A back-to-back test platform for energy storage converters

Abstract

The present invention discloses a back-to-back test platform for an energy storage converter, which relates to the technical field of converter testing. It includes a data acquisition module that uses a voltage source converter to access the power grid and collects data under different load conditions; a capacitor health prediction module that constructs a deep learning model to extract features and calculates a fitness function for predicting the capacitor state value; a redundant control path module that constructs a redundant control path for predicting the backup capacitor state; and a load balancing module. The method of the present invention can respond in real time when the capacitor state changes through the CNN-PSO-SVR model, ensuring that the capacitor attenuation state can be quickly fed back to the control system, guaranteeing the continuous and stable operation of the capacitor under complex load conditions. It compensates the power output in a timely manner through the redundant path to avoid the shutdown of the overall system due to the failure of a single component. Through the dual-path capacitor health monitoring and Droop-MPC dynamic power distribution, it can adjust the power output in a timely manner when the capacitor shows early attenuation, avoiding the shutdown of the system caused by a single point of failure.

Description

Technical Field

[0001] The present invention relates to the technical field of converter testing, and particularly to a back-to-back test platform for an energy storage converter. Background Art

[0002] With the transformation of the global energy structure towards low-carbon and sustainable development, the importance of energy storage technology in the power system has become increasingly prominent. As an important part of power electronic equipment, the energy storage converter plays a key role in scenarios such as power grid frequency modulation, peak shaving, dynamic reactive power compensation, and new energy grid connection. In recent years, the energy storage converter technology has made great progress and has gradually developed from the traditional unidirectional current conversion mode to a more flexible two-way energy flow control direction. Among them, the back-to-back converter architecture is widely used in the energy storage system test and the power grid AC-DC interconnection field due to its two-way energy flow characteristics;

[0003] Traditional energy storage converter test platforms mainly rely on constant loads or single-direction energy flow tests, and it is difficult to comprehensively reflect the dynamic response characteristics of the converter under complex working conditions and the health status of capacitors. First, traditional test systems usually use fixed loads or discrete loads to verify the performance of the converter, and it is difficult to achieve a comprehensive monitoring of the capacitor state under dynamic load changes, resulting in limited system test coverage. Second, the existing capacitor health monitoring mostly relies on single-model prediction or simple empirical formulas, lacking an in-depth analysis of the attenuation law of capacitors in complex load environments, and it is easy to cause inaccurate capacitor failure prediction due to model deviation, thereby triggering system failures. In addition, in terms of converter power distribution, the existing technologies have not fully utilized deep learning algorithms for multi-dimensional load feature extraction and real-time power flow optimization scheduling, resulting in uneven load distribution and low overall system operation efficiency. Especially during the process of large-power load mutation or power cycle test, the traditional system is difficult to adjust the converter output power in a timely manner, easily causing capacitor overload and shortening the equipment life. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a back-to-back test platform for energy storage converters to solve the problems that traditional test systems usually use fixed loads or discrete loads to verify the performance of converters, making it difficult to comprehensively monitor the state of capacitors under dynamic load changes, resulting in limited system test coverage. Secondly, existing capacitor health monitoring mostly relies on single-model prediction or simple empirical formulas, lacking in-depth analysis of the attenuation law of capacitors in complex load environments, and being prone to inaccurate capacitor failure prediction due to model deviation, which may further lead to system failures. In addition, in terms of converter power distribution, existing technologies have not fully utilized deep learning algorithms for multi-dimensional load feature extraction and real-time power flow optimization scheduling, resulting in uneven load distribution and low overall system operation efficiency. Especially during the process of large-power load mutation or power cycle testing, traditional systems are difficult to adjust the converter output power in a timely manner, easily causing capacitor overload and shortening the equipment lifespan.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a back-to-back test platform for energy storage converters, which includes:

[0008] A data acquisition module that accesses the power grid using a voltage source converter and acquires data under different load conditions;

[0009] A capacitor health prediction module that constructs a deep learning model to extract features and calculates a fitness function to predict the capacitor state value;

[0010] A redundant control path module that constructs a redundant control path for backup capacitor state prediction;

[0011] A load balancing module that optimizes and balances the load of the converter based on the prediction of the capacitance value according to the real-time power flow;

[0012] A data synchronization module that conducts data synchronization monitoring through a power monitoring unit;

[0013] A data simulation module that constructs a virtual simulation model for data simulation;

[0014] A temperature management module that dynamically adjusts the heat dissipation strategy according to temperature changes;

[0015] An energy storage module that conducts energy feedback storage to achieve energy cycling.

[0016] As a preferred solution of the back-to-back test platform for energy storage converters of the present invention, the following applies: The access to the power grid using a voltage source converter means using a voltage source back-to-back converter VSC composed of two bidirectional voltage source inverters. The AC output terminal of the first bidirectional voltage source inverter is connected to the grid side, and the AC output terminal of the second bidirectional voltage source inverter is connected to the load side.

[0017] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: collecting data under different load conditions is to collect data based on different load conditions, and record data of voltage, current, temperature and resistance through a voltage sensor, a current sensor, a temperature sensor and a capacitance equivalent series resistance (ESR) sensor;

[0018] Introduce a high-frequency sampling circuit on the DC bus to transmit the voltage pulsation signal to the ADC sampler;

[0019] Perform a fast Fourier transform on the DC bus voltage, extract the high-frequency voltage spectrum feature data, and store it through a sliding window mechanism;

[0020] Data storage adopts a dual-channel backup mechanism, where the main storage channel is the internal memory and the backup channel is an external storage device.

[0021] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: constructing a deep learning model to extract features and calculate a fitness function for predicting the capacitance state value, constructing a deep learning model based on a convolutional neural network, including an input layer, a convolutional layer and an output layer;

[0022] The data collected under different load conditions is used as training data for model training. Select the cross-entropy loss function to calculate the calculation loss between the class probability predicted by the CNN and the actual label, use the Adam optimizer for gradient descent optimization, update the weight parameters of the CNN model, and stop the iteration when the loss of the model no longer decreases significantly during continuous iteration;

[0023] Based on the newly collected converter test data, input it into the deep learning model, and perform dimensionality reduction on the model output data using principal component analysis (PCA);

[0024] Use the particle swarm optimization (PSO) algorithm to optimize the hyperparameters of the support vector regression (SVR) model, and use the feature matrix after PCA dimensionality reduction as the input of the SVR model;

[0025] Based on the particle swarm optimization (PSO) algorithm, set the particle swarm size to include three categories, and determine and represent the hyperparameters of the SVR model based on historical data, including the penalty factor, the kernel function parameter and the allowable error range;

[0026] Train the SVR model based on the hyperparameters of randomly initialized particles, and calculate the root mean square error (RMSE) of the current particle corresponding SVR model on the test set as the fitness function;

[0027] At the same time, update the particle position and velocity, and perform iterative calculations. During continuous iteration When the decline loss is no longer obvious, stop the iteration and select the SVR model corresponding to the particle with the smallest RMSE as the final model;

[0028] Based on the newly collected converter operation data, the equivalent capacitance value of the capacitor at the corresponding time point is output through the SVR model, and the multiple capacitance prediction values within a continuous time are fitted into an attenuation curve to predict the capacitance state value.

[0029] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: a redundant control path is constructed for backup capacitance state prediction, the output of the SVR model is used as the main control path, and a dual control path is introduced to add a redundant control path;

[0030] Based on the attenuation curve data of the converter health prediction, a one-dimensional linear regression model is constructed by least squares fitting, expressed as:

[0031] ;

[0032] ;

[0033] ;

[0034] where represents the linear capacitance prediction value, represents time, a represents the slope of the regression line, b represents the intercept of the regression line, represents the number of data points of the attenuation curve, represents the data at the i-th time point, represents the output value of the SVR model at the i-th time point, and respectively represent and the average values of

[0035] Compare the SVR model output with the redundant control path prediction output result in real time. Based on the sum of the mean and standard deviation of the historical error values as the deviation threshold, if the real-time comparison result is greater than or equal to the deviation threshold, it is judged that the deviation is large, and the prediction output result of the redundant control path is switched to be used.

[0036] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: based on the prediction of the capacitance value, optimize the load of the balance converter according to the real-time power flow, dynamically allocate the output power of the converter according to the prediction of the capacitance value, and monitor the bus voltage and the current output power of the converter in real time through voltage sensors and current sensors;

[0037] The basic Droop control calculates the current output power of the converter according to the difference between the actual voltage and the set voltage, expressed as:

[0038] ;

[0039] where represents the output power of the converter i at time t, represents the maximum output power of the converter i, and k represents the Droop coefficient, represents the bus voltage at time t, represents the set value of the bus voltage;

[0040] The model predictive control MPC is introduced, and a load dynamic model is established by fitting the historical power curve, which is expressed as:

[0041] ;

[0042] where represents the load power at time t, represents the trend curvature of the load change, represents the linear change speed of the load with time, represents the base load power;

[0043] Based on the load dynamic model, the power demand for future time steps is calculated, and an adaptive dynamic compensation mechanism is introduced. The difference between the actual load power and the total output power of the converters is calculated at each time step as the power distribution error, which is expressed as:

[0044] ;

[0045] ;

[0046] where represents the power distribution error, represents the total number of data, represents the corrected Droop coefficient, represents the compensation gain;

[0047] After correcting the Droop coefficient, the output power of the converter is dynamically calculated.

[0048] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: the data synchronization monitoring is performed through the power monitoring unit. A power monitoring unit PMU is embedded in each converter to real-time monitor the output power, voltage, and current state parameters, and communicate with other converters through the CAN bus to form a distributed monitoring network;

[0049] One converter is set as the master converter, which is responsible for summarizing the power states of each converter and distributing the load, and other converters act as slave converters to respond to the instructions of the master converter;

[0050] Based on the monitoring period, the master converter collects the output power data of all slave converters, calculates the total load, and distributes it to each converter according to the calculated load ratio to balance the load.

[0051] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: a virtual simulation model is constructed for data simulation. Through digital twin technology, a virtual simulation model of the converter and capacitor is constructed based on MATLAB.

[0052] Through IoT devices and sensors, the data of the physical system is synchronously transmitted to the digital twin in real time, and scenarios such as sudden load increase, overload, and capacitor attenuation are simulated in the digital environment to optimize the parameters of the Droop controller and MPC.

[0053] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: according to the temperature change, the heat dissipation strategy is dynamically adjusted. A temperature sensor is installed on the surface of the capacitor to monitor the temperature change in real time. The safe temperature range of the capacitor is set, and the temperature data is transmitted to the master converter to dynamically adjust the heat dissipation strategy according to the temperature change.

[0054] Based on the device heat dissipation temperature standard, a high temperature threshold is determined. If the monitored temperature data exceeds the high temperature threshold, the cooling fan is started to execute the heat dissipation strategy.

[0055] As a preferred solution of the back-to-back test platform for the energy storage converter described in the present invention, wherein: the realization of energy feedback storage and energy cycle means that during the back-to-back test process, the excess energy is fed back to the energy storage device. Through the electric energy metering device, the energy flow direction between the DC bus and the power grid is monitored. When the load decreases and the system has excess power, the inverter automatically switches to the energy feedback mode, and the remaining power is fed back to the power grid and the energy storage device to realize the recycling of energy.

[0056] The beneficial effects of the present invention are as follows: through the CNN-PSO-SVR model, it can respond in real time when the capacitance state changes, ensure that the capacitance attenuation state can be quickly fed back to the control system, ensure the continuous and stable operation of the capacitor under complex load conditions, compensate the power output in time through redundant paths, and avoid the shutdown of the overall system due to the failure of a single component. Through the dual-path capacitor health monitoring and Droop-MPC dynamic power distribution, the power output can be adjusted in time when the capacitor shows early attenuation, and the shutdown of the system caused by a single point of failure can be avoided. Description of the Drawings

[0057] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0058] Figure 1 It is a schematic structural diagram of the back-to-back test platform for the energy storage converter in Embodiment 1.

[0059] Figure 2 It is a schematic flow diagram of the back-to-back test platform for the energy storage converter in Embodiment 1. Specific Embodiments

[0060] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention with reference to the drawings in the specification.

[0061] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0062] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0063] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a back-to-back test platform for an energy storage converter, including the following steps:

[0064] S1, Connect to the power grid using a voltage source converter and collect data under different load conditions;

[0065] Preferably, connecting to the power grid using a voltage source converter means using a voltage source back-to-back converter VSC, which includes two bidirectional voltage source inverters. The AC output terminal of the first bidirectional voltage source inverter is connected to the grid side, and the AC output terminal of the second bidirectional voltage source inverter is connected to the load side;

[0066] The DC sides of two bidirectional voltage source inverters share the same DC bus. A supercapacitor energy storage module (rated voltage 1000V, capacity 100F) is connected in parallel to the bus to provide short-term power support, ensure the dynamic response speed, and use a bidirectional DC-DC converter to connect the DC bus and the supercapacitor to control the bidirectional flow of energy between the DC bus and the energy storage module.

[0067] The supercapacitor energy storage module provides short-term power support. When power fluctuations occur on the grid side or the load side, it responds quickly to make up for the power gap, avoiding the bus voltage drop caused by sudden load increase or the operation of the inverter overcurrent protection, reducing the risk of system shutdown, improving the ability of the converter to handle impact loads (such as motor startup or large-power equipment input), ensuring the continuous power supply of critical loads. By controlling the energy flow between the supercapacitor and the DC bus through a bidirectional DC-DC converter, it allows energy to be fed back to the supercapacitor when the load decreases, realizing energy recovery and storage. Through the power complementarity between the grid side and the load side, the bus power balance is achieved, reducing the operating pressure of the inverter, extending the service life of the equipment, supporting the parallel operation of multiple converters, improving the system power redundancy, reducing the impact of a single converter failure on the system, and enhancing the overall system fault tolerance.

[0068] Furthermore, data is collected under different load conditions. Based on different load conditions, data is collected, and the voltage, current, temperature, and resistance data are recorded through voltage sensors, current sensors, temperature sensors, and capacitance equivalent series resistance (ESR) sensors.

[0069] A high-frequency sampling circuit (bandwidth above 10kHz) is introduced on the DC bus to transmit the voltage pulsation signal to the ADC sampler, and the DSP performs digital signal processing to extract the high-frequency components.

[0070] Perform a fast Fourier transform on the DC bus voltage to extract the high-frequency voltage spectrum characteristic data and store it through a sliding window mechanism.

[0071] The data storage adopts a dual-channel backup mechanism. The main storage channel is the internal memory, and the backup channel is an external storage device.

[0072] Through voltage sensors, current sensors, temperature sensors, and ESR sensors, comprehensively monitor core parameters such as capacitance, current, voltage, and temperature, realizing all-round data acquisition of the system operation status. Through ESR (equivalent series resistance) monitoring, it can directly reflect the degree of capacitor aging, early warning of capacitor attenuation and failure risks. By performing a fast Fourier transform (FFT) on the DC bus voltage, convert the time-domain signal to the frequency domain, extract the high-frequency voltage spectrum characteristics of the capacitor state, extract the frequency-domain characteristics in the capacitor health state, compare with the standard spectrum, and quickly locate the source of abnormal capacitors or bus voltage fluctuations. Through the sliding window mechanism, it can dynamically update the continuous data stream, always retain the latest key data, and delete expired data at the same time to ensure the efficient use of storage space. By adopting a dual-channel backup mechanism, ensure the redundant storage of key monitoring data, and avoid data loss caused by single-channel failures or memory damage.

[0073] S2, construct a deep learning model to extract features, and calculate the fitness function to predict the capacitor state value;

[0074] Preferably, construct a deep learning model to extract features, and calculate the fitness function to predict the capacitor state value. Based on a convolutional neural network, construct a deep learning model, including an input layer, a convolutional layer, and an output layer;

[0075] Use the data collected under different load conditions as training data for model training. Select the cross-entropy loss function to calculate the calculation loss between the class probabilities predicted by the CNN and the actual labels. Use the Adam optimizer for gradient descent optimization to update the weight parameters of the CNN model. Stop the iteration when the loss of the model no longer decreases significantly during continuous iteration;

[0076] Based on the newly collected converter test data, input it into the deep learning model, and perform dimensionality reduction on the model output data using principal component analysis PCA;

[0077] The dimensionality reduction target can be set to retain 95% of the feature information volume, and finally reduce it to a feature matrix of 5 principal components;

[0078] Use the particle swarm optimization PSO algorithm to optimize the hyperparameters of the support vector regression SVR model, and use the feature matrix after PCA dimensionality reduction as the input of the SVR model;

[0079] Based on the particle swarm optimization PSO algorithm, set the particle swarm size to include three categories, and determine and represent the hyperparameters of the SVR model based on historical data, including the penalty factor, kernel function parameter, and allowable error range;

[0080] Based on the randomly initialized hyperparameters of the particles, train the SVR model, and calculate the root mean square error RMSE of the current particle corresponding SVR model on the test set as the fitness function, expressed as:

[0081] ;

[0082] where represents the root mean square error value, N represents the total number of samples, and represent the predicted value of the SVR model and the actual measured value respectively;

[0083] At the same time, update the particle position and velocity, and perform iterative calculation. During the continuous iterative process when the decreasing loss is no longer obvious, stop the iteration, and select the SVR model corresponding to the particle with the minimum RMSE as the final model;

[0084] Based on the newly collected converter operation data, output the equivalent capacitance value of the capacitor at the corresponding time point through the SVR model, and fit the multiple capacitance prediction values within the continuous time into an attenuation curve to predict the capacitance state value.

[0085] Extract complex capacitance data features through a convolutional neural network (CNN), combine principal component analysis (PCA) for dimensionality reduction and particle swarm optimization (PSO) to optimize the support vector regression model (SVR) to achieve high-precision prediction of the capacitance state. PSO can quickly search for the optimal solution in the multi-dimensional parameter space to ensure the stable operation of the SVR model under complex load and environmental conditions. Use PCA to reduce the dimensionality of the high-dimensional features extracted by the CNN. On the premise of maintaining high prediction accuracy, reduce the computational cost and storage requirements of the model, which is suitable for embedded devices or industrial control systems. Through the SVR model, predict the capacitance state at continuous time points, fit the capacitance attenuation curve, and realize the early prediction of the capacitance life, avoid system failures caused by sudden capacitor failures. The capacitance state attenuation curve can intuitively reflect the capacitance health status, provide intuitive warning information for maintenance personnel, and perform capacitor replacement or maintenance in advance. Based on the CNN-PSO-SVR model, it can respond in real time when the capacitance state changes, ensure that the capacitance attenuation state can be quickly fed back to the control system, ensure the continuous and stable operation of the capacitor under complex load conditions, be suitable for dynamic load fluctuation scenarios, quickly adjust the power flow distribution when the load changes violently, prevent capacitor overload or failure, improve the anti-interference ability of the converter system, and reduce system fluctuations or bus voltage instability phenomena caused by abnormal capacitance states.

[0086] S3. Construct a redundant control path for backup capacitance state prediction;

[0087] Preferably, construct a redundant control path for backup capacitance state prediction, use the output of the SVR model as the main control path, and introduce a dual control path to add a redundant control path;

[0088] Based on the decay curve data of the converter health prediction, a one-dimensional linear regression model is constructed by least squares fitting, which is expressed as:

[0089] ;

[0090] ;

[0091] ;

[0092] where represents the predicted value of the linear capacitance, represents time, a represents the slope of the regression line, b represents the intercept of the regression line, represents the number of data points of the decay curve, represents the data at the i-th time point, represents the output value of the SVR model at the i-th time point, and respectively represent and the average values of;

[0093] The output of the SVR model is compared with the predicted output result of the redundant control path in real time. Based on the sum of the mean and standard deviation of the historical error values as the deviation threshold, if the real-time comparison result is greater than or equal to the deviation threshold, it is judged that the deviation is large, and the predicted output result of the redundant control path is switched to be adopted.

[0094] Through the dual-path architecture of the SVR model main control path and the linear regression redundant path, it is ensured that when an abnormality or large deviation occurs in the main path prediction, it will automatically switch to the redundant path to ensure the continuous and stable operation of the system. The linear regression model fits the decay curve based on the least squares method, and can provide stable prediction results even in complex environments, reducing the system failure rate. When a calculation abnormality or short-term prediction distortion occurs in the main path, the redundant path can quickly take over to prevent the interruption of the capacitor health monitoring system and ensure the uninterrupted operation of the system, improving the test effect. The linear regression model has a low computational complexity and can provide effective capacitor health prediction even under the condition of limited hardware performance. By comparing the output of the SVR model with the predicted value of the redundant path in real time, the prediction accuracy of the main path is continuously monitored to ensure that abnormal deviations are identified and corrected in a timely manner. When abnormal fluctuations occur in the main control path prediction, the redundant path can intervene in a timely manner to avoid system overload or capacitor failure caused by misjudgment of the capacitor state, and early warning of potential risks. Using the least squares method to fit the capacitor decay curve, the change trend of the capacitor state is transformed into an intuitive mathematical model, providing long-term reliable data support for capacitor maintenance and replacement. The linear regression model accurately fits the long-term decay characteristics, enabling maintenance personnel to intuitively predict the capacitor failure time point and optimize the maintenance strategy;

[0095] Under different load conditions, the system can continuously monitor the health status of capacitors and dynamically adjust the output power of the converter according to the real-time power demand. When the load suddenly increases or decreases, the system can respond quickly to prevent large fluctuations in the bus voltage and ensure the continuous and stable operation of the entire system under complex working conditions. Even if some capacitors decay or fail, the power output can be compensated in a timely manner through redundant paths to avoid the shutdown of the overall system due to the failure of a single component.

[0096] S4, based on the prediction of capacitance value, optimize and balance the load of the converter according to the real-time power flow;

[0097] Preferably, based on the prediction of capacitance value, optimize and balance the load of the converter according to the real-time power flow, dynamically allocate the output power of the converter according to the prediction of capacitance value, and continuously monitor the bus voltage and the current output power of the converter through voltage sensors and current sensors.

[0098] The basic Droop control calculates the current output power of the converter according to the difference between the actual voltage and the set voltage, expressed as:

[0099] ;

[0100] where represents the output power of converter i at time t, represents the maximum output power of converter i, k represents the Droop coefficient, which is determined by experiments, represents the bus voltage at time t, represents the set value of the bus voltage;

[0101] Adjust the output power of the converter by reducing the output voltage to achieve automatic power distribution. The core of Droop control is to adjust the output power of the converter according to the real-time load condition, so that the parallel converters share the load together to prevent uneven power distribution;

[0102] Introduce model predictive control MPC, and establish a load dynamic model by fitting the historical power curve, expressed as:

[0103] ;

[0104] where represents the load power at time t, represents the trend curvature of the load change, represents the linear change speed of the load with time, represents the basic load power, , and are solved by the least squares method, by fitting the known load power data and minimizing the error between the predicted power and the actual power.

[0105] Although Droop control can respond quickly to load fluctuations, it mainly relies on real-time feedback and there is a certain delay. To further improve the dynamic response speed, model predictive control (MPC) is introduced. MPC can predict future loads in advance based on the load change trend and add feedforward control in the power distribution process to reduce the hysteresis of Droop control, allocate power in advance, and avoid capacitor overload;

[0106] Calculate the power demand for future time steps based on the load dynamic model, and introduce an adaptive dynamic compensation mechanism. Calculate the difference between the actual load power and the total output power of the converters at each time step as the power distribution error, expressed as:

[0107] ;

[0108] ;

[0109] where represents the power distribution error, represents the total number of data, represents the corrected Droop coefficient, represents the compensation gain. Based on experiments, it is usually set between 0.05 and 0.1;

[0110] Although Droop control and MPC can optimize power distribution under most working conditions, due to the uncertainty and non-linear fluctuations of the load, the Droop coefficient k may deviate after long-term operation, resulting in inaccurate power distribution. Therefore, the system introduces an adaptive dynamic compensation mechanism to dynamically adjust the Droop coefficient by real-time monitoring of the power distribution error to achieve a more accurate power flow distribution;

[0111] After correcting the Droop coefficient, dynamically calculate the output power of the converter.

[0112] By combining Droop control with MPC (Model Predictive Control), it is possible to dynamically allocate the output power of converters according to the real-time fluctuations of the bus voltage and the future load change trend, prevent individual converters from being overloaded or unloaded, improve the operation efficiency and stability of the overall system. After long-term operation, the Droop coefficient may deviate due to capacitor aging or load fluctuations. The adaptive compensation mechanism can automatically correct the error, ensure the power distribution accuracy, and prevent uneven power distribution. Especially in the environment of parallel operation of multiple converters, it helps to optimize the power flow distribution and maintain the stable operation state of the system. Through dual-path capacitor health monitoring and Droop-MPC dynamic power distribution, it is possible to adjust the power output in a timely manner when the capacitor shows early attenuation, avoid system shutdown caused by single-point failures. This strategy has a high degree of redundancy, ensuring that even if the main path fails, the redundant path can still maintain the basic operation of the system and reduce the unplanned downtime. Through the optimized control strategy of power flow, it effectively avoids the capacitor being in a high-load state for a long time, reduces the phenomena of overcurrent and overheating of the capacitor, and reduces the capacitor loss;

[0113] The real-time monitoring of the capacitor health status is achieved through the main path (PSO-SVR model) and the redundant path (linear regression model). However, the monitoring of the health status only provides feedback on the system operation situation and does not directly intervene in the power flow distribution of the system. Therefore, in order to ensure that the capacitor participates in the system operation reasonably under different health states, balance the loads of each converter, prevent local equipment from being overloaded or having uneven operation efficiency, it is necessary to introduce a real-time power flow optimization control strategy. According to the real-time health status of the capacitor, dynamically adjust the output power of each converter to match the system load demand, and evenly distribute the power according to the load conditions of each converter module to avoid overloading of individual converters;

[0114] When the health status of the capacitor gradually deteriorates, the system can issue an early warning signal in advance through the capacitor attenuation curve, enabling the maintenance personnel to perform necessary maintenance or replacement before the capacitor fails. Avoid the impact on the system caused by the sudden failure of the capacitor, reduce the possibility of unexpected shutdowns, and ensure the continuity and stability of production and power supply.

[0115] S5, perform data synchronization monitoring through the power monitoring unit;

[0116] Preferably, perform data synchronization monitoring through the power monitoring unit. Embed the power monitoring unit PMU in each converter to monitor the output power, voltage, and current status parameters in real time, and communicate with other converters through the CAN bus to form a distributed monitoring network;

[0117] Set one converter as the master converter, which is responsible for summarizing the power status of each converter and allocating loads. Other converters act as slave converters and respond to the instructions of the master converter;

[0118] Based on the monitoring period, the master converter collects the output power data of all slave converters, calculates the total load, and distributes it to each converter according to the calculated load ratio to balance the load.

[0119] Through the power monitoring unit (PMU), the master converter can monitor each converter in real time, dynamically grasp the operating status and output power of each converter, and allocate power in real time according to the load conditions to ensure that the output power of each converter remains balanced. When the load suddenly increases or is unevenly distributed, the master converter can quickly reallocate the load to prevent a certain converter from being overloaded for a long time, reducing the risk of overheating or damage to local converters. Even in the case of instantaneous load changes, the system can quickly adjust the power output of each converter to prevent bus voltage fluctuations or short-term overload of the converter. Real-time monitoring enables the system to have higher dynamic response capabilities and meet the operating requirements of fast load change scenarios.

[0120] S6, construct a virtual simulation model for data simulation;

[0121] Preferably, construct a virtual simulation model for data simulation. Through digital twin technology, a virtual simulation model of the converter and capacitor is constructed based on MATLAB;

[0122] Through IoT devices and sensors, the data of the physical system is synchronized to the digital twin in real time, and scenarios such as sudden load increase, overload, and capacitor attenuation are simulated in the digital environment to optimize the parameters of the Droop controller and MPC.

[0123] By improving the digital twin system, the operating status of the capacitor and converter can be simulated in the virtual environment. Through real-time monitoring and simulation, the capacitor attenuation process and potential failure risks can be predicted. Before the actual deployment of the converter and capacitor system, a complete debugging and load test can be completed in the digital twin, reducing the on-site equipment debugging time and accelerating the equipment commissioning speed. Various complex working conditions are simulated in the virtual simulation to optimize the control strategy and hardware configuration in advance, reducing the on-site repeated debugging and experimental costs.

[0124] S7, dynamically adjust the heat dissipation strategy according to the temperature change;

[0125] Preferably, dynamically adjust the heat dissipation strategy according to the temperature change. Install temperature sensors on the surface of the capacitor to monitor the temperature change in real time, set the safe temperature range of the capacitor, transmit the temperature data to the master converter, and dynamically adjust the heat dissipation strategy according to the temperature change;

[0126] Based on the equipment heat dissipation temperature standard, determine the high-temperature threshold. If the monitored temperature data exceeds the high-temperature threshold, start the cooling fan to execute the heat dissipation strategy.

[0127] By monitoring the capacitor temperature in real time, when the capacitor approaches or exceeds the safe temperature range, the cooling system can be quickly activated to prevent the capacitor from being damaged or its performance from degrading due to overheating. By dynamically adjusting the cooling strategy, the loss of capacitor materials caused by long-term high-temperature operation can be avoided, ensuring the stable operation of the system, reducing the risk of capacitor failure caused by overheating, guaranteeing the continuous and stable operation of the converter system, and reducing system downtime or unexpected failures caused by capacitor failure.

[0128] S8, perform energy feedback storage to achieve energy cycling;

[0129] Preferably, performing energy feedback storage to achieve energy cycling means that during the back-to-back test, the excess energy is fed back to the energy storage device. Through the electric energy metering device, the energy flow direction between the DC bus and the power grid is monitored. When the load decreases or there is excess power in the system, the inverter automatically switches to the energy feedback mode, and the remaining power is fed back to the power grid or the energy storage device to achieve the recycling of energy.

[0130] Through the energy cycling mechanism, the excess energy is stored and released for reuse when the load increases, ensuring the full utilization of energy and improving the overall energy efficiency ratio. When the load suddenly decreases, the excess electric energy is fed back to the energy storage device or the power grid, avoiding overload or damage to the converter caused by the increase in bus voltage, effectively protecting the safety of system equipment, and reducing the impact of load fluctuations on the equipment. In industrial scenarios with large fluctuations in energy consumption, the system can quickly adjust the direction of electric energy flow, avoiding power shortages or surpluses, and improving the load adaptability and stability of the system.

[0131] In summary, through the CNN-PSO-SVR model, the present invention can respond in real time when the capacitor state changes, ensure that the capacitor attenuation state can be quickly feedback to the control system, guarantee the continuous and stable operation of the capacitor under complex load conditions, compensate the power output in a timely manner through redundant paths, avoid the shutdown of the overall system caused by the failure of a single component, and through the dual-path capacitor health monitoring and Droop-MPC dynamic power distribution, can adjust the power output in a timely manner when the capacitor shows early attenuation, avoiding system shutdown caused by single-point failures.

[0132] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A back-to-back test platform for energy storage converters, characterized in that: include: The data acquisition module uses a voltage source converter to access the power grid and collect data under different load conditions; The capacitor health prediction module builds a deep learning model to extract features and calculates the fitness function to predict the capacitor status value; A redundant control path module, which constructs a redundant control path to predict the state of the backup capacitor; A load balancing module, based on the prediction of the capacitance value, optimizes and balances the load of the converter according to the real-time power flow; Data synchronization module, which performs data synchronization monitoring through a power monitoring unit; Data simulation module, building a virtual simulation model for data simulation; The temperature management module dynamically adjusts the heat dissipation strategy according to temperature changes; The energy storage module performs energy feedback storage to realize energy circulation.

2. The energy storage converter back-to-back test platform according to claim 1, characterized in that: The use of a voltage source converter to access the grid refers to the use of a voltage source back-to-back converter VSC, which includes two bidirectional voltage source inverters, the AC output end of the first bidirectional voltage source inverter is connected to the grid side, and the AC output end of the second bidirectional voltage source inverter is connected to the load side.

3. The energy storage converter back-to-back test platform according to claim 2, characterized in that: The collecting of data under different load conditions is based on collecting data under different load conditions, and recording the data of voltage, current, temperature and resistance through a voltage sensor, a current sensor, a temperature sensor and a capacitor equivalent series resistance (ESR) sensor; A high-frequency sampling circuit is introduced on the DC bus to transmit the voltage pulsation signal to the ADC sampler; Perform fast Fourier transform on the DC bus voltage, extract high-frequency voltage spectrum feature data, and store it through a sliding window mechanism; Data storage adopts a dual-channel backup mechanism, the main storage channel is the internal memory, and the backup channel is the external storage device.

4. The energy storage converter back-to-back test platform according to claim 3, characterized in that: The deep learning model is constructed to extract features and calculate the fitness function to predict the capacitance state value. The deep learning model is constructed based on a convolutional neural network, including an input layer, a convolutional layer and an output layer; The data collected under different load conditions are used as training data for model training. The cross entropy loss function is selected to calculate the computational loss between the category probability predicted by CNN and the actual label. The Adam optimizer is used for gradient descent optimization to update the weight parameters of the CNN model. If the model loss no longer decreases significantly during continuous iterations, the iteration is stopped. Based on the newly collected converter test data, the deep learning model is input, and the principal component analysis (PCA) is used to reduce the dimension of the model output data; The particle swarm optimization (PSO) algorithm is used to optimize the hyperparameters of the support vector regression (SVR) model, and the feature matrix after PCA dimension reduction is used as the input of the SVR model. The particle swarm size setting based on the particle swarm optimization PSO algorithm includes three categories, which are determined based on historical data and represent the hyperparameters of the SVR model, including penalty factors, kernel function parameters, and allowable error range; The SVR model is trained based on the hyperparameters of randomly initialized particles, and the root mean square error RMSE of the SVR model corresponding to the current particle on the test set is calculated as the fitness function; At the same time, the particle position and velocity are updated and iterative calculations are performed. When the drop in loss is no longer obvious, the iteration is stopped, and the SVR model corresponding to the particle with the smallest RMSE is selected as the final model; Based on the newly collected converter operation data, the equivalent capacitance value of the capacitor at the corresponding time point is output through the SVR model, and multiple capacitance prediction values ​​in continuous time are fitted into a decay curve to predict the capacitance state value.

5. The energy storage converter back-to-back test platform according to claim 4, characterized in that: The redundant control path is constructed to predict the state of the backup capacitor, the output of the SVR model is used as the main control path, and a dual control path is introduced to add a redundant control path; Based on the attenuation curve data of the converter health prediction, a one-dimensional linear regression model is constructed by least squares fitting, which is expressed as: ； ； ； in represents the predicted value of linear capacitance, represents time, a represents the slope of the regression line, b represents the intercept of the regression line, represents the number of data points of the decay curve, represents the data at the i-th time point, represents the output value of the SVR model at the i-th time point, and Respectively and The average value of The SVR model output is compared with the redundant control path prediction output result in real time. The sum of the mean and standard deviation of the historical error values ​​is used as the deviation threshold. If the real-time comparison result is greater than or equal to the deviation threshold, it is judged that the deviation is large and the prediction output result of the redundant control path is switched.

6. The energy storage converter back-to-back test platform according to claim 5, characterized in that: The load of the converter is balanced based on the prediction of the capacitance value, according to the real-time power flow optimization, the output power of the converter is dynamically allocated according to the prediction of the capacitance value, and the bus voltage and the current output power of the converter are monitored in real time through the voltage sensor and the current sensor; Basic Droop control calculates the current output power of the converter based on the difference between the actual voltage and the set voltage, expressed as: ； in represents the output power of converter i at time t, represents the maximum output power of converter i, k represents the Droop coefficient, represents the bus voltage at time t, Indicates the bus voltage setting value; Model predictive control (MPC) is introduced, and the load dynamic model is established by fitting the historical power curve, which is expressed as: ； in represents the load power at time t, Indicates the trend curvature of load change, Indicates the linear change rate of load over time, Indicates the base load power; The power demand of future time steps is calculated based on the load dynamic model, and an adaptive dynamic compensation mechanism is introduced. The difference between the actual load power and the total output power of the converter is calculated at each time step as the power allocation error, which is expressed as: ； ； in represents the power allocation error, Indicates the total number of data. represents the modified Droop coefficient, represents the compensation gain; After correcting the Droop coefficient, the converter output power is dynamically calculated.

7. The back-to-back test platform for energy storage converters according to claim 6, characterized in that: The data is synchronously monitored by the power monitoring unit, and a power monitoring unit PMU is embedded in each converter to monitor the output power, voltage, and current state parameters in real time, and communicate with other converters through the CAN bus to form a distributed monitoring network; One converter is set as the master converter, which is responsible for summarizing the power status of each converter and distributing the load. Other converters are set as slave converters, which respond to the instructions of the master converter. Based on the monitoring cycle, the master converter collects the output power data of all slave converters, calculates the total load, and distributes it to each converter according to the calculated load ratio to balance the load.

8. The back-to-back test platform for energy storage converters according to claim 7, characterized in that: The virtual simulation model is constructed to perform data simulation, and a virtual simulation model of the converter and the capacitor is constructed based on MATLAB through the digital twin technology; Through IoT devices and sensors, the data of the physical system is synchronized to the digital twin in real time, and load surge, overload and capacitance attenuation scenarios are simulated in the digital environment to optimize the Droop controller and MPC parameters.

9. The energy storage converter back-to-back test platform according to claim 8, characterized in that: According to the temperature change, the heat dissipation strategy is dynamically adjusted, a temperature sensor is installed on the surface of the capacitor, the temperature change is monitored in real time, a safe temperature range of the capacitor is set, the temperature data is transmitted to the main control converter, and the heat dissipation strategy is dynamically adjusted according to the temperature change; A high temperature threshold is determined based on the device heat dissipation temperature standard. If the monitored temperature data exceeds the high temperature threshold, the heat dissipation fan is started to execute the heat dissipation strategy.

10. The back-to-back test platform for energy storage converters according to claim 9, characterized in that: The energy feedback storage to achieve energy circulation refers to feeding back excess energy to the energy storage device during the back-to-back test, monitoring the energy flow direction between the DC bus and the power grid through the electric energy metering device, and when the load is reduced and the system has excess power, the inverter automatically switches to the energy feedback mode, feeding back the excess power to the power grid and the energy storage device, thereby achieving energy recycling.

Images