Inverter current control method based on model predictive control

Abstract

The present application relates to the technical field of power electronic converter control, and particularly relates to an inverter current control method based on model predictive control, which first establishes a discrete state space model with the filter inductance current as a state variable, then obtains current vector and other information, predicts the predicted current vector corresponding to the candidate switching state, dynamically adjusts the weight coefficient, safely screens and calculates the evaluation function value, determines the optimal switching state to drive the inverter and rolls optimization, the present application adjusts the weight coefficient in the difference of transient, transition and steady state mode, and gives consideration to dynamic response and steady state quality; and identifies the ratchet tendency state based on the weight moving average value and the intermittent disturbance density, triggers the ratchet inhibition enable signal and the rapid recovery and smooth back migration, and avoids long-term depression or recovery overshoot of the weight.

Description

Technical Field

[0001] This invention relates to the field of power electronic converter control technology, and more specifically, to an inverter current control method based on model predictive control. Background Technology

[0002] Inverters, as core power conversion devices in new energy power generation, motor drives, and energy storage systems, directly impact the system's power quality, dynamic response speed, and operating efficiency through their current control performance. With the increasing demands for control accuracy and response speed in applications such as distributed generation, electric vehicle traction, and uninterruptible power supplies (UPS), model predictive control (MMC) has become a research hotspot in inverter current control due to its intuitive concept, ease of handling multi-objective constraints, and fast dynamic response. In a finite control set MMC framework, the controller traverses all possible switching states in each sampling cycle. An evaluation function comprehensively evaluates the current tracking effect and switching action cost of each candidate state, selecting the optimal switching state to apply to the inverter. However, the evaluation function typically consists of a weighted sum of a current tracking error term and a switching state change penalty term. The weighting coefficients between these two terms directly determine the trade-off between tracking accuracy and switching losses in the control system. Larger weighting coefficients result in more suppressed switching actions but potentially slower current tracking response; smaller weighting coefficients lead to more agile current tracking but may increase switching frequency and losses.

[0003] A search revealed that patent CN116388527B discloses a multi-objective control method for a three-level inverter based on model prediction. This method weights multiple objective functions, including current error, number of switch switching operations, midpoint potential offset, and the difference between the selected switch state and the optimized pulse width modulation switch state, to obtain an optimization function. Through rolling optimization, the optimal vector that balances multiple optimization objectives is selected. This patent introduces multi-objective optimization concepts into model predictive control, providing a more comprehensive performance evaluation framework for inverter control.

[0004] A search revealed a patent with publication number CN116954072A, which discloses a method for dynamically optimizing predictive control weight factors based on deep residual networks. This method involves establishing a predictive mathematical model and designing a cost function to obtain data samples, constructing a deep residual network, and using weight factors as the network input layer. The trained network predicts the optimal combination of weight factors, and the dynamic optimization of the weight factors is achieved through a lookup table method. This patent attempts to use deep learning to solve the problem of weight factor adaptation under different operating conditions, realizing online adjustment of the weight factors.

[0005] However, the aforementioned existing technologies still face a series of unresolved technical problems in practical applications. In the traditional multi-objective model predictive control method represented by CN116388527B, the weight coefficients of each objective in the evaluation function are usually preset by the designer based on experience and remain fixed throughout the operation of the control system. The limitation of this fixed-weight strategy is that when the inverter operates in a steady-state condition with a small current tracking error, appropriate switching state change penalties help reduce the switching frequency, switching losses, and output current harmonics; however, when transient conditions such as load changes, grid voltage drops, or step changes in reference current occur, the current tracking error increases sharply. If the weight of the switching state change penalty term remains unchanged at this time, it will severely restrict the controller's ability to respond to errors, resulting in slower current tracking speed, increased overshoot, and even potentially triggering system overcurrent protection. In other words, fixed weights cannot achieve a dynamic autonomous balance between transient response speed and steady-state operating quality. Although CN116954072A proposes a dynamic optimization method for weight factors based on deep residual networks, adjusting the weight factors through offline training and online table lookup, this method essentially relies on training the deep network with a large amount of offline simulation data. Its prediction accuracy is significantly affected by the coverage of the training data—when the actual operating conditions deviate from the range covered by the training samples, the weight factors output by the network may deviate from their optimal values. More importantly, the inference computation of deep residual networks requires considerable computing resources and time overhead. In real-time control systems of inverters, where the sampling period is typically only tens of microseconds, this computational burden may be unbearable. Furthermore, this method, based on a table lookup mechanism for weight adjustment, has an adjustment granularity limited by the size of the pre-built database, making it difficult to provide a fine and smooth response to continuously changing operating conditions.

[0006] The aforementioned deficiencies in existing technologies can cause a series of specific and serious impacts in practical applications. In applications such as traction inverters for new energy vehicles, where dynamic response requirements are extremely high, fixed-weight strategies may lead to motor current tracking lag when the vehicle accelerates rapidly or encounters sudden load changes such as steep inclines. This manifests as delayed torque response or even jitter, directly affecting driving experience and safety. In photovoltaic grid-connected inverters, when drastic fluctuations in sunlight intensity cause frequent changes in DC bus voltage and reference current, the inability of the weighting coefficients to adapt autonomously can result in excessively high switching frequencies, increased losses, and decreased efficiency during certain periods, while in other periods, dynamic response is sluggish and power quality deteriorates. More importantly, when the system experiences intermittent disturbances, such as periodic load fluctuations or frequent small drops in grid voltage, the weighting coefficient adjustment strategies in existing technologies often exhibit asymmetrical adjustment characteristics: the weights decrease rapidly during transients to prioritize tracking performance, while the weights recover slowly after the transient ends. This asymmetric regulation, characterized by rapid decreases and slow increases, can lead to a unidirectional, cumulative, and continuous drop in the weighting coefficient under repeated intermittent disturbances. This is known as the "ratchet effect," where the weighting coefficient is gradually suppressed by repeated disturbances and cannot be effectively recovered, ultimately remaining at a level far below the design value for an extended period. The direct consequence of this persistently low weighting coefficient is that the penalty term for switching state changes becomes almost ineffective, the inverter switching frequency increases significantly, switching losses rise sharply, and the junction temperature of power devices rises. This not only reduces system efficiency but also accelerates the aging and failure of power semiconductor devices, seriously threatening the long-term operational reliability of the inverter. Summary of the Invention

[0007] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an inverter current control method based on model predictive control.

[0008] To achieve the above objectives, the present invention provides the following technical solution: Inverter current control methods based on model predictive control include: S1: Based on the inverter main circuit parameters and the characteristics of the connected objects, establish a discrete state-space model with the filter inductor current as the state variable. The connected objects are the load or the power grid, and the voltage corresponding to the connected objects is the object voltage. S2: At the k-th sampling time, acquire the current current vector, current object voltage vector, current reference current vector, DC bus voltage, the final applied switching state of the previous control cycle, and the operating mode of the previous sampling cycle in the two-phase stationary coordinate system. S3: Based on the discrete state-space model, the current current vector, the current object voltage vector, and the final applied switching state in the previous control cycle, obtain the predicted current vector corresponding to each candidate switching state in the candidate switching state set; S4: Based on the current current vector, the current reference current vector, the operating mode of the previous sampling period, and the historical information of the weight coefficients, determine the operating mode of the current sampling period, and dynamically adjust the weight coefficients of the switch state change penalty term in the evaluation function to obtain the dynamically adjusted weight coefficients. S5: Based on the predicted current vector, current safety constraints, current reference current vector, dynamically adjusted weighting coefficients, and the final applied switching state in the previous control cycle, perform safety screening on each candidate switching state in the candidate switching state set and calculate the evaluation function value to obtain the candidate switching states that meet the safety constraints and their corresponding evaluation function values. S6: Determine the optimal switching state based on the candidate switching states that meet the safety constraints and their corresponding evaluation function values, and generate a drive signal based on the optimal switching state to act on the inverter, then return to step S2 for rolling optimization control.

[0009] In one embodiment, the discrete state-space model uses the sampling period, filter inductance value, equivalent resistance, DC bus voltage, and object voltage as parameters. The object voltage is either the load voltage or the grid voltage. The discrete state-space model outputs a difference equation to describe the current dynamics, so that the current vector at the next sampling moment can be explicitly expressed by the current vector at the current sampling moment, the inverter output voltage vector corresponding to the current sampling moment, and the object voltage vector at the current sampling moment.

[0010] In one embodiment, the acquisition of the current current vector, the current object voltage vector, the current reference current vector, and the DC bus voltage includes: at the k-th sampling time, acquiring the instantaneous values ​​of the inverter's three-phase output current and the object voltage, acquiring the DC bus voltage, and acquiring the reference current for the current sampling period; performing coordinate transformation on the instantaneous values ​​of the inverter's three-phase output current, the object voltage, and the reference current for the current sampling period to obtain the current current vector, the current object voltage vector, and the current reference current vector in a two-phase stationary coordinate system.

[0011] In one embodiment, the step of obtaining the predicted current vector includes: firstly, calculating the current estimate at the current moment based on the final applied switching state in the previous control cycle and the current current vector; then, starting from the current estimate, calculating the predicted current vector corresponding to each candidate switching state in the candidate switching state set at the (k+1)th sampling moment.

[0012] In one embodiment, in step S4, the operating mode of the current sampling period includes a transient mode, a steady-state mode, and a transition mode. The dynamic adjustment method of the weight coefficient includes: updating the weight coefficient according to the operating mode of the current sampling period: in the transient mode, decreasing the weight coefficient according to a preset first attenuation coefficient; in the steady-state mode, increasing the weight coefficient according to a preset second growth coefficient; in the transition mode, keeping the weight coefficient unchanged; the absolute value of the first attenuation coefficient is greater than the absolute value of the second growth coefficient; and applying a preset upper limit value and a preset lower limit value constraint to the updated weight coefficient, the lower limit value being a positive constant, so that the switch state change penalty term maintains a non-zero constraint force.

[0013] In one embodiment, the method for determining the operating mode of the current sampling period includes: determining the current tracking error amplitude of the current sampling period based on the current current vector and the current reference current vector, and determining the error change rate based on the current tracking error amplitude of the current sampling period and the current tracking error amplitude of the previous sampling period. Based on the preset transient entry threshold, transient exit threshold, steady-state entry threshold, and operating mode of the previous sampling period, the operating mode of the current sampling period is determined according to the following priority order: when the current tracking error amplitude is greater than the transient entry threshold, it is determined to be in transient mode; when the operating mode of the previous sampling period was transient mode and the current tracking error amplitude is greater than the transient exit threshold, the transient mode is maintained; when the operating mode of the previous sampling period was transient mode and the current tracking error amplitude is less than or equal to the transient exit threshold, the transient mode is exited and it is determined to be in transition mode; when the operating mode of the previous sampling period was not transient mode, the current tracking error amplitude is less than the steady-state entry threshold, and the error change rate indicates that the current tracking error amplitude has converged, it is determined to be in steady-state mode; when none of the conditions are met, it is determined to be in transition mode.

[0014] In one embodiment, the dynamic adjustment method for the weighting coefficients further includes: The weight coefficients are averaged using a long time window covering multiple sampling periods to obtain the weighted moving average; and the proportion of the number of sampling periods in the transient mode within the preset evaluation period to the total number of sampling periods within the preset evaluation period is used as the intermittent disturbance density. The weighted moving average is compared with the preset steady-state optimal weight lower limit, and the intermittent disturbance density is compared with the preset disturbance density threshold. When the weighted moving average is lower than the steady-state optimal weight lower limit and the intermittent disturbance density is greater than the disturbance density threshold, the current sampling period is determined to be in a ratchet tendency state, and a ratchet suppression enable signal is generated. After the ratchet suppression enable signal is effective, when the weighted moving average rises back to above the preset recovery safety threshold and the intermittent disturbance density is less than the disturbance density threshold, the ratchet suppression enable signal is deactivated.

[0015] In one embodiment, in step S5, phase advance compensation is performed on the current reference current vector to obtain a compensated reference current vector; for each candidate switch state and its predicted current vector obtained in step S3, the amplitude of the predicted current vector is compared with a preset current safety threshold, and candidate switch states whose amplitude exceeds the current safety threshold are eliminated to obtain candidate switch states that meet the safety constraints; for candidate switch states that meet the safety constraints, an evaluation function value is calculated based on the compensated reference current vector, the dynamically adjusted weight coefficient, and the switch state finally applied in the previous control cycle.

[0016] In one embodiment, the evaluation function value is obtained by weighted summation of the current tracking error term and the switching state change penalty term. The current tracking error term is calculated by quadratic form of the error weight matrix through the difference between the predicted current vector and the compensation reference current vector. The switching state change penalty term is obtained by weighting the switching difference between the corresponding candidate switching state and the switching state finally applied in the previous control cycle through dynamically adjusted weight coefficients.

[0017] In one embodiment, in step S6, among all candidate switch states that meet the safety constraints, the candidate switch state with the smallest corresponding evaluation function value is selected as the optimal switch state. The drive signal corresponding to the optimal switch state is applied to the inverter at the beginning of the next control cycle, and the process returns to step S2 for rolling optimization in subsequent sampling cycles.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention divides the inverter's operating state into transient mode, transition mode, and steady-state mode, and applies weight adjustment strategies of rapid decay, constant change, and slow increase to respectively. This allows the switching constraints to be rapidly weakened when the current tracking error suddenly increases in order to prioritize dynamic response speed. In the steady-state region after the error converges, the switching constraints are gradually restored to reduce switching frequency and output current harmonics. Thus, a time-varying balance between tracking accuracy and switching losses is achieved within a single evaluation function framework, avoiding the shortcomings of traditional fixed weights that cannot balance speed and steady-state quality when switching between transient and steady states. The system obtains the weight moving average by performing a moving average on the weight coefficients using a long time window that is much slower than the sampling period. At the same time, the proportion of transient mode sampling periods within a preset evaluation period is counted as the intermittent disturbance density. When the weight moving average is continuously lower than the steady-state optimal weight lower limit and the intermittent disturbance density exceeds the disturbance density threshold, the system is determined to enter a ratchet tendency state, and a ratchet suppression enable signal is generated. This mechanism can identify the phenomenon of unidirectional cumulative drop in weight caused by insufficient slow recovery due to frequent triggering of rapid weight drop by short-term disturbances. It is different from traditional threshold oscillation or data drift detection and provides an accurate triggering basis for subsequent rapid recovery intervention. After determining that the system has entered a ratchet tendency state, the slow second growth coefficient in steady-state mode is temporarily replaced with a third growth coefficient whose absolute value is greater than or equal to the rapid decay coefficient. This forces the weight coefficient to quickly rise to the steady-state design range at a rate not less than the decay rate after each exit from the transient mode, thereby interrupting the ratchet-like decline of the weight in the long term. After the ratchet suppression enable signal is removed, a fourth growth coefficient with a rate between the second and third growth coefficients is introduced during the transition period. This allows the weight coefficient to smoothly transition from a high state back to the steady-state optimal range before restoring the original second growth coefficient. This avoids the weight overshoot after recovery, which would cause the subsequent transient response to be overly constrained, and achieves a seamless switch between weight adjustment and system protection. Attached Figure Description

[0019] Figure 1 This is a flowchart of an inverter current control method based on model predictive control. Figure 2 This is a schematic diagram of the operation mode recognition and determination process; Figure 3 This is a schematic diagram illustrating the dynamic updating of weight coefficients and ratchet intervention adjustment. Figure 4 This is a schematic diagram of the ratchet tendency state detection and suppression enable logic. Figure 5 A schematic diagram illustrating the safe screening and rolling optimization selection of candidate switch states. Detailed Implementation

[0020] Reference Figures 1 to 5 An inverter current control method based on model predictive control includes the following steps: S1: Based on the inverter main circuit parameters and the characteristics of the connected load or grid, a discrete state-space model with the filter inductor current as the state variable is established. The model takes the sampling period, filter inductor value, equivalent resistance, DC bus voltage and load / grid voltage as parameters, and outputs a difference equation to describe the current dynamics, so that the current vector at the next sampling moment can be explicitly expressed by the current vector at the current moment, the inverter output voltage vector at the current moment and the load / grid voltage vector. The establishment of the discrete state-space model described in S1 includes: assuming the nominal value of the filter inductor is... The nominal equivalent series resistance is The sampling period is , No. The DC bus voltage at each sampling time is Filter inductance estimate and equivalent resistance estimate Take the nominal value or the online estimated value, and take the first sampling period. When taking the online estimate, the estimated value of the filter inductance is limited to a fraction of the nominal inductance. Doubled Within a range of times, the estimated equivalent resistance is limited to a fraction of the nominal resistance. Doubled Within a certain range; when exceeding the boundary, the effective value of the previous sampling period is taken. The discrete state-space model in the two-phase stationary coordinate system is as follows: in, For the first The filter inductor current vector at each sampling time. The inverter output voltage vector. The voltage vector of the object.

[0021] S2: At the kth sampling moment, the instantaneous values ​​of the inverter's three-phase output current and the load / grid voltage are collected, the DC bus voltage is obtained, and the current current vector and current voltage vector in the two-phase stationary coordinate system are obtained through coordinate transformation. The switching state applied in the last control cycle is also read. The current sampling information also includes the object voltage vector. In grid-connected inverters, Take the grid connection point voltage vector; in the motor drive inverter, Take the back electromotive force vector or the equivalent voltage vector obtained from the observer; when there is no external voltage term in the resistive-inductive load model. Set to zero. The object voltage vector of the previous sampling period is denoted as... .

[0022] S3: Using the discrete state space model, the current current vector, the current voltage vector, and the previous cycle switching state, execute the prediction process that takes into account the single-step delay of digital control: First, calculate the current estimate at the current moment using the previous cycle switching state and the current current vector, and then, starting from the current estimate, calculate the predicted current vector at the k+1 sampling moment corresponding to all candidate switching states one by one. The candidate switch state set mentioned in S3 adopts the full candidate set of a two-level three-phase inverter. The state of any candidate switch is denoted as: in, Assign a candidate switch state number. Each value can be either 0 or 1. Candidate switch state. The corresponding phase voltage vector is: The inverter output voltage vector in a two-phase stationary coordinate system is: in, In the At each sampling time, the switching state applied by the previous control cycle. Obtain the inverter output voltage vector Calculate the correction current: For candidate switch states Calculate the predicted current vector: S4: Construct a dynamic weighted adjustment factor, and based on this dynamic weighted adjustment factor, form the weight coefficients of the switch state change penalty term in the evaluation function, specifically including: In S4, the historical information of the weighting coefficients includes the weighting coefficients from the previous sampling period. The weighted moving average of the previous sampling period The most recent evaluation window's operating mode identifier sequence, and the ratchet suppression enable signal from the previous sampling period. The transition phase state stored at the end of the previous sampling period and the current tracking error amplitude of the previous sampling period. The initial state is... The initial state during the migration phase is invalid, the initial operating mode identifier is the steady-state mode, and the first sampling period is... .

[0023] No. Within each sampling period, based on the current operating mode, The state of the backtracking phase stored at the end of the previous sampling period and Calculate the dynamically adjusted weighting coefficients ;by renew and ;in accordance with and renew Status during the relocation phase. The state of the return phase in the first Each sampling period participates in weight updates.

[0024] S4a: Based on the current tracking error amplitude and its rate of change at the current moment, combined with preset transient entry threshold, transient exit threshold, steady-state entry threshold, and the operating mode of the previous sampling period, the operating mode of the current sampling period is determined in the following priority order: When the current tracking error amplitude is greater than the transient entry threshold, it is determined to be a transient mode; when the previous period was a transient mode and the current tracking error amplitude is greater than the transient exit threshold, the transient mode is maintained; when the previous period was a transient mode and the current tracking error amplitude is less than or equal to the transient exit threshold, the transient mode is exited and it is determined to be a transition mode; when the previous period was a non-transient mode and the current tracking error amplitude is lower than the steady-state entry threshold and the error rate of change indicates that the error has converged, it is determined to be a steady-state mode; if none of the above conditions are met, it is determined to be a transition mode. In S4a, the current tracking error magnitude is defined as: The error increment is defined as: in, For the first The reference current vector at each sampling time. For the first The current vector at each sampling time. The transient entry threshold is taken as the rated current amplitude. of Doubled The transient exit threshold is taken as the rated current amplitude. of Doubled The steady-state threshold is taken as the rated current amplitude. of Doubled The allowable error increment threshold is set to the rated current amplitude, and the steady-state entry threshold is less than the transient exit threshold, while the transient exit threshold is less than the transient entry threshold. of Doubled Times; measured current sampling noise standard deviation When the allowable error increment threshold is taken as the above proportional value plus three times, The larger value in the range.

[0025] when When the transient threshold is reached, the current sampling period is determined to be in transient mode; when the previous sampling period was in transient mode and When the value exceeds the transient exit threshold, the current sampling period remains in transient mode; when Not higher than the steady-state entry threshold and When the sampling period does not exceed the allowable error increment threshold, it is determined to be in steady-state mode; otherwise, it is determined to be in transition mode. The current operating mode identifier is written to the controller memory.

[0026] S4b: In transient mode, the weight coefficient of the switch state change penalty term is rapidly reduced according to a preset first decay update amount; in steady-state mode, the weight coefficient is slowly increased according to a preset second growth update amount; in transition mode, the weight coefficient is kept unchanged; the absolute value of the first decay update amount is greater than the absolute value of the second growth update amount; and a preset upper limit and lower limit constraint is applied to the weight coefficient, the lower limit being a preset positive constant, to ensure that the switch state change penalty term maintains a non-zero basic constraint under any operating condition; In S4b, the steady-state optimal weight value Obtained through offline calibration. Selected at geometric intervals within the range of 0.01 to 5. There are candidate weights, denoted as . ,in For each candidate weight Calculate the steady-state comprehensive index: in, Candidate weights The corresponding total harmonic distortion of the current, The average switching frequency, The root mean square of the steady-state current error. Take 0.05, Take the allowable average switching frequency of the power device. This is the rated current amplitude. .Pick The minimum corresponding candidate weight is used as When multiple minimum values ​​exist, the smaller one is selected. The lower limit of the weight is denoted as... ,Pick The upper limit of the weight is denoted as ,Pick .

[0027] The first decay update amount is: The second growth update is: in, Take a value of 0.05 to 0.30. Take values ​​from 0.005 to 0.05, and Less than .

[0028] S4c: The weight coefficients are averaged using a long time window that is much slower than the sampling period to obtain the weighted moving average; at the same time, the proportion of the total number of sampling periods in transient mode within the preset evaluation period to the total number of sampling periods in the evaluation period is calculated as the intermittent disturbance density. In S4c, the weighted moving average is calculated using the exponential moving average method. in, As the smoothing coefficient, according to calculate; according to Converted to the number of sampling points, take the nearest integer that is not less than 1. The value is 5 to 20, preferably 10, with the reference current fundamental frequency period. Pick .in, The reference current fundamental angular frequency is used; the phase-locked loop output angular frequency is used in grid-connected inverters, and the power angular velocity command or rotor position observer output angular frequency is used in motor drive inverters.

[0029] The intermittent disturbance density is: in, To evaluate window length, by Converted to the number of sampling points, take the nearest integer that is not less than 1. Choose from 1 to 10, with 5 being the preferred choice; For the most recent The operating mode identifier within each sampling period is equal to the number of sampling periods in the transient mode. The circular buffer length is... Each sampling period writes the current operating mode identifier and removes the oldest operating mode identifier.

[0030] S4d: Compare the weighted moving average obtained in S4c with a preset steady-state optimal weight lower limit, and compare the intermittent disturbance density with a preset disturbance density threshold. When the weighted moving average is lower than the steady-state optimal weight lower limit and the intermittent disturbance density exceeds the disturbance density threshold, it is determined to be a ratchet tendency state, and a ratchet suppression enable signal is generated. When the ratchet suppression enable signal is effective, if the weighted moving average rises back to above a preset recovery safety threshold and the intermittent disturbance density falls back to below the disturbance density threshold, the ratchet suppression enable signal is canceled. In S4d, the lower bound of the steady-state optimal weight is taken as Restoring the safety threshold The perturbation density threshold is set to 0.30. The number of de-jitter confirmations is based on the evaluation window length. of Round up to the nearest whole number, and not less than 100%. Each sampling period; the number of release confirmations is based on the evaluation window length. of Round up to the nearest whole number, and the result must be greater than the number of debounce confirmations.

[0031] when Below and Greater than When the state continuously meets the debouncing confirmation number, the ratchet suppression enable signal is activated. Set to valid. When If it is valid, Not less than and Less than If the state continuously satisfies the release confirmation number of times, then Set to invalid; otherwise, Keep The value of .

[0032] In S4e, the third boost update amount is: in, Take a value between 0.05 and 0.40, and not less than 0.40. The fourth relocation and renewal volume is: in, Take values ​​between 0.01 and 0.10, and greater than [amount missing]. Less than The steady-state optimal weight interval is determined by... Confirmed, the range is to ,in The value should be between 0.03 and 0.08, with 0.05 being preferred.

[0033] No. The weight update for each sampling period is based on the current operating mode. The state of the transition phase stored at the end of the previous sampling period is determined. When the current sampling period is in transient mode, The current sampling period is in steady-state mode and When it is valid, The current sampling period is in steady-state mode and When it is invalid and in the resettlement phase, The current sampling period is in steady-state mode and When it is invalid and not in the migration phase, When the current sampling period is in transition mode, .

[0034] The dynamically adjusted weighting coefficients are: when When valid, the transition phase state is set to invalid; when From effective to When invalid, the controller records entry into the migration phase; when in the migration phase and When the system enters the steady-state optimal weight range, the transition phase ends; otherwise, the controller maintains the original transition phase state.

[0035] S5: Before calculating the evaluation function, perform phase lead compensation on the current reference current vector. The compensation angle is: The compensation reference current vector is: in, For the total delay of digital control, take to Single-step predictive control In the controller containing sample-and-hold and calculation delay Pick to .

[0036] The penalty metric for switch state changes is: in, Candidate switch state In the The switching function value of the phase bridge arm, The switching state applied in the previous control cycle In the Switching function values ​​of the phase bridge arms. Current safety threshold: in, For the peak current allowed by power devices, This is the saturation current of the filter inductor. The system overcurrent protection activation threshold, Pick The value should be 0.95, preferably 0.90. This ensures the predicted current amplitude does not exceed [a certain value]. The candidate switch states form a safety candidate set, denoted as . . When empty, in Chinese Press Sort the switches in ascending order, and take the first candidate switch state as the emergency candidate state; if there are ties, then sort them in order of priority. Sort the candidate switch states from smallest to largest and select the first one.

[0037] In S5, candidate switch state The evaluation function value is: in, To compensate for the reference current vector, Candidate switch state The corresponding predicted current vector. Error weight matrix. A second-order identity matrix is ​​preferred; when it is necessary to distinguish the error weights of the two axes, Take a positive definite diagonal matrix. for The penalty metric for switch state changes defined in the previous section. After normalization, both the current tracking error term and the switching state change penalty term are dimensionless.

[0038] S6: When not empty, in Selecting evaluation function values The candidate switch state with the smallest minimum value is selected as the optimal switch state; if multiple minimum evaluation function values ​​exist, then the optimal switch state is chosen. The smallest candidate switching state; if multiple candidate switching states still exist, and the switching state applied in the previous control cycle... If it belongs to the same set of candidates, then keep ;like If a switch state does not belong to the set of parallel candidates, then the candidate switch state with the smallest binary number is selected. When empty, use Determine the emergency candidate switch states. The drive signal corresponding to the selected switch state will be assigned in the [number]th [phase / step]. The control cycle begins by acting on the inverter and returns in the next sampling cycle. Perform scrolling optimization.

[0039] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An inverter current control method based on model predictive control, characterized in that, include: S1: Based on the inverter main circuit parameters and the characteristics of the connected objects, establish a discrete state-space model with the filter inductor current as the state variable. The connected objects are the load or the power grid, and the voltage corresponding to the connected objects is the object voltage. S2: At the k-th sampling time, acquire the current current vector, current object voltage vector, current reference current vector, DC bus voltage, the final applied switching state of the previous control cycle, and the operating mode of the previous sampling cycle in the two-phase stationary coordinate system. S3: Based on the discrete state-space model, the current current vector, the current object voltage vector, and the final applied switching state in the previous control cycle, obtain the predicted current vector corresponding to each candidate switching state in the candidate switching state set; S4: Based on the current current vector, the current reference current vector, the operating mode of the previous sampling period, and the historical information of the weight coefficients, determine the operating mode of the current sampling period, and dynamically adjust the weight coefficients of the switch state change penalty term in the evaluation function to obtain the dynamically adjusted weight coefficients. S5: Based on the predicted current vector, current safety constraints, current reference current vector, dynamically adjusted weighting coefficients, and the final applied switching state in the previous control cycle, perform safety screening on each candidate switching state in the candidate switching state set and calculate the evaluation function value to obtain the candidate switching states that meet the safety constraints and their corresponding evaluation function values. S6: Determine the optimal switching state based on the candidate switching states that meet the safety constraints and their corresponding evaluation function values, and generate a drive signal based on the optimal switching state to act on the inverter, then return to step S2 for rolling optimization control.

2. The inverter current control method based on model predictive control according to claim 1, characterized in that, The discrete state-space model takes the sampling period, filter inductance value, equivalent resistance, DC bus voltage and object voltage as parameters. The object voltage is the load voltage or grid voltage. The discrete state-space model outputs a difference equation to describe the current dynamics, so that the current vector at the next sampling time can be explicitly expressed by the current vector at the current sampling time, the inverter output voltage vector corresponding to the current sampling time, and the object voltage vector at the current sampling time.

3. The inverter current control method based on model predictive control according to claim 1, characterized in that, The acquisition methods for the current current vector, current object voltage vector, current reference current vector, and DC bus voltage include: at the k-th sampling time, acquiring the instantaneous values ​​of the inverter's three-phase output current and the object voltage, obtaining the DC bus voltage, and acquiring the reference current for the current sampling period; performing coordinate transformation on the instantaneous values ​​of the inverter's three-phase output current, the object voltage, and the reference current for the current sampling period to obtain the current current vector, current object voltage vector, and current reference current vector in a two-phase stationary coordinate system.

4. The inverter current control method based on model predictive control according to claim 1, characterized in that, The steps for obtaining the predicted current vector include: first, calculating the current estimate at the current moment based on the final applied switching state and the current current vector in the previous control cycle; then, starting from the current estimate, calculating the predicted current vector corresponding to each candidate switching state in the candidate switching state set at the (k+1)th sampling moment.

5. The inverter current control method based on model predictive control according to claim 1, characterized in that, In step S4, the current sampling period's operating mode includes transient mode, steady-state mode, and transition mode. The dynamic adjustment method of the weight coefficient includes: updating the weight coefficient according to the current sampling period's operating mode: in transient mode, decreasing the weight coefficient according to a preset first attenuation coefficient; in steady-state mode, increasing the weight coefficient according to a preset second growth coefficient; in transition mode, keeping the weight coefficient unchanged; the absolute value of the first attenuation coefficient is greater than the absolute value of the second growth coefficient; and applying preset upper and lower limits to the updated weight coefficient, with the lower limit being a positive constant, so that the switch state change penalty term maintains a non-zero constraint.

6. The inverter current control method based on model predictive control according to claim 5, characterized in that, The method for determining the operating mode of the current sampling period includes: determining the current tracking error amplitude of the current sampling period based on the current current vector and the current reference current vector, and determining the error change rate based on the current tracking error amplitude of the current sampling period and the current tracking error amplitude of the previous sampling period; Based on the preset transient entry threshold, transient exit threshold, steady-state entry threshold, and operating mode of the previous sampling period, the operating mode of the current sampling period is determined according to the following priority order: when the current tracking error amplitude is greater than the transient entry threshold, it is determined to be in transient mode; when the operating mode of the previous sampling period was transient mode and the current tracking error amplitude is greater than the transient exit threshold, the transient mode is maintained; when the operating mode of the previous sampling period was transient mode and the current tracking error amplitude is less than or equal to the transient exit threshold, the transient mode is exited and it is determined to be in transition mode; when the operating mode of the previous sampling period was not transient mode, the current tracking error amplitude is less than the steady-state entry threshold, and the error change rate indicates that the current tracking error amplitude has converged, it is determined to be in steady-state mode; when none of the conditions are met, it is determined to be in transition mode.

7. The inverter current control method based on model predictive control according to claim 6, characterized in that, The dynamic adjustment methods for weighting coefficients also include: The weight coefficients are averaged using a long time window covering multiple sampling periods to obtain the weighted moving average; and the proportion of the number of sampling periods in the transient mode within the preset evaluation period to the total number of sampling periods within the preset evaluation period is used as the intermittent disturbance density. The weighted moving average is compared with the preset steady-state optimal weight lower limit, and the intermittent disturbance density is compared with the preset disturbance density threshold. When the weighted moving average is lower than the steady-state optimal weight lower limit and the intermittent disturbance density is greater than the disturbance density threshold, the current sampling period is determined to be in a ratchet tendency state, and a ratchet suppression enable signal is generated. After the ratchet suppression enable signal is effective, when the weighted moving average rises back to above the preset recovery safety threshold and the intermittent disturbance density is less than the disturbance density threshold, the ratchet suppression enable signal is deactivated.

8. The inverter current control method based on model predictive control according to claim 1, characterized in that, In step S5, phase lead compensation is performed on the current reference current vector to obtain the compensated reference current vector; For each candidate switch state and its predicted current vector obtained in step S3, the magnitude of the predicted current vector is compared with the preset current safety threshold. Candidate switch states whose magnitude exceeds the current safety threshold are eliminated to obtain candidate switch states that meet the safety constraints. For candidate switch states that meet the safety constraints, the evaluation function value is calculated based on the compensation reference current vector, the dynamically adjusted weight coefficient, and the switch state finally applied in the previous control cycle.

9. The inverter current control method based on model predictive control according to claim 8, characterized in that, The evaluation function value is obtained by weighted summation of the current tracking error term and the switching state change penalty term. The current tracking error term is calculated by quadratic form of the error weight matrix through the difference between the predicted current vector and the compensation reference current vector. The switching state change penalty term is obtained by weighting the switching difference between the corresponding candidate switching state and the final applied switching state in the previous control cycle through dynamically adjusted weight coefficients.

10. The inverter current control method based on model predictive control according to claim 9, characterized in that, In step S6, among all candidate switch states that meet the safety constraints, the candidate switch state with the smallest corresponding evaluation function value is selected as the optimal switch state. The drive signal corresponding to the optimal switch state is applied to the inverter at the beginning of the next control cycle, and the process returns to step S2 for rolling optimization in subsequent sampling cycles.

Images