

[0285] Finally, the system realizes the parallel balance of IGBT modules. By precisely controlling the turn-on timing of each IGBT, the conduction delay difference of each module is controlled within 50 ns, and combined with the dynamic current distribution strategy, current impact and parasitic oscillation are effectively avoided.

[0286] Step S7: Energy restoration stage. In the energy restoration stage, when the arc is extinguished, the system controls the coupling circuit to upload the data waveform to the dispatching center. The uploaded data includes arc characteristic parameters, suppression process records, and system status information. The data compression ratio is 10:1, and the transmission delay is less than 10 ms.

[0287] Then, the system receives the control instructions issued by the dispatching center. The control instructions go through a strict verification and decryption process to ensure the effectiveness and security of the instructions, and the response time is less than 5 ms.

[0288] Finally, the system starts the energy recovery process. The energy recovery adopts a soft start method, with the voltage rise rate controlled within the range of 10V / ms to 50V / ms and the current rise rate controlled within the range of 100A / ms to 500A / ms, to avoid new faults caused by instantaneous large current shocks.

[0289] Embodiment 3: Application effect of the system of the present invention in a DC distribution network

[0290] The intelligent DC circuit breaker arc energy suppression system of the present invention was tested in a DC distribution network with a rated voltage of 1000V and a rated current of 1000A. The test results show that compared with traditional arc suppression methods, the present invention has significant advantages:

[0291] 1. Arc extinction time: 15 - 30ms for the traditional method, 0.5 - 2ms for the present invention, an improvement of about 90%;

[0292] 2. Arc energy loss: 10 - 20kJ for the traditional method, 1 - 3kJ for the present invention, a reduction of about 85%;

[0293] 3. System voltage fluctuation: ±15% for the traditional method, ±5% for the present invention, with the stability improved by about 67%;

[0294] 4. Imbalance degree of IGBT module current distribution: 20% - 30% for the traditional method, 3% - 5% for the present invention, an improvement of about 85%;

[0295] 5. Ability to adapt to different types of arcs: The present invention can identify 5 different types of arcs and adopt the optimal suppression strategy accordingly.

[0296] The above test data fully demonstrate the significant advantages of the present invention in terms of arc suppression efficiency, system stability and reliability, providing important technical support for the safe operation of DC distribution networks.

[0297] The above is only the specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. Intelligent DC circuit breaker arc energy suppression system, characterized in that, Including: A high-frequency signal injection module, configured to: Generate a high-frequency carrier signal; Inject the high-frequency carrier signal into the arc path of the DC circuit breaker; An arc feature recognition module, electrically connected to the high-frequency signal injection module, configured to: Collect the arc voltage signal and the arc current signal; Based on the arc voltage signal and the arc current signal, identify the arc type using the VMD-HHT algorithm; Generate arc feature information; A magnetic flux disturbance observation module, electrically connected to the arc feature recognition module, configured to: Detect the current magnetic flux in the circuit in real time; Convert the current magnetic flux into a voltage signal; Based on the voltage signal, calculate the magnetic flux disturbance value; Generate the phase difference between the observation signal and the desired signal; A deep learning control module, electrically connected to the magnetic flux disturbance observation module and the arc feature recognition module, configured to: Receive the arc feature information and the phase difference; Based on the arc feature information and the phase difference, generate a control strategy through a deep reinforcement learning algorithm; Output the IGBT control timing signal; A multi-port commutation control module, electrically connected to the deep learning control module, configured to: Receive the IGBT control timing signal; Control the switching frequency of the multi-level IGBT parallel sub-module; Based on the PWM control method, achieve efficient suppression of arc energy; A dual heuristic algorithm control module, electrically connected to the multi-port commutation control module, configured to: Construct a dual heuristic intelligent algorithm based on the combination of an improved genetic algorithm and a particle swarm algorithm; Adaptive allocate the drive current of each port; Achieve parallel balance of the IGBT module; [[ID=2 ​ ​ ​ 2. The intelligent DC circuit breaker arc energy suppression system according to claim 1, wherein ​ ​ ​ ​ ​ 3. The intelligent DC circuit breaker arc energy suppression system according to claim 1, wherein ​ ​ ​ ​ ​ ​ The arc type recognition unit, electrically connected to the arc feature extraction unit, is configured to analyze the arc spectrum features based on the VMD-HHT algorithm and identify the arc as one of a continuous DC short arc, a DC small-gap arc, a DC long-gap arc, and a pure resistive or inductive arc.

4. The intelligent DC circuit breaker arc energy suppression system according to claim 1, characterized in that, The magnetic flux disturbance observation module includes: A magnetic flux observer for sampling the magnetic flux by detecting the current in the circuit; An integrator, electrically connected to the magnetic flux observer, for converting the magnetic flux into a voltage; An inverter, electrically connected to the integrator, for inverting the voltage to obtain a magnetic flux disturbance value; A phase comparator, electrically connected to the inverter, for calculating the phase difference between the observed signal and the desired signal.

5. The intelligent DC circuit breaker arc energy suppression system according to claim 1, characterized in that, The deep learning control module includes: A feature input unit for receiving the arc feature information and the phase difference; A neural network processing unit, electrically connected to the feature input unit, for: Constructing a training sample based on the arc feature information and the phase difference; Training a deep neural network to generate a frequency estimation model; A reinforcement learning unit, electrically connected to the neural network processing unit, for: Training a deep reinforcement learning algorithm based on the frequency estimation model; Optimizing the control strategy through a reward function; Among them, the reward function is as follows: , is the current of the multi-way composite branch, is the output voltage of the high-frequency transformer, is the loss of the high-frequency transformer.

6. The intelligent DC circuit breaker arc energy suppression system according to claim 1, wherein The multi-port commutation control module includes: A multi-stage commutation structure unit for: Constructing multiple commutation channels; Each stage of the commutation branch includes 2 parallel IGBTs and a series-connected thyristor; A fast recovery diode is connected in parallel between the IGBT and the thyristor; A PWM signal control unit, electrically connected to the multi-stage commutation structure unit, for: Receiving the IGBT control timing signal; Generating a PWM control signal; Controlling the switching frequency of the IGBT; An intelligent drive unit, electrically connected to the PWM signal control unit, for: Implementing intelligent drive with variable conduction angles; Dynamically adjusting the conduction angle of the IGBT according to system requirements; Changing the commutation timing.

7. The intelligent DC circuit breaker arc energy suppression system according to claim 1, characterized in that, The dual heuristic algorithm control module includes: A genetic algorithm unit for globally optimizing the port current distribution scheme; A particle swarm algorithm unit for local precise optimization; An algorithm fusion unit, electrically connected to the genetic algorithm unit and the particle swarm algorithm unit, for: Combining the genetic algorithm and the particle swarm algorithm to construct a dual heuristic intelligent algorithm; Dynamically adjusting the optimization parameters according to the arc type and system state; A current distribution unit, electrically connected to the algorithm fusion unit, for: Allocating the drive current of each port based on the optimization result of the dual heuristic intelligent algorithm; Achieving parallel balance of the IGBT modules.

8. The intelligent DC circuit breaker arc energy suppression system according to claim 1, wherein The communication interface module includes: A 5G communication unit for realizing high-speed data transmission; A data processing unit, electrically connected to the 5G communication unit, for: Encoding the arc feature information and the control strategy; Decoding the control instruction returned by the dispatching center; A security encryption unit, electrically connected to the data processing unit, for encrypting and protecting the transmitted data; A cloud interface unit, electrically connected to the security encryption unit, for data interaction with the dispatching center.

9. The intelligent DC circuit breaker arc energy suppression system according to claim 1, characterized in that The system further includes: A multi-stage variable FPI filter module, electrically connected to the high-frequency signal injection module, is used for: Suppressing high-order harmonics in the voltage signal; Separating the high-frequency signal from the power frequency signal during arc occurrence; A variable high-frequency transformer module, electrically connected to the high-frequency signal injection module, is used for: Absorbing the impact of arc energy on the system; Appearing inductive in the case of an arc and capacitive in the case of an AC signal; Wherein, the variable high-frequency transformer module includes a primary coil and multiple secondary coils, and the secondary coils are respectively connected to variable resistive switches at different levels.

10. Method for suppressing arc energy of intelligent DC circuit breaker, using the system described in any one of claims 1-9, characterized in that, Including the following steps: Arc discharge stage: Detecting DC breaker fault information through a sensor; Starting the inverter of the energy storage system to provide power frequency current; Based on high-frequency signal injection technology, performing closed-loop control on the DC breaker; Arc feature recognition stage: Collecting arc voltage signals and arc current signals; Performing time-frequency analysis on the arc voltage signals and the arc current signals using the fast Fourier algorithm; Identifying arc types based on the VMD-HHT algorithm; Magnetic flux disturbance observation stage: Real-time detecting the current in the circuit to sample the magnetic flux; Converting the magnetic flux into a voltage signal; Inverting the voltage signal to obtain a magnetic flux disturbance value; Calculating the phase difference between the observed signal and the desired signal; Deep learning control stage: Constructing training samples to train a deep neural network; Generating a frequency estimation model; Training a deep reinforcement learning algorithm based on the frequency estimation model; Optimizing the control strategy through a reward function; Arc suppression stage: Generating IGBT control timing signals based on the control strategy; Controlling the switching frequency of the multi-stage IGBT parallel sub-module; Realizing the suppression of arc energy based on the PWM control method; Current balancing stage: Constructing a dual heuristic intelligent algorithm based on the improved genetic algorithm and particle swarm algorithm; Adaptive allocating the drive current of each port; Realizing IGBT module parallel balancing; Power restoration stage: When the arc goes out, controlling the coupling circuit to upload the data waveform to the dispatching center; Receiving the control instructions issued by the dispatching center; Starting the power restoration process.

Images