Artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization method and system

Abstract

The present invention provides an artificial intelligence-driven carrier phase shift optimization method and system for cascaded, directly mounted SVGs. The method comprises the following steps: constructing a sensor input layer, a slow decision layer, and a fast execution layer; transmitting multidimensional grid state information collected by the sensor input layer to the slow decision layer and the fast execution layer; the slow decision layer outputting modulation control parameters based on a reinforcement learning neural network; and the fast execution layer outputting a modulation signal based on the modulation control parameters and the multidimensional grid state information, for driving a cascaded H-bridge module. The system has a simple structure and low control and construction costs. By constructing a coordinated fast and slow dual-layer architecture and incorporating a reinforcement learning neural network at the method level, the system achieves nonlinear fitting capabilities, enabling dynamic parameter adjustment and collaborative optimization for multiple dimensions and multiple objectives, thereby enhancing the stability of the cascaded structure and improving the overall performance of the SVG.

Description

Technical Field

[0001] The present invention relates to the field of power supply technology, and in particular to an artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method and system. Background Art

[0002] As the demand for dynamic power quality regulation in medium and high voltage power grids increases, the shortcomings of traditional control strategies and modulation methods in terms of dynamic adaptability, multi-parameter collaborative optimization, and real-time control efficiency are becoming increasingly prominent. The specific shortcomings are as follows:

[0003] 1. Traditional control strategies lack staticity and adaptability

[0004] Dynamic response hysteresis: Traditional SVG control relies on PID or fixed rules, and parameters must be manually preset. When the grid load suddenly changes or a fault occurs, adjustment lags occur, and the response time exceeds 10ms, resulting in reduced reactive compensation accuracy.

[0005] Limited harmonic suppression capability: In nonlinear load scenarios, it is difficult to track harmonic spectrum changes in real time, and the output current THD exceeds 5%, affecting power quality.

[0006] Poor robustness: Fixed control rules are prone to instability when faced with changes in grid impedance or voltage fluctuations, requiring frequent manual intervention.

[0007] 2. Challenges of CPS-SPWM modulation parameter fixation and coupling

[0008] Complex parameter coupling: Parameters such as phase shift angle and modulation ratio have nonlinear interactions, making it difficult for traditional empirical adjustment methods to balance multiple conflicts such as harmonic distribution and switching frequency.

[0009] Dynamic modulation failure: When the SVG output power changes rapidly, fixed parameters can easily lead to over-modulation or under-modulation, causing DC-side capacitor voltage fluctuations. In some scenarios, the amplitude can reach ±15% of the rated value.

[0010] Insufficient harmonic optimization: The fixed phase shift strategy cannot adaptively suppress specific subharmonics, increasing the filter design cost.

[0011] 3. The contradiction between computational complexity and real-time performance of multi-parameter collaborative optimization

[0012] High-dimensional search space: Cascading SVGs requires coordinating dozens of parameters, such as the H-bridge modulation wave phase and current loop gain. Traditional exhaustive or gradient descent methods take over 100ms for a single optimization, failing to meet real-time requirements.

[0013] Local optimal trap: Traditional optimization methods such as genetic algorithms are prone to falling into local optimality, resulting in reduced system efficiency;

[0014] Lack of online updates: Relying on offline optimization, it is impossible to dynamically adjust parameters based on real-time operating data such as temperature and device aging.

[0015] In addition to the problems with the above-mentioned traditional control methods, the cascaded direct-mounted structure also has unique technical difficulties, specifically: capacitor voltage balancing: The traditional sorting voltage balancing method relies on high-frequency switching, and switching losses account for more than 30% of the total system losses, with a high risk of voltage balancing failure in dynamic scenarios; poor module fault tolerance: When a single H-bridge fails, the control strategy cannot be reconfigured through parameter adaptiveness, resulting in reduced system availability; common-mode voltage and electromagnetic interference: The fixed modulation strategy exacerbates the common-mode voltage, causing electromagnetic compatibility issues, requiring additional hardware suppression measures. At the hardware level, there is also the problem of insufficient trade-off between system-level energy efficiency and lifespan, such as the loss-harmonic paradox: Reducing switching losses requires reducing the switching frequency, but this will exacerbate harmonics, and traditional methods cannot dynamically balance this; thermal stress accumulation: Under heavy load, the fixed parameter strategy causes excessive fluctuations in the IGBT junction temperature, shortening the device life; and capacitor aging is ignored: The voltage balancing strategy is not adjusted according to the capacitor's health status, accelerating the degradation of the capacitor's capacitance.

[0016] Therefore, in the face of the above problems, existing technologies have introduced artificial intelligence technology, which has developed rapidly in recent years. However, it also has corresponding application difficulties, such as strong data dependence: neural networks require a large amount of high-quality data, actual power grid fault data is scarce, and the model generalization ability is insufficient; another example is low online learning efficiency: traditional reinforcement learning algorithms converge slowly, and even require thousands of iterations, which is difficult to meet the millisecond-level decision-making requirements of SVG. Furthermore, the introduction of artificial intelligence technology also has the problem of lack of security and explanation: black box models may output over-limit control instructions, such as over-modulation problems, and it is difficult to pass the power grid compliance verification. Furthermore, although the simple mounting of artificial intelligence technology at the hardware level can solve some problems, it still faces the inherent factors and requirements of the scene, especially: (1) Strong coupling of multiple parameters: CPS-SPWM modulation ratio, carrier frequency, phase difference and other parameters have nonlinear interactions, which increases the difficulty of optimization. (2) Control delay constraints: From state acquisition to parameter update, it must be completed within 50μs, which requires extremely high real-time computing capabilities. (3) Memory limitations: The neural network model needs to be compressed into the DSP on-chip RAM, which restricts the model complexity and accuracy.

[0017] In summary, traditional methods have significant defects in dynamic adaptability, multi-parameter coupling optimization, real-time computing efficiency and system-level multi-objective trade-offs. There is an urgent need to achieve dynamic optimization of cascaded SVG modulation parameters and improve system performance through improved control methods and systems. Summary of the Invention

[0018] In response to the problems existing in the prior art, the present invention provides a system with simple structure and low cost. The method can be combined with a neural network of reinforcement learning to achieve dynamic optimization of cascaded SVG modulation parameters and improve system performance.

[0019] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0020] In one aspect, the present invention provides an artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method, which mainly includes the following steps:

[0021] Construct sensor input layer, slow decision layer and fast execution layer;

[0022] The sensor input layer collects multi-dimensional grid state information and transmits the multi-dimensional grid state information to the slow decision layer and the fast execution layer;

[0023] Constructing a strategy mathematical model for the slow decision layer, and training and deploying a reinforcement learning neural network based on the strategy network in the strategy mathematical model, wherein the reinforcement learning neural network outputs a modulation control parameter based on the multi-dimensional power grid state information;

[0024] A signal generation mathematical model is constructed for the fast execution layer, and the signal generation mathematical model outputs a modulation signal based on the modulation control parameter and the multi-dimensional grid state information, so as to drive a cascaded H-bridge module.

[0025] Optionally, the following steps are also included:

[0026] Updating the reinforcement learning neural network based on knowledge distillation technology, and lightweighting the reinforcement learning neural network model through dynamic weights;

[0027] The dynamic weight adjustment mechanism is based on the training gradient amplitude of the reinforcement learning neural network, and the model compression adopts an iterative pruning algorithm.

[0028] Optionally, the training of the reinforcement learning neural network includes the following steps:

[0029] Use full-precision training on the server side to generate a full-precision training model and optimize the multi-objective loss function;

[0030] The full-precision training model simulates the inference error of the embedded device through pseudo-quantization nodes, and maintains full-precision training for the quantization-sensitive layers in the model to obtain a reinforcement learning neural network after quantization-aware training.

[0031] Optionally, the deployment of the reinforcement learning neural network includes the following steps:

[0032] Before deployment, the reinforcement learning neural network is quantized to 8-bit fixed-point operations, and the critical path retains floating-point calculations;

[0033] After deployment, the reinforcement learning neural network accelerates matrix operations through a coprocessor.

[0034] Optionally, the multi-dimensional grid state information includes total harmonic distortion, DC bus voltage change rate, thermal stress coefficient and grid frequency offset;

[0035] The modulation control parameters include a modulation ratio and a carrier frequency.

[0036] Optionally, the calculation of the modulation signal includes the following steps:

[0037] Based on the multi-dimensional grid state information and the modulation control parameters, solving the carrier phase value by a direct digital frequency synthesis algorithm;

[0038] Mapping the carrier phase value to a triangular wave lookup table to obtain a carrier waveform, and outputting a carrier signal;

[0039] The phase difference is calculated by a THD real-time algorithm, and a multi-channel phase difference is generated by a nonlinear algorithm; each channel of the multi-channel phase difference corresponds to one channel of the cascaded H-bridge module;

[0040] Based on the carrier signal and the multi-path phase difference, a plurality of pairs of complementary PWM signals are calculated by a modulation signal generation algorithm;

[0041] The multiple pairs of complementary PWM signals are the modulation signals.

[0042] Optionally, the THD real-time algorithm includes a sliding window Fourier transform algorithm, and the sliding window Fourier transform algorithm is used to calculate the total harmonic distortion rate of the current power grid system in real time.

[0043] On the other hand, the present invention also provides an artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization system, which is used to implement the above method;

[0044] The system includes a sensor input layer, a slow decision layer, and a fast execution layer;

[0045] The sensor input layer is provided with multiple types of sensors for collecting current power grid status information; the sensor input layer is connected to the slow decision layer through an analog-to-digital conversion unit;

[0046] The slow decision layer is provided with a digital signal processor, and the slow decision layer is deployed with a reinforcement learning neural network; the slow decision layer is also connected to the fast execution layer;

[0047] The fast execution layer is provided with a field programmable gate array, and the fast execution layer is connected to the cascade H-bridge module of the current power grid;

[0048] The digital signal processor and the field programmable gate array form a collaborative architecture and are connected via a data bus.

[0049] Optionally, a server is further included, the server being connected to the slow decision layer and configured to deploy and online update the reinforcement learning neural network;

[0050] The slow decision layer is also configured with a coprocessor for accelerating calculations.

[0051] Optionally, the fast execution layer is further configured with a multi-phase DDS generator, a real-time THD calculation unit, a phase difference adjustment module and a PWM generation module;

[0052] The multi-phase DDS generator is used to independently generate carrier signals of different phases;

[0053] The real-time THD calculation unit is used to calculate the phase difference;

[0054] The phase difference adjustment module is used to generate multiple phase differences, and each of the multiple phase differences corresponds to one of the cascaded H-bridge modules;

[0055] The PWM generation module is used to generate a PWM signal;

[0056] A triangle wave lookup table is stored in the block random access memory of the field programmable gate array.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] The system structure of the present invention is simple and the control and construction costs are low. By constructing a fast and slow double-layer coordinated architecture and combining it with a reinforcement learning neural network at the method level, it achieves nonlinear fitting capabilities, can perform dynamic parameter adjustment and collaborative optimization for multiple dimensions and multiple objectives, thereby improving the stability of the cascade structure and enhancing the overall performance of SVG. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0060] Figure 1 This is a system architecture diagram in a specific embodiment of the present invention;

[0061] Figure 2 This is a flow chart of the pre-training and deployment phases in a specific embodiment of the present invention;

[0062] Figure 3 This is a comparison of THD recovery curves under load mutation between the solution of the present invention and the traditional solution in comparative example 1;

[0063] Figure 4 This is a comparison chart of switching losses between the solution of the present invention and the traditional solution under sudden load changes in comparative example 1;

[0064] Figure 5 This is a comparison chart of the dynamic response time under load mutation of the solution of the present invention and the traditional solution in comparative example 1. DETAILED DESCRIPTION

[0065] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0066] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings.

[0067] In the description of the present invention, “plurality” means two or more, unless otherwise clearly defined.

[0068] It is worth noting that the methods used in the present invention are all conventional methods unless otherwise specified; the raw materials and devices used are all conventional commercially available products, and their sources are not specifically limited unless otherwise specified.

[0069] On the one hand, this embodiment provides an artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method, which constructs a sensor input layer, a slow decision layer, and a fast execution layer.

[0070] Among them, Figure 1 As shown, the sensor input layer is built based on existing detection sensors, connecting voltage / current sensors, multiple temperature sensors, grid frequency detection sensors, and DC bus voltage sensors to other layers. Specifically, the sensor input layer collects the aforementioned multidimensional grid status information; correspondingly, this multidimensional grid status information includes total harmonic distortion, DC bus voltage change rate, thermal stress coefficient, and grid frequency offset. This multidimensional grid status information is then transmitted to the slow decision layer and the fast execution layer.

[0071] The slow decision layer is deployed with a reinforcement learning neural network, which outputs modulation control parameters based on multi-dimensional grid state information.

[0072] In order to formally describe the control decision-making process of the cascaded SVG system at this layer, a theoretical framework needs to be constructed. Its function is to convert the operating state, control strategy, and optimization goal of the physical system into computable mathematical expressions. It has never been possible to standardize the input and output of the reinforcement learning neural network model and set constraints that meet the requirements. Therefore, a strategic mathematical model is constructed for the slow decision-making layer, including:

[0073] Establish state observation phasors based on multi-dimensional power grid state information , the formula is:

[0074] ;

[0075] Where THD is the total harmonic distortion (THD), is the DC bus voltage change rate, is the thermal stress coefficient, is the grid frequency offset, is a 4-dimensional set of real numbers.

[0076] The strategy network mapping is established based on the state observation phasor, which is expressed as:

[0077] ;

[0078] Where, is a parameterized policy network, is the network weight, is the action space constraint, m is the modulation ratio, is the carrier signal frequency, are all carrier phase offsets between cascaded H-bridge modules. Furthermore, cascaded SVG is usually composed of multiple H-bridge modules connected in series. For example, in this embodiment, a 7-way design is taken as an example, which includes a 7-level structure, corresponding to 7 H-bridges. The core of CPS-SPWM is to cancel each other out the harmonics of the total output voltage through the orderly offset of the carrier phase of each H-bridge. For N cascaded modules, the traditional method requires setting N-1 independent phase differences (phase offsets between adjacent modules), but in this embodiment, the independent variables are simplified to 2 through a symmetrical grouping strategy. Specifically, the symmetrical grouping strategy divides the 7 H-bridges into two groups, such as the first 3 groups and the last 3 groups, and the middle group is used as a benchmark. The modules in each group share the same base phase difference. , reducing independent parameters through symmetry.

[0079] State observation vector in mathematical modeling It is the model input of the reinforcement learning neural network, which clarifies the physical quantities that need to be collected. It is the output range of the model, which constrains the feasible solution space of the parameters.

[0080] The policy network mapping serves as the standard decision-making logic structure, employing a reinforcement learning neural network as the execution structure. The mathematical form of the policy network, such as the number of neural network layers and activation function, determines the model architecture. For example, a three-layer fully connected network with a 4-dimensional input, 32-dimensional, 16-dimensional, and 4-dimensional output is used. Combined with the ReLU activation function, this achieves a nonlinear mapping from state to parameters. Constrained optimization objectives: Multiple objectives in mathematical modeling, such as THD minimization and loss balancing, are achieved through model training. For example, the loss function in online transfer learning directly corresponds to the optimization objective in the mathematical modeling, ensuring that the model output parameters meet the dual requirements of harmonic suppression and efficiency improvement.

[0081] Therefore, if Figure 2 As shown in Figure 2, the training and deployment of a reinforcement learning neural network consists of the following steps:

[0082] First, during the pre-training phase, the server uses full-precision training to generate a full-precision training model and optimizes a multi-objective loss function. That is, initial model training is completed at FP32 precision, while simultaneously optimizing the multi-objective loss function. Specifically, the model training adopts a mixed-precision training strategy, using a GPU / CPU cluster to train the model at FP32 full precision during the server-side pre-training phase, optimizing the multi-objective loss function including THD and loss.

[0083] Before deployment, the reinforcement learning neural network is quantized to 8-bit fixed-point operations, and floating-point calculations are retained in the critical path. Optionally, the full-precision training model simulates the inference error of the embedded device through pseudo-quantization nodes, and the quantization-sensitive layers in the model are trained with full precision to obtain the reinforcement learning neural network after quantization-aware training. Specifically, pseudo-quantization nodes are then inserted for quantization-aware training to simulate INT8 inference error, while quantization-sensitive layers such as the policy network output layer are trained with FP32 precision. The FP32 weights are then converted to INT8 and verified to ensure that the THD error is less than 2%.

[0084] After deployment, the reinforcement learning neural network accelerates matrix operations through the coprocessor, specifically using the CLA coprocessor to accelerate INT8 inference, improves computing efficiency through fixed-point format data and dynamic quantization scaling, and maintains FP32 fixed-point operations for key modules such as thermal stress prediction at the next level. The quantization parameters are stored in BRAM and support dynamic precision switching. Finally, different quantization models such as high-frequency dynamics and thermal safety are dynamically loaded through code according to working conditions such as the grid frequency change rate and thermal stress gradient, realizing deep coordinated optimization of control strategy and hardware architecture. The reinforcement learning neural network model meets the RAM ≤ 256KB storage limit and ≤ 35μs inference delay in the slow decision-making layer hardware, while ensuring THD error < 2% and strategy similarity of 91.3%.

[0085] In this embodiment, in order to improve the accuracy and adaptability of the model to multiple working conditions, an online transfer learning mechanism is introduced. The server dynamically adjusts the student network through the knowledge distillation framework, as follows:

[0086] Based on the knowledge distillation technology, the reinforcement learning neural network is updated and the model is lightweighted through dynamic weights. A lightweight knowledge distillation framework is constructed during training. The loss function consists of a Q-value matching term (mean square error) to ensure the consistency of the value function and a strategy distribution alignment term to retain the diversity of the teacher's strategy decision. The formula is as follows:

[0087] ;

[0088] Where, are the Q-value functions of the teacher and student networks respectively; is the strategy distribution of teachers and students, that is, in the state s Next select action probability; is the dynamic weight coefficient; is the Jensen-Shannon divergence, which is used to measure the difference in strategy distribution; State-Action Distribution D Find the expectation, that is, the statistical average based on historical data or real-time sampling data.

[0089] Dynamic Weight The formula is:

[0090] ;

[0091] Where, For the Sigmoid function, the input value is compressed to [0,1], ensuring Dynamic changes within the effective range; τA sliding time window, such as a typical value of 100ms, is used to calculate the past τ The average gradient amplitude over time avoids the influence of instantaneous noise and smoothes the weight adjustment process; Network parameters for students θ The gradient amplitude reflects the current learning difficulty. The larger the gradient, the more difficult the learning.

[0092] The dynamic weight adjustment mechanism is based on the training gradient amplitude of the reinforcement learning neural network. In the dynamic weight, when the gradient amplitude of the student network increases (learning difficulty), the Q value item weight is automatically increased. , strengthen supervised learning. In the mathematical property verification, Lyapunov stability analysis proves that dynamic weight adjustment can ensure:

[0093] ;

[0094] Where, is a non-negative constant.

[0095] During knowledge distillation training, dynamic weights are adjusted in real time based on the student network's learning progress. Lyapunov stability analysis theoretically demonstrates that under this dynamic adjustment mechanism, the student network's training process converges without oscillation or divergence, thus ensuring the effectiveness and stability of model training.

[0096] Therefore, in the optimization of the multi-objective loss function, THD and loss are minimized simultaneously through gradient descent, achieving a balance between "harmonic suppression" and "energy efficiency optimization".

[0097] The weight is adjusted according to the gradient amplitude of the student network, and the weight of the Q-value item is increased to strengthen supervised learning when learning is difficult. Among them, the model compression adopts an iterative pruning algorithm, that is, iterative amplitude pruning is used to make the number of student model parameters ≤ 30% of the teacher model to meet the lightweight constraint. Among them, in this embodiment, in order to achieve lightweight design, it is also necessary to establish corresponding hardware resource constraints to ensure that the student network can run efficiently on embedded hardware and balance model performance with hardware resource limitations, such as computing speed, storage capacity, etc. Its core includes model complexity constraints and computational delay constraints, and the formula is as follows:

[0098] Model complexity constraints: ;

[0099] Calculate the delay constraint: ;

[0100] Where, is the parameter set of the student network, is the parameter set of the teacher network; It is the single inference time of the model, which is the computation time from input state to output action.

[0101] In model compression, we construct a sparse constraint and use iterative magnitude pruning to satisfy constraint:

[0102] ;

[0103] Where, is a dynamic threshold, which increases in each round; is the mask matrix, are the elements of the mask matrix.

[0104] The above method steps target the reinforcement learning neural network in the slow decision layer, and together form a complete chain of "model training → deployment adaptation → real-time optimization". The core is to achieve efficient operation of the reinforcement learning strategy on resource-constrained embedded platforms through a combination of static compression before deployment and dynamic tuning through online migration after deployment, ultimately achieving performance improvements such as reduced THD and switching loss.

[0105] As can be seen from the above, the modulation control parameters output by the slow decision layer in this embodiment include parameters such as modulation ratio and carrier frequency, and are input into the fast execution layer together with the aforementioned multi-dimensional grid status information.

[0106] Since the slow decision layer and the fast execution layer are a collaborative architecture, a corresponding signal generation mathematical model is constructed in the fast execution layer. This model is a high-frequency dynamic optimization model that focuses on quickly adjusting the aforementioned phase difference parameters of CPS-SPWM modulation to achieve real-time harmonic suppression.

[0107] Furthermore, the modulation signal generation function as follows:

[0108] ;

[0109] In the formula, input: modulation parameters Output: 7 pairs of complementary PWM signals, a 7x2 binary matrix. This function converts the continuous modulation parameters into a discrete PWM pulse train, achieving harmonic optimization through carrier phase shifting.

[0110] Therefore, the calculation of the modulation signal includes the following steps:

[0111] (1) Carrier phase generation;

[0112] Based on the modulation control parameters and multi-dimensional grid status information output by the upper layer, the carrier phase is generated through a multi-phase direct digital frequency synthesis (DDS) algorithm: a 17-bit phase accumulator is adapted, a 200MHz clock is used, and 2000 phase accumulations are completed within a 10μs fast adjustment layer update cycle to ensure phase adjustment resolution.

[0113] The high 7 bits are extracted from the 17-bit accumulator as the base phase and combined with the offset to generate 7 sets of independent carrier phase offsets. Each path corresponds to an H-bridge module to achieve precise phase shifting between adjacent carriers.

[0114] (2) Carrier signal generation;

[0115] The carrier phase value of each H-bridge is mapped to a 12-bit resolution triangle wave lookup table, which is pre-stored in the BRAM of the fast execution layer. The carrier waveform is generated by looking up the table through the phase offset index; wherein, the carrier frequency Given by the output of the upper strategy network, the phase accumulation step is based on Dynamic adjustment.

[0116] Output 7 carrier signals, and the phase difference of each signal is bound to the hardware channel of the corresponding H-bridge module one by one.

[0117] (3) Real-time THD calculation and phase difference feedback;

[0118] Synchronously sample the voltage / current signal, use the fast Fourier transform (FFT) algorithm to perform harmonic analysis on the waveform data in the sliding window, and calculate the amplitude of each harmonic. The formula is:

[0119] ;

[0120] Where, is the real-time THD value, For the time window [ t-τ,t ] in the waveform data.

[0121] The THD calculation results are fed back to the upper layer as the input of the reinforcement learning neural network. The phase difference parameters are generated by the policy network and dynamically adjusted through multi-objective optimization.

[0122] (4) PWM signal generation;

[0123] Based on the carrier signal and the modulation ratio of the output, a unipolar PWM signal is generated by comparing the modulation wave with the carrier wave. When the modulation wave amplitude is greater than the carrier wave amplitude, a high level (1) is output, otherwise a low level (0) is output. A pair of complementary PWM signals is generated for each H-bridge, i.e., the upper tube / lower tube drive. Hardware dead time is inserted to avoid direct conduction of the bridge arm. Finally, 7 pairs of complementary PWM signals are output to drive the cascaded H-bridge module.

[0124] On the other hand, this embodiment also provides an artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization system, which is used to implement the aforementioned artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method.

[0125] The system of this embodiment includes a sensor input layer, a slow decision layer, and a fast execution layer;

[0126] The sensor input layer is equipped with multiple types of sensors to collect current power grid status information; the sensor input layer is connected to the slow decision layer through an analog-to-digital conversion unit;

[0127] The slow decision layer includes a digital signal processor (DSP) with a reinforcement learning neural network deployed. The slow decision layer is also connected to the fast execution layer. Optionally, the DSP in this embodiment uses a TMS320F28337 with a built-in 16-bit ADC and a sampling rate of 1MSPS, which synchronously collects voltage, current, temperature, grid frequency, and DC bus voltage. This layer also integrates a 50MHz CLA coprocessor, which performs 16×16 fixed-point multiplication and accumulation in a single cycle. On-chip RAM is set to ≤ 256KB for storing lightweight neural network models. Furthermore, the DSP also has a built-in ADC for synchronously collecting multi-sensor signals, with a sampling rate of 1MHz and synchronized with the underlying clock domain.

[0128] The fast execution layer features a field-programmable gate array (FPGA) connected to the cascaded H-bridge modules of the current power grid. The FPGA can optionally use an XC7K70T processor operating in the 200MHz clock domain, with 58% LUT resource usage and 72% BRAM resource usage. The BRAM stores a 12-bit resolution triangle wave lookup table and quantization parameters, and uses a DSP48E1 slice for floating-point operations. Furthermore, the FPGA implements phase generation using Verilog. A 17-bit phase accumulator uses a sign-extended 16-bit input phase difference to output seven phase offsets. The BRAM stores the triangle wave lookup table to generate seven carrier signals. After modulated-carrier comparison, seven pairs of complementary PWM signals are output, each driving the upper and lower arms of an H-bridge module with a 2μs dead time inserted. The FPGA also generates carrier phase using multiphase DDS technology. A 128-point FFT is implemented using the DSP48E1 slice to calculate the total distortion (THD) with a 50μs sliding window. The results are transmitted back to the DSP via GPIO.

[0129] The fast execution layer's FPGA also includes a multiphase DDS generator, a real-time THD calculation unit, a phase difference adjustment module, and a PWM generation module. The multiphase DDS generator independently generates carrier signals of varying phases; the real-time THD calculation unit calculates phase differences; the phase difference adjustment module generates multiple phase differences, each corresponding to a phase of the cascaded H-bridge module; and the PWM generation module generates PWM signals.

[0130] Optionally, the FPGA also integrates a hardware protection module that directly shuts down the PWM output within a preset time when overcurrent or overtemperature is detected.

[0131] DSP+FPGA form a collaborative architecture and are connected through the EMIF bus to achieve a bandwidth of 100MB / s for interactive modulation parameters and status data, supplemented by a 10MHz SPI communication rate to transmit configuration instructions.

[0132] The system also includes a server on-premises or in the cloud, which is connected to the slow decision layer and is used to deploy and / or update the reinforcement learning neural network online.

[0133] Comparative Example 1;

[0134] In a specific embodiment, a system using the method of the present invention and a corresponding system using the traditional method are simulated for the same power grid environment, thereby forming a comparison between the solution of the present invention and the traditional solution. The simulation simulates a sudden load change, with the time uniformly set at 0.5ms and a 50%→100% step change. The THD recovery performance is then tested using the simulation results, as shown in the following example. Figure 3 and as shown in Table 1 below.

[0135] Table 1

[0136]

[0137] The graphs and tables show that, for the conventional solution, the THD peak reaches 5.0% at 0.7ms after the mutation, and it takes 2.1ms to recover to 3%. For the solution of the present invention, the THD peak reaches 3.5% at 0.8ms after the mutation, and it takes 0.8ms to recover to 2.8%. This demonstrates that the solution of the present invention has extremely strong harmonic suppression capabilities and dynamic response speed, significantly exceeding those of the conventional solution, achieving a 30% reduction in THD peak and a 61.9% reduction in recovery time.

[0138] Furthermore, the switching loss and response time are tested in the simulation to form Figure 4 、 Figure 5 and Table 2 below.

[0139] Table 2

[0140]

[0141] The graphs and tables show that, compared to the traditional solution, the proposed solution reduces dynamic losses by 18.3% and peak losses by 53%. This demonstrates the proposed solution's dynamic carrier frequency regulation and loss suppression during sudden load changes. Furthermore, through FPGA hardware acceleration, the proposed solution reduces dynamic response time from 120μs to 28μs, a 76.7% reduction.

[0142] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Simple modifications or equivalent substitutions of the technical solution of the present invention by ordinary technicians in this field do not deviate from the essence and scope of the technical solution of the present invention.

Claims

1. An artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method, characterized by: The steps include: Construct sensor input layer, slow decision layer and fast execution layer; The sensor input layer collects multi-dimensional grid state information and transmits the multi-dimensional grid state information to the slow decision layer and the fast execution layer; The multi-dimensional grid status information includes total harmonic distortion, DC bus voltage change rate, thermal stress coefficient and grid frequency offset; Constructing a strategy mathematical model for the slow decision layer, and training and deploying a reinforcement learning neural network based on the strategy network in the strategy mathematical model, wherein the reinforcement learning neural network outputs a modulation control parameter based on the multi-dimensional power grid state information; The modulation control parameters include modulation ratio and carrier frequency; A signal generation mathematical model is constructed for the fast execution layer. The signal generation mathematical model outputs a modulation signal based on the modulation control parameter and the multi-dimensional grid state information to drive the cascaded H-bridge module. The calculation of the modulation signal includes the following steps: Based on the multi-dimensional grid state information and the modulation control parameters, solving the carrier phase value by a direct digital frequency synthesis algorithm; Mapping the carrier phase value to a triangular wave lookup table to obtain a carrier waveform, and outputting a carrier signal; The phase difference is calculated by a THD real-time algorithm, and a multi-channel phase difference is generated by a nonlinear algorithm; each channel of the multi-channel phase difference corresponds to one channel of the cascaded H-bridge module; Based on the carrier signal and the multi-path phase difference, a modulation signal generation algorithm is used to calculate and obtain multiple pairs of complementary PWM signals; The multiple pairs of complementary PWM signals are the modulation signals.

2. The artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method according to claim 1 is characterized by: The following steps are also included: Updating the reinforcement learning neural network based on knowledge distillation technology, and lightweighting the reinforcement learning neural network model through dynamic weights; The dynamic weight adjustment mechanism is based on the training gradient amplitude of the reinforcement learning neural network, and the model compression adopts an iterative pruning algorithm.

3. The artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method according to claim 1 is characterized by: The training of the reinforcement learning neural network includes the following steps: Use full-precision training on the server side to generate a full-precision training model and optimize the multi-objective loss function; The full-precision training model simulates the inference error of the embedded device through pseudo-quantization nodes, and maintains full-precision training for the quantization-sensitive layers in the model to obtain a reinforcement learning neural network after quantization-aware training.

4. The artificial intelligence-driven cascaded direct-mounted SVG carrier phase shift optimization method according to claim 3 is characterized by: The deployment of the reinforcement learning neural network includes the following steps: Before deployment, the reinforcement learning neural network is quantized to 8-bit fixed-point operations, and the critical path retains floating-point calculations; After deployment, the reinforcement learning neural network accelerates matrix operations through a coprocessor.

5. The artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization method according to claim 1 is characterized by: The THD real-time algorithm includes a sliding window Fourier transform algorithm, which is used to calculate the total harmonic distortion rate of the current power grid system in real time.

6. An artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization system, characterized by: Used to implement the artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization method according to any one of claims 1 to 5; The system includes a sensor input layer, a slow decision layer, and a fast execution layer; The sensor input layer is provided with multiple types of sensors for collecting current power grid status information; the sensor input layer is connected to the slow decision layer through an analog-to-digital conversion unit; The slow decision layer is provided with a digital signal processor, and the slow decision layer is deployed with a reinforcement learning neural network; the slow decision layer is also connected to the fast execution layer; The fast execution layer is provided with a field programmable gate array, and the fast execution layer is connected to the cascade H-bridge module of the current power grid; The digital signal processor and the field programmable gate array form a collaborative architecture and are connected via a data bus.

7. The artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization system according to claim 6 is characterized by: Also included is a server, connected to the slow decision layer, for deploying and online updating the reinforcement learning neural network; The slow decision layer is also configured with a coprocessor for accelerating calculations.

8. The artificial intelligence driven cascaded direct-mounted SVG carrier phase shift optimization system according to claim 6 is characterized by: The fast execution layer is also configured with a multi-phase DDS generator, a real-time THD calculation unit, a phase difference adjustment module and a PWM generation module; The multi-phase DDS generator is used to independently generate carrier signals of different phases; The real-time THD calculation unit is used to calculate the phase difference; The phase difference adjustment module is used to generate multiple phase differences, and each of the multiple phase differences corresponds to one of the cascaded H-bridge modules; The PWM generation module is used to generate a PWM signal; A triangle wave lookup table is stored in the block random access memory of the field programmable gate array.

Images