Fault-tolerant control method and system for a three-level inverter

Abstract

The application discloses a kind of three-level inverter fault-tolerant control method and system, belong to power electronic control technical field.The application is inquired in the operation condition-weight mapping table of all operation conditions of three-level inverter covered by pre-collection, as candidate weight triplet, the optimal weight triplet corresponding to the actual operation condition of three-level inverter;Then based on different real-time monitoring index, different weight in candidate weight triplet is dynamically revised, and thus the optimal weight triplet matching actual operation condition is obtained.The application determines a better search seed based on prior knowledge in advance, and dynamically revises according to actual monitoring index based on this, avoids artificial repeated trial and error, significantly shortens the debugging cycle, reduces cost, and is suitable for different operation conditions, and still can guarantee global optimization under different operation conditions, and the robustness of fault-tolerant control is stronger, and precision is higher.

Description

Technical Field

[0001] This invention belongs to the field of power electronic control technology, and more specifically, relates to a fault-tolerant control method and system for a three-level inverter. Background Technology

[0002] The overexploitation of fossil fuels has exacerbated the energy crisis and environmental degradation, making new energy power generation technologies, such as photovoltaics and wind power, the core direction for future energy structure transformation. Three-level inverters, with their advantages of low output harmonics, low switching losses, and high bus utilization, have become a key component in new energy power conversion. Their fault-tolerant control quality directly affects energy conversion efficiency, power quality, and system safety; therefore, researching a fault-tolerant control method for three-level inverters is of great significance.

[0003] In existing technologies, fixed-weight model predictive control (MPC) is commonly used for fault-tolerant control of three-level inverters. However, under multiple operating conditions such as grid-connected / off-grid dual-mode, wide-range load, and device parameter drift, traditional fixed-weight model predictive control faces the challenge of multi-objective coupling, including current tracking, midpoint potential, common-mode voltage, and switching losses. Different operating conditions have significantly different emphases on various performance aspects, and manual adjustment of weights requires repeated trial and error, resulting in long debugging cycles, high costs, and difficulty in guaranteeing global optimization, leading to poor robustness and accuracy of fault-tolerant control. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a fault-tolerant control method and system for a three-level inverter, which solves the technical problems of the existing technology, which relies on manual adjustment of weights under different operating conditions, resulting in long debugging cycles, high costs, and difficulty in guaranteeing global optimization, thus leading to poor robustness and accuracy of fault-tolerant control.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a fault-tolerant control method for a three-level inverter, comprising: performing a fault-tolerant operation on the three-level inverter; wherein the fault-tolerant operation performed in the k-th control cycle includes: Obtain the candidate weight triplet corresponding to the k-th control cycle. , cost function Weights in , , Each one is set as a candidate weight. , , The MPC algorithm is used to solve the three-phase level switching state in the k-th control cycle with the goal of minimizing the cost value, so as to perform fault-tolerant control on the three-level inverter. Where k is a positive integer; This is the index of the control window corresponding to the k-th control cycle; k is a positive integer. This is the index of the control window corresponding to the k-th control cycle; , , , These correspond to the tracking error of the three-level inverter output current, the average common-mode voltage of the three-level inverter output, and the switching overhead, respectively, in the k-th control cycle. ; ; x It can be a, b, or c; For the k-th control cycle x Phase level switch state; During the fault-tolerant operation of the three-level inverter, each M control cycles constitute a control window, and weight adjustment operations are continuously performed within each control window; M is a positive integer. The weight adjustment operation performed in the t-th control window includes: obtaining the current operating condition of the three-level inverter, and querying the optimal weight triplet corresponding to the current operating condition from the pre-established operating condition-weight mapping table, as a candidate weight triplet for the t-th control window. In the last control cycle of the t-th control window, calculate the harmonic distortion rate of the current output current of the three-level inverter, the DC bus capacitor voltage difference, and the average common-mode voltage of the three-level inverter output, and denote them accordingly as follows: , , ;when or When the corresponding preset threshold is exceeded, increase ;when When the corresponding preset threshold is exceeded, increase ; The mapping table mentioned above covers all operating conditions of the three-level inverter, including the optimal weight triplet corresponding to each operating condition of the three-level inverter collected in advance; all operating conditions cover all combinations of different operating modes, different load states and different operating states of the three-level inverter.

[0006] More preferably, the optimal weight triplet corresponding to each operating condition s in the above mapping table To incorporate the output current, output voltage, three-phase level switching states, upper and lower bus capacitor voltages, and DC bus voltage of the three-level inverter under operating condition s into the model predictive control cost function The weight triplet calculated to minimize the predictive control cost of the model. ; in, This is a preset baseline coefficient; Preset operating mode coefficients; Preset load state coefficient; Preset operating state coefficients; For the average tracking error The normalization result; ; N The preset number of control cycles; For preset control cycle index; The preset reference current amplitude; Average DC bus capacitor voltage difference The normalization result; ; This is the static DC bus voltage; This represents the overall average common-mode voltage. ; This represents the average switching overhead.

[0007] More preferably, the optimal weight triplet corresponding to each operating condition s in the above mapping table To calculate the weighted triplet that minimizes the model predictive control cost by substituting the output current, output voltage, three-phase level switching states, DC bus voltage, and upper and lower bus capacitor voltages of the three-level inverter under operating condition s into the model predictive control cost function under constraints. ; The constraints include: Hard constraints include: the harmonic distortion rate of the three-level inverter output current in the k-th control cycle is less than the corresponding preset threshold. The soft constraint condition includes: the average common-mode voltage of the three-level inverter output in the k-th control cycle is less than the corresponding preset threshold.

[0008] More preferably, the optimal weight triplet corresponding to each operating condition s in the above mapping table It was calculated using the Bayesian optimization algorithm.

[0009] More preferably, in the above Bayesian optimization algorithm, the initial search space of the weight triples is obtained by Latin hypercube sampling from the preset search space of the weight triples. The data was obtained through sampling; the probabilistic surrogate model is a Gaussian process; the acquisition function is the EI acquisition function. Among them, the preset search space for:

[0010] and Weights The minimum and maximum values; and Weights The minimum and maximum values; and Weights The minimum and maximum values.

[0011] More preferably, the tracking error of the three-level inverter output current in the kth control cycle is... DC bus capacitor voltage difference under the kth control cycle The average common-mode voltage of the three-level inverter output during the kth control cycle. They are respectively:

[0012]

[0013]

[0014] in, This is the dynamic reference current in the kth control cycle; This represents the output current of the three-level inverter during the kth control cycle. This is the dynamic reference current for phase x in the kth control cycle; Let x be the x-phase output current of the three-level inverter in the kth control cycle; and These are the upper and lower bus capacitor voltages under the kth control cycle, respectively; ; Let x be the x-phase output voltage of the three-level inverter in the kth control cycle; This represents the dynamic DC bus voltage during the k-th control cycle.

[0015] More preferably, the method for determining the current operating condition of the three-level inverter includes: An islanding detection algorithm is used to detect the current operating mode of the three-level inverter; the operating modes include: grid-connected mode and off-grid mode. Calculate the current output current value of the three-level inverter Calculate the current load rate ;like If the load rate is less than the preset load rate, the three-level inverter is determined to be in a light load state; otherwise, it is in a heavy load state. Rated current; A fault diagnosis algorithm is used to detect the current operating status of the three-level inverter.

[0016] In a second aspect, the present invention provides a three-level inverter system, comprising: a three-level inverter and a controller; The controller is used to execute the fault-tolerant control method provided in the first aspect of the present invention.

[0017] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory stores a computer program, and the processor executes the fault-tolerant control method provided in the first aspect of the present invention when executing the computer program.

[0018] Fourthly, the invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the fault-tolerant control method provided in the first aspect of the invention.

[0019] In summary, the above-described technical solutions conceived in this invention can achieve the following beneficial effects: 1. This invention provides a fault-tolerant control method for a three-level inverter. It involves querying a pre-collected operating condition-weight mapping table covering all operating conditions of the three-level inverter to find the optimal weight triplet corresponding to the actual operating condition, using this as a candidate weight triplet. Then, based on different real-time monitoring indicators, the weights in the candidate weight triplet are dynamically adjusted accordingly, thereby obtaining the optimal weight triplet matching the actual operating condition. This invention pre-determines a relatively optimal search seed based on prior knowledge and dynamically adjusts it according to actual monitoring indicators, avoiding repeated manual trial and error, significantly shortening the debugging cycle, reducing costs, and adapting to different operating conditions. It can still guarantee global optimality under different operating conditions, exhibiting strong robustness and high accuracy in fault-tolerant control.

[0020] 2. Furthermore, in the fault-tolerant control method for a three-level inverter provided by this invention, the optimal weight triplet corresponding to each operating condition s in the operating condition-weight mapping table... To incorporate the output current, output voltage, three-phase level switching states, upper and lower bus capacitor voltages, and DC bus voltage of the three-level inverter under operating condition s into the model predictive control cost function The weight triplet calculated to minimize the predictive control cost of the model. ;in, These are adaptive coefficients corresponding to the operating mode, load state, and operating condition, respectively. This invention considers the dynamic priority differences of multiple control objectives under different operating boundaries of a three-level inverter, and specifically decouples the conflicts of multi-objective optimization. Since the system has significantly different performance priorities for current tracking, midpoint balancing, common-mode rejection, and switching losses under different operating conditions (such as grid-connected / off-grid, light load / heavy load, healthy / fault), the introduction of the above-mentioned physically meaningful adaptive coefficients can "target and weight" the core control requirements under different operating conditions (such as assigning different weights according to healthy / fault scenarios or grid-connected / off-grid modes), thereby dynamically reshaping the optimization space of the model prediction cost function. This effectively overcomes the performance degradation and compromises that traditional fixed weights are prone to cause under wide operating conditions, and can obtain a more accurate optimal weight triplet.

[0021] 3. Furthermore, in the fault-tolerant control method for a three-level inverter provided by this invention, when determining the optimal weight triplet corresponding to each operating condition s in the operating condition-weight mapping table, hard constraints and soft constraints are introduced. Among them, the hard constraints guarantee the bottom line of control performance, while the soft constraints ensure EMI suppression without excessively sacrificing current tracking performance, so that the entire operation does not exceed the standard or compromise, and meets the dual requirements of grid connection and safety regulations.

[0022] 4. Further, in the fault-tolerant control method for a three-level inverter provided by this invention, the optimal weight triplet corresponding to each operating condition s in the operating condition-weight mapping table is calculated using a Bayesian optimization algorithm; wherein, the initial search space of the weight triplet is obtained by Latin hypercube sampling from the preset search space of the weight triplet. The sampling method ensures coverage of high-probability regions, improving convergence robustness. The probabilistic surrogate model is a Gaussian process; the acquisition function is an EI acquisition function, which balances "exploring high-uncertainty regions" with "developing known favorable regions," further improving the accuracy of fault-tolerant control. Attached Figure Description

[0023] Figure 1 The topology of the grid-connected T-type inverter provided in the embodiments of the present invention; Figure 2 This is a flowchart of the Bayesian optimization weight adaptation process provided in an embodiment of the present invention. Figure 3 A flowchart for searching the optimal weights using the Bayesian optimization algorithm is provided as an embodiment of the present invention; Figure 4 A flowchart illustrating the online operating condition detection process provided in this embodiment of the invention; Figure 5 The control flowchart of the MPC algorithm provided in the embodiments of the present invention; Figure 6 The above is a simulation waveform diagram of the entire process of "grid connection → fault → fault tolerance" based on the fault-tolerant control method provided in the embodiments of the present invention. Figure 7 This is a schematic diagram of the performance after control is performed using the fault-tolerant control method provided in the embodiments of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0025] To achieve the above objectives, in a first aspect, the present invention provides a fault-tolerant control method for a three-level inverter, comprising: performing a fault-tolerant operation on the three-level inverter; wherein the fault-tolerant operation performed in the k-th control cycle includes: Obtain the candidate weight triplet corresponding to the k-th control cycle. The weights in the cost function , , Each one is set as a candidate weight. , , The MPC algorithm is used to solve the three-phase level switching state in the k-th control cycle with the goal of minimizing the cost value, so as to perform fault-tolerant control on the three-level inverter. Where k is a positive integer; Here is the index of the control window corresponding to the k-th control cycle; the cost function is:

[0026] The tracking error of the three-level inverter output current in the k-th control cycle; The DC bus capacitor voltage difference during the kth control cycle; The average common-mode voltage output of the three-level inverter during the kth control cycle; This represents the switching overhead during the k-th control cycle. ; ; x It can be a, b, or c; For the k-th control cycle x Phase level switch state; During the fault-tolerant operation of the three-level inverter, each M control cycles constitute a control window, and weight adjustment operations are continuously performed within each control window; M is a positive integer. The weight adjustment operation performed in the t-th control window includes: obtaining the current operating condition of the three-level inverter, and querying the optimal weight triplet corresponding to the current operating condition from the pre-established operating condition-weight mapping table, as a candidate weight triplet for the t-th control window. In the last control cycle of the t-th control window, calculate the harmonic distortion rate of the current output current of the three-level inverter, the DC bus capacitor voltage difference, and the average common-mode voltage of the three-level inverter output, and denote them accordingly as follows: , , ;when or When the corresponding preset threshold is exceeded, increase ;when When the corresponding preset threshold is exceeded, increase ; The mapping table mentioned above covers all operating conditions of the three-level inverter, including the optimal weight triplet corresponding to each operating condition of the three-level inverter collected in advance; all operating conditions cover all combinations of different operating modes, different load states and different operating states of the three-level inverter.

[0027] Preferably, in one optional implementation, the optimal weight triplet corresponding to each operating condition s in the above mapping table is... The weighted triplet, calculated by substituting the output current, output voltage, three-phase level switching states, upper and lower bus capacitor voltages, and DC bus voltage of the three-level inverter under operating condition s into the model predictive control cost function, minimizes the model predictive control cost. ; The predictive control cost function of the above model is:

[0028] This is a preset baseline coefficient; Preset operating mode coefficients; Preset load state coefficient; Preset operating state coefficients; For the average tracking error The normalization result; ; N The preset number of control cycles; For preset control cycle index; The preset reference current amplitude; Average DC bus capacitor voltage difference The normalization result; ; This is the static DC bus voltage; This represents the overall average common-mode voltage. ; This represents the average switching overhead.

[0029] Preferably, in one optional implementation, the optimal weight triplet corresponding to each operating condition s in the above mapping table is... To incorporate the output current, output voltage, three-phase level switching states, DC bus voltage, and upper and lower bus capacitor voltages of the three-level inverter under operating condition s into the model predictive control cost function, the weighted triplet that minimizes the model predictive control cost is calculated under constraints. ; The constraints include: Hard constraints include: the harmonic distortion rate of the three-level inverter output current in the k-th control cycle is less than the corresponding preset threshold. The soft constraint condition includes: the average common-mode voltage of the three-level inverter output in the k-th control cycle is less than the corresponding preset threshold.

[0030] Preferably, in one optional implementation, the optimal weight triplet corresponding to each operating condition s in the above mapping table is... It was calculated using the Bayesian optimization algorithm.

[0031] Preferably, in an optional implementation, in the above-described Bayesian optimization algorithm, the initial search space of the weight triples is obtained by Latin hypercube sampling from the preset search space of the weight triples. The data was obtained through sampling; the probabilistic surrogate model is a Gaussian process; the acquisition function is the EI acquisition function. Among them, the preset search space for:

[0032] and Weights The minimum and maximum values; and Weights The minimum and maximum values; and Weights The minimum and maximum values.

[0033] It should be noted that the method for obtaining the initial search space of the aforementioned weighted triples is not the only method; other methods can also be used, such as by using a preset search space. The samples are obtained through sampling; where each sampled weighted triplet in the initial search space satisfies a log-uniform distribution. .

[0034] In one alternative implementation, the tracking error of the three-level inverter output current during the kth control cycle is... DC bus capacitor voltage difference under the kth control cycle The average common-mode voltage of the three-level inverter output during the kth control cycle. They are respectively:

[0035]

[0036]

[0037] in, This is the dynamic reference current in the kth control cycle; This represents the output current of the three-level inverter during the kth control cycle. This is the dynamic reference current for phase x in the kth control cycle; For the three-level inverter in the kth control cycle x Phase output current; and These are the upper and lower bus capacitor voltages under the kth control cycle, respectively; ; Let x be the x-phase output voltage of the three-level inverter in the kth control cycle; This represents the dynamic DC bus voltage during the k-th control cycle.

[0038] In one optional implementation, the method for determining the current operating condition of the three-level inverter includes: An islanding detection algorithm is used to detect the current operating mode of the three-level inverter; the operating modes include: grid-connected mode and off-grid mode. Calculate the current output current value of the three-level inverter Calculate the current load rate ;like If the load rate is less than the preset load rate, the three-level inverter is determined to be in a light load state; otherwise, it is in a heavy load state. Rated current; A fault diagnosis algorithm is used to detect the current operating status of the three-level inverter.

[0039] In one optional implementation, the operating status is identified in real time through multi-layer discrimination logic:

[0040]

[0041]

[0042] in Indicates grid connection, This indicates disconnection from the network, determined by the island detection algorithm. For health, For horizontal pipe failure, The fault is a vertical pipe fault, as indicated by the fault diagnosis algorithm; load rate. Classified as light load For heavy load, Based on.

[0043] Construct a runtime condition-weight mapping table Operating status key in After that, with Complexity retrieval map Get the corresponding weighted triplet This enables plug-and-play functionality across various scenarios; if the operating condition is not matched (e.g., medium load), Then rollback is performed according to priority: (1) same mode, same fault, different load; (2) different mode, same fault, same load; (3) default weight, to ensure robust operation.

[0044] In one alternative implementation, when the three-level inverter is in a fault state, When the three-level inverter is in a non-faulty state, Where -1 indicates that the level switch is connected to the lower bridge arm, 0 indicates that the level switch is connected to the midpoint, and +1 indicates that the level switch is connected to the upper bridge arm.

[0045] To further illustrate the fault-tolerant control method for a three-level inverter provided in the first aspect of the present invention, a detailed description is provided below with reference to a specific embodiment: This embodiment uses a T-type three-level inverter as an example for illustration. Figure 1 The diagram shows a T-type three-level inverter topology in one embodiment of the present invention. The present invention requires real-time acquisition of grid voltage. , , Output voltage , , and the voltage of the upper and lower bus capacitors , It achieves multi-condition fault-tolerant control based on scene detection module, offline weight optimization module, online weight mapping and adaptive weight correction module, and MPC vector solution module.

[0046] Figure 2 A flowchart of the fault-tolerant control method provided in this invention is given. The control period Ts = 50µs, and each period executes the following sequentially: ① Online scenario identification; ② Weighted triplet. ③ Retrieve; ④ Fine-tune online if performance indicators exceed limits; ⑤ Select vector set based on health / fault flags; ⑥ Minimize the two-step prediction cost function to obtain the optimal switching state. Where -1 indicates the level switch is connected to the lower bridge arm, 0 indicates the level switch is connected to the midpoint, and +1 indicates the level switch is connected to the upper bridge arm; ⑥ Drive PWM update. Detailed explanations of key steps are provided.

[0047] 1. Construction of a complete runtime set for the scenario This embodiment uses Cartesian combinations of parallel / off-grid, light / heavy load, and health / horizontal pipe fault / vertical pipe fault to construct a complete operational set consisting of 12 discrete scenarios, denoted as:

[0048] in, Characterizes the operating mode (grid for grid-connected, island for off-grid). Load factor (0.5 corresponds to light load, 1.2 corresponds to heavy load). The operating status type is specified (0 indicates healthy, 1 indicates horizontal tube open circuit, 2 indicates vertical tube open circuit); the three-phase grid voltage (i.e., the three-phase output voltage of the inverter) is collected in real time at a sampling frequency of 20kHz. , , ), three-phase output current ( , , DC bus voltage (upper and lower capacitor voltages) , Define a multi-objective weighted cost function that includes current tracking error, midpoint balance, common-mode rejection, and switching losses:

[0049] in For the weight vector to be optimized, each sub-cost , , , The performance of midpoint offset, common-mode voltage, switching frequency, and current tracking is characterized separately; offline Bayesian optimization is performed on the operating condition set. For each operating condition s, the optimal weight triplet is solved independently:

[0050] and with scene key Create a weighted lookup table (i.e., mapping table) for the index. To achieve "offline global search optimization and online constant time optimization" The efficient weight supply mechanism for "retrieval" avoids the computational burden of online optimization; the lookup table supports rapid switching across scenarios, ensuring real-time control and robustness.

[0051] 2. Offline Bayesian weight optimization In the offline phase, Bayesian optimization is used to search for the optimal weight triplet for each running condition. and with scene key Store the index in the lookup table To achieve "offline global optimization, online..." The system provides efficient weighted search; scene keys and weights are mapped one-to-one, supporting millisecond-level scene recognition and seamless switching.

[0052] Bayesian optimization with Gaussian process regression Constructing the weight space To the cost The probabilistic proxy model uses an improved expectation-enhanced acquisition function:

[0053] The mean function With covariance kernel All use the squared exponent kernel (RBF kernel) to characterize smoothness. For the current optimal cost, To explore parameters, indicator functions Under the limited assessment budget Candidate weights are sampled iteratively within the inner loop, and the true cost is obtained through closed-loop simulation. The Gaussian process posterior is dynamically updated to balance "exploring regions with high uncertainty" and "developing known favorable regions." The exploration rate is set to 0.5 to suppress premature convergence. Initial samples are either log-uniformly distributed or around the default weights. Latin hypercube sampling ensures coverage of high-probability regions and improves convergence robustness.

[0054] Specifically, Figure 3 This demonstrates the process of using Bayesian optimization to search for the optimal weight triplet in the offline phase, scene by scene. Bayesian optimization uses Gaussian process regression (GPR) to construct a probabilistic surrogate model from the weight space to the cost, avoiding exhaustive evaluation. Step 1: For a given operating condition Assuming the cost function Prior to a Gaussian process:

[0055] The mean function Initially set to zero (or a constant), covariance kernel The squared exponential kernel RBF is used to characterize the smoothness of the function:

[0056] in This represents the signal variance (typically 1.0). The length scale parameter (typically 0.5) controls the rate of correlation decay of the function; given the evaluated observation dataset. ( The posterior distribution of a Gaussian process is a conditional Gaussian distribution.

[0057] Their predicted mean and variance are as follows:

[0058]

[0059] in ( ) is a kernel matrix and ( ), Observation noise variance (used for regularization, typical value) The acquisition function employs an enhanced expectation improvement EI+ strategy to balance exploration and development.

[0060] in Given the current known optimal cost, To explore parameters (typical value 0.01, encouraging exploration of high uncertainty regions), , and These are the standard normal cumulative distribution function and the probability density function, respectively. In each iteration, the solution is obtained.

[0061] Step 2: Evaluation through closed-loop simulation and will new observations join in Update the Gaussian process posterior; set the evaluation budget. It converges to the global optimum or near-optimal solution within a finite number of attempts; the exploration rate is set to 0.5 (i.e., (or multi-point sampling strategy), effectively suppressing local optima traps; initial samples are arranged around preset weights. Latin hypercube sampling (LHS) is used to cover high-probability, high-quality areas and improve initialization quality. Each sub-item of the cost function is evaluated through the evaluation window. Inner cumulative normalization calculation, where The initial time after warm-up (skipping the first 40ms transient state) is defined as follows: 1) Average tracking error: measures the sum of squared errors between the three-phase output current and the reference current;

[0062] in For reference current, The measured output current is; the normalized form is ,in This is the preset reference current amplitude.

[0063] 2) Average DC bus capacitor voltage difference: Characterizes the midpoint potential shift to describe the fluctuation of the voltage difference between the upper and lower bus capacitors;

[0064] in This is the midpoint potential offset, ideally zero; the midpoint RMS is defined as... The normalized form is ; 3) Overall common-mode voltage average: the absolute average of the three-phase voltage center point relative to ground potential;

[0065] in This is the voltage at the midpoint of the three-phase bridge arm; (The three-level switch state is: 1 connected to the lower bridge arm, 0 connected to the midpoint, and +1 connected to the upper bridge arm). The normalized form is... .

[0066] 4) Average switching overhead: The switching frequency is indirectly constrained by accumulating the number of switching transitions.

[0067] After normalization (This is already a dimensionless quantity); Based on this, the model predictive control cost function is:

[0068] The following constraints are introduced: a) Hard constraint penalty mechanism: To ensure the bottom line of control performance, a hard constraint is introduced: the harmonic distortion rate of the output current of the three-level inverter in the k-th control cycle is less than the corresponding preset threshold. The harmonic distortion rate of the three-level inverter output current in the kth control cycle is:

[0069] The corresponding preset threshold is denoted as the hard threshold; in this embodiment, a hard threshold is set. (National standard requirements); Hard suppression in this embodiment:

[0070] like Then, a high cost is imposed to eliminate inferior solutions:

[0071] b) Soft constraint conditions, including: the average common-mode voltage of the three-level inverter output in the k-th control cycle is less than the corresponding preset threshold; In this embodiment, a quadratic penalty function is used to achieve soft suppression of the common-mode voltage.

[0072] in , Based on the operating mode setting (grid connection) Offline , Soft penalty ensures EMI suppression without excessively sacrificing current tracking performance, smoothly suppressing electromagnetic interference (EMI) while taking current tracking performance into account.

[0073] After introducing constraints, the final model predictive control cost function is:

[0074] in (Benchmark weight) For scene adaptation coefficients (healthy scene) Fault scenarios ; grid connection Offline Accurate evaluation is achieved through closed-loop simulation.

[0075] 3. Online scene detection Figure 4 The multi-layered discrimination logic of the scene detection module is presented. First, an islanding detection step involving "over / underfrequency + over / undervoltage" is used to compare and determine grid connection / disconnection. Second, the load factor is calculated based on the ratio of the output current RMS to the rated value. Finally, a fault flag from an external fault diagnosis module is received. The detection delay is less than 50µs, meeting the single-cycle switching requirements. The specific implementation is as follows: a. Islanding detection using "over / under frequency + over / under voltage (OUF-OUV)" is used to distinguish between grid-connected and off-grid modes. This step uses the sampled values ​​of the three-phase capacitor voltage. , and three-phase power grid voltage (i.e., three-phase output current) , , Using as input, first calculate the RMS value Upcc of the common point of coupling (PCC) voltage and the system frequency. ,like or Exceeding the preset normal range and continuously exceeding the clearing time If the grid fails, it is determined that the grid has lost power, and the off-grid flag island=1 is immediately set; otherwise, the grid-connected flag island=0 is maintained.

[0076] The effective value of the common coupling point voltage in the kth power frequency cycle By analyzing N instantaneous sampled values ​​within this period The square root of the square of the product is obtained by averaging the squares of the products.

[0077] System frequency By measuring the time interval between two consecutive positive zero crossings of the phase A voltage. Taking the reciprocal, we get:

[0078] This embodiment uses 230V and 50Hz as the rated voltage. With rated frequency The baseline value:

[0079] The voltage threshold is set to 88% and 110% of the rated value, i.e., the lower limit. upper limit

[0080] The frequency thresholds are set to 49.5Hz and 50.5Hz, corresponding to the rated frequency, respectively. 1% vs. +1%

[0081] If M' consecutive control cycles satisfy Not here Within the interval, or Not here If the interval is within a certain range, it is determined to be an island state:

[0082] b. Calculate the RMS value of the three-phase output current. and with rated current Compare to obtain the load rate ,like If it is classified as light load, then it is classified as heavy load; The root mean square average of the three-phase currents is calculated using 3N sampling points. As the current actual load current:

[0083] c. The open-circuit detection method in patent 202211443044.2 is used to detect open-circuit faults, including the following steps: (1) Assuming the current control cycle is 20kHz, according to the sliding window length Continuous acquisition of three-phase current And normalize each phase. ; (2) Calculate the reference values ​​of the current in each phase within 2L sliding windows. And extract extreme values , .

[0084] (3) Calculate the Hausdorff distance between phases: ( , Similarly, take the median. As fault detection information.

[0085] (4) If Greater than the threshold If so, it is determined that a power transistor open circuit fault has occurred.

[0086] (5) Fault phase location: Using the fundamental frequency amplitude of the last sliding window, the phase corresponding to the minimum value is the fault phase.

[0087]

[0088] (6) Fault tube location: 6a) Obtain the first group of candidate transistors based on the polarity of the final sliding window current.

[0089] 6b) The second group of candidate transistors is obtained based on the fundamental frequency comparison within the L sliding window.

[0090] Find the intersection The faulty power transistor is uniquely identified and a fault flag is output. The diagnosis was completed.

[0091] 4. Weight Mapping and Adaptive Correction The online adaptive weight adjustment module dynamically fine-tunes the weights based on real-time performance metrics to address model uncertainty or scene boundary drift. The specific strategy is as follows: Each M (in this embodiment) Each control period (corresponding to 10ms) is used as a control window, and weight adjustment operations are continuously performed in each control window.

[0092] Specifically, in the t-th control window, the current operating conditions of the three-level inverter (operating mode + load status + operating status) are obtained, and the above three pieces of information are combined into the scene key Skey: Query the pre-stored lookup table Obtain the corresponding optimal weight triplet and use it as the candidate weight triplet under the t-th control window. If an exact match fails (e.g., the load rate is at the boundary), the nearest matching principle or interpolation strategy will be used:

[0093] in For nearest neighbor load rate, This is an alternative mode (such as rollback from grid connection to off-grid); the mechanism ensures robustness of table lookup and cross-scenario generalization ability, with a retrieval time complexity of O(log n). (Hash table implementation).

[0094] In the last control cycle of the t-th control window, calculate the harmonic distortion rate of the current output current of the three-level inverter, the DC bus capacitor voltage difference (i.e., the midpoint RMS), and the average common-mode voltage of the three-level inverter output. These are used as the harmonic distortion rate, DC bus capacitor voltage difference, and average common-mode voltage in the t-th control window, respectively denoted as: , , The specific calculation formula is as follows:

[0095]

[0096]

[0097]

[0098] in, This is the index of the last control cycle in the t-th control window.

[0099] And based on this, adaptive adjustment rules are triggered: when or When the corresponding preset threshold is exceeded, aggressively increase. To strengthen constraints

[0100] Wherein, the indicator function It is a Boolean value (1 if it exceeds the limit, 0 otherwise); The THD over-limit gain is set to 0.15 in this embodiment; The gain for midpoint exceedance is set to 0.15 in this embodiment; gain accumulation ensures sufficient weighting when multiple constraints exceed limits simultaneously; an upper limit is set. To avoid being overly conservative, in this embodiment, , Take aggressive gains to respond quickly to hard constraint defaults.

[0101] when When the corresponding preset threshold is exceeded, the increase is moderated. To suppress EMI:

[0102] in The gain is adjusted for CMV and is significantly smaller than In this embodiment, the value is set to 0.1125, which is a mild gain to smoothly adjust the soft constraint; an upper limit is set. Soft constraints use a smaller gain to avoid sacrificing current tracking performance, suppress EMI, and take efficiency into account; they can also further limit the mean and peak CMV drift to meet safety requirements.

[0103] In this embodiment, a saturation upper limit is set. To prevent excessive weighting from causing oscillations or instability, specifically, .

[0104] 5. MPC vector solution using the MPC algorithm. like Figure 5 As shown, the MPC vector solving module enumerates the candidate switch vector set in each control cycle. Predict the system state in the next one (or two) steps and select the optimal vector that minimizes the cost function; specifically implemented as follows: (1) Health scenario Employing a complete three-level 27-vector set The reference voltage vector is calculated using the Clarke transform. And based on this, determine the sector in question. The sector pruning strategy retains the target sector and its adjacent sector vectors (approximately 8-12), forming a simplified candidate set. This reduces computational load while maintaining modulation density; for each candidate vector The current in the next control cycle is predicted based on the discrete model of the system.

[0105] Discrete model of midpoint potential:

[0106]

[0107] in , , , ; Zero level indication, It is used to balance the bus voltage and suppress DC offset; the optional "warm-up steps" skip the initial transient and reduce evaluation noise.

[0108] Calculate the overall cost of each candidate and select the optimal vector:

[0109]

[0110] (2) Fault Scenario: When a fault is detected in the switching device (such as an open circuit in the upper or middle transistor), the three-level operation is degraded to two-level operation through topology reconstruction (such as shorting the SCR or remapping the switch state), and the candidate vector set is reduced to There is no midpoint balance issue in two-level mode. The cost function simplifies to:

[0111] The number of vectors is reduced to 8, reducing the computational load by about 70% and improving real-time performance; a zero-vector buffer (lasting 1 to 2 cycles) is introduced during fault switching to ensure a smooth transition and avoid current surges; (3) Delay compensation: Considering the inherent one-step delay of digital control, a two-step prediction method is adopted: first, the switch state of the previous step is used to predict the delay. Then predict using candidate vectors and with The cost of predicting values ​​at any time is used to ensure that the system state matches the actual application's switching state.

[0112] In this embodiment, a smooth transition is achieved when switching scenes, such as the instant the fault flag is triggered: and After the fault is cleared, the system switches back via zero vector buffer to reduce voltage jumps and torque shocks, achieving a seamless transition from fault to recovery and supporting control law switching within a single cycle.

[0113] In summary, this embodiment achieves adaptive adjustment of weights for the MPC fault-tolerant control of a three-level inverter. The core modules include: Offline optimization module: Generates a mapping table for storage ; MPC solver module: for beg have to ; Operating condition detection module (islanding detection + fault diagnosis): Output ; Weight mapping and adaptive module: Based on scene detection results, it provides... And fine-tune it online; Topology reconfiguration module: drives SCR / equivalent switch in and Seamless switching between modes, achieving a smooth transition.

[0114] In summary, this embodiment constructs a complete set of operating scenarios covering grid-connected / off-grid, multiple load levels, and health / fault states. It employs Bayesian optimization to offline search for the globally optimal solution of control weights in each scenario and establishes a lookup table. In the online phase, a scenario identification mechanism retrieves the corresponding weights at millisecond levels, and implements hierarchical dynamic corrections based on the degree of hard constraint exceedance and soft constraint deviation, achieving adaptive control parameters under various operating conditions. Compared to fixed-weight control and single-point parameter optimization methods, this invention solves the performance degradation problem caused by weight mismatch under wide operating conditions, improving the system's operational quality during multi-condition switching and fault states. Simultaneously, this invention consumes less CPU resources and has low computational complexity.

[0115] like Figure 6 The diagram shows the simulation waveforms of the entire process of "grid connection → fault → fault tolerance" running in the Simulink environment at a sampling frequency of 20 kHz. The four vertical dashed lines correspond to the following times in sequence: t1 = 0.025 s: grid connection switch closed; t2 = 0.08 s: model automatically connects to the grid and sends PWM; t3 = 0.15 s: Sa2 fault injected in Simulink through the "IGBT open circuit" module; t4 = 0.20 s: topology reconstruction completed and fault-tolerant MPC switched in. After t2, equal amplitude and equal phase difference are quickly established; near t3, a fault trigger causes a short-term amplitude / phase disturbance; after t4, equal amplitude tracking is restored within approximately one power frequency cycle, indicating that the fault-tolerant MPC re-converges under the new weights without generating DC bias. Figure 7 As shown, the Bayesian-optimized weights outperform manually adjusted weights in most operating conditions in terms of THD bias, line voltage balance bias, midpoint voltage offset, and common-mode voltage offset (smaller negative bias is better, and larger positive bias is better), showing an overall "positive improvement" trend. Even in demanding scenarios such as faults / heavy loads, the Bayesian-optimized weights can still reduce midpoint offset and CMV bias, avoiding overfitting or imbalance common in manually adjusted weights. In some scenarios where THD is slightly higher, it remains within acceptable thresholds, demonstrating the weight search's priority on robustness.

[0116] In a second aspect, the present invention provides a three-level inverter system, comprising: a three-level inverter and a controller; The controller is used to execute the fault-tolerant control method provided in the first aspect of the present invention.

[0117] The related technical solutions are the same as the fault-tolerant control method provided in the first aspect of this invention, and are not limited here.

[0118] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory stores a computer program, and the processor executes the fault-tolerant control method provided in the first aspect of the present invention when executing the computer program.

[0119] The related technical solutions are the same as the fault-tolerant control method provided in the first aspect of this invention, and are not limited here.

[0120] Fourthly, the invention also provides a computer program product, including a computer program / instruction that, when executed by a processor, implements the fault-tolerant control method provided in the first aspect of the invention.

[0121] The related technical solutions are the same as the fault-tolerant control method provided in the first aspect of this invention, and are not limited here.

[0122] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A fault-tolerant control method for a three-level inverter, characterized in that, include: Perform fault-tolerant operations on the three-level inverter; wherein, the fault-tolerant operations performed in the k-th control cycle include: Obtain the candidate weight triplet corresponding to the k-th control cycle. , cost function Weights in , , Each one is set as a candidate weight. , , The MPC algorithm is used to solve the three-phase level switching state in the k-th control cycle with the goal of minimizing the cost, so as to perform fault-tolerant control on the three-level inverter; where k is a positive integer. This is the index of the control window corresponding to the k-th control cycle; , , , These correspond to the tracking error of the three-level inverter output current, the average common-mode voltage of the three-level inverter output, and the switching overhead, respectively, in the k-th control cycle. ; ; x It can be a, b, or c; For the k-th control cycle x Phase level switch state; During the fault-tolerant operation of the three-level inverter, each M control cycles constitute a control window, and a weight adjustment operation is continuously performed within each control window; M is a positive integer; the weight adjustment operation performed in the t-th control window includes: obtaining the current operating condition of the three-level inverter, and querying the optimal weight triplet corresponding to the current operating condition from the pre-established operating condition-weight mapping table, which is then used as the candidate weight triplet for the t-th control window. In the last control cycle of the t-th control window, calculate the harmonic distortion rate of the current output current of the three-level inverter, the DC bus capacitor voltage difference, and the average common-mode voltage of the three-level inverter output, and denote them accordingly as follows: , , ;when or When the corresponding preset threshold is exceeded, increase ;when When the corresponding preset threshold is exceeded, increase ; The mapping table covers all operating conditions of the three-level inverter, including the optimal weight triplet corresponding to each operating condition of the three-level inverter collected in advance; all operating conditions cover all combinations of different operating modes, different load states and different operating states of the three-level inverter.

2. The fault-tolerant control method according to claim 1, characterized in that, The optimal weight triplet corresponding to each operating condition s in the mapping table To incorporate the output current, output voltage, three-phase level switching states, upper and lower bus capacitor voltages, and DC bus voltage of the three-level inverter under operating condition s into the model predictive control cost function The weight triplet calculated to minimize the predictive control cost of the model. ;in, This is a preset baseline coefficient; Preset operating mode coefficients; Preset load state coefficient; Preset operating state coefficients; For the average tracking error The normalization result; ; N The preset number of control cycles; For preset control cycle index; The preset reference current amplitude; Average DC bus capacitor voltage difference The normalization result; ; This is the static DC bus voltage; This represents the overall average common-mode voltage. ; This represents the average switching overhead.

3. The fault-tolerant control method according to claim 2, characterized in that, The optimal weight triplet corresponding to each operating condition s in the mapping table To incorporate the output current, output voltage, three-phase level switching states, DC bus voltage, and upper and lower bus capacitor voltages of the three-level inverter under operating condition s into the model predictive control cost function, the weighted triplet that minimizes the model predictive control cost is calculated under constraints. ; The constraints include: Hard constraints include: the harmonic distortion rate of the three-level inverter output current in the k-th control cycle is less than the corresponding preset threshold. The soft constraint condition includes: the average common-mode voltage of the three-level inverter output in the k-th control cycle is less than the corresponding preset threshold.

4. The fault-tolerant control method according to claim 2, characterized in that, The optimal weight triplet corresponding to each operating condition s in the mapping table It was calculated using the Bayesian optimization algorithm.

5. The fault-tolerant control method according to claim 4, characterized in that, In the Bayesian optimization algorithm, the initial search space of the weight triples is obtained by Latin hypercube sampling from the preset search space of the weight triples. The data was obtained through sampling; the probabilistic surrogate model is a Gaussian process; the acquisition function is the EI acquisition function. Wherein, the preset search space for: and Weights The minimum and maximum values; and Weights The minimum and maximum values; and Weights The minimum and maximum values.

6. The fault-tolerant control method according to any one of claims 1-5, characterized in that, Tracking error of the three-level inverter output current in the kth control cycle DC bus capacitor voltage difference under the kth control cycle The average common-mode voltage of the three-level inverter output during the kth control cycle. They are respectively: in, This is the dynamic reference current in the kth control cycle; This represents the output current of the three-level inverter during the kth control cycle. This is the dynamic reference current for phase x in the kth control cycle; Let x be the x-phase output current of the three-level inverter in the kth control cycle; and These are the upper and lower bus capacitor voltages under the kth control cycle, respectively; ; Let x be the x-phase output voltage of the three-level inverter in the kth control cycle; This represents the dynamic DC bus voltage during the k-th control cycle.

7. The fault-tolerant control method according to any one of claims 1-5, characterized in that, The methods for determining the current operating condition of a three-level inverter include: An islanding detection algorithm is used to detect the current operating mode of the three-level inverter; the operating modes include: grid-connected mode and off-grid mode. Calculate the current output current value of the three-level inverter Calculate the current load rate ;like If the load rate is less than the preset load rate, the three-level inverter is determined to be in a light load state; otherwise, it is in a heavy load state. Rated current; A fault diagnosis algorithm is used to detect the current operating status of the three-level inverter.

8. A three-level inverter system, characterized in that, include: Three-level inverter and controller; The controller is used to execute the fault-tolerant control method according to any one of claims 1-7.

9. An electronic device, characterized in that, include: A memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the fault-tolerant control method according to any one of claims 1-7.

10. A computer program product, characterized in that, It includes a computer program / instruction that, when executed by a processor, implements the fault-tolerant control method according to any one of claims 1-7.

Images