Series thyristor assembly dynamic voltage sharing adaptive adjustment method and system

Abstract

The application relates to the technical field of thyristor voltage-sharing control, and particularly discloses a series thyristor assembly dynamic voltage-sharing adaptive adjustment method and system, which comprises the following steps: constructing an extreme transient scene generation model based on power grid topology parameters and historical fault data, and generating a transient scene set; performing electromagnetic transient simulation on each transient scene in the transient scene set, and extracting a scene core feature vector; calculating the optimal emergency control strategy of each flexible alternating current power transmission system valve group under each transient scene through a double-target collaborative optimization model, and establishing a scene-strategy mapping library. The technical scheme provided by the application pre-constructs the transient scene set and the strategy mapping library offline, pre-processes the time-consuming electromagnetic transient simulation and strategy optimization calculation process, and only needs to directly match and call the corresponding optimal emergency control strategy according to the real-time collected transient characteristics in actual power grid operation, thereby avoiding the processing delay problem caused by online calculation.

Description

Technical Field

[0001] This invention relates to the field of thyristor voltage equalization control technology, and in particular to a dynamic voltage equalization adaptive adjustment method and system for series thyristor assemblies. Background Technology

[0002] Series-connected thyristor assemblies are widely used in flexible AC transmission systems to meet the demands of high-voltage, high-capacity power transmission. In actual operation, to prevent the series-connected thyristors from experiencing high reverse voltages due to inconsistent dynamic characteristics, existing technologies typically employ passive RC snubber circuits or conventional calculation algorithms relying on local controllers to achieve voltage distribution balance among the thyristors. Under normal operating conditions where the power grid is in a steady state or has a low transient voltage change rate, these methods can maintain the voltage distribution of the series-connected thyristor assemblies.

[0003] However, when the power grid experiences rapid transient conditions, the transient voltage rise time is short and accompanied by high-frequency oscillations. Existing voltage balancing control methods rely on online simulation and real-time parameter solving, which, limited by the hardware computing power of the controller, makes it difficult to generate and deploy response strategies within the time window of the transient occurrence.

[0004] Meanwhile, the existing distributed control architecture lacks a global time synchronization constraint mechanism between valve groups, resulting in unpredictable physical timing misalignment between the local controller receiving commands and the actual execution of gate trigger actions. The computational delay and execution timing differences make it difficult for the series thyristor components to achieve synchronized voltage equalization regulation responses during transient events, causing some thyristors to withstand transient overvoltages exceeding design margins, posing a potential hardware damage risk. Summary of the Invention

[0005] This invention aims to at least partially solve one of the technical problems in related technologies. Therefore, the objective of this invention is to propose a dynamic voltage equalization adaptive adjustment method and system for a series thyristor assembly to ensure the stability of the series thyristor during the response process.

[0006] To achieve the above objectives, a first aspect of the present invention provides a dynamic voltage equalization adaptive adjustment method for a series thyristor assembly, comprising:

[0007] An extreme transient scenario generation model is constructed based on power grid topology parameters and historical fault data to generate a transient scenario set; electromagnetic transient simulation is performed on each transient scenario in the transient scenario set to extract the core feature vector of the scenario; the optimal emergency control strategy of each flexible AC transmission system valve group under each transient scenario is calculated through a dual-objective collaborative optimization model to establish a mapping library between scenarios and strategies.

[0008] Real-time acquisition of bus voltage and operating parameters of valve groups in each flexible AC transmission system; acquisition of transient voltage rise time; and automatic switching of control mode based on the transient voltage rise time.

[0009] When the transient voltage rise time is less than the first time threshold, switch to the first control mode, extract the real-time transient core feature vector through the feature extraction algorithm; calculate the scene similarity between the real-time transient core feature vector and the scene core feature vector of each scene, and call the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold.

[0010] A global synchronization timestamp is attached to the invoked optimal emergency control strategy and sent to the local controller of the corresponding flexible AC transmission system valve group;

[0011] The local controller, based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, synchronously starts the controllable negative pressure source module at the determined absolute trigger time, and performs gate pulse reconstruction based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, thereby realizing dynamic voltage equalization regulation of the series thyristors.

[0012] The operating status of each thyristor is monitored in real time, and the feedback signal is input into the electrical stress prediction model for online correction of model parameters. The mapping library is updated based on the correction results.

[0013] To achieve the above objectives, a second aspect of the present invention provides a dynamic voltage equalization adaptive adjustment system for a series thyristor assembly, comprising:

[0014] The mapping library establishment module is used to construct an extreme transient scenario generation model based on power grid topology parameters and historical fault data, and generate a transient scenario set; perform electromagnetic transient simulation on each transient scenario in the transient scenario set, and extract the core feature vector of the scenario; calculate the optimal emergency control strategy of each flexible AC transmission system valve group under each transient scenario through a dual-objective collaborative optimization model, and establish a mapping library between scenarios and strategies.

[0015] The parameter acquisition and mode switching module is used to acquire the bus voltage and the operating parameters of each flexible AC transmission system valve group in real time, obtain the transient voltage rise time, and automatically switch the control mode according to the transient voltage rise time.

[0016] The feature matching and strategy invocation module is used to switch to the first control mode when the transient voltage rise time is less than the first time threshold, extract the real-time transient core feature vector through the feature extraction algorithm, calculate the scene similarity between the real-time transient core feature vector and the scene core feature vector of each scene, and invoke the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold.

[0017] The timestamp appending and distribution module is used to append a globally synchronized timestamp to the invoked optimal emergency control strategy and distribute it to the local controller of the corresponding flexible AC transmission system valve group.

[0018] The local execution and regulation module, configured in the local controller, is used to synchronously start the controllable negative pressure source module at the determined absolute trigger time according to the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, and to reconstruct the gate pulse based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, so as to realize the dynamic equalization regulation of the series thyristors.

[0019] The monitoring and correction update module is used to monitor the operating status of each thyristor in real time, input the feedback signal into the electrical stress prediction model to correct the model parameters online, and update the mapping library according to the correction results.

[0020] To achieve the above objectives, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, it implements the above-described dynamic voltage equalization adaptive adjustment method for series thyristor assemblies.

[0021] The technical solution provided by this invention pre-constructs a transient scenario set and strategy mapping library offline, pre-processing the time-consuming electromagnetic transient simulation and strategy optimization calculation process. In actual power grid operation, it is only necessary to directly match and call the corresponding optimal emergency control strategy based on the transient characteristics collected in real time, avoiding the processing delay problem caused by online calculation and meeting the control timeliness requirements under transient conditions.

[0022] Meanwhile, by adding a global synchronization timestamp to the control strategy for matching calls, and combining the timing offset with the gate pulse delay compensation matrix, this scheme can control different local controllers to synchronously start the negative pressure source module for reconfiguration at a determined absolute trigger time. This eliminates the timing misalignment phenomenon in the distributed hardware during instruction execution, ensures the consistency of multiple valve groups and each series thyristor in the transient response process, and prevents the occurrence of local thyristor transient overvoltage. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the dynamic voltage equalization adaptive adjustment method for series thyristor assemblies provided by the present invention.

[0024] Figure 2 This is a voltage waveform and characteristic parameter extraction and calibration diagram of the dynamic voltage equalization adaptive adjustment method for series thyristor components provided by the present invention in extreme transient scenarios;

[0025] Figure 3This is a Pareto front distribution diagram of the dynamic voltage equalization coefficient and the secondary overvoltage suppression dual-objective collaborative optimization in the dynamic voltage equalization adaptive adjustment method for series thyristor components provided by the present invention.

[0026] Figure 4 This is a wavelet transform high- and low-frequency scale energy distribution bitmap of the extremely fast transient voltage signal in the dynamic voltage equalization adaptive adjustment method for series thyristor components provided by this invention;

[0027] Figure 5 This is a nonlinear response surface mapping diagram from the feature dimension comprehensive difference index to scene similarity in the dynamic voltage equalization adaptive adjustment method of the series thyristor assembly provided by the present invention.

[0028] Figure 6 This is a comparison diagram of the voltage distribution at the terminals of the series thyristors before and after gate pulse delay reconstruction control in the dynamic voltage equalization adaptive adjustment method for series thyristor assemblies provided by this invention.

[0029] Figure 7 This is a schematic diagram of the safety generalization boundary based on multi-source information fusion and margin amplification in the dynamic voltage equalization adaptive adjustment method for series thyristor components provided by the present invention under unknown transient scenarios;

[0030] Figure 8 This is a time delay estimation convergence curve of the Kalman filter tracking model under strong electromagnetic interference environment in the dynamic voltage equalization adaptive adjustment method of series thyristor assembly provided by the present invention.

[0031] Figure 9 This is a schematic diagram illustrating the implementation of the dynamic voltage equalization and adaptive adjustment system for series thyristor assemblies provided by the present invention.

[0032] Figure 10 This is a schematic diagram of the electronic device provided by the present invention. Detailed Implementation

[0033] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0034] The following description, with reference to the accompanying drawings, describes a dynamic voltage equalization adaptive adjustment system, method, and electronic device for a series thyristor assembly according to embodiments of the present invention.

[0035] Example 1:

[0036] like Figure 1As shown, this embodiment provides a dynamic voltage equalization adaptive adjustment method and system for series thyristor assemblies, which is mainly applied in high-voltage direct current transmission systems or multiple flexible alternating current transmission systems (FACTS) containing a large number of high-power power electronic devices.

[0037] Specifically, the dynamic voltage equalization adaptive adjustment method for series thyristor components in this embodiment includes: constructing an extreme transient scenario generation model based on power grid topology parameters and historical fault data to generate a transient scenario set; performing electromagnetic transient simulation on each transient scenario in the transient scenario set to extract the core feature vector of the scenario; calculating the optimal emergency control strategy for each flexible AC transmission system valve group under each transient scenario through a dual-objective collaborative optimization model, and establishing a mapping library between scenarios and strategies;

[0038] The system collects the bus voltage and operating parameters of each flexible AC transmission system valve group in real time, obtains the transient voltage rise time, and automatically switches the control mode according to the transient voltage rise time. When the transient voltage rise time is less than the first time threshold, it switches to the first control mode and extracts the real-time transient core feature vector through the feature extraction algorithm.

[0039] Calculate the scene similarity between the real-time transient core feature vector and the scene core feature vectors of each scene, and call the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold; add a global synchronization timestamp to the called optimal emergency control strategy and send it to the local controller of the corresponding flexible AC transmission system valve group; the local controller synchronously starts the controllable negative pressure source module at the determined absolute trigger time according to the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, and performs gate pulse reconstruction based on the gate pulse delay compensation matrix included in the optimal emergency control strategy to realize dynamic voltage equalization regulation of the series thyristors;

[0040] The operating status of each thyristor is monitored in real time, and the feedback signal is input into the electrical stress prediction model for online correction of model parameters. The mapping library is updated based on the correction results.

[0041] To achieve the above technical solution, this embodiment divides it into the following five steps, specifically:

[0042] 1. Offline pre-playing and mapping library construction phase.

[0043] Specifically, before the system is officially put into operation on the power grid, it is first necessary to rely on a high-performance computing cluster to prepare offline data. During this stage, the system constructs an extreme transient scenario generation model based on power grid topology parameters and historical fault data, generating a transient scenario set.

[0044] The power grid topology parameters mainly refer to the physical and electrical parameters of the target substation and its surrounding transmission and distribution network, including but not limited to the bus admittance matrix parameters of each node, the distributed capacitance and distributed inductance parameters of transmission lines, the subtransient reactance parameters of synchronous generators, and the leakage inductance parameters of transformers. Historical fault data consists of actual power grid disturbance waveform data recorded by fault recorders during past operating cycles, such as voltage and current sequences under conditions like lightning overvoltage, single-phase ground fault, and converter valve failure. The extreme transient scenario generation model utilizes the Monte Carlo sampling algorithm to randomly combine and traverse uncertain parameters such as the physical location of the fault, the initial phase angle of the fault, and the fault impedance, thereby generating a set of transient scenarios with extremely wide coverage in virtual space.

[0045] It is important to note that the system subsequently performs electromagnetic transient simulations on each transient scenario in the transient scenario set, extracting the core feature vectors of the scenarios. Electromagnetic transient simulations typically rely on numerical integrators with simulation step sizes in the microsecond or even nanosecond range, such as algorithms based on the implicit trapezoidal rule, to accurately reproduce the traveling wave reflection phenomenon in distributed parameter circuits and the nonlinear and nonideal characteristics of carrier recombination processes within semiconductor devices. After each simulation, the system extracts key physical quantities from the simulation data pool. The core feature vectors of the scenarios include: transient voltage peak value, transient voltage rise time, transient voltage oscillation frequency, transient voltage polarity, and transient voltage initial phase angle.

[0046] like Figure 2 This demonstrates the system's execution basis for performing electromagnetic transient simulations and extracting core feature vectors in extreme transient scenarios during the offline pre-simulation and mapping library construction phases.

[0047] The horizontal axis in the figure represents time (microseconds), with a value ranging from 0 to 5, reflecting the extremely short physical time window in which ultrafast transients occur; the vertical axis represents transient voltage amplitude (kilovolts), with a value ranging from 0 to 1500.

[0048] The solid blue curve in the figure represents the transient voltage waveform. The curve exhibits an extremely high voltage change rate at the beginning stage and rapidly climbs to a peak value of nearly 1140 kV at about 0.6 microseconds. Subsequently, it shows a decay trend with high-frequency fluctuations, which truly reflects the physical characteristics of extreme and critical conditions such as lightning intrusion waves or ultra-fast transient overvoltages mentioned in the embodiment.

[0049] To accurately extract characteristic parameters, a red dashed curve was generated in the figure as the 10% peak baseline, which intersects the transient voltage waveform at a height of approximately 114 kV; at the same time, a green dashed curve was generated as the 90% peak baseline, which intersects the transient voltage waveform at a height of approximately 1026 kV.

[0050] The blue solid line spans the time period between these two baselines, which is the transient voltage rise time extracted by the system. In the figure, this time span is significantly smaller than the one-microsecond first time threshold defined in the first control mode in the specification.

[0051] In addition, the peak amplitude of the blue waveform directly quantifies the peak characteristics of the transient voltage, while the undulation pattern during its decay process intuitively characterizes the oscillation frequency characteristics of the transient voltage.

[0052] Through this waveform analysis and multi-baseline calibration extraction method, the system can accurately construct the core feature vector of the scenario, thus providing a solid data foundation for establishing a scenario and strategy mapping library that covers transient voltage polarity and transient voltage initial phase angle. This ensures that the subsequent ultra-fast transient pre-simulation matching mode has high pertinence and execution effectiveness when facing severe impacts on the actual power grid.

[0053] Here's an explanation of the aforementioned characteristic parameters: the transient voltage peak value characterizes the absolute maximum amplitude of the surge voltage, directly determining the avalanche breakdown risk of the thyristor silicon wafer; the transient voltage rise time is defined as the time it takes for the voltage to increase sharply from 10% to 90% of the nominal amplitude, this parameter determines the magnitude of the displacement current generated in the parasitic capacitance, thus affecting the dynamic voltage imbalance of the thyristor at the moment of turn-on; the transient voltage oscillation frequency reflects the resonance characteristics of the system's stray inductance and capacitance parameters; the transient voltage polarity indicates whether the first overvoltage peak is forward or reverse; and the transient voltage initial phase angle records the specific electrical angle at the moment the fault occurs within the power frequency AC cycle. These five characteristic parameters together uniquely and accurately define a specific power grid impact environment in multidimensional space.

[0054] For example, after obtaining the feature vectors, the system calculates the optimal emergency control strategy for each flexible AC transmission system valve group under each transient scenario using a dual-objective collaborative optimization model, and establishes a mapping library between scenarios and strategies. In this optimization process, the objective function of the dual-objective collaborative optimization model is to minimize the system cost function; the system cost function is composed of a weighted sum of a dynamic voltage equalization coefficient penalty term and a secondary overvoltage suppression penalty term; wherein, the dynamic voltage equalization coefficient penalty term is the product of the maximum dynamic voltage equalization coefficient of all flexible AC transmission system valve groups and a first weighting coefficient, and the secondary overvoltage suppression penalty term is the product of the maximum secondary overvoltage amplitude of all flexible AC transmission system valve groups and a second weighting coefficient.

[0055] The system cost function for this stage is calculated using the following mathematical model:

[0056] Formula for calculating the objective function:

[0057] ;

[0058] in, Represents the system cost function; This represents the first weighting coefficient, which is used to adjust the proportion of the dynamic equal pressure index in the overall optimization objective; This represents the maximum dynamic voltage equalization coefficient of all valve groups in a flexible AC transmission system. The dynamic voltage equalization coefficient is defined as the ratio between the actual voltage value of the thyristor bearing the highest voltage in the series valve group and the theoretical average voltage division value. The closer this ratio is to 1, the better the voltage equalization effect. This represents the second weighting coefficient, used to adjust the proportion of the secondary overvoltage suppression index; This represents the maximum secondary overvoltage amplitude of all valve groups in the flexible AC transmission system. Secondary overvoltage refers to the derivative overvoltage pulse generated on local devices due to the charging and discharging of the RC network and the freewheeling effect of the line inductance during the application of equalization control. Using the above objective function, optimization algorithms (such as offline advanced genetic algorithms or differential evolution algorithms) search for the optimal solution in the vast control parameter solution space. The set of parameters with the smallest values.

[0059] like Figure 3 The Pareto front distribution of the dual-objective collaborative optimization model in handling dynamic voltage equalization and overvoltage suppression is demonstrated during the offline pre-simulation and mapping library construction phases of the system.

[0060] The horizontal axis in the graph is named the maximum dynamic equalization voltage coefficient, and the corresponding unit is dimensionless, with a value range between 1.0 and 1.2; the vertical axis is named the maximum secondary overvoltage amplitude, and the corresponding unit is kilovolt, with a value range between 0 and 200.

[0061] The gray scatter plot in the figure represents the large number of optimization process solutions generated by the optimization algorithm as it explores the vast solution space of control parameters, while the red solid line and circular markers delineate the Pareto front solution set that converges after multiple iterations.

[0062] The red curve objectively presents the physical constraint relationship between the two control objectives. When the control strategy attempts to bring the maximum dynamic equalization coefficient close to 1.02 in order to pursue a higher equalization level, the strong negative pressure source injection action will cause the maximum secondary overvoltage amplitude to rise to more than 150 kV.

[0063] Conversely, if the maximum secondary overvoltage amplitude is suppressed to around 50 kV, the maximum dynamic voltage averaging coefficient will increase accordingly to around 1.15. This distribution pattern, exhibiting a decreasing curve characteristic, intuitively demonstrates the necessity of using a weighted summation of the dynamic voltage averaging coefficient penalty term and the secondary overvoltage suppression penalty term in the system cost function of the embodiment.

[0064] By selecting the equilibrium point on the red leading edge curve according to the set weight coefficient, the system can effectively calculate the optimal emergency control strategy of the valve group of the flexible AC transmission system under the corresponding transient scenario. Then, the determined parameters such as the output voltage amplitude, output frequency and pre-calculated timing offset of the controllable negative pressure source are solidified and stored in the mapping library, providing reliable data support for ultra-fast transient protection in actual power grid operation.

[0065] The final set of parameters obtained constitutes the emergency control strategy. The optimal emergency control strategy includes: the output voltage amplitude of the controllable negative voltage source of the target flexible AC transmission system valve group, the output frequency of the controllable negative voltage source, the pre-calculated timing offset, and the gate pulse delay compensation matrix. The output voltage amplitude of the controllable negative voltage source determines the total reverse charge energy that needs to be injected into the gate drive circuit or buffer circuit; the output frequency of the controllable negative voltage source determines the compatibility between the reverse energy injection rate and the high-frequency resonant mode of the system; the pre-calculated timing offset is used to compensate for the physical phase difference time at the moment of impact caused by the spatial physical position difference of each valve group when the transient electromagnetic traveling wave of the power grid propagates in the transmission line, i.e., the spatial flight time difference of the electromagnetic wave; the gate pulse delay compensation matrix is ​​a nanosecond-level time adjustment table for the individual differences of each thyristor in the series component, used to actively offset the dynamic uneven voltage caused by the discreteness of the device carrier lifetime by fine-tuning the trigger or turn-off time of each device. These strategy parameters are bound one-to-one with the corresponding scene feature vectors, packaged and stored in a high-speed random access memory, forming a mapping library that supports microsecond-level retrieval.

[0066] 2. Real-time data acquisition and adaptive switching of multi-level control modes.

[0067] Specifically, when the system is in online operation, the sensing and decision-making unit intervenes. The system collects the bus voltage and operating parameters of each flexible AC transmission system valve group in real time, obtains the transient voltage rise time, and automatically switches the control mode based on the transient voltage rise time. Data acquisition here mainly relies on wideband capacitive voltage transformers and Rogowski coil current transformers installed in the primary circuit. The collected analog signals are converted into digital sequences by an analog-to-digital converter with extremely high conversion rates. Within the high-speed field-programmable gate array (FPGA) chip of the main control logic controller, the system performs continuous differentiation operations on the incoming voltage digital sequence to extract the rate of change of the grid voltage in real time and calculate the real-time transient voltage rise time.

[0068] Optionally, to ensure reasonable allocation of system resources under different urgency levels, the system pre-acquires a second time threshold, which is greater than the first time threshold. When the transient voltage rise time is greater than or equal to the second time threshold, the system switches to a third control mode; when the transient voltage rise time is greater than or equal to the first time threshold and less than the second time threshold, the system switches to a second control mode. In practical engineering, the first time threshold is typically set to an extremely short time window at the microsecond level to define extreme critical conditions caused by lightning strikes or ultra-fast transient overvoltages. The second time threshold is typically set to the hundreds of microseconds or milliseconds level to define ordinary disturbance conditions caused by generator low-frequency oscillations or routine load switching. The third control mode corresponds to the system's steady-state or extremely slow transient environment, where system computing resources are mainly focused on reducing operating losses. The second control mode corresponds to the normal response state, requiring a certain amount of computing power for local closed-loop feedback adjustment.

[0069] It should be noted that when switching to the second or third control mode, the following steps are performed:

[0070] Based on the electrical coupling degree between valve groups in the flexible AC transmission system, valve groups are divided into clusters. With the optimization objectives of minimizing the dynamic voltage equalization coefficient within each valve group and minimizing the interactive interference between valve groups, a particle swarm optimization algorithm is used to solve for the optimal phase control angle command and voltage equalization control priority. The local controller automatically switches between low-loss operation, active voltage equalization operation, and emergency protection operation based on the transient overvoltage amplitude prediction value output by the electrical stress prediction model, performing multi-level active energy management.

[0071] The process of grouping valve groups based on electrical coupling primarily examines the electrical distance between valve groups in multiple flexible AC transmission systems on the electrical topology, such as the number of transformer stages and the equivalent impedance of the transmission lines. Shorter electrical distances indicate higher coupling, leading to grouping them into the same cluster for centralized coordination. The subsequent particle swarm optimization algorithm, simulating the information-sharing mechanism of bird flocks foraging, allows particles representing the solution space to iterate in a multi-dimensional parameter space, ultimately converging to the optimal phase angle command. When managing the underlying actuators, the local controller performs proactive energy management based on the predicted transient overvoltage amplitude: in low-loss operation, additional voltage equalization energy-consuming branches are disabled to improve energy conversion efficiency; in proactive voltage equalization operation, buffer resistors and capacitors are appropriately added for impedance matching suppression; and in emergency protection operation, all suppression modules are activated at full power. This multi-level management mechanism achieves a scientific and objective balance between equipment lifespan reduction, energy dissipation, and electrical safety.

[0072] 3. Feature extraction and scene matching mechanism under extremely fast transient conditions.

[0073] Specifically, when the power grid suffers a severe transient short circuit or lightning surge, the transient voltage change rate is extremely high. When the rise time of the transient voltage is less than a first time threshold, the system switches to the first control mode, and extracts the real-time transient core feature vector through a feature extraction algorithm. The first control mode represents the highest response authority of the system. At this time, since the decision time left for the controller is usually no more than a few hundred nanoseconds, traditional online integral iterative algorithms will cause the controller to be inactive or lag behind due to computing power bottlenecks. Therefore, the system directly adopts a hardware-accelerated architecture of feature extraction and database matching.

[0074] The step of extracting the real-time transient core feature vector through a feature extraction algorithm includes: obtaining a real-time voltage sampling sequence containing multiple sampling points; performing convolution and summation calculations on the real-time voltage sampling sequence and a preset wavelet basis function to extract multiple real-time transient feature values; and combining the multiple real-time transient feature values ​​to generate the real-time transient core feature vector.

[0075] Since extremely fast transient signals have wide bandwidth and non-stationary distortion characteristics, traditional Fourier transform based on fixed time windows is difficult to obtain good resolution in both the time and frequency domains. Therefore, wavelet transform algorithm is adopted.

[0076] The specific formula for feature extraction convolution summation is expressed as follows:

[0077] ;

[0078] in, This represents the j-th extracted real-time transient feature value; This represents the total number of sampling points in the real-time voltage sampling sequence, which is determined by the sampling frequency and the fixed truncation window length. Indicates the integer sequence number of the current sampling point in the sequence; This represents the digital voltage amplitude corresponding to the nth sampling point in the real-time voltage sampling sequence; This represents the spatial or temporal basis function coefficient value corresponding to the j-th feature dimension and the n-th sampling point in a preset wavelet basis function (e.g., a Daubechies wavelet function of a specific order).

[0079] Through this convolution summation operation, the high-frequency and low-frequency scale coefficients that characterize the energy distribution are extracted and then combined to transform into a real-time transient core feature vector.

[0080] like Figure 4This diagram illustrates the multidimensional energy space distribution process of the system when obtaining real-time transient core feature vectors through feature extraction algorithms in an ultrafast transient pre-simulation matching mode. The horizontal axis represents time, with the corresponding unit being microseconds, ranging from 0 to 5, objectively corresponding to the extremely short control time window during extreme power grid faults. The vertical axis represents the wavelet feature scale, with the corresponding unit being dimensionless, ranging from 1 to 64, representing the feature dimension expansion from small-scale high-frequency signals to large-scale low-frequency trends.

[0081] The color bitmap in the image, combined with the color bars on the right, maps the energy density of the signal at different times and spaces. The color gradient from dark blue to dark red represents the energy coefficient increasing from 0 to around 80. In terms of the graphic distribution characteristics, at a time axis of approximately 0.6 microseconds and a wavelet feature scale of around 40, a dark red energy concentration region, as shown in the legend, appears. This low-frequency, large-scale, high-energy patch accurately captures the main surge energy contained in the broadband transient impulse waveform.

[0082] At approximately 1.2 microseconds on the time axis and around 10 wavelet feature scales, another yellow-red secondary energy concentration region appears, with an energy coefficient of around 60. This clearly extracts the high-frequency transient oscillation component caused by the distributed inductance and distributed capacitance of the primary circuit.

[0083] By transforming a one-dimensional real-time voltage sampling sequence into a two-dimensional energy distribution bitmap with distinct high- and low-frequency scale boundaries, the digital signal processing logic unit inside the local controller can accurately extract multiple real-time transient feature values ​​and combine them to generate a real-time transient core feature vector using convolution summation operations.

[0084] This feature extraction mechanism based on bitmap transformation effectively overcomes the resolution limitation problem of traditional fixed-time-window frequency domain analysis in non-stationary distortion signal processing. It enables the system to reliably calculate scene similarity and call the optimal emergency control strategy in the mapping library even under harsh conditions with extremely short transient voltage rise times. This provides an accurate source of feature data for the underlying nanosecond-level gate pulse reconstruction and dynamic voltage equalization of series thyristors in multi-flexible AC transmission system valve groups.

[0085] After extraction, the system calculates the scene similarity between the real-time transient core feature vector and each scene core feature vector, and calls the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold.

[0086] The calculation of the scene similarity between the real-time transient core feature vector and each scene core feature vector includes: calculating the normalized difference between the real-time transient core feature vector and the scene core feature vector in each corresponding feature dimension; performing a weighted summation of the squares of the normalized differences in each corresponding feature dimension to obtain a comprehensive difference index; and substituting the comprehensive difference index into a Gaussian kernel function to perform a negative exponential operation to obtain the scene similarity.

[0087] The mathematical implementation of this calculation process includes two steps:

[0088] 1. Formula for calculating the comprehensive difference index:

[0089] ;

[0090] 2. Scene similarity calculation formula:

[0091] ;

[0092] in, Indicates the comprehensive difference index; Indicates the total number of feature dimensions; This indicates the current feature dimension index, which increases from 1 to the total number. This represents the preset weight value corresponding to the m-th feature dimension, which is pre-tuned based on the sensitivity of this dimension to the pressure equalization control effect; This represents the normalized difference between the real-time transient core feature vector and the scene core feature vector on the m-th feature dimension. The normalization process eliminates the interference of different physical dimensions such as voltage amplitude, time, and frequency on numerical calculation. This represents scene similarity, with values ​​ranging from 0 to 1.

[0093] By performing the above multiply-accumulate operations in parallel in a programmable gate array, the system can select the target transient scene with the closest similarity to the real-time environment from a huge mapping library within tens of nanoseconds, and directly retrieve the emergency control strategy bound to it to the buffer.

[0094] like Figure 5 The system demonstrates a three-dimensional spatial response surface in which the normalized difference of the feature dimension is nonlinearly mapped to scene similarity through a comprehensive difference index during ultrafast transient feature matching.

[0095] The horizontal axis in the figure represents the normalized difference of the first feature dimension, and the vertical axis represents the normalized difference of the second feature dimension. Both are dimensionless and range from -1 to +1, representing the degree of deviation between the real-time transient core feature vector and the scene core feature vector in the mapping library in a specific physical dimension. The vertical axis represents the scene similarity, which is also dimensionless and ranges from 0 to 1.

[0096] The nonlinear response surface shown in the figure presents a bell-shaped three-dimensional geometric shape that smoothly decays from the center outwards. The color smoothly transitions from dark blue at the edges to dark red at the center apex. The more red the color, the higher the scene similarity value.

[0097] When the normalized difference between the horizontal and vertical coordinates approaches 0, it means that the real-time features and the pre-stored scene are highly consistent. At this time, the comprehensive difference index is the smallest, the surface reaches the vertex at the center position, and the scene similarity is close to the maximum value of 1.

[0098] As the absolute value of the difference in any dimension increases, the comprehensive difference index accumulates rapidly. Through the negative exponential operation of the Gaussian kernel function, the surface height and color depth exhibit a non-linear and rapid decreasing trend.

[0099] This three-dimensional nonlinear response surface objectively reflects the system's matching algorithm's keen ability to identify minute feature deviations and its smooth convergence characteristics of the matching results. This enables the system to effectively amplify the numerical weights of high-matching scenarios within an extremely short computation time of tens of nanoseconds, thereby accurately selecting the target transient scenario that most closely approximates the real-time power grid environment and directly calling the optimal emergency control strategy bound to it. This provides scientific and reliable mathematical support for the underlying nanosecond-level gate pulse reconstruction of valve groups in multi-flexible AC transmission systems.

[0100] 4. Precise execution phase of control strategy based on global synchronization timestamp.

[0101] For example, after the strategy is invoked, it needs to be distributed to the underlying execution mechanisms located throughout the large substation. The system attaches a global synchronization timestamp to the invoked optimal emergency control strategy and sends it to the local controller of the corresponding flexible AC transmission system valve group. The global synchronization timestamp is an absolute clock reference marker generated jointly by a high-precision satellite timing system (such as the BeiDou Navigation Satellite System) and the Precision Time Protocol (PTP) network clock, with an accuracy down to the nanosecond level. This timestamp is broadcast to each local controller via a high-speed fiber optic local area network along with the control command data packet.

[0102] The local controller synchronously starts the controllable negative pressure source module at the determined absolute trigger time based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy.

[0103] The calculation process for absolute time is explained here: The local controller, based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, synchronously starts the controllable negative pressure source module at the determined absolute trigger time. This includes: adding the time value corresponding to the global synchronization timestamp to the pre-calculated timing offset to obtain the absolute trigger time for starting emergency control of the target flexible AC transmission system valve group; and synchronously triggering the controllable negative pressure source module and the gate drive unit of the target flexible AC transmission system valve group when the absolute trigger time is reached.

[0104] The formula for calculating absolute trigger time is as follows:

[0105] ;

[0106] in, This indicates the absolute trigger time for the emergency control of the valve group in the target flexible AC transmission system. This represents the time value corresponding to the global synchronization timestamp parsed from the control policy data packet; This represents the pre-calculated timing offset specific to this valve assembly. This offset comprehensively includes the space fiber optic transmission delay calibration value and the tactical-level time difference value required for the valve assembly to act earlier or later for a specific transient scenario.

[0107] By requiring all controllers to reach their own hardware clock... The control level is released only at this precise moment, effectively eliminating asynchronous execution problems caused by software factors such as communication network jitter and task queuing delays at each node.

[0108] In terms of execution, the local controller not only activates the controllable negative voltage source at a specified time, but also reconstructs the gate pulse based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, thereby achieving dynamic voltage equalization regulation of the series thyristors. The controllable negative voltage source module is essentially a high-frequency isolated DC-DC converter circuit with fast response capability. At startup, it can inject a reverse bias voltage into the RC buffer network or cathode-gate junction of a specific thyristor, thereby forcibly extracting the internally accumulated excess minority carriers. Simultaneously, the gate pulse reconstruction, based on the gate pulse delay compensation matrix, artificially introduces individual delays of several nanoseconds to tens of nanoseconds into the trigger turn-on or turn-off pulse signals that should originally be sent simultaneously to fifty or one hundred thyristors in the series assembly. Because of the slight physical discrepancies in the turn-on delay time and reverse recovery charge of thyristors in the manufacturing process, the thyristors that are originally slow to act receive the trigger signal earlier, while the thyristors that act faster receive the trigger signal later. This forces all series-connected devices to complete the impedance state transition at the same physical instant in the external macroscopic performance, which greatly reduces the dynamic voltage imbalance.

[0109] like Figure 6 The system demonstrates a dynamic comparison of the voltage distribution at the terminals of the series thyristors before and after the gate pulse delay reconstruction control is executed.

[0110] The horizontal axis in the graph represents time, with the unit being microseconds and the value range from 0 to 15; the vertical axis represents the thyristor terminal voltage, with the unit being volts and the value range from 0 to 5500.

[0111] In the pre-reconstruction waveform comparison in the upper half, the red solid line, blue dashed line, and green dotted line representing different thyristors exhibit severe dispersion at the transient turn-off moment around 5 microseconds. Due to the physical differences in the recombination rate of charge carriers within semiconductor devices, the first thyristor, which reacts the fastest before reconstruction, experiences an extremely high voltage change rate, causing its terminal voltage to rise rapidly and generate a severe overvoltage spike as high as 4800 volts. In contrast, the third thyristor, which reacts more slowly, only shares a very low transient voltage. This severe voltage imbalance in macroscopic performance can easily lead to overvoltage damage to local devices.

[0112] In the reconstructed waveform comparison in the second half, the local controller issued a gate pulse delay compensation matrix based on the optimal emergency control strategy, performing nanosecond-level individual delay compensation for the triggering timing of each thyristor. After compensation, the three curves representing the terminal voltages of different devices highly overlapped during the transient turn-off period. The original voltage spike of up to 4800 volts was significantly flattened, and the maximum peak voltage of all devices was effectively suppressed to around 3150 volts, eventually smoothly transitioning to a steady-state average voltage level of 3000 volts.

[0113] The red, blue, and green curves in the graph change from severe divergence before reconstruction to tight convergence after reconstruction, objectively and intuitively proving that the control method of the present invention can force all series devices to complete the impedance state transition at the same macroscopic physical instant, significantly reducing the dynamic voltage imbalance and significantly improving the electrical safety and operational reliability of the converter valve group when facing extreme high-voltage impacts.

[0114] 5. Online monitoring and feedback correction closed-loop stage of operational status.

[0115] Specifically, to ensure the system's safety and reliability during long-term service and to maintain high-precision voltage equalization performance even when devices undergo physical aging and degradation, an adaptive adjustment mechanism is introduced. The system monitors the operating status of each thyristor in real time and inputs the feedback signal into the electrical stress prediction model for online correction of model parameters, updating the mapping library based on the correction results.

[0116] The real-time monitoring of the operating status of each thyristor and the input of feedback signals into the electrical stress prediction model for online correction of model parameters include: extracting abnormal feature vectors corresponding to the operating status of each thyristor, wherein the abnormal feature vectors include changes in turn-on delay time, changes in turn-off delay time, and communication bit error rate; using an anomaly detection algorithm to identify and analyze the abnormal feature vectors; when a minor state degradation fault is identified, adaptively adjusting the gate pulse delay compensation matrix; when a severe isolation fault is identified, triggering a bypass switch to isolate the faulty device, and recalculating the gate pulse delay compensation matrix for the remaining normal thyristors.

[0117] During operation, high-bandwidth isolated current sensors and voltage probes continuously record the voltage and current slopes at each switching instant, and calculate the actual turn-on delay time and turn-off delay time of the thyristor based on this. These values ​​are then compared with the initial factory calibration values ​​to generate the changes in turn-on and turn-off delay times. Simultaneously, the communication bit error rate statistically analyzed by the underlying optical transceiver module is extracted. These three key indicators constitute the anomaly feature vector.

[0118] Anomaly detection algorithms typically employ isolated forest algorithms or support vector machine models based on high-dimensional spatial distance metrics. If the model output indicates that the turn-on delay time is merely due to normal thermal drift of the silicon wafer's internal carrier lifetime over long-term operation (judged as a minor state degradation fault), the main control unit will automatically modify the corresponding entry in the gate pulse delay compensation matrix for that device, increasing or decreasing a certain nanosecond-level compensation amount, and simultaneously apply this correction to all relevant strategies in the mapping library to achieve adaptive evolution of control parameters.

[0119] If the model output indicates that a thyristor has lost its voltage blocking capability or that the communication control link has experienced an irreversible hard interruption (determined as a severe isolation fault), the system will trigger the high-speed thyristor bypass switch (or mechanical bypass switch) connected in parallel with the faulty device within microseconds. This will allow the load current to flow through the bypass, safely removing the faulty device from the high-voltage main circuit. At this time, because the total number of stages in the series components has decreased by one stage, the base steady-state DC voltage that each remaining normal thyristor needs to share has changed. The system will immediately call the reconstruction algorithm to recalculate and issue a new version of the gate pulse delay compensation matrix applicable to the remaining normal thyristors, ensuring that the reconstructed topology can still safely withstand the next extreme transient impact.

[0120] In summary, considering the limitations prevalent in existing technologies, conventional passive RC snubber circuits, while simple in structure and requiring no complex software control, suffer from limited voltage absorption capacity due to the charging and discharging inertia of their capacitors when facing lightning transients with extremely fast rising edges, often leading to local overvoltage breakdown. Furthermore, traditional active gate drivers rely solely on local voltage sampling for closed-loop feedback. When facing multi-valve group coordinated transient responses at the entire station level, the lack of independent hardware clock operation for each driver and insufficient computing power inevitably result in timing misalignments.

[0121] This embodiment transfers the extensive and complex electromagnetic transient calculation process to an offline space for preprocessing before system commissioning, greatly reducing the computational burden on the real-time controller. After achieving online operation, it directly invokes the high-quality control strategy optimized by dual objectives based on rapid feature matrix similarity matching. Combined with global timing management based on nanosecond-level absolute time reference, this technical solution not only breaks down the instruction asynchrony barrier existing in conventional distributed control networks, but also smooths out the unbalanced voltage distribution at the microscale of device physical characteristics through microsecond-level negative pressure injection and gate pulse delay reconstruction at the underlying level. This mechanism effectively reduces the probability of chain avalanche breakdown caused by devices subjected to extremely severe electrical stress, significantly improving the overall insulation safety margin of the converter valve group, and providing a new generation of operational support system with strong adaptive and anti-aging correction capabilities for large-scale power electronic clusters in substations.

[0122] Example 2:

[0123] This embodiment is based on the hardware system and basic control logic of Embodiment 1. It focuses on explaining the technical solution for how the system can maintain safe operation when facing unknown extreme transient conditions that are unprecedented and exceed the coverage of the pre-simulation dataset.

[0124] Specifically, the step of calculating the scene similarity between the real-time transient core feature vector and each scene core feature vector, and calling the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold, also includes an online safety generalization step when no strategy is matched.

[0125] In the actual operation environment of modern high-voltage power grids, with the grid connection of massive new energy units and frequent network topology reconfiguration, the equivalent impedance and resonant frequency of the power grid exhibit highly time-varying and nonlinear characteristics. When a chain fault occurs, such as lightning strikes combined with commutation failure or multiple DC lines experiencing simultaneous ground faults, the extremely fast transient voltage waveform excited on the bus may exhibit a novel spatiotemporal characteristic in its rise time, oscillation frequency, and peak polarity combination. Since the computing power and storage space of offline constructed transient scenario sets are limited, this rare operating condition generated by the superposition of multiple accidental factors will inevitably have some feature vectors that wander outside the boundaries of the known mapping library, i.e., the out-of-distribution scenario defined in engineering.

[0126] For series thyristor assemblies, if matching control parameters cannot be obtained under such unknown and high-risk transient conditions, the underlying actuator of the system will face a situation where there are no instructions to execute or the wrong instructions are executed. This will lead to a serious mismatch of the parameters of the RC buffer circuit, which can easily cause one or more thyristors to undergo irreversible physical breakdown due to transient overvoltages that exceed the insulation design margin.

[0127] For example, in order to address the problems that exist in the actual engineering, the online security generalization step in this embodiment specifically includes: when the scene similarity between the real-time transient core feature vector and all scene core feature vectors in the mapping library is lower than the scene similarity threshold, extracting a preset number of alternative transient scenes and their corresponding alternative emergency control strategies that rank first in scene similarity.

[0128] During the matching retrieval cycle of the main control field-programmable gate array (FPGA) chip, the parallel comparison module traverses all pre-stored vectors in the mapping library. If the system sets the scene similarity threshold to 90%, and after traversal, it finds that the highest scene similarity returned in the hardware register is only 82%, the system logic will immediately determine that the current power grid has encountered a new transient state not covered by offline pre-simulation. Under this logic branch, the system abandons the single highest similarity call strategy and instead starts a multi-source information fusion mechanism.

[0129] The system relies on a hardware sorting tree to quickly extract the top few pre-stored scenarios with the highest scene similarity, defining them as a preset number of candidate transient scenarios. The preset number is usually tuned based on the density of the mapping library and the multiply-accumulate unit resources of the FPGA, typically selected as 3 to 5. Although these candidate transient scenarios do not completely match the current real-time transient in terms of overall characteristics, their distribution in the multi-dimensional feature space represents the known operating experience closest to the current unknown operating condition in terms of physical distance, and their corresponding candidate emergency control strategies contain highly valuable control action benchmarks.

[0130] It is important to note that after obtaining the aforementioned set of candidate scenarios, the system cannot directly perform a simple arithmetic average, because the similarity between different candidate scenarios and the current unknown transient state still varies significantly. Therefore, the system calculates the normalized weight coefficient of each candidate transient scenario based on the scenario similarity of each candidate transient scenario. This calculation process aims to quantify and allocate the influence of each candidate strategy in the subsequent fusion process. Scenarios closer to the current transient state should have their weight increased exponentially to ensure the physical targeting of the fusion strategy.

[0131] The specific calculation process for the normalized weighting coefficient is demonstrated using the following formula:

[0132] ;

[0133] In the formula, This represents the normalized weight coefficient corresponding to the p-th alternative transient scenario calculated and output by the system; This represents a pre-set scaling constant, which is used to adjust the sensitivity of similarity differences. By taking an appropriate value, the weight difference between high and low similarity can be widened, avoiding the failure of the strategy caused by egalitarianism. This represents the specific numerical value of the scene similarity of the p-th candidate transient scene extracted from the mapping library; This represents the total number of preset alternative transient scenarios extracted by the system; This is the index for the summation operation, with a starting value of 1 and an ending value of 1. ; This indicates the number of iterations in the denominator of the summation. Scene similarity among the candidate transient scenarios.

[0134] Through this formula, the sum of the normalized weight coefficients of all candidate transient scenarios is strictly equal to 1, thus providing a rigorous mathematical basis for the subsequent parameter-weighted projection.

[0135] Specifically, after weight allocation, the system uses the normalized weight coefficients to perform a weighted summation of the control parameters in the multiple alternative emergency control strategies to generate an initial fusion control strategy. The control parameters include the output voltage amplitude of the controllable negative voltage source and the pre-calculated timing offset. The voltage equalization process of the series thyristor assembly under extremely fast transient conditions is essentially a continuous adjustment process of a multidimensional physical quantity in the time domain.

[0136] Therefore, the control parameters have the physical feasibility of linear interpolation in a continuous space. For example, regarding the parameter of the output voltage amplitude of a controllable negative voltage source, if the first alternative strategy requires an output of 600 volts with a weight of 60%, the second alternative strategy requires an output of 800 volts with a weight of 30%, and the third alternative strategy requires an output of 500 volts with a weight of 10%, the system can quickly generate a new instruction with an output of 650 volts through the multiplicative summation hardware unit.

[0137] Similarly, for the pre-calculated timing offset, the system interpolates and calculates a comprehensive timing compensation node based on the pre- or post-action time requirements of each alternative scenario. This fusion process dynamically extends the originally discretely stored static policy library into a continuous control plane covering the entire state space. This allows the system to quickly deduce a comprehensive initial fusion control strategy based on known experience boundaries when facing unknown transient characteristics, thus expanding the effective application boundary of the control strategy.

[0138] Alternatively, although the initial fusion control strategy is the statistically and spatially optimal estimate, in the field of extreme high-voltage power transmission, any interpolation-based prediction carries inherent uncertainty. If the initial strategy is directly implemented at this time, and the interpolation result is too weak to extract the excessive reverse recovery charge generated inside the thyristor due to transient impact, the local device will still be at risk of thermal damage.

[0139] Therefore, the difference between the scene similarity threshold and the highest scene similarity among the candidate transient scenes is calculated, and this difference is mapped to an uncertainty margin coefficient. Here, the highest scene similarity represents the degree to which the current real-time transient signal is closest to the system's known experience base. The further this value is from the preset safety limit (i.e., the scene similarity threshold), the more unfamiliar the current transient condition is to the system, the greater its inherent risk of overshooting, and the corresponding increase in the level of protection required by the system.

[0140] The specific mathematical calculation logic for this uncertainty margin coefficient is demonstrated using the following formula:

[0141] ;

[0142] In the formula, This represents the uncertainty margin coefficient generated in the final calculation, which is a scalar multiplier that is always greater than or equal to 1; This represents the preset margin gain adjustment constant, which is usually determined by the insulation coordination margin during the substation design phase and the conservatism of the thyristor safe operating area (SOA) curve. This represents the system's preset threshold for defining the similarity between known and unknown scenarios; This represents the highest scene similarity among the multiple candidate transient scenes extracted so far.

[0143] As can be seen from the formula, the further the real-time scene deviates from the mapping library (i.e., the lower the highest similarity), the greater the difference between the two. After gain amplification, the uncertainty margin coefficient added to the base value of 1 also increases accordingly. This establishes a security prevention feedback mechanism that grows linearly with the level of unknown risk from a mathematical perspective.

[0144] For example, after obtaining the uncertainty margin coefficient, the system needs to physicalize it and perform compensation actions. The uncertainty margin coefficient is used to product amplify and compensate the output voltage amplitude of the controllable negative voltage source in the initial fusion control strategy, generating a safe generalized emergency control strategy. At the actual physical level, the series thyristor withstands extremely high [temperature / temperature] during turn-off transients. When the voltage change rate is high, if the minority carriers remaining in the base region cannot be quickly swept out, it will cause a surge in the reverse recovery current of the device, which in turn will induce a severe secondary overvoltage on the distributed inductance. The function of a controllable negative voltage source is to actively apply a reverse electric field to forcibly extract these charges.

[0145] like Figure 7 The system demonstrates how, when faced with unknown extreme transient conditions that exceed the coverage of the pre-simulated dataset, it generates a three-dimensional spatial mapping relationship for safety control commands through multi-source information fusion and uncertainty margin amplification.

[0146] The horizontal axis in the figure represents the first feature dimension, and the vertical axis represents the second feature dimension. Both are dimensionless and range from 0 to 2.0, used to characterize the spatial distribution characteristics of transient signals. The vertical axis represents the output voltage amplitude of the controllable negative voltage source, in volts, with a range of 300 to 850.

[0147] The blue scatter clusters in the figure represent known scenarios and their corresponding control policy amplitudes in the mapping library, mainly concentrated in the low-to-mid-dimensional region of space. When the system identifies a high-risk unknown transient scenario with feature dimension coordinates around 1.5, since it deviates from the distribution of blue known scenarios, the system extracts multiple candidate transient scenarios with the highest similarity ranking, performs a weighted summation, and calculates the initial fusion control policy point (yellow in the figure), whose corresponding voltage amplitude is approximately 650 volts.

[0148] Subsequently, the system calculates an uncertainty margin coefficient based on the scene similarity difference. This coefficient is then used to perform product amplification compensation on the initial yellow fusion point, extrapolating it along the vertical black dashed line trajectory to the red safety generalization emergency control strategy point in the figure, where its voltage amplitude is amplified to 780 volts. A semi-transparent red safety generalization boundary extrapolation surface is generated around this red target point. This geometric envelope surface demonstrates the defense space set by the system in the three-dimensional feature space to cope with unknown operating conditions.

[0149] This geometric extrapolation process, from known discrete points to unknown target points and then to the safety envelope, objectively reflects the system's defense logic of increasing the insulation safety margin of equipment by adding a small amount of energy loss, enabling the series thyristor assembly to have a reliable voltage equalization protection reference when encountering distributed extreme impacts.

[0150] When the system faces an unknown transient, the initially fused voltage amplitude is multiplied by a factor greater than 1. A coefficient (for example, if the calculated coefficient is 1.2, then the original 650 volt voltage is forcibly amplified to 780 volts) means that the system commands the negative voltage source circuit (usually composed of an insulated gate bipolar transistor full-bridge topology) to inject more reverse energy into the thyristor gate network instantaneously.

[0151] This product amplification compensation mechanism, which introduces a conservative safety margin, is essentially a classic engineering strategy that sacrifices minor energy loss to ensure the absolute probability of equipment survival under unknown operating conditions. Through this step, the probabilistic estimation strategy is transformed into a safety-generalized emergency control strategy with robust bottom-line defense capabilities, ensuring that even in the face of shocks beyond our comprehension, the thyristor voltage stress remains effectively limited within the safe insulation margin.

[0152] It is also important to note that after the security generalization of the strategy is generated, the system's action sequence and subsequent maintenance process must also be closed-loop. The security generalization emergency control strategy is used as the optimal emergency control strategy corresponding to the final target transient scenario, with the global synchronization timestamp attached and distributed. At the same time, the current real-time transient is marked as an unknown high-risk scenario, triggering a priority update of the mapping library.

[0153] At the underlying execution level, this generalized strategy is completely consistent with the normal pre-stored strategy in terms of data structure. After being packaged and appended with a global synchronization timestamp timed by the satellite clock system, it is synchronously distributed to the valve base controllers at all levels along the wide area fiber optic network for execution. This ensures that the actuator does not need to care about the source of the strategy, but only needs to strictly follow the timing offset and absolute timestamp to complete the gate triggering or shutdown.

[0154] In the upper-level architecture, the capture of an unknown transient event is itself extremely valuable. By labeling this real-time transient as an "unknown high-risk scenario," the high-performance offline computing cluster of the scheduling center will be automatically triggered in the next idle cycle to retrieve the complete waveform recording data of the transient and perform a full-scale, refined electromagnetic transient simulation. After the simulation converges, the optimal phase control angle and resistance-capacitance parameters for this specific transient are solved, and this new scenario feature vector, along with the true optimal strategy, is packaged as a new knowledge entry and inserted into the existing mapping library. This update mechanism enables the voltage equalization control system to have continuous online evolution capabilities, and as the power grid operates for longer periods, its coverage of extreme operating conditions will become increasingly comprehensive.

[0155] In summary, considering the practical problems faced in the existing technology field, the main limitation of traditional flexible AC transmission valve group control schemes based on table lookup or simple offline calculation is their insufficient generalization ability to unknown transient characteristics.

[0156] When the physical quantity dimension of the power grid impact slightly exceeds the extreme value of the preset table, traditional controllers often directly trigger hardware fault lock-up signals, leading to large-scale power outages; or they output default zero values ​​or old instructions from the previous moment. Direct application of such instructions in the microsecond-level high-voltage drastic change environment can easily lead to control failure, potentially causing overcurrent damage to RC components or the series valve group to stop operating due to voltage instability.

[0157] This embodiment constructs a logically rigorous and highly engineered degradation defense barrier. Utilizing pure mathematical weighting and spatial distance mapping, it generates a control command highly consistent with physical operating trends through multi-source feature interpolation within an extremely short computation time window, achieving a technological leap from discrete known control points to a continuous unknown safety surface. Simultaneously, an uncertainty margin amplification mechanism is introduced, employing conservative energy injection at the physical level to offset the risk of insulation collapse due to unknown operating conditions. Finally, an adaptive learning closed loop is incorporated, enabling the entire voltage equalization regulation system to possess both extremely fast underlying hardware-level transient response capabilities and scenario adaptive evolution capabilities, significantly improving the robustness and survivability of the core converter equipment in complex power grid environments.

[0158] Example 3:

[0159] This embodiment focuses on explaining the technical solution for maintaining absolute time synchronization of multiple valve groups in AC / DC hybrid power grids or multi-flexible AC transmission systems under extreme operating environments, where timing deviations occur due to physical environmental interference in the underlying hardware communication links.

[0160] In ultra-high voltage substations containing numerous flexible AC transmission system valve groups, the cluster controller that issues strategies is typically connected to the various local controllers distributed in the field via a wide-area fiber optic communication network spanning hundreds or even thousands of meters. In actual operation, substations face drastic diurnal temperature variations, seasonal climate changes, and strong electromagnetic interference. The physical medium of optical fiber (such as the quartz glass core) expands and contracts with temperature changes, and its refractive index also undergoes microscopic shifts with temperature. This causes dynamic changes in the propagation speed and transmission distance of optical signals within the fiber.

[0161] Meanwhile, the high-frequency electromagnetic pulses generated by high-power thyristors during switching can easily couple into the power supply and signal links of the photoelectric conversion module, causing random jitter in the hardware timestamp recording. If a fixed theoretical transmission delay is used directly to calculate the synchronization trigger time, the aforementioned environmentally induced delay drift and random jitter will directly translate into timing misalignment of the gate trigger pulse. For series thyristor assemblies, nanosecond-level timing misalignment will cause some thyristors to bear the entire bus transient voltage prematurely or lag behind, thereby disrupting the dynamic voltage equalization effect and causing overvoltage damage to the devices.

[0162] Specifically, in this embodiment, before the step of synchronously starting the controllable negative pressure source module and gate pulse reconstruction at the determined absolute trigger time based on the global synchronization timestamp and the pre-calculated timing offset, the local controller further includes a timing synchronization dynamic compensation step to eliminate communication delay errors. This step aims to build a dynamic defense line within the gap between policy issuance and local execution, tracking and offsetting time-varying errors in the physical communication link in real time to ensure absolute accuracy of the final execution time.

[0163] For example, the timing synchronization dynamic compensation step specifically includes: periodically sending hardware diagnostic pulses between the strategy-issuing controller and each of the local controllers via a wide-area fiber optic communication network to obtain actual round-trip time delay data with timestamps. In physical implementation, the strategy-issuing controller has an independent high-precision hardware timer internally, which is unaffected by operating system software task scheduling. During non-transient control data idle periods, the controller directly injects a specific format hardware diagnostic pulse data frame into the wide-area fiber optic communication network through the Media Access Control (MAC) layer. The data frame is automatically timestamped by the underlying hardware the instant it leaves the main control chip's transmit pin; when the physical layer chip detects the frame header start delimiter of the data frame, each receiving end's local controller also automatically records the receive timestamp and sends back a response data frame with the receive timestamp within microseconds. After receiving the response frame, the strategy-issuing controller records the final return timestamp. By eliminating the fixed hardware processing delay of the local controller, the system can accurately calculate the optical signal spatial flight time for each probe cycle, which is the actual round-trip time delay data. This detection method, based on the pure hardware link layer, effectively avoids the uncertainty and delay caused by protocol stack parsing and software interrupt response.

[0164] Optionally, since single measurement data often contains sudden high-frequency noise caused by electromagnetic interference, directly using single measurement data for compensation can lead to oscillations in the control system. Therefore, the system further arranges multiple historical, consecutive actual communication round-trip delay data in chronological order to construct a delay state time series. A fixed-length circular buffer is allocated in the random access memory of the local controller as a sliding data window. Whenever a new hardware diagnostic pulse probe is completed, the latest acquired actual communication round-trip delay data is pushed into this circular buffer, while the oldest data is removed and discarded. This data set arranged in strict chronological order constitutes a time series containing the recent evolution trend of communication link delay. This delay state time series not only reflects the current delay state but also contains the low-frequency drift characteristics of the optical fiber medium caused by slowly changing environmental factors such as temperature.

[0165] It is important to note that, in order to extract the true drift trend reflecting the physical nature from the noisy observation sequence, the system inputs the time delay state sequence into a preset Kalman filter tracking model for state-space estimation. This filters out random time jitter components caused by environmental electromagnetic interference and predicts the dynamic phase shift error of the underlying clock at the current moment due to variations in the physical medium of the optical fiber. This is the residual synchronous clock drift that the standard PTP protocol cannot completely filter out under extreme strong magnetic environments. The Kalman filter tracking model is an efficient recursive mathematical algorithm that does not require storing all historical data. It only needs to calculate the optimal state estimate for the current moment based on the state estimate from the previous moment and the observation value at the current moment. In this application scenario, the physical time delay variation of the optical fiber affected by temperature is modeled as a low-frequency Markov process with a stationary increment, while the timestamp recording deviation caused by electromagnetic interference is modeled as Gaussian white noise superimposed on it.

[0166] In this Kalman filter tracking model, the system's state recursion and correction process is illustrated by the following two core formula algorithms:

[0167] State time update formula (prior estimate):

[0168] ;

[0169] State observation update formula (posterior correction):

[0170] ;

[0171] In the formula: The model represents the first time. The prior delay state estimation vector calculated in discrete time steps represents the theoretical prediction of the communication delay at the current moment based on physical inertia, before the latest measurement data is introduced. This represents the preset system state transition matrix, which describes the physical laws governing the natural evolution of fiber delay over time. The model represents the first time. The posterior optimal time delay state estimation vector output after correction for each discrete time step (i.e. the previous time step); The model represents the first time. The final output of each discrete time step is the posterior optimal time delay state estimation vector, which is the true time delay state after filtering out high-frequency jitter. Indicates the first The Kalman gain matrix is ​​dynamically calculated at discrete time steps. This matrix balances the uncertainty of the theoretical prediction of the model with the uncertainty of the actual measurement of the sensor in real time, and determines the trust ratio of the current new observation data. Indicates the first Each discrete time step is an observation actually measured by the underlying hardware, that is, the actual round-trip delay data of communication obtained in the preceding steps; This represents a preset observation matrix used to map internal states to externally observable physical quantities.

[0172] Through the above recursive calculations, the Kalman filter tracking model iterates continuously at each time step. When environmental electromagnetic interference is severe, leading to... When a sudden, large jump occurs, the algorithm's internal error covariance matrix will adaptively adjust, causing the Kalman gain matrix to... By reducing reliance on current observations, random time jitter caused by environmental electromagnetic interference is mathematically filtered out. The filtered and smoothed posterior optimal delay state estimate vector is then reduced by the system, subtracted from the factory-calibrated baseline static delay, and divided by 2 to accurately obtain the real-time one-way communication delay drift at the fiber optic physical layer at the current moment. This drift accurately reflects the additional delay in one-way signal transmission caused by fiber stretching or refractive index changes.

[0173] like Figure 8 This paper demonstrates the dynamic estimation and convergence process of the system using a Kalman filter tracking model to handle communication delay errors under strong electromagnetic interference. The horizontal axis in the figure represents time, in milliseconds, with values ​​ranging from 0 to 100; the vertical axis represents the one-way communication delay drift, in nanoseconds, with values ​​ranging from -30 to 80.

[0174] The light blue lines and dots in the figure represent the actual measured drift obtained through the periodic pulse diagnosis of the underlying hardware. Affected by high-frequency radiation interference from the strong electromagnetic field environment of the substation, this data sequence contains high-amplitude random time jitter components, exhibiting characteristics of violent fluctuations and sudden jumps. The green dashed line in the figure represents the true physical time delay drift baseline. Its gently rising curve objectively reflects the low-frequency evolution law of the optical fiber physical medium caused by slowly changing factors such as ambient temperature.

[0175] The red solid line in the figure is the Kalman filter optimal estimation convergence curve. At the beginning of the time axis, due to the objective deviation of the initial state estimation, there is a certain distance between the red solid line and the green baseline. However, as the state space estimation is iterated cycle by cycle, the system adaptively adjusts the internal gain matrix, so that the red solid line converges quickly and closely fits the green baseline.

[0176] Throughout the state tracking process, the red solid line maintained a high degree of smoothness, effectively removing high-frequency noise and sudden interference from the light blue data. Through this data filtering process, the model can accurately extract the real-time one-way communication delay drift at the fiber optic physical layer. The local controller adaptively superimposes this high-precision calibration value with the pre-calculated timing offset, effectively offsetting the hardware synchronization error caused by the nonlinear variation of the communication medium in the distributed nodes. This provides a rigorous timing reference guarantee for the underlying gate pulse reconstruction of valve groups in multi-flexible AC transmission systems.

[0177] Specifically, after acquiring the high-precision drift amount, the system dynamically corrects the command parameters. The system adaptively superimposes the underlying clock dynamic phase shift error with the pre-calculated timing offset to generate an actual execution timing offset with underlying clock calibration attributes. The pre-calculated timing offset is a control strategy parameter generated offline by the dual-objective collaborative optimization model in Example 1. It only considers the spatial distance difference between valve groups and the static response difference of the thyristors themselves, and belongs to the theoretical benchmark value. Algebraically adding the real-time one-way communication delay drift amount with this theoretical benchmark value is equivalent to injecting dynamic field environment perception attributes into the static control commands. The generated actual execution timing offset includes both the power grid control tactical intent for specific transient scenarios and absorbs the microscopic changes in the current underlying physical fiber optic link, thereby eliminating blind zone errors on the command delivery path.

[0178] For example, after completing the adaptive calculation of the above parameters, the underlying control logic will take over the final physical action execution. The local controller adds the global synchronization timestamp to the actual execution timing offset to determine the corrected absolute trigger time, thereby controlling the valve groups of different flexible AC transmission systems to perform synchronous triggering operations.

[0179] Inside the local controller's Field Programmable Gate Array (FPGA) chip, a 64-bit absolute time counter driven by a high-precision cryogenic crystal oscillator runs. This counter continuously maintains nanosecond-level clock synchronization with the global satellite timing system via Precise Time Protocol (PTP). The local controller latches a new, corrected absolute trigger time threshold in a register by summing the parsed global synchronization timestamp with the dynamically generated actual execution timing offset. The FPGA's internal hardware comparator module performs a hardware-level comparison of the current local real-time clock value with the corrected absolute trigger time in the register every nanosecond clock cycle. The comparator outputs a high-level transition signal only if the two values ​​are completely equal within a single clock cycle. This signal is directly hardwired to the digital signal output pin, instantly activating the insulated-gate bipolar transistor (IGBT) driver circuit of the controllable negative voltage source module, and simultaneously issuing gate trigger or turn-off pulses for all thyristors according to the reconstructed matrix.

[0180] For example, the revised formula for calculating the absolute trigger time is as follows:

[0181] ;

[0182] in, This indicates the absolute trigger time that the local controller finally latches after clock phase shift compensation; The time value corresponding to the globally synchronized timestamp; For pre-calculated timing offsets; Indicates the first The underlying clock dynamic phase shift error is predicted by the Kalman filter output.

[0183] This hardware-based synchronous triggering operation, based on corrected absolute time, enables highly unified timing reference alignment for valve groups of various flexible AC transmission systems distributed over hundreds of meters within a substation and connected to different fiber optic cable lengths. Regardless of how drastic fluctuations in external ambient temperature cause changes in the physical characteristics of the fiber optic cable, or how strong electromagnetic radiation interference to communication is generated by thyristor operation, each valve group can perform pressure equalization actions with high precision and synchronization at a predetermined microscopic physical instant.

[0184] In summary, considering the limitations prevalent in existing technologies, current distributed power electronic control systems, when handling multi-machine collaboration or multi-valve group synchronization tasks, often assume that the delay of the fiber optic communication network is a scalar constant, and after a one-time static compensation is completed during the engineering commissioning phase, it remains unchanged. This static compensation scheme performs reasonably well in laboratory constant temperature and humidity environments, but once deployed to extremely harsh high-voltage transmission sites, the slow time-varying drift of the fiber optic physical characteristics and the high-frequency jitter of electromagnetic interference will easily exceed the tolerance limit of nanosecond-level synchronous control. This directly results in a tiny time difference in the sequential action of each device in the series thyristor assembly during transient control, causing the slower-acting devices to experience extremely high voltage change rates. Transient reverse overvoltages significantly reduce the reliability and lifespan of equipment.

[0185] The timing synchronization dynamic compensation mechanism described in this embodiment establishes an independent communication environment perception and calibration system at the bottom layer of the original transient feature matching and adaptive voltage equalization strategy framework. This scheme uses pure hardware probe pulses to extract the most original timestamp data and combines it with a Kalman filter tracking model to remove physical interference noise in the mathematical space, achieving online tracking and prediction of optical signal propagation delay at the sub-nanosecond level.

[0186] By superimposing this dynamic drift amount into the offline pre-simulated static control strategy, this solution effectively bridges the timing gap between the upper-level power grid transient control algorithm and the underlying semiconductor device gate drive physical execution. This overall solution not only endows the valve group of the flexible AC transmission system with high robustness against harsh operating environments, but also eliminates the synchronization error caused by the nonlinear drift of distributed control nodes due to communication delays at its physical root. It ensures a high degree of voltage stress balance among hundreds of series-connected thyristors under extreme transient shocks, providing solid equipment and technical support for the safe and stable operation of the next-generation large-capacity AC / DC hybrid power grid.

[0187] Example 4:

[0188] In existing technologies, flexible AC transmission systems widely employ series thyristor assemblies. However, when faced with rapid transient conditions such as lightning strikes and short-circuit faults, existing passive RC snubber circuits struggle to suppress localized high voltage spikes caused by inconsistent dynamic characteristics of the thyristors. (Voltage change rate), and traditional distributed controllers are limited by the computing power bottleneck of online computing and lack a global rigid time synchronization mechanism, which inevitably leads to timing misalignment of control commands, causing series thyristors to face the risk of breakdown due to transient overvoltage.

[0189] To address the aforementioned technical issues, this embodiment provides a system architecture that integrates software and hardware.

[0190] like Figure 9 As shown, this embodiment provides a dynamic voltage equalization adaptive adjustment system for a series thyristor assembly, including: a mapping library establishment module, a parameter acquisition and mode switching module, a feature matching and strategy invocation module, a timestamp appending and distribution module, a local execution and adjustment module, and a monitoring and correction update module. Specifically:

[0191] First, the system includes a mapping library establishment module, which is used to construct an extreme transient scenario generation model based on power grid topology parameters and historical fault data, and generate a transient scenario set; perform electromagnetic transient simulation on each transient scenario in the transient scenario set, and extract the core feature vector of the scenario; calculate the optimal emergency control strategy of each flexible AC transmission system valve group under each transient scenario through a dual-objective collaborative optimization model, and establish a mapping library of scenarios and strategies.

[0192] In practical large-scale AC / DC hybrid power grid projects, this module is typically deployed in the high-performance computing (HPC) server cluster of the power grid dispatch center or in the advanced industrial control host at the station control layer. Because electromagnetic transient simulation and dual-objective collaborative optimization require enormous computing power, this module operates offline before the system is officially put into operation or after significant changes to the power grid structure. The generated mapping library data is distributed and stored in the static random access memory (SRAM) or double data rate synchronous dynamic random access memory (DDR4 SDRAM) of the field master control equipment to provide microsecond-level non-blocking data reading in the event of an emergency transient.

[0193] Secondly, the system includes a parameter acquisition and mode switching module, used to acquire the bus voltage and operating parameters of each flexible AC transmission system valve group in real time, obtain the transient voltage rise time, and automatically switch the control mode based on the transient voltage rise time. At the physical hardware level, this module relies on a wideband capacitive voltage transformer (CVT) and Rogowski coil installed on the primary high-voltage bus. The acquired analog electrical signals are converted into digital sequences by an analog-to-digital converter (ADC) with a high sampling rate (e.g., 100 MS / s) and input to the field-programmable gate array (FPGA) of the central controller. The differential operation unit inside the FPGA continuously calculates the slope of the voltage sequence to obtain the transient voltage rise time. When a sharp shortening of the voltage rise time is detected, indicating that the power grid is under impact, the module quickly switches the system from the conventional steady-state control mode to an emergency control mode to cope with extreme conditions through a hardware interrupt mechanism.

[0194] Next, the system includes a feature matching and strategy invocation module, which is used to switch to the first control mode when the transient voltage rise time is less than a first time threshold, extract the real-time transient core feature vector through a feature extraction algorithm, calculate the scene similarity between the real-time transient core feature vector and the scene core feature vectors of each scene, and invoke the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold.

[0195] In the actual device, the core of this module is a set of parallel digital signal processing (DSP) logic slices configured inside the main control FPGA. When the system enters the first control mode (i.e., when facing extremely fast transients), these DSP slices use multiply-accumulate units (MACs) to execute wavelet transform algorithms at high speed to extract real-time features. Subsequently, through a hardware comparison tree array, the real-time features are compared with pre-stored scene features read from SRAM to perform matrix distance calculation. This "feature extraction-similarity calculation-address addressing" process based on pure hardware logic can locate and retrieve the optimal emergency control strategy instruction frame within hundreds of nanoseconds, effectively avoiding the millisecond-level latency caused by traditional CPU online solution of differential equations.

[0196] Subsequently, the instructions need to be distributed. The system includes a timestamp appending and distribution module, which appends a globally synchronized timestamp to the invoked optimal emergency control strategy and distributes it to the local controller of the corresponding flexible AC transmission system valve group. In actual substations, this module is directly connected to the high-precision clock synchronization network within the substation (such as an IEEE 1588PTP precision time protocol switch based on a BeiDou / GPS dual-mode receiver). Before issuing control instructions, this module adds an absolute time stamp with nanosecond precision to the header of the data frame and broadcasts the timestamped strategy data packet to various valve base control units distributed in different physical locations via the substation's gigabit wide-area fiber optic ring network.

[0197] As the core terminal for executing actions in this system, the local execution and adjustment module is configured in the local controller. It is used to synchronously start the controllable negative pressure source module at a determined absolute trigger time based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy. Furthermore, it reconstructs the gate pulse based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, thereby achieving dynamic voltage equalization adjustment of the series thyristors. After receiving the fiber optic signal, the local controller (i.e., the valve-based electronic device, VBE) adds the parsed global synchronization timestamp to the pre-calculated timing offset provided by the strategy using its internal hardware timer, locking in a future absolute physical moment. When the local hardware clock reaches this absolute moment, the drive trigger circuit is instantly turned on.

[0198] In actual electrical circuits, the controllable negative voltage source module manifests as an insulated-gate bipolar transistor (IGBT) full-bridge high-frequency converter and energy storage capacitor connected in parallel with each stage of the thyristor. At the absolute trigger time, the IGBT conducts according to the microscopic time difference set by the gate pulse delay compensation matrix, forcibly injecting a reverse extraction current into the gate-cathode junction of the thyristor to compensate for the physical differences in carrier recombination velocities within different silicon wafers. Through this nanosecond-level multi-valve absolute time alignment and underlying nanosecond-level pulse reconstruction, all series-connected devices are forced to complete impedance transformation at the same macroscopic physical instant, thereby maintaining a balanced voltage distribution at the hardware physical level.

[0199] Finally, the system also includes a monitoring and correction update module, used to monitor the operating status of each thyristor in real time, input the feedback signals into the electrical stress prediction model for online correction of model parameters, and update the mapping library based on the correction results. In actual equipment operation, this module relies on the opto-isolation status monitoring circuit on the board of each thyristor level. The actual delay time of each thyristor switch and the bit error status of the cascaded communication link are sent to the data server of the station control layer via backhaul fiber. The machine learning algorithm (i.e., the electrical stress prediction model) running in the industrial host performs a health assessment on these status feedbacks. If it is found that a thyristor has a characteristic deviation due to thermal aging, the system will recalculate its delay compensation parameters and overwrite the updated parameters into the front-end SRAM mapping library, thereby giving the entire hardware system the ability to adaptively evolve and self-correct during long-term operation.

[0200] In summary, the dynamic voltage equalization adaptive regulation system for series thyristor components provided in this embodiment rationally divides the complex control algorithm with the specific substation hardware. By deploying the high-computing-power-requirement pre-simulation optimization module on an offline server, and embedding the high-time-sensitivity-requirement feature matching and synchronization execution module into the underlying FPGA hardware, and by introducing a rigid alignment mechanism based on absolute timestamps in command transmission, the system effectively solves the dynamic voltage equalization failure problem caused by computing power lag and communication asynchrony in existing technologies. Without altering the existing main primary power grid topology, this system architecture significantly improves the electrical safety and operational reliability of core equipment for flexible AC transmission under extreme high-voltage impacts, demonstrating high engineering application value.

[0201] Example 5:

[0202] Corresponding to the above embodiments, the present invention also proposes an electronic device.

[0203] like Figure 10The diagram shows a structural schematic of an electronic device according to the present invention. The electronic device 100 includes a processor 101 and a memory 103. The processor 101 and the memory 103 are connected, for example, via a bus 102. Optionally, the electronic device 100 may further include a transceiver 104. It should be noted that in practical applications, the transceiver 104 is not limited to one unit, and the structure of this electronic device 100 does not constitute a limitation on the embodiments of the present invention.

[0204] Processor 101 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA, or other programmable logic device, transistor logic device, hardware component, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 101 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0205] Bus 102 may include a pathway for transmitting information between the aforementioned components. Bus 102 may be a PCI bus or an EISA bus, etc. Bus 102 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 10 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0206] The memory 103 stores a computer program corresponding to the dynamic voltage equalization adaptive adjustment method for a series thyristor assembly according to the above embodiments of the present invention. This computer program is executed under the control of the processor 101. The processor 101 executes the computer program stored in the memory 103 to implement the content shown in the aforementioned method embodiments.

[0207] Among them, electronic devices 100 include, but are not limited to: mobile terminals such as laptops and PADs (tablet computers) and fixed terminals such as desktop computers. Figure 10 The electronic device 100 shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of the present invention.

[0208] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for dynamic voltage equalization and adaptive adjustment of a series thyristor assembly, characterized in that, include: An extreme transient scenario generation model is constructed based on power grid topology parameters and historical fault data to generate a transient scenario set; Electromagnetic transient simulations are performed on each transient scenario in the transient scenario set to extract the core feature vectors of the scenarios; the optimal emergency control strategy for each valve group of the flexible AC transmission system under each transient scenario is calculated through a dual-objective collaborative optimization model, and a mapping library between scenarios and strategies is established. Real-time acquisition of bus voltage and operating parameters of valve groups in each flexible AC transmission system; acquisition of transient voltage rise time; and automatic switching of control mode based on the transient voltage rise time. When the transient voltage rise time is less than the first time threshold, switch to the first control mode, extract the real-time transient core feature vector through the feature extraction algorithm; calculate the scene similarity between the real-time transient core feature vector and the scene core feature vector of each scene, and call the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold. A global synchronization timestamp is attached to the invoked optimal emergency control strategy and sent to the local controller of the corresponding flexible AC transmission system valve group; The local controller, based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, synchronously starts the controllable negative pressure source module at the determined absolute trigger time, and performs gate pulse reconstruction based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, thereby realizing dynamic voltage equalization regulation of the series thyristors. The operating status of each thyristor is monitored in real time, and the feedback signal is input into the electrical stress prediction model for online correction of model parameters. The mapping library is updated based on the correction results.

2. The method according to claim 1, characterized in that, The core feature vector of the scene includes: Transient voltage peak value, transient voltage rise time, transient voltage oscillation frequency, transient voltage polarity, and transient voltage initial phase angle.

3. The method according to claim 1, characterized in that, The objective function of the dual-objective collaborative optimization model is to minimize the system cost function; The system cost function is composed of a weighted sum of a dynamic equalization coefficient penalty term and a secondary overvoltage suppression penalty term; The dynamic voltage equalization coefficient penalty term is the product of the maximum dynamic voltage equalization coefficient of all flexible AC transmission system valve groups and the first weighting coefficient, and the secondary overvoltage suppression penalty term is the product of the maximum secondary overvoltage amplitude of all flexible AC transmission system valve groups and the second weighting coefficient.

4. The method according to claim 1, characterized in that, The automatic switching of control modes based on the transient voltage rise time includes: Obtain a second time threshold, wherein the second time threshold is greater than the first time threshold; When the transient voltage rise time is greater than or equal to the second time threshold, switch to the third control mode; When the transient voltage rise time is greater than or equal to the first time threshold and less than the second time threshold, the system switches to the second control mode.

5. The method according to claim 4, characterized in that, When switching to the second or third control mode, the following steps are performed: Valve clusters are defined based on the electrical coupling degree between valve groups in the flexible AC transmission system. With the optimization objectives of minimizing the dynamic pressure equalization coefficient within the valve cluster and minimizing the interactive interference between valve clusters, the particle swarm optimization algorithm is used to solve for the optimal phase control angle command and pressure equalization control priority. The local controller automatically switches between low-loss operation, active voltage equalization operation and emergency protection operation based on the transient overvoltage amplitude prediction value output by the electrical stress prediction model, and performs multi-level active energy management.

6. The method according to claim 1, characterized in that, The optimal emergency control strategy includes: The target flexible AC transmission system valve group's controllable negative pressure source output voltage amplitude, controllable negative pressure source output frequency, the pre-calculated timing offset, and the gate pulse delay compensation matrix.

7. The method according to claim 1, characterized in that, The local controller, based on the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, synchronously starts the controllable negative pressure source module at the determined absolute trigger time, including: Add the time value corresponding to the global synchronization timestamp to the pre-calculated timing offset to obtain the absolute trigger time for the emergency control of the valve group of the target flexible AC transmission system. When the absolute trigger time is reached, the controllable negative pressure source module and the gate drive unit of the target flexible AC transmission system valve group are synchronously triggered.

8. The method according to claim 1, characterized in that, The extraction of real-time transient core feature vectors through feature extraction algorithms includes: Obtain a real-time voltage sampling sequence containing multiple sampling points; The real-time voltage sampling sequence is convolved with a preset wavelet basis function to extract multiple real-time transient feature values. The real-time transient core feature vector is generated by combining multiple real-time transient feature values.

9. The method according to claim 1, characterized in that, The calculation of the scene similarity between the real-time transient core feature vector and each of the scene core feature vectors includes: Calculate the normalized difference between the real-time transient core feature vector and the scene core feature vector in each corresponding feature dimension; The weighted summation of the squares of the normalized differences for each of the corresponding feature dimensions yields the comprehensive difference index. The scene similarity is obtained by substituting the comprehensive difference index into the Gaussian kernel function and performing a negative exponentiation operation.

10. A dynamic voltage equalization adaptive adjustment system for a series thyristor assembly, characterized in that, include: The mapping library creation module is used to build extreme transient scenario generation models based on power grid topology parameters and historical fault data, and generate transient scenario sets. Electromagnetic transient simulations are performed on each transient scenario in the transient scenario set to extract the core feature vectors of the scenarios; the optimal emergency control strategy for each valve group of the flexible AC transmission system under each transient scenario is calculated through a dual-objective collaborative optimization model, and a mapping library between scenarios and strategies is established. The parameter acquisition and mode switching module is used to acquire the bus voltage and the operating parameters of each flexible AC transmission system valve group in real time, obtain the transient voltage rise time, and automatically switch the control mode according to the transient voltage rise time. The feature matching and strategy invocation module is used to switch to the first control mode when the transient voltage rise time is less than the first time threshold, extract the real-time transient core feature vector through the feature extraction algorithm, calculate the scene similarity between the real-time transient core feature vector and the scene core feature vector of each scene, and invoke the optimal emergency control strategy corresponding to the target transient scene that meets the scene similarity threshold. The timestamp appending and distribution module is used to append a globally synchronized timestamp to the invoked optimal emergency control strategy and distribute it to the local controller of the corresponding flexible AC transmission system valve group. The local execution and regulation module, configured in the local controller, is used to synchronously start the controllable negative pressure source module at the determined absolute trigger time according to the global synchronization timestamp and the pre-calculated timing offset in the optimal emergency control strategy, and to reconstruct the gate pulse based on the gate pulse delay compensation matrix included in the optimal emergency control strategy, so as to realize the dynamic equalization regulation of the series thyristors. The monitoring and correction update module is used to monitor the operating status of each thyristor in real time, input the feedback signal into the electrical stress prediction model to correct the model parameters online, and update the mapping library according to the correction results.

Images