High-efficiency frequency converter control method and system based on multi-level topology

Abstract

The invention belongs to the technical field of power electronics, and provides a multi-level topology-based high-efficiency frequency converter control method, which comprises the following steps of: acquiring current and voltage data through a sensor, predicting load current by using a Kalman filter, dynamically adjusting a sub-module switching sequence and optimizing a switching frequency; then collecting historical data for preprocessing, predicting a load current by using an LSTM model, further adjusting a switching sequence and a switching frequency in combination with a Kalman filter optimization result, and finally optimizing the switching frequency by using the model.

Description

Technical Field

[0001] The present invention belongs to the technical field of electronic power, and in particular relates to a high-efficiency frequency converter control method based on a multi-level topology. Background Art

[0002] Frequency converters (VFDs) play a vital role in modern industrial production and power transmission. Traditional multilevel topology control methods, such as modular multilevel converters, suffer from numerous issues. Their dynamic response is insufficient. When the load changes suddenly or renewable energy fluctuates, traditional PI control struggles to quickly adjust the voltage level, leading to voltage and current distortion. For harmonic suppression, traditional methods rely on multi-stage filtering or complex modulation strategies, which increases hardware costs and system size. Furthermore, traditional multilevel topologies have weak fault tolerance. A single module failure can easily lead to cascading failures, severely impacting system reliability.

[0003] To this end, the present invention provides a high-efficiency inverter control method based on a multi-level topology. Summary of the Invention

[0004] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] In a first aspect, the present invention provides a high-efficiency inverter control method based on a multi-level topology, comprising:

[0007] Step 1: Deploy sensors to monitor the load current and capacitor voltage in the load circuit in real time;

[0008] Step 2: Predict the load current in the next k control cycles based on the real-time current data obtained by the sensor;

[0009] Step 3: Based on the predicted load current and the capacitor voltage deviation, the capacitor voltage deviation analysis results are determined. The load current trend is then analyzed to determine the switching action for the submodules. A priority list is then generated, and the switching order of the submodules is dynamically adjusted based on the priority list.

[0010] Step 4: Collect the DC bus current and voltage, introduce the switching loss model, constrain and optimize the model, and use the switching loss model to optimize the switching frequency in real time;

[0011] Step 5: Collect historical data, clean and preprocess the collected historical data;

[0012] Step 6: Select the long short-term memory network (LSTM) as the deep learning model, use the processed data to train the model, and optimize the switching frequency.

[0013] In a second aspect, the present invention provides a high-efficiency inverter control system based on a multi-level topology, comprising:

[0014] Hardware deployment module: deploys sensors to monitor the load current and capacitor voltage in the load circuit in real time;

[0015] Current data prediction module: predicts the load current in the next k control cycles based on the real-time current data obtained by the sensor;

[0016] Dynamic adjustment module: Based on the predicted load current and capacitor voltage deviation, it determines the capacitor voltage deviation analysis results, then analyzes the load current trend to determine the switching action of the submodules. It then generates a priority list and dynamically adjusts the switching order of the submodules based on the priority list.

[0017] Real-time optimization module: collects DC bus current and voltage, introduces a switching loss model, constrains and optimizes the model, and uses the switching loss model to optimize the switching frequency in real time;

[0018] Data processing module: collects historical data, cleans and preprocesses the collected historical data;

[0019] Model training and application module: Select the long short-term memory network (LSTM) as the deep learning model, use the processed data to train the model, and optimize the switching frequency.

[0020] The present invention has the following beneficial effects: The proposed method for coordinated control of dynamic capacitor voltage balancing and switching frequency, combined with a Kalman filter and LSTM model, can effectively address rapid power fluctuations in scenarios such as wind, solar, and energy storage. It maintains capacitor voltage stability, improving overall system performance, power quality, and efficiency, while also optimizing switching frequency and extending device life. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The present invention will be further described below with reference to the accompanying drawings.

[0022] Figure 1 is a flowchart of the steps of Example 1 of the present invention;

[0023] Figure 2 This is a flow chart of the steps for obtaining the gap impact value in Example 1 of the present invention. DETAILED DESCRIPTION

[0024] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0025] Example 1

[0026] like Figure 1 As shown, an efficient inverter control method based on a multi-level topology according to an embodiment of the present invention includes:

[0027] Step 1: A current sensor is provided at the current load to monitor the load current in the load circuit in real time; a capacitor voltage sensor is provided to obtain the capacitor voltage;

[0028] Step 2: Based on the real-time current data obtained by equipping the current load with a current sensor, use the Kalman filter to predict the load current I in the next k control cycles. L (t+Δt),I L (t+2Δt),...,I L (t+kΔt), where Δt is the control period;

[0029] Specifically, the load current is predicted using a Kalman filter;

[0030] The Kalman filter is a commonly used optimal estimation method. It is based on the state space model of the system and continuously updates the estimation of the system state by observing the input and output of the system.

[0031] In the present invention, the state space model of the load current can be expressed as: Among them, X k is the system state vector at time k, U k-1 is the input vector at time k-1, Y k is the observation vector at time k, A is the natural dynamic characteristic matrix of the load current, B is the matrix of the influence of the associated control input switch signal on the current, and C is the matrix mapping the internal state to the actual measurement value; w k-1 is process noise, which comes from external interference; v k is the observation noise, the error from the sensor;

[0032] Step 3: Based on the predicted load current, dynamically adjust the switching order of the submodules; the switching order of the submodules of the multilevel converter directly affects the balance of the capacitor voltage; the traditional method relies on a fixed threshold or rotation strategy, which is difficult to adapt to dynamic load changes; the present invention uses a Kalman filter to predict the load current I in the next k control cycles L (t+Δt),I L (t+2Δt),...,I L(t+kΔt), where Δt is the control period; combined with the capacitor voltage deviation, the switching sequence is dynamically optimized. The specific steps are as follows:

[0033] Specifically, the switching order of the submodules is dynamically adjusted according to the predicted load current. Assume that there are n submodules in the multilevel converter, and the capacitor voltage of each submodule is Vi, (i = 1, 2, ..., n). Define the capacitor voltage deviation ΔVi = Vi - V ref ; Among them, V ref is the reference value of the capacitor voltage, Vi is the capacitor voltage of the submodule;

[0034] If the capacitor voltage deviation is greater than 0, it means that the capacitor is overcharged and needs to be discharged;

[0035] If the capacitor voltage deviation is less than 0, it means that the capacitor is undercharged and needs to be charged;

[0036] Specifically, real-time data collection and prediction; input the current load current I L (t), the submodule capacitor voltage Vi at the current moment, and the load current I in the next k control cycles L (t+Δt),I L (t+2Δt),...,I L (t+kΔt), where Δt is the control period; determine the load change trend and calculate the slope F of the predicted current K ; The formula used is:

[0037] When F K >0; the load current increases and the submodule needs to be charged;

[0038] When F K <0; the load current decreases and the submodule needs to be cut off for discharge;

[0039] Generate a priority list; in charging mode: sort in ascending order by ΔVi; in discharging mode: sort in descending order by ΔVi;

[0040] Based on the load current prediction results, the sub-modules with larger capacitor voltage deviation are switched on and off first to reduce the overall voltage deviation;

[0041] Specifically, for example, when it is predicted that the load current will increase, the submodule with a lower capacitor voltage is prioritized to charge the capacitor, and conversely, the submodule with a higher capacitor voltage is prioritized to be disconnected;

[0042] When the load current increases, that is, I L (t+Δt)>I L At (t), the submodule is put into operation to charge the capacitor and replenish energy; the submodule with the lowest capacitor voltage deviation is put into operation first;

[0043] When the load current decreases, that is, I L (t+Δt)<I L (t); then the submodule is removed to discharge the capacitor and release excess energy; the submodule with the highest ΔVi is removed first to avoid voltage exceeding the limit;

[0044] Step 4: Collect DC bus current and voltage, and optimize the switching frequency in real time: Introduce a switching loss model; switching loss mainly includes turn-on loss and turn-off loss;

[0045] Specifically, taking a typical IGBT switch as an example, its turn-on loss E on and turn-off loss E off The calculation formula is as follows:

[0046] Energy loss E caused by voltage-current overlap during the turn-off to turn-on process on for: The loss E from turn-on to turn-off off for: Among them, V DC is the DC bus voltage; I L is the DC bus current, t r and t f They are the turn-on rise time and turn-off fall time, f s is the switching frequency; the total switching loss P S For: P S =(E on +E off )f s ;

[0047] Specifically, the switching frequency is optimized in real time with minimizing the total loss as the objective function; the total loss P co for: Where R is the parasitic resistance; with the minimum total loss minP to The objective function is: By solving the objective function, the optimal switching frequency f is obtained s ';

[0048] Specifically, the above model is simplified to minP to Taking the derivative and setting it to zero gives: Zero frequency means no switching action, but it will lead to worse harmonics;

[0049] Therefore, it is necessary to introduce constraints and transform it into a constrained optimization problem;

[0050] Specifically, the harmonic limit constraint: the total harmonic distortion rate THD of the output current must meet the standard, that is, THD ≤ THD MAX , where THD is related to the switching frequency fs The relationship can be approximated by Fourier analysis as: THD MAX The maximum value of total harmonic distortion is preset. The converter needs to quickly recover to steady state when the load changes suddenly. s Not less than a certain lower limit f min , that is, f s ≥f min , the switching frequency is too high, which will cause the device temperature to rise beyond the limit, so f s The upper limit f max , that is, f s ≤f max ;

[0051] The objective function is calculated in [f min ,f max ] interval to generate N candidate frequencies; for each candidate frequency f s Calculate P to , and check the harmonic limit constraint conditions, select P to The smallest f′ s ;

[0052] The technical solution of this embodiment is as follows: through the above-mentioned dynamic capacitor voltage balancing and switching frequency coordinated control method, in fluctuating scenarios such as wind, solar and storage, this method can effectively cope with rapid changes in power, maintain capacitor voltage stability, and improve the overall performance of the system. For example, in an example where a modular multi-level converter comprising multiple submodules is applied to a photovoltaic power generation system, when drastic changes in light intensity cause power generation fluctuations, the control method of the present invention can be used to quickly adjust the capacitor voltage and maintain it within a stable range, significantly improving the quality of the power output of the system and significantly increasing system efficiency compared to traditional methods.

[0053] Example 2

[0054] Based on Example 1, the present invention provides a high-efficiency inverter control method based on a multi-level topology, further comprising the following steps:

[0055] Step 5: Collect historical data, clean and preprocess the collected historical data;

[0056] It should be noted that the sensors include but are not limited to: capacitor voltage sensors, DC bus voltage and current sensors; historical data include: load current, DC bus current, DC bus voltage and capacitor voltage;

[0057] The collected data is cleaned and preprocessed, including removing noise and filling missing values. Filtering can be used to extract useful features from the raw data, such as calculating the rate of change of load current and the deviation of capacitor voltage. Fourier transforms are performed on current and voltage signals to extract frequency domain features. The difference between capacitor voltages in different submodules is calculated.

[0058] The submodule capacitor voltage sensor based on the multilevel converter collects the capacitor voltage data Vi(t) (i=1, 2, 3, ..., n) of each submodule at different times; where n is the number of submodules; and simultaneously collects the voltage data V of the DC bus. DC (t) and current data I DC (t);

[0059] Data preprocessing: For current and voltage signals, a low-pass filter is used to remove high-frequency noise. Taking the load current signal as an example, assuming it is interfered with by high-frequency noise, a smoother signal can be obtained after passing through a low-pass filter.

[0060] Filling missing values: Check whether there are missing values in the data. If so, use linear interpolation, polynomial interpolation, or machine learning-based methods to fill the missing values. For example, for missing values in capacitor voltage data, linear interpolation can be performed based on the voltage values at adjacent moments to obtain an estimated value.

[0061] Calculate the load current change rate: Extract useful features from the original data and calculate the load current change rate; calculate the change rate of the load current in adjacent time intervals Δt to obtain the load current change rate feature rI L (t);

[0062] Based on the collected capacitor voltage, according to the formula ΔVi(t)=Vi(t)-V ref (t) The capacitor voltage deviation characteristic ΔVi(t) is calculated, and the capacitor voltage of each submodule is compared with the reference value V ref (t), forming the deviation feature vector [ΔV1(t), ΔV2(t), ..., ΔVn(t)], where n is the number of submodules;

[0063] Frequency domain feature extraction: Perform discrete Fourier transform on signals such as load current and DC bus current to convert time domain signals into frequency domain signals, and extract the amplitude and phase information of the signals at different frequency components as features;

[0064] Specifically, for example, the load current signal I L (t) is transformed by discrete Fourier transform to obtain I L (t)(f), where f represents frequency;

[0065] Submodule capacitor voltage difference characteristics: Calculate the difference between the capacitor voltages of different submodules, using the formula ΔV cj =V c (t)-V j (t) Calculate ΔV cj , where c is not equal to j; forming a submodule capacitance-voltage difference feature set;

[0066] Step 6: Select a deep learning model. Since the data is time series data with long-term dependencies, a long short-term memory (LSTM) network is selected as the deep learning model. LSTM networks can effectively process and memorize long-term information in time series through a gating mechanism, making them suitable for tasks such as load current prediction.

[0067] The data processed in step 5 is divided into 70% as the training set, 15% as the validation set, and 15% as the test set. Assuming that the total number of data samples is N, the number of training set samples is about 0.7N, and the number of validation set and test set samples is about 0.15N respectively.

[0068] For each sample, its related features are combined into an input feature vector, and the corresponding future load current value is used as the target output;

[0069] For example, an LSTM network is constructed, including an input layer, several LSTM layers, and an output layer; the number of neurons in the input layer is determined according to the dimension of the input feature vector;

[0070] For example, if the dimension of the input feature vector X is m, then the number of neurons in the input layer is m;

[0071] Assume that two LSTM layers are set up. The number of neurons in each LSTM layer can be determined based on experience or through hyperparameter adjustment, such as 64 neurons. The number of neurons in the output layer is 1, which is used to output the predicted load current value.

[0072] Select the Adam optimizer, which updates the model parameters by adaptively adjusting the learning rate to minimize the loss function;

[0073] Training model: Use the training set to train the LSTM model; in each training batch, input the input feature vector X into the model and calculate the predicted load current value through forward propagation

[0074] The error between the predicted value and the true value is calculated based on the loss function, and then the error is passed back to each layer of the network through the back propagation algorithm to update the parameters of the model;

[0075] During training, use the validation set to monitor model performance. Calculate the loss on the validation set every 10 training rounds. If the loss on the validation set stops decreasing or starts to increase, the model may be overfitting, and measures such as stopping training early or increasing regularization can be taken.

[0076] Combine deep learning models with existing solutions to predict load current;

[0077] Use the trained LSTM model to predict load currents for multiple time periods in the future. Use these predicted values as inputs to the Kalman filter, which is then used to further optimize the load current predictions. This results in a more accurate prediction, allowing the submodule switching sequence to be dynamically adjusted based on the predicted load currents obtained from the LSTM model and the Kalman filter, combined with capacitor voltage deviations.

[0078] If the capacitor voltage deviation is greater than 0, it means that the submodule capacitor is overcharged and needs to be discharged; if the capacitor voltage deviation is less than 0, it means that the submodule capacitor is undercharged and needs to be charged;

[0079] Real-time data acquisition and prediction: Input the current load current and submodule capacitor voltage, and predict the load current for the next k steps. Determine the change trend of the predicted load current and calculate the slope of the predicted load current. The formula is: in, It indicates the slope of the load current over time, which reflects the rate of change of the load current. is the predicted load current value at the current moment t, k is the number of control cycles, is the time interval of a single control cycle;

[0080] When S>0, the load current is predicted to rise and the submodule needs to be charged;

[0081] When S<0, the load current is predicted to decrease and the submodule needs to be cut off for discharge;

[0082] Generate a priority list: in charging mode, sort by capacitor voltage deviation in ascending order; in discharging mode, sort by capacitor voltage deviation in descending order;

[0083] Based on the load current prediction results, the submodules with larger capacitor voltage deviations are prioritized for switching on and off to reduce the overall voltage deviation. For example, when the load current is predicted to increase, the submodules with lower capacitor voltages are prioritized for switching on and off to charge the capacitors; otherwise, the submodules with higher capacitor voltages are prioritized for switching off.

[0084] Real-time switching frequency optimization uses the predicted load current and DC bus voltage predicted by the LSTM model as input to optimize the switching frequency; calculates switching losses, including the energy loss caused by voltage-current overlap during the turn-off to turn-on process; and calculates losses during the turn-on to turn-off process.

[0085] The total switching loss is: P SW =E on +E off +I L (t) 2 ×f s ×R, where t r and t f They are the switching rise time and the turn-off fall time, f s is the switching frequency, R is the parasitic resistance;

[0086] Calculation of conduction loss: Conduction loss is proportional to the square of the current and can be expressed as where R eq is the equivalent resistance, is the predicted value of the load current;

[0087] Objective function: Minimizing the total loss is the objective function, and the total loss is P to (t) = P SW (t)+P co (t); By solving the objective function, the optimal switching frequency is obtained; the above model is simplified to f s The derivative is taken and set to zero, but zero frequency means no switching action, which will lead to worsening harmonics, so constraints need to be introduced to convert it into a constrained optimization problem;

[0088] Harmonic limit constraint: The total harmonic distortion (THD) of the output current must meet the standard, that is, THD ≤ THD MAX , and the relationship with the switching frequency can be approximated by Fourier analysis as THD MAX The maximum value of the total harmonic distortion rate is preset. The converter needs to quickly recover to steady state when the load changes suddenly. This requires that the switching frequency is not lower than a certain lower limit, that is, f s ≥f min

[0089] Too high a switching frequency will cause the device temperature to rise beyond the standard, so the upper limit of the switching frequency needs to be limited, i.e., f s ≤f max ;

[0090] Objective function calculation: In the interval [f min ,f max ] Generate N candidate frequencies; calculate P for each candidate frequency to , and check the constraints; choose to make P toThe minimum candidate frequency f s ;

[0091] Evaluation and validation: Use the test set to evaluate model performance;

[0092] The test set is fed into the solution combined with the deep learning model to evaluate the accuracy of the load current prediction; metrics such as root mean square error can be used. where N test is the number of test set samples is the predicted load current, yi is the actual load current;

[0093] To evaluate the improvement of capacitor voltage balance, the standard deviation of all submodule capacitor voltages can be calculated as the average value of all submodule capacitor voltages, and the changes in standard deviation before and after combining the model can be compared;

[0094] Example 3

[0095] like Figure 2 As shown, the present invention provides a high-efficiency inverter control system based on a multi-level topology, comprising:

[0096] Hardware deployment module: A current sensor is equipped at the current load to monitor the current in the load circuit in real time; a current sensor is equipped at the current input terminal to monitor the current at the input terminal in real time; a capacitor voltage sensor is used to obtain the capacitor voltage; a DC bus voltage and current sensor is used to obtain the DC bus voltage and current sensor;

[0097] Current data prediction module: Based on the real-time current data obtained by the current sensor equipped at the current load, the Kalman filter is used to predict the load current I in the next k control cycles. L (t+Δt), where Δt is the control period;

[0098] Specifically, the load current is predicted using a Kalman filter;

[0099] The Kalman filter is a commonly used optimal estimation method. It is based on the state space model of the system and continuously updates the estimation of the system state by observing the input and output of the system.

[0100] Dynamic adjustment module: Based on the predicted load current, the switching order of the submodules is dynamically adjusted; the switching order of the submodules of the multilevel converter directly affects the balance of the capacitor voltage; the traditional method relies on a fixed threshold or rotation strategy, which is difficult to adapt to the changes in dynamic load; the present invention predicts the load current I by using a Kalman filter L (t), combined with the capacitor voltage deviation, the switching sequence is dynamically optimized. The specific steps are as follows:

[0101] Specifically, the switching order of the submodules is dynamically adjusted according to the predicted load current. Assume that there are n submodules in the multilevel converter, and the capacitor voltage of each submodule is Vi, (i = 1, 2, ..., n). Define the capacitor voltage deviation ΔVi = Vi - V ref ; Among them, V ref is the reference value of the capacitor voltage, Vi is the capacitor voltage of the submodule;

[0102] If the capacitor voltage deviation is greater than 0, it means that the capacitor is overcharged and needs to be discharged;

[0103] If the capacitor voltage deviation is less than 0, it means that the capacitor is undercharged and needs to be charged;

[0104] Specifically, real-time data collection and prediction; input the current load current I L (t), the submodule capacitor voltage Vi(t) at the current moment, Vi(t) represents the capacitor voltage of the submodule at time t; predict the load current sequence I in the next k steps L (t+Δt),I L (t+2Δt),...,I L (t+kΔt), where Δt is the control period; determine the load change trend and calculate the slope F of the predicted current K ; The formula used is:

[0105] When F K >0; the load current increases and the submodule needs to be charged;

[0106] When F K <0; the load current decreases and the submodule needs to be cut off for discharge;

[0107] Generate a priority list; in charging mode: sort in ascending order by ΔVi; in discharging mode: sort in descending order by ΔVi;

[0108] Based on the load current prediction results, the sub-modules with larger capacitor voltage deviation are switched on and off first to reduce the overall voltage deviation;

[0109] Specifically, for example, when it is predicted that the load current will increase, the submodule with a lower capacitor voltage is prioritized to charge the capacitor, and conversely, the submodule with a higher capacitor voltage is prioritized to be disconnected;

[0110] When the load current increases, that is, I L (t+Δt)>I L At (t), the submodule is put into operation to charge the capacitor and replenish energy; the submodule with the lowest capacitor voltage deviation is put into operation first; it is charged to approach V ref ;

[0111] When the load current decreases, that is, IL (t+Δt)<I L (t); then the submodule is removed to discharge the capacitor and release excess energy; the submodule with the highest ΔVi is removed first to avoid voltage exceeding the limit;

[0112] Real-time optimization module: collects DC bus current and voltage, and optimizes switching frequency in real time: introduces a switching loss model; switching loss mainly includes turn-on loss and turn-off loss;

[0113] Specifically, taking a typical IGBT switch as an example, its turn-on loss E on and turn-off loss E off The calculation formula is as follows:

[0114] Energy loss caused by voltage-current overlap during the turn-off to turn-on process: Losses from turn-on to turn-off: Among them, V DC is the DC bus voltage; I L is the load current, t r and t f They are the turn-on rise time and turn-off fall time, f s is the switching frequency; the total switching loss is: P S =(E on +E off )f s ;

[0115] Specifically, the switching frequency is optimized in real time with minimizing the total loss as the objective function; let the total loss be: P to Conduction loss P co It is proportional to the square of the current and can be expressed as: Where R is the parasitic resistance; then the objective function is: By solving the objective function, the optimal switching frequency f is obtained. s ';

[0116] Data processing module: collect historical data: obtain other relevant data from current sensors, including but not limited to capacitor voltage sensors, DC bus voltage and current sensors; collect historical data, including load current, capacitor voltage, DC bus voltage and current sensors;

[0117] The collected data is cleaned and preprocessed, including noise removal and missing value filling. Filtering can be used to extract useful features from the raw data, such as calculating the rate of change of load current and the deviation of capacitor voltage. Fourier transforms are performed on current and voltage signals to extract frequency domain features. The differences between capacitor voltages in different submodules are calculated.

[0118] Model Training and Application Module: Selecting a Deep Learning Model. Since the data is time series data with long-term dependencies, a long short-term memory (LSTM) network is selected as the deep learning model. The LSTM network, through its gating mechanism, can effectively process and memorize long-term information in time series, making it suitable for tasks such as load current prediction.

[0119] The preprocessed and feature-engineered data is divided into 70% as a training set, 15% as a validation set, and 15% as a test set. Assuming the total number of data samples is N, the number of training set samples is approximately 0.7N, and the number of validation set and test set samples is approximately 0.15N respectively.

[0120] For each sample, its related features are combined into an input feature vector X, and the corresponding future load current value is used as the target output y;

[0121] The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-efficiency inverter control method based on a multi-level topology, characterized by: include: Step 1: Deploy sensors to monitor the load current and capacitor voltage in the load circuit in real time; Step 2: Predict the load current in the next k control cycles based on the real-time current data obtained by the sensor; Step 3: Based on the predicted load current and the capacitor voltage deviation, the capacitor voltage deviation analysis results are determined. The load current trend is then analyzed to determine the switching action for the submodules. A priority list is then generated, and the switching order of the submodules is dynamically adjusted based on the priority list. Step 4: Collect the DC bus current and voltage, introduce the switching loss model, constrain and optimize the model, and use the switching loss model to optimize the switching frequency in real time; Step 5: Collect historical data, clean and preprocess the collected historical data; Step 6: Select the long short-term memory network (LSTM) as the deep learning model, use the processed data to train the model, and optimize the switching frequency.

2. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of predicting the load current in the next k control cycles based on the real-time current data obtained by the sensor is as follows: based on the real-time current data obtained by the current sensor equipped at the current load; using the Kalman filter to predict the load current in the next k control cycles.

3. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of combining the capacitance voltage deviation is as follows: define capacitance voltage deviation ΔVi=Vi-V ref ; Among them, V ref is the reference value of the capacitor voltage, and Vi is the capacitor voltage of the submodule.

4. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of determining the capacitor voltage deviation analysis result is as follows: if the capacitor voltage deviation is greater than 0, it indicates that the capacitor is overcharged and needs to be discharged; if the capacitor voltage deviation is less than 0, it indicates that the capacitor is undercharged and needs to be charged.

5. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of analyzing the load current variation trend to determine the switching action of the submodule is: predicting the future load current sequence; determining the load variation trend, and calculating the slope of the predicted current.

6. The high-efficiency inverter control method based on multi-level topology according to claim 5, characterized in that: The specific process of switching the submodule is as follows: if the slope of the predicted current is greater than the zero-load current, the submodule needs to be switched on for charging; if the slope of the predicted current is less than the zero-load current, the submodule needs to be switched off for discharging.

7. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of dynamically adjusting the switching order of the submodules according to the priority list is as follows: Generate a priority list; in charging mode: sort in ascending order by capacitor voltage deviation; in discharging mode: sort in descending order by capacitor voltage deviation; when the load current increases, the submodule is put into operation to charge the capacitor; when the load current decreases, the submodule is removed to discharge the capacitor.

8. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of introducing the switching loss model and constraining and optimizing the model is as follows: the total harmonic distortion rate THD of the output current must meet the standard, that is, THD≤THD MAX , where THD is related to the switching frequency f s The relationship can be approximated by Fourier analysis as: THD MAX is the preset maximum value of total harmonic distortion rate. The converter needs to quickly recover to steady state when the load suddenly changes. s Not less than a certain lower limit f min , that is, f s ≥f min , the switching frequency is too high, causing the device temperature to exceed the standard, and f s The upper limit f max , that is, f s ≤f max .

9. The high-efficiency inverter control method based on multi-level topology according to claim 1, characterized in that: The specific process of optimizing the switching frequency in real time by using the switching loss model is: optimizing the switching frequency by selecting a frequency that minimizes the switching loss model.

10. A high-efficiency inverter control system based on a multi-level topology, characterized by: include: Hardware deployment module: deploys sensors to monitor the load current and capacitor voltage in the load circuit in real time; Current data prediction module: predicts the load current in the next k control cycles based on the real-time current data obtained by the sensor; Dynamic adjustment module: Based on the predicted load current and capacitor voltage deviation, it determines the capacitor voltage deviation analysis results, then analyzes the load current trend to determine the switching action of the submodules. It then generates a priority list and dynamically adjusts the switching order of the submodules based on the priority list. Real-time optimization module: collects DC bus current and voltage, introduces a switching loss model, constrains and optimizes the model, and uses the switching loss model to optimize the switching frequency in real time; Data processing module: collects historical data, cleans and preprocesses the collected historical data; Model training and application module: Select the long short-term memory network (LSTM) as the deep learning model, use the processed data to train the model, and optimize the switching frequency.

Images