An ac solid state circuit breaker adaptive turn-off method and system

Abstract

The application discloses an AC solid-state circuit breaker adaptive turn-off method and system, the core of the method is to construct a two-dimensional phase space geometric evolution manifold according to the current amplitude sequence of the fault current and the hardware differential rate signal, realize geometric extrapolation of the short-circuit current peak value containing the non-periodic component through the trajectory curvature characteristics; combined with the cumulative equivalent operation times, the radiator temperature and the standing time and other physical state parameters, input them into the endogenous physical information neural network model, define the safe operation interval with physical certainty by using the hard-coded embedded Cauer thermal network operator; and in the interval, an analytical projection algorithm based on stress cost manifold is adopted to execute constant level optimization of the optimal turn-off time by using the control barrier function. The application can effectively solve the distortion problem of the traditional model under the magnetic saturation working condition and the uncertainty of the soft constraint prediction, while ensuring the breaking reliability, significantly reducing the online calculation complexity.

Description

Technical Field

[0001] This invention relates to the field of intelligent electrical appliances and overcurrent protection technology for power systems, and more specifically to an adaptive turn-off method and system for AC solid-state circuit breakers. In particular, it relates to an intelligent control method that dynamically predicts and performs hard-constraint optimization decisions on the active turn-off timing of IGBTs by integrating real-time transient geometric features, equipment physical boundaries, and historical operational evolution data. Background Technology

[0002] Solid-state AC circuit breakers (SSDs) have demonstrated great potential in the protection and control of medium- and low-voltage distribution networks due to their advantages such as arc-free operation, fast breaking speed, and long lifespan. Their core principle is to utilize the rapid controllability of fully controllable semiconductor devices (such as IGBTs) to interrupt fault currents. However, AC system short-circuit currents contain both decaying aperiodic components and power frequency periodic components, resulting in complex waveforms, and the breaking process must withstand extremely high voltages. and This poses a serious challenge to the safe and reliable shutdown of IGBTs.

[0003] Existing AC solid-state circuit breaker turn-off control strategies are mostly based on fixed delay or current threshold criteria, failing to fully consider dynamic evolution factors such as device thermal state, drive characteristics, and chip aging. Premature turn-off may lead to interruption failure due to incomplete commutation process, while excessively late turn-off may cause irreversible thermal breakdown damage to the IGBT due to junction temperature exceeding safety limits.

[0004] To address this problem, existing research has attempted to introduce parameter identification or machine learning methods. However, the following fundamental technical shortcomings remain in practical applications:

[0005] 1. Physical distortion of prediction models: Existing methods mostly rely on the transient differential equations of the line (RL model) combined with the Euler difference approximation for parameter identification. However, in actual high-current faults, core magnetic saturation can cause a nonlinear drop in line inductance. Linear prediction models based on fixed parameter assumptions will produce huge prediction errors when waveform distortion occurs. 2. Uncertainty of safety constraints: Although some solutions employ physical information neural networks, they mostly introduce junction temperature boundary constraints through regularization terms in the loss function. This "soft constraint" mechanism essentially only "penalizes" the algorithm output when it deviates from the safe zone, and cannot mathematically guarantee that the instructions will absolutely not violate the physical limits of semiconductor devices. Under extreme operating conditions, there is a risk of unreliable output. 3. Real-time bottleneck of optimization algorithm: At the turn-off time optimization level, the weight coefficients of the evaluation function are mostly fixed values, which cannot be dynamically adjusted according to the severity of the fault and the real-time status of the device. It is difficult to achieve the optimal balance between the breaking speed and the device stress throughout the entire life cycle. Alternatively, traditional algorithm iterative optimization methods are used, but the computation time is highly correlated with the number of iterations and is subject to time uncertainty due to the initial search interval, which makes it difficult to meet the microsecond-level deterministic response requirements of power electronic systems.

[0006] There is an urgent need in this field for an intelligent shutdown decision-making method for AC solid-state circuit breakers that can accurately adapt to multi-dimensional changes in equipment status, and has high real-time performance and low computational complexity, so as to achieve a balance between disconnection reliability, semiconductor device safety and control engineering feasibility. Summary of the Invention

[0007] In view of the above problems, the present invention provides an adaptive shutdown method and system for AC solid-state circuit breakers, aiming to overcome the protection inaccuracies caused by the neglect of the evolution of the physical state of the equipment in the existing AC solid-state circuit breaker shutdown control strategy, as well as the defects of the existing intelligent optimization method in terms of calculation uncertainty and difficulty in real-time implementation at the microsecond level on embedded platforms.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide an adaptive shutdown method for an AC solid-state circuit breaker, comprising the following steps: Extract the current amplitude sequence and hardware differential rate of change signal of the fault current, map them to the phase space manifold, and predict the full-wave peak value of the short-circuit current based on the trajectory curvature characteristics in the phase space manifold. The multi-source feature vector, including the full-wave peak value, is input into a trained endogenous physical information neural network model. The Cauer sixth-order thermal network physical operator embedded in the model is used to transform the thermodynamic laws of the semiconductor device into hard topological constraints of the neurons, and output a safe operating time interval with physical determinism. Within the safe operation time interval, the optimal turn-off time is solved by analytical geometric projection based on a pre-constructed dynamic stress cost manifold. When the optimal shutdown time is reached, the actuator is triggered to shut down the AC solid-state circuit breaker.

[0009] Preferably, the full-wave peak value of the short-circuit current is estimated based on the trajectory curvature characteristics in the phase space manifold, including: Based on the direction of the motion vector and the second derivative of the current time point on the manifold trajectory, calculate the maximum envelope radius of the current trajectory on the coordinate axes; The instantaneous radius of curvature is obtained. When it is detected that the instantaneous radius of curvature decreases nonlinearly with the increase of current, it is identified as the magnetic saturation phenomenon of the iron core. At the same time, the geometric projection operator is called to extrapolate and correct the maximum envelope radius to obtain the full-wave peak value of the short-circuit current.

[0010] Preferably, the multi-source feature vector includes at least: the full-wave peak value, the heat sink temperature of the semiconductor device, the cumulative equivalent number of break-off operations, the settling time since the last operation, and the instantaneous curvature of the trajectory in the phase space manifold.

[0011] Preferably, a hardware-based physical boundary projection operator is embedded after the output layer of the endogenous physical information neural network model to map the junction temperature safety limit of the semiconductor device into a physically feasible region of safe operating time. The safe operation time interval is projected into the physical feasible region by forcibly truncating it.

[0012] Preferably, the safe operation time interval is: ,in, This is the predicted value for the current commutation readiness time. This is the predicted value for the safe flow window width.

[0013] Preferably, the dynamic stress cost manifold is a two-dimensional analytic manifold that includes a voltage stress cost dimension and a thermal stress cost dimension. The geometric curvature of the two-dimensional analytic manifold is determined by a dynamic weight tensor, which is a function that changes nonlinearly with the full-wave peak value and the heat sink temperature of the semiconductor device.

[0014] Preferably, the optimal turn-off time is solved by analytical geometric projection, which includes: defining the ideal control target as an ideal point on the two-dimensional analytical manifold, using the safety manifold defined by the control barrier function as a constraint condition, and taking the time corresponding to the projection point of the ideal point on the stress cost manifold as the optimal turn-off time.

[0015] Preferably, it further includes: Record the actual physical response after each break-off operation to generate a physical residual flow that characterizes the deviation between the model prediction and the actual value; The probability confidence of the physical residual flow belonging to structural aging shift is calculated by a pre-trained Bayesian confidence evaluator; when the structural aging shift characteristics are met, the physical parameters of the Cauer sixth-order thermal network physical operator are automatically fine-tuned.

[0016] In a second aspect, embodiments of the present invention provide an adaptive shutdown system for an AC solid-state circuit breaker, used to implement the adaptive shutdown method for an AC solid-state circuit breaker as described in any of the preceding claims, including: The main circuit of the AC solid-state circuit breaker includes a parallel thyristor main current-carrying branch and an IGBT turn-off branch, which is used to turn off the thyristor main current-carrying branch and turn on the IGBT turn-off branch when a short circuit occurs. A high-speed sensing and acquisition unit is used to extract the current amplitude sequence and hardware differential rate of change signal of the fault current, map them to the phase space manifold, and predict the full-wave peak value of the short-circuit current based on the trajectory curvature characteristics in the phase space manifold. An embedded intelligent decision-making unit is used to input multi-source feature vectors, including the full-wave peak value, into a trained endogenous physical information neural network model. Using the Cauer sixth-order thermal network physical operator embedded in the model, the thermodynamic laws of the semiconductor device are transformed into hard topological constraints of neurons, and the output is a physically deterministic safe operating time interval. Within the safe operating time interval, the optimal turn-off time is solved by analytical geometric projection based on a pre-constructed dynamic stress cost manifold. A high-precision drive unit is used to trigger the actuator to turn off the AC solid-state circuit breaker when the optimal turn-off time is reached.

[0017] Preferably, the system further includes a multi-source state sensing unit for sensing multi-source feature vectors.

[0018] This invention provides an adaptive shutdown method and system for AC solid-state circuit breakers. Addressing the shortcomings of existing technologies, it mainly offers an intelligent decision-making scheme that combines physical information-driven and geometric mapping, aiming to achieve deep "hard coupling" between the control algorithm and the underlying physical characteristics of power electronics. Under the premise of ensuring the absolute physical safety of semiconductor devices, it significantly improves the real-time performance of decision-making and the adaptive shutdown capability throughout the entire life cycle.

[0019] Compared with the prior art, the above-mentioned technical solution provided by the present invention has at least the following beneficial effects: 1) Enhanced physical consistency: Through phase space geometric mapping and hard coding of physical operators, high-fidelity perception of fault conditions and semiconductor thermal state is achieved, ensuring that the decision logic naturally conforms to the thermodynamic laws of power electronics and improving the reliability of the disconnection.

[0020] 2) Improved decision certainty: By using stress manifold analytical projection instead of iterative optimization, the computational complexity of decision-making is reduced to the constant level, eliminating the uncertainty of control delay and meeting the microsecond-level response requirements of high-power systems.

[0021] 3) Hardened safety boundaries: The hardware-based projection operator ensures that the system will never output instructions that violate the device's safe operating area under extreme unknown conditions, thus achieving "structure-level" safety defense.

[0022] 4) Full life cycle adaptive: Through Bayesian confidence perception and incremental learning, the system has the ability to automatically correct aging deviations as the number of years of operation increases, which significantly extends the service life of solid-state circuit breakers and reduces maintenance costs. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a flowchart of the fault current feature extraction and peak value prediction based on phase space geometric trajectory mapping in an embodiment of the present invention. Figure 2 This is a schematic diagram of a neural network architecture with hard-coded embedded Cauer sixth-order thermal network physical operators and output hard projection layers in an embodiment of the present invention. Figure 3 A schematic diagram of the overall architecture of the intelligent shutdown system for AC solid-state circuit breakers based on physical information neural operators and energy flow observation provided in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the interrupt handling process for software task scheduling and time complexity decision-making in the embedded intelligent decision-making unit of this invention. Figure 5 A schematic diagram of analytical projection of stress cost manifold and distribution of optimal decision points based on control barrier function provided in an embodiment of the present invention; Figure 6 This is a waveform comparison diagram of the current trajectory between the physical safety shutdown window definition and the optimal time of shutdown under the condition of magnetic saturation distortion, according to an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In response to the shortcomings of existing technologies: 1) Short-circuit current prediction is mostly based on linear parameter identification, which is difficult to cope with waveform distortion and aperiodic component offset caused by magnetic saturation under high current; 2) Machine learning models are mostly "black box" architectures, lacking hard constraints of physical laws, and there is a risk of output violating junction temperature limits under extreme operating conditions; 3) Optimization of turn-off time mostly adopts iterative search methods, which have large fluctuations in calculation time and are difficult to meet the requirements of power electronic protection for "deterministic response time". This invention discloses an adaptive turn-off method and system for AC solid-state circuit breakers, which aims to achieve deep "hard coupling" between control algorithms and the underlying physical characteristics of power electronics, and significantly improve the real-time performance of decision-making and the adaptive turn-off capability throughout the entire life cycle while ensuring the absolute physical safety of semiconductor devices.

[0027] Example 1 This embodiment provides an adaptive turn-off method for an AC solid-state circuit breaker. The AC solid-state circuit breaker includes a parallel thyristor main current-carrying branch and an IGBT turn-off branch. During normal operation, current flows through the low-conduction-loss main current-carrying branch (thyristor on, IGBT off). In the event of a short-circuit fault, after intelligent decision-making, the thyristor is first commanded to turn off (by applying a reverse pulse), and simultaneously the IGBT is triggered to turn on, completing current commutation. The method then operates at the predicted optimal time. The controller sends a precise signal to turn off the IGBT, forcing the current to be transferred to the energy absorption unit for dissipation, and finally disconnecting the fault current.

[0028] This application is intended to achieve optimal timing. The optimization process includes the following steps: S1. Extract the current amplitude sequence and hardware differential rate of change signal of the fault current, map them to the phase space manifold, and estimate the full-wave peak value of the short-circuit current based on the trajectory curvature characteristics in the phase space manifold. S2. Input the multi-source feature vector, including the full-wave peak value, into the trained endogenous physical information neural network model. Utilize the Cauer sixth-order thermal network physical operator embedded in the model to transform the thermodynamic laws of the semiconductor device into hard topological constraints of the neurons, and output a safe operating time interval with physical determinism. S3. Within the safe operation time interval, the optimal turn-off time is solved by analytical geometric projection based on the pre-constructed dynamic stress cost manifold. S4. When the optimal shutdown time is reached, the actuator is triggered to shut down the AC solid-state circuit breaker.

[0029] The following specific examples illustrate this.

[0030] In an optional embodiment, in step S1, an observation model based on phase space geometric trajectory mapping is constructed. This model is used to acquire real-time current amplitude sequences and their synchronized hardware differential rate of change signals through high-speed sampling. These sequences are then mapped onto a two-dimensional phase space manifold with current as the horizontal axis and current rate of change as the vertical axis. Utilizing the instantaneous curvature of the manifold's trajectory in phase space, the nonlinear evolution trend of the system caused by core magnetic saturation is identified without performing linear differential equation identification. Based on this, a geometric extrapolation is performed to obtain the predicted full-wave peak value of the short-circuit current, including the influence of aperiodic components. .

[0031] Traditional extrapolation methods often rely on expressions Fitting methods are prone to significant errors when parameters drift nonlinearly. This application uses a geometric projection method to achieve prediction based on the geometric closure constraint of the phase space trajectory. Figure 1 As shown, the specific steps are as follows: Based on the direction of the motion vector and the second derivative of the current time point on the manifold trajectory, calculate the maximum envelope radius of the current trajectory on the coordinate axes; When constructing an observation model based on phase space geometric trajectory mapping, the system will collect instantaneous current. Hardware Differentiated Signal Defined as phase space state vector The maximum envelope radius The computational logic originates from the geometric closure constraint of an ideal AC transient process in a two-dimensional coordinate system. Under ideal linear impedance conditions, the system current follows a sinusoidal evolution law. Its corresponding differential signal is Therefore, it can be deduced that the trajectory in phase space presents as... For long axis, A standard ellipse with a minor axis has the following analytical expression:

[0032] Therefore, the real-time calculation formula for the maximum envelope radius is derived as follows:

[0033] in ω is the angular frequency.

[0034] Instantaneous curvature is acquired. When a nonlinear decrease in instantaneous curvature with increasing current is detected, it is identified as core magnetic saturation. Simultaneously, a geometric projection operator is invoked to extrapolate and correct the maximum envelope radius to obtain the full-wave peak value of the short-circuit current. When handling magnetic saturation conditions, the projection operator... It is not a simple algebraic function, but a geometric extrapolation function with nonlinear compensation characteristics. When the system detects instantaneous curvature... When the value deviates from the linear reference value, the saturation flag is triggered. This is due to the inductance caused by the magnetic saturation of the iron core. The current decreases dynamically as it increases, and the originally smooth elliptical trajectory will shift towards... The axial direction produces nonlinear expansion, resulting in trajectory curvature. Distortion occurs. At this point, the projection operator... By calculating the geometric deviation between the current radius vector direction and the ideal envelope trajectory, and using Lagrange interpolation or a preset saturation sensitivity tensor, the initially calculated... Radial extrapolation correction is performed. Specifically, this operator maps the instantaneous increment of the trajectory eccentricity to the compensation amount of the current peak value. Thus, without solving complex transient differential equations, it can accurately predict the peak value of the entire wave, including the non-periodic component, based solely on the geometric slope characteristics of the fault initiation segment, thereby achieving geometric prediction of the fault current waveform.

[0035] This approach transforms the solution of complex transient differential equations into instantaneous geometric mappings, allowing the microcontroller to obtain the estimated peak value without performing time-consuming exponential function calculations. When processing waveforms containing severe aperiodic components (DC offset), the eccentricity of the geometric trajectory directly maps the decay rate of the DC component, thereby achieving "holographic" geometric prediction of the full-wave waveform of short-circuit current.

[0036] In one optional embodiment, S2 includes first constructing a multi-source feature vector. In this embodiment, the multi-source feature vector At least including: full peak value The heat sink temperature of semiconductor devices Cumulative equivalent number of segmentation operations Settling time since the last operation and the instantaneous curvature of the trajectory in the phase space manifold. ; indicates ; In this application, the instantaneous curvature of the trajectory The trend of impedance variation in a system used for real-time decoupling is expressed as follows:

[0037] In the formula, Represents current Regarding time The first derivative, in phase space, characterizes the state point along... The instantaneous velocity components of the shaft; Hardware differentiated signal This corresponds to the horizontal axis rate of change term in the formula; Represents current Regarding time The second derivative The instantaneous acceleration vector representing the trajectory in phase space is obtained by numerical differentiation after rapid differential or high-frequency sampling of the hardware differential signal by the embedded intelligent decision unit. Components along the axial direction; This represents the rate of change of current with respect to time. The first derivative is the instantaneous velocity component of the state point along the vertical axis.

[0038] By using the cross product of the aforementioned higher-order derivative terms and the modulus ratio, this formula can quantitatively characterize the curvature of the phase space trajectory in real time. Its numerator reflects the cross product modulus of the displacement and velocity vectors, while the denominator provides normalization of the kinetic energy reference, thus reflecting the instantaneous curvature. It becomes the core geometric characteristic parameter for decoupling the dynamic change trend of system impedance.

[0039] On the two-dimensional state plane, the instantaneous curvature of the trajectory The inductance is positively correlated with the dynamic ratio of the inductive and resistive components of the system. When a large current causes the iron core to enter the saturation region, the sudden drop in inductance will be directly reflected in the curvature. The distortion. Furthermore, this application will accumulate the number of operations. The aging attenuation operator, defined as a phase space manifold, can compensate for the contact resistance drift caused by contact wear in the peak prediction model, thereby... It becomes a dynamic state vector with a sense of physical evolution, rather than a simple static data set.

[0040] Then, the multi-source feature vectors are input into the endogenous physical information neural network model, such as... Figure 2 As shown, the input layer is The model input layer consists of two channels: a time-series feature channel and a state-series feature channel. The time-series feature channel contains three input nodes, which receive the real-time current sequence extracted by the sensor. Hardware Differentiated Signal and the calculated phase space trajectory curvature The status characteristic channel contains four input nodes, each receiving the estimated peak current. Cumulative number of operations Radiator temperature and settling time .

[0041] In the intermediate hidden layer processing flow, the timing channel first connects to a recurrent neural network layer (LSTM or GRU topology) containing 32 hidden units, and uses its internal gating mechanism and hidden state propagation characteristics to extract the dynamic features of the aperiodic component decay rate and slope of the transient current; the state channel simultaneously connects to a branch fully connected layer containing 16 neurons and using the ReLU activation function to achieve nonlinear compression and normalization of the device aging parameters and environmental boundaries.

[0042] Subsequently, the output features of the two branches undergo tensor fusion at the feature splicing node to generate a feature vector with 48-dimensional comprehensive information. This feature vector is then used as an excitation source and input to the physical operator layer of the Cauer sixth-order thermal network. This operator layer differs from traditional weighted neurons in that its internal transfer function is hard-coded as the analytical solution of the sixth-order RC equivalent thermal impedance differential equation of the IGBT chip junction-shell path. This transforms the thermodynamic laws of semiconductor devices into hard topological constraints of neurons, giving the intermediate features of the model a clear dynamic physical meaning of junction temperature. The physically constrained intermediate feature vector undergoes nonlinear decision mapping through a dense fully connected layer containing eight neurons. Finally, the final layer, containing two output nodes, outputs a preliminary predicted value, i.e., the current commutation readiness time. With the width of the safety flow window .

[0043] As a preferred embodiment of this invention, a hardware-based physical boundary projection operator is further embedded after the output layer of the endogenous physical information neural network model. This operator uses the analytical solution of the Cauer sixth-order transient thermal network equation of the IGBT as the "forced saturation region" of the algorithm to map the junction temperature safety limit of the device to the physically feasible region of the algorithm output space. By forcibly truncating the model output analytically, the correction value for the current commutation ready moment is obtained. Correction value for safety flow window width This allows for the definition of physically deterministic safe operating time intervals. .

[0044] Finally, a hardware-based physical boundary projection operator is connected in series to the model output. This operator uses the Lagrange multiplier method to perform projection truncation of the physical feasible region on the predicted value based on the junction temperature manifold calculated in real time by the Cauer operator, ensuring that the output meets the physical safety limits of power electronic devices.

[0045] Specifically, in calculating the width of the safe flow window At that time, the operator will adjust according to the measured heatsink temperature. Real-time solution of junction temperature manifold boundary. If the predicted value The logically derived junction temperature will exceed the junction temperature safety boundary. The projection operator uses the Lagrange multiplier method to force the output value to be "truncated" to the physical safety boundary. This limiting mechanism ensures that the algorithm output naturally conforms to the laws of electroelectronic thermodynamics in its logical structure, solving the problem of unpredictable and uninterpretable prediction results of machine learning models under extreme conditions.

[0046] In an alternative embodiment, S3 includes performing an analytical geometric projection decision based on the stress-cost manifold during the safe operation time interval.

[0047] In this embodiment, the dynamic stress cost manifold is a two-dimensional analytic manifold containing both voltage stress cost and thermal stress cost dimensions. The geometric curvature of the two-dimensional analytic manifold is determined by a dynamic weight tensor, which varies with the full-wave peak value. and the heat sink temperature of semiconductor devices A function that exhibits nonlinear variation, in the form of: ,in This is a preset stress sensitivity matrix.

[0048] When solving for the optimal turn-off time, the ideal control objective is defined as an ideal point on the two-dimensional analytic manifold, using the control barrier function. Using the defined safety manifold as a constraint, the time corresponding to the projection of the ideal point onto the stress-cost manifold is taken as the optimal turn-off time; this application can solve for the optimal turn-off time without iterative optimization. The analytical closed-form solution is obtained, and the above projection process only involves fixed-point matrix multiplication of the microcontroller, eliminating the iteration step fluctuations that may occur when the search algorithm processes non-convex evaluation functions, and achieving a decision time complexity of [missing information]. This approach ensures a high degree of temporal determinism in the triggering of the turn-off command, and achieves an adaptive trade-off between turn-off speed and device physical damage under high stress conditions.

[0049] Furthermore, in an optional embodiment, in S4, at the optimal turn-off time... At that time, the high-precision timer of the microcontroller performs timing matching, triggering the action command of the shutdown branch semiconductor device to complete the interruption of the fault current.

[0050] To further optimize the above technical solution, the adaptive shutdown method of this application also includes: recording the parameters of this shutdown operation, and constructing a Bayesian reliability estimator based on historical operation data to distinguish between random disturbances and structural aging. The specific process is as follows: Record the actual physical response (such as actual arc energy, actual junction temperature rise, etc.) after each breakup operation, and generate the physical residual flow that characterizes the deviation between the model prediction and the actual value. The probability reliability of the physical residual flow in the historical aging trend is calculated by a pre-trained Bayesian reliability evaluator. When it is determined that the structural aging shift characteristics are met (such as the continuous increase in thermal resistance due to bond wire fatigue), the parameter fine-tuning of the "shadow physical kernel" of the background asynchronous kernel is initiated. By automatically fine-tuning the physical parameters of the Cauer sixth-order thermal network physical operator, the prediction model can automatically adapt to the aging state of the semiconductor device according to the total runtime, thereby realizing the closed-loop evolution of the control strategy and the maintenance of accuracy within the life cycle.

[0051] Example 2 Based on the same inventive concept, this invention also provides an adaptive shutdown system for an AC solid-state circuit breaker, the overall structure of which is as follows: Figure 3 ,include: The main circuit of the AC solid-state circuit breaker consists of a parallel low-loss thyristor main current-carrying branch and an IGBT turn-off branch with a low-inductance busbar structure. This is used to turn off the thyristor main current-carrying branch and turn on the IGBT turn-off branch during a short circuit. A high-speed sensing and acquisition unit is used to extract the current amplitude sequence and hardware differential rate of change signal of the fault current. In some implementations, this unit includes a hardware differential channel composed of a high-frequency Rogowski coil and an active differentiating circuit for synchronously acquiring the current. With rate of change The original signal; The embedded intelligent decision-making unit is centered on a microcontroller integrating a hardware floating-point arithmetic unit and a dual-core processing architecture. The microcontroller's real-time kernel is pre-configured with a phase-space geometric observation module, used to map extracted information onto a phase-space manifold. Based on the trajectory curvature characteristics of the phase-space manifold, the full-wave peak value of the short-circuit current is estimated. A high-performance computing kernel is pre-configured with an endogenous physical information neural network model and an analytical projection decision-making module. This module inputs multi-source feature vectors, including the full-wave peak value, into the trained endogenous physical information neural network model. Utilizing the Cauer sixth-order thermal network physical operator embedded in the model, the thermodynamic laws of the semiconductor device are transformed into hard topological constraints of neurons, outputting a physically deterministic safe operating time interval. Within this safe operating time interval, the optimal turn-off time is solved using analytical geometric projection based on a pre-constructed dynamic stress-cost manifold. This embedded intelligent decision-making unit utilizes a dedicated computing kernel to independently run a phase-space-based analytical projection algorithm, eliminating the jitter impact of task scheduling on decision response time.

[0052] In some implementation schemes, the overall data processing process in the embedded intelligent decision-making unit refers to Figure 4 ,include: The main program's real-time kernel (Cortex-M4) continuously performs high-frequency sampling of the ADC, and once a decision is made... If the instantaneous value exceeds the preset safety threshold, the external interrupt service routine (ISR) will be triggered immediately.

[0053] In the ISR, the real-time kernel performs deterministic interrupt response operations: first, it locks the current time frame. and The coordinate points are used as the initial observation data. Then, the thyristor turn-off command is issued and the IGBT branch is simultaneously triggered to complete the current commutation. Finally, the event flag is set through shared memory and the high-performance computing core (Cortex-M7) is woken up.

[0054] Upon receiving the wake-up command, the high-performance kernel immediately invokes the preset phase space geometry observation module within a single clock cycle. Based on the instantaneous radius of curvature of the trajectory, it analyzes and calculates the maximum envelope radius, thereby quickly locking the full-wave peak prediction value of the fault current within 0.5ms.

[0055] Next, the kernel reads multi-source physical parameters from the FRAM and temperature sensor, inputting them into the PI-NODE prediction model, which is trained offline and deployed in a fixed-point manner. Unlike traditional optimization algorithms that require multiple iterations, this application directly solves the analytical closed-form solution of the control barrier function (CBF) in the current state through stress-cost manifold analytical projection decision. This calculation process only involves basic matrix multiplication and addition operations and fixed-point geometric projection mapping, greatly reducing the time fluctuations that may occur during online optimization. Finally, the calculated optimal turn-off time is written into the comparator counter of a high-precision timer. When the hardware timer reaches the set value, the hardware logic directly triggers a nanosecond-precision drive signal to drive the IGBT to perform an adaptive turn-off action.

[0056] In this embodiment, a multi-source state sensing unit is also provided. This unit integrates non-volatile memory (FRAM) and a high-precision real-time clock to maintain the physical meta-tag library, including accumulating the number of equivalent break-off operations. Radiator temperature and settling time ; High-precision drive execution unit, including having and A smart drive circuit with closed-loop correction capability is used to receive the... It instructs and drives the semiconductor device to operate.

[0057] Furthermore, taking a solid-state circuit breaker used in a 10kV medium-voltage AC distribution network as an example, its rated current is 1000A, its short-circuit breaking capacity is 50kA (peak value), and its maximum breaking time requirement is less than 5ms.

[0058] I. System Hardware: Designed to provide high deterministic data support and execution accuracy for millisecond-level transient processes, the specific structure is as follows: 1. AC Solid State Circuit Breaker Main Circuit: The main circuit consists of a low-loss main current-carrying branch and a semiconductor turn-off branch connected in parallel. The main current-carrying branch includes a pair of anti-parallel fast thyristor modules for carrying the full current during normal operation, which have extremely low on-state voltage drop to reduce operating losses. The semiconductor turn-off branch includes a series-parallel matrix composed of multiple high-power IGBT modules. To suppress overvoltages generated during turn-off transients, each IGBT module is connected in parallel with a resistor. and capacitor The circuit is an RC active buffer circuit, and a varistor (MOV) is configured as the final overvoltage absorption unit.

[0059] 2. High-speed sensing and synchronous acquisition unit: This embodiment employs a composite sensing architecture consisting of a high-frequency Rogowski coil and a wideband low-inductance shunt to capture the high-frequency dynamics of the main circuit current. The Rogowski coil is equipped with a high-frequency active differential-integral circuit, capable of directly outputting a hardware differential signal proportional to the first-order rate of change of the current. Its measurement bandwidth covers DC to 2MHz. The sensor signal is input to a multi-channel synchronous sampling ADC module to ensure the current amplitude... With rate of change Strict alignment on the time axis provides raw data with zero phase lag for subsequent phase space geometry modeling.

[0060] 3. Multi-source physical state sensing unit: This unit is responsible for collecting long-term time-varying parameters that affect the thermodynamic characteristics of the device, including the cumulative number of equivalent break-off operations. Stored in a non-volatile ferroelectric RAM (FRAM) external to the microcontroller, the system updates it in real time based on the breaking current energy after each breaking task to quantify the aging degree of the device. Heatsink temperature. The temperature is measured in real time by a high-precision digital temperature sensor mounted on the ceramic substrate of the IGBT module, reflecting the initial thermal boundary of the device. Settling time. The system utilizes the hardware real-time clock (RTC) module inside the main controller to calculate the time interval between the current fault and the previous circuit breaker operation or preheating, which is used to assist in correcting the transient thermal resistance model.

[0061] 4. Embedded Intelligent Decision Unit: Its core is a heterogeneous multi-core microprocessor. One high-performance computing core (such as a Cortex-M7) is dedicated to executing the endogenous physical information neural network model and analytical projection decision algorithm; another ultra-real-time core (such as a Cortex-M4 or dedicated DMA logic) is responsible for managing ADC sequence acquisition, initial fault identification, and high-speed communication. This architecture eliminates the risk of preemption between complex algorithms and basic monitoring tasks, ensuring a high degree of determinism in instruction generation response time.

[0062] 5. High-precision drive execution unit: including those with... and An intelligent IGBT drive circuit and isolated power supply with closed-loop regulation capability. The drive circuit receives a comparison output command from the microcontroller's high-precision timer (HRTIM). Upon fault detection, the system first controls the thyristor branch to turn off (applying a reverse turn-off pulse) and simultaneously triggers the IGBT branch to turn on to complete commutation. Finally, at the optimal moment calculated analytically... The controller sends a drive signal with nanosecond precision to turn off the IGBT via HRTIM, forcibly cutting off the current.

[0063] II. Predictive Model Training and Embedded Deployment The offline training process for the prediction model includes: deeply optimizing the neural topology embedded with physical operators using a dataset generated from a high-fidelity digital twin model. First, a high-fidelity digital twin model is built in the simulation software PLECS, incorporating electromagnetic transients, IGBT dynamic loss characteristics, a Cauer-type sixth-order transient thermal network, parasitic parameters of the drive circuit, and stray inductance of the busbar. Batch simulations are performed by executing parametric scans, where the scan range of key variables is set as follows: equivalent resistance of the line. The equivalent inductance varies between 0.1 and 10 mΩ. The temperature range is 10-500 μH to accurately simulate the evolution of system impedance at different fault locations in the distribution network; simultaneously, the fault closing angle is set to 0-360°, and the initial temperature of the radiator is... The temperature range is 40-100°C, and the cumulative number of equivalent dissection operations is [number missing]. The range is 0-10000 times, and the rest time since the last action is recorded. The duration is 0.1-100 hours. Unlike traditional methods that only record the current amplitude sequence, this embodiment simultaneously records the phase space characteristics, i.e., discrete current. Hardware Differentiated Signal The coordinate sequence provides the original dimension for the model to perceive the trajectory distortion caused by the magnetic saturation of the iron core.

[0064] The model training phase employs an endogenous physical information neural network model architecture, which features a dual-path concurrent feature extraction mechanism. The first path, the temporal feature channel, is responsible for sensing and identifying the slope and DC component attenuation rate of the transient waveform; the second path, the state feature channel, introduces the peak value estimate initially determined by phase space geometric mapping. And multiple physical parameters. The essential difference between this model and known techniques lies in the "structured constraint" design of the output layer: the analytical expression of the Cauer thermal network differential equation is used as the hidden layer transfer function of the neural network, so that the internal state evolution conforms to the laws of thermodynamics. Then, a physical boundary projection operator is constructed to calculate the training loss. This operator defines the device junction temperature safety boundary. The analytical mapping is used to define the physical envelope of the current carrying time and peak current. If the model predictions tend to exceed the limits, the projection operator uses the Lagrange multiplier method to force them to be truncated to the physical safety boundary. This "hard constraint" architecture ensures that even under extreme stress conditions not covered by the training samples, the algorithm logic cannot output illegal instructions that would cause thermal breakdown of the device. After offline training, the system converts the model into a C language kernel for fixed-point arithmetic for embedded deployment. To optimize the microcontroller's execution efficiency, a pre-computed lookup table (LUT) method is used to store the nonlinear physical boundary functions, enabling the Cortex-M7 core with integrated hardware floating-point arithmetic units to... It completes a single reasoning decision within the system, meeting the real-time requirement for rapid segmentation in communication systems.

[0065] III. Implementation of Control Software Process During the fault detection and interrupt response phase, the system main program uses the ADC module to perform high-frequency continuous sampling of the line current. Once the rate of change of current is determined... If the preset safety threshold is exceeded, the microcontroller immediately triggers an external interrupt service routine and simultaneously sends a shutdown signal to the main current-carrying branch. Upon response, the interrupt routine immediately locks the current in the two-dimensional state plane. With rate of change Coordinate points. Peak short-circuit current. The geometric extrapolation prediction is the key to this invention's replacement of traditional linear identification. This algorithm calculates the instantaneous radius of curvature of the current point in phase space. This is used to characterize transient energy. According to the principle of energy conservation, the transient trajectory of an ideal lossless system in phase space is represented as an ellipse with its major axis radius corresponding to the current peak value. The system uses hardware differential signals to capture the rate of change of the trajectory radius in real time. When curvature is detected... When the current decreases nonlinearly as it increases, the system automatically identifies this as a magnetic saturation phenomenon in the iron core and calls the analytical geometric projection operator to correct the extrapolation value of the major axis of the ellipse in real time.

[0066] This geometric mapping method can complete the full-wave peak value mapping within 0.5ms after a fault occurs, using only a very small portion of the riser waveform information. The prediction accuracy is significantly better than that of linear extrapolation models that rely on the fixed impedance assumption. In the safety window prediction and analytical decision-making stage, the eigenvector... It is fed into the deployed PI-NODE model to obtain the safe operating range constrained by physical hard projection. .

[0067] This embodiment does not choose the traditional optimization search method, but instead adopts an analytical projection decision based on the stress-cost manifold: the system projects the ideal zero-stress shutdown target onto the current stress-cost manifold surface. This projection point is obtained instantaneously by solving the analytical closed-form solution of the control barrier function (CBF). This analytical decision mechanism eliminates the time uncertainty in the online optimization process, ensuring the absolute timeliness of shutdown command issuance in high-power protection tasks.

[0068] After shutdown, the physical residuals generated by recording the actual arcing time and estimating the junction temperature are used to identify trend deviations such as thermal resistance drift caused by device aging using a Bayesian reliability evaluator. By fine-tuning the physical core parameters, the system achieves adaptive evolution of the control strategy as the service life increases.

[0069] IV. Simulation Verification and Result Analysis To verify the effectiveness of the method of this invention, a detailed model of an AC solid-state circuit breaker, including its nonlinear characteristics, was built in the MATLAB / Simulink and Simplier co-simulation environment. Regarding stress manifold analytical decision analysis, Figure 5 This demonstrates the normalized sub-objective functions and the overall cost function. The evolution process depending on the turn-off time selection. Unlike traditional algorithms that blindly search in non-convex space, the analytical projection method of this invention directly locks the global optimum with the lowest overall cost (located at 35ms). The results clearly show that when the heat sink temperature... When the weighting factor is dynamically adjusted upwards due to the increase in stress manifold, the decision point automatically shifts to a region with a smaller rate of change in current, verifying the active protection logic of the physical driving algorithm for device safety. Regarding the breaking waveform analysis under magnetic saturation conditions... Figure 6 The entire fault disconnection process, including severe aperiodic components and magnetic saturation distortion, is fully demonstrated. At 5ms of the fault occurrence, even though the current rise waveform deviates from the sinusoidal characteristics due to core saturation, the geometric extrapolation algorithm based on the phase space trajectory curvature accurately locks the predicted peak value, avoiding the risk of overcurrent protection failure that might occur with linear prediction. Within the predicted safe turn-off window (blue area in the figure), the system triggers IGBT operation at the analytically optimal time of 35ms. At this time, the semiconductor branch current is forcibly cut off in a very short time, and the overvoltage peak and device junction temperature rise at the moment of turn-off are strictly limited within the physical safety manifold defined by the hard projection operator. Simulation comparisons show that, compared with traditional fixed-time turn-off, this invention not only improves the reliability of the disconnection but also significantly reduces the device temperature rise under high-stress conditions, fully verifying the effectiveness of the adaptive control strategy in complex AC transient processes.

[0070] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0071] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An adaptive turn-off method for an AC solid-state circuit breaker, characterized in that, include: Extract the current amplitude sequence and hardware differential rate of change signal of the fault current, map them to the phase space manifold, and predict the full-wave peak value of the short-circuit current based on the trajectory curvature characteristics in the phase space manifold. The multi-source feature vector, including the full-wave peak value, is input into a trained endogenous physical information neural network model. The Cauer sixth-order thermal network physical operator embedded in the model is used to transform the thermodynamic laws of the semiconductor device into hard topological constraints of the neurons, and output a safe operating time interval with physical determinism. Within the safe operation time interval, the optimal turn-off time is solved by analytical geometric projection based on a pre-constructed dynamic stress cost manifold. When the optimal shutdown time is reached, the actuator is triggered to shut down the AC solid-state circuit breaker.

2. The method as described in claim 1, characterized in that, Based on the trajectory curvature characteristics in the phase space manifold, the full-wave peak value of the short-circuit current is estimated, including: Based on the direction of the motion vector and the second derivative of the current trajectory on the manifold trajectory, calculate the maximum envelope radius of the current trajectory on the coordinate axes; The instantaneous radius of curvature is obtained. When it is detected that the instantaneous radius of curvature decreases nonlinearly with the increase of current, it is identified as the magnetic saturation phenomenon of the iron core. At the same time, the geometric projection operator is called to extrapolate and correct the maximum envelope radius to obtain the full-wave peak value of the short-circuit current.

3. The method as described in claim 1, characterized in that, The multi-source feature vector includes at least: the full-wave peak value, the heat sink temperature of the semiconductor device, the cumulative equivalent number of break-off operations, the settling time since the last operation, and the instantaneous curvature of the trajectory in the phase space manifold.

4. The method as described in claim 1, characterized in that, The output layer of the endogenous physical information neural network model is followed by a hardware-based physical boundary projection operator, which is used to map the junction temperature safety limit of the semiconductor device to the physical feasible region of safe operating time. The safe operation time interval is projected into the physical feasible region by forcibly truncating it.

5. The method as described in claim 1 or 4, characterized in that, The safe operation time interval is: ,in, This is the predicted value for the current commutation readiness time. This is the predicted value for the safe flow window width.

6. The method as described in claim 1, characterized in that, The dynamic stress cost manifold is a two-dimensional analytic manifold that includes voltage stress cost dimension and thermal stress cost dimension. The geometric curvature of the two-dimensional analytic manifold is determined by a dynamic weight tensor, which is a function that changes nonlinearly with the full-wave peak value and the heat sink temperature of the semiconductor device.

7. The method as described in claim 1, characterized in that, The optimal turn-off time is determined by analytical geometric projection, which includes: defining the ideal control target as an ideal point on the two-dimensional analytical manifold, using the safety manifold defined by the control barrier function as a constraint, and taking the time corresponding to the projection point of the ideal point on the stress cost manifold as the optimal turn-off time.

8. The method as described in claim 1, characterized in that, Also includes: Record the actual physical response after each break-off operation to generate a physical residual flow that characterizes the deviation between the model prediction and the actual value; The probability confidence that the physical residual flow belongs to the structural aging shift is calculated using a pre-trained Bayesian confidence evaluator. When the structural aging offset characteristics are met, the physical parameters of the Cauer sixth-order thermal network physical operator are automatically fine-tuned.

9. An adaptive shutdown system for an AC solid-state circuit breaker, characterized in that, The method for implementing the adaptive shutdown method of the AC solid-state circuit breaker according to any one of claims 1-8 includes: The main circuit of the AC solid-state circuit breaker includes a parallel thyristor main current-carrying branch and an IGBT turn-off branch, which is used to turn off the thyristor main current-carrying branch and turn on the IGBT turn-off branch when a short circuit occurs. A high-speed sensing and acquisition unit is used to extract the current amplitude sequence and hardware differential rate of change signal of the fault current, map them to the phase space manifold, and predict the full-wave peak value of the short-circuit current based on the trajectory curvature characteristics in the phase space manifold. An embedded intelligent decision-making unit is used to input multi-source feature vectors, including the full-wave peak value, into a trained endogenous physical information neural network model. Using the Cauer sixth-order thermal network physical operator embedded in the model, the thermodynamic laws of the semiconductor device are transformed into hard topological constraints of neurons, and the output is a physically deterministic safe operating time interval. Within the safe operating time interval, the optimal turn-off time is solved by analytical geometric projection based on a pre-constructed dynamic stress cost manifold. A high-precision drive unit is used to trigger the actuator to turn off the AC solid-state circuit breaker when the optimal turn-off time is reached.

10. The AC solid-state circuit breaker adaptive shutdown system as described in claim 9, characterized in that, It also includes a multi-source state sensing unit, used to sense multi-source feature vectors.

Images