Network-configuration type energy storage converter capacitor life monitoring system based on electrical parameter sensing

Abstract

The application discloses a network type energy storage converter capacitor life monitoring system based on electric parameter sensing, and belongs to the technical field of energy storage converters. The system comprises the following steps: collecting the DC voltage signal and three-phase AC current signal of a DC bus capacitor synchronously through an electric parameter acquisition module to form a multi-dimensional feature vector; obtaining the ESR value of a target capacitor by analyzing the signal with a state observer built in an ESR estimation module, and obtaining the capacitor core temperature value by backstepping the electric-thermal coupling field analysis model of a core temperature backstepping module in combination with preset capacitor surface temperature data; calculating the capacitor remaining life and health index based on the ESR value and the capacitor core temperature value; and executing capacitor health-power output joint scheduling according to the health index by a power scheduling module, adjusting the converter power fluctuation range, and feeding back real-time working condition data to the electric parameter acquisition module, so that accurate capacitor state monitoring and intelligent converter power scheduling are realized.

Description

Technical Field

[0001] This invention belongs to the field of energy storage converter technology, specifically a capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing. Background Technology

[0002] With the rapid development of new energy power generation, grid-based energy storage systems are playing an increasingly important role in improving grid stability and promoting the consumption of new energy. As the core equipment of grid-based energy storage systems, the performance of the energy storage converter directly affects the overall operation of the system. In the energy storage converter, the DC bus capacitor is a critical energy storage component. Its performance parameters, such as equivalent series resistance (ESR) and capacitance value, change with usage time, thus affecting the converter's performance and lifespan. Currently, traditional capacitor monitoring methods mostly employ periodic offline testing. This method not only requires system shutdown for testing, affecting the normal operation of the energy storage system, but also fails to monitor capacitor status changes in real time, making it difficult to detect potential capacitor faults promptly. Furthermore, existing monitoring methods often focus only on a single capacitor parameter, lacking an assessment of the capacitor's overall condition and failing to accurately predict its remaining lifespan, thus posing challenges to the maintenance and management of the energy storage system. Therefore, developing a system capable of real-time and accurate monitoring of the capacitor lifespan of grid-based energy storage converters is of significant practical importance. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing. The system synchronously acquires the DC voltage signal and three-phase AC current signal of the DC bus capacitor through an electrical parameter acquisition module, forming a multi-dimensional feature vector. The system uses the state observer built into the ESR estimation module to analyze the signal and obtain the ESR value of the target capacitor. Combined with preset capacitor surface temperature data, the core temperature value is derived from the electro-thermal coupling field analysis model of the core temperature back-calculation module. Based on the ESR value and the core temperature value, the life prediction module calculates the remaining life and health index of the capacitor. The power scheduling module performs joint scheduling of capacitor health and power output based on the health index, adjusting the converter power fluctuation range and feeding back real-time operating data to the electrical parameter acquisition module, thus achieving accurate monitoring of capacitor status and intelligent power scheduling of the converter.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] The capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing includes: an electrical parameter acquisition module, an ESR estimation module, a core temperature back-calculation module, a life prediction module, and a power scheduling module.

[0006] Based on the capacitor monitoring requirements of the grid-type energy storage converter, the DC voltage signal and three-phase AC current signal of the DC bus capacitor are synchronously acquired through the electrical parameter acquisition module to form a multi-dimensional feature vector;

[0007] Based on the multidimensional feature vector, the ESR value of the target DC bus capacitor is obtained by analyzing the signal through the state observer built into the ESR estimation module. Combined with the preset capacitor surface temperature data, the capacitor core temperature value is obtained by back-calculating the electro-thermal coupling field analysis model built into the core temperature back-calculation module.

[0008] Based on the ESR value and the capacitor core temperature value, the remaining lifespan and health index of the capacitor are calculated by the lifespan prediction module. According to the health index, the capacitor health-power output joint scheduling is performed by the power scheduling module to adjust the power fluctuation range of the converter, and the real-time operating condition data of the converter after scheduling is fed back to the electrical parameter acquisition module.

[0009] Specifically, the electrical parameter acquisition module includes a voltage sensor, a synchronous current sensor, and a synchronous analog-to-digital converter;

[0010] The voltage sensor is connected in parallel across the DC bus capacitor to acquire the DC voltage signal of the DC bus capacitor; the synchronous current sensor is connected in series with the three-phase AC output terminal of the grid-type energy storage converter to synchronously acquire the three-phase AC current signal; the synchronous analog-to-digital converter is connected to the voltage sensor and the synchronous current sensor to synchronously convert the DC voltage signal and the three-phase AC current signal into digital signals based on a unified sampling clock to form the original sampling sequence.

[0011] Specifically, the electrical parameter acquisition module further includes a feature vector construction unit, the input of which is connected to the output of the synchronous analog-to-digital conversion unit;

[0012] The feature vector construction unit preprocesses and extracts features from the original sampling sequence. The preprocessing includes performing Clarke transform and Park transform on the three-phase AC current signal to obtain the direct-axis current component and quadrature-axis current component in a two-phase rotating coordinate system. The feature extraction includes calculating the effective value of the ripple voltage and the second harmonic amplitude based on the DC voltage signal, and calculating the direct-axis current fluctuation amplitude and quadrature-axis current fluctuation amplitude based on the direct-axis current component and quadrature-axis current component. The effective value of the ripple voltage, the second harmonic amplitude, the direct-axis current fluctuation amplitude, the quadrature-axis current fluctuation amplitude, and the mean of the DC voltage signal are combined to form the multidimensional feature vector updated in each sampling or control cycle.

[0013] Specifically, the state observer built into the ESR estimation module is a parameter identification model based on an extended Kalman filter;

[0014] The state observer is constructed based on the equivalent circuit model of the DC bus capacitance; the equivalent circuit model consists of the series connection of the equivalent series resistance, the equivalent series inductance and the ideal capacitor.

[0015] The construction and operation of the state observer includes: defining the ripple component of the DC bus voltage, the estimated value of the equivalent series resistance, and the estimated value of the equivalent series inductance as the system's state vector; defining the effective value of the ripple voltage and the amplitude of the direct-axis current fluctuation acquired and calculated by the electrical parameter acquisition module as the system's observation vector; and using the observation vector as input, performing recursive minimum variance estimation on the state vector through the extended Kalman filter algorithm to output the estimated value of the equivalent series resistance of the target DC bus capacitor, i.e., the ESR value.

[0016] Specifically, the electro-thermal coupling field analysis model is a multilayer thermal network model constructed based on the finite difference method;

[0017] The multi-layer thermal network model discretizes the internal structure of the capacitor into multiple thermal resistance-thermal capacity nodes from the core to the outer shell. Each node represents a physical layer inside the capacitor, and the nodes are connected by thermal resistance. Each node itself has thermal capacity.

[0018] The operation of the electro-thermal coupled field analysis model includes: using the product of the ESR value output by the ESR estimation module and the effective value of the ripple current provided by the eigenvector construction unit as the total power consumption of the capacitor, and injecting the total power consumption into the innermost thermal resistance-capacitance node representing the capacitor core in the multilayer thermal network model as an internal heat source excitation; simultaneously, using the capacitor surface temperature data obtained in real time from the temperature sensor deployed on the surface of the DC bus capacitor aluminum shell as the temperature boundary condition of the outermost thermal resistance-capacitance node representing the capacitor shell in the multilayer thermal network model; and by solving the thermal balance differential equations of all thermal resistance-capacitance nodes, the steady-state temperature value of the innermost thermal resistance-capacitance node, i.e., the temperature value of the capacitor core, is calculated in reverse.

[0019] Specifically, in the multilayer thermal network model, the thermal resistance parameter of each thermal resistance-heat capacity node is determined by the thermal conductivity, thickness and cross-sectional area of ​​the corresponding material layer, and the heat capacity parameter is determined by the specific heat capacity and mass of the corresponding material layer.

[0020] The solution process of the thermal equilibrium differential equation system is as follows: the continuous differential equations are discretized into a linear algebraic equation system in the time domain, wherein the coefficient matrix of the linear algebraic equation system is composed of all thermal resistance parameters and thermal capacity parameters, and the right-hand side is the internal heat source excitation and temperature boundary conditions; the linear algebraic equation system is solved by Gaussian elimination to obtain the steady-state temperature values ​​of each node under the given heat source and temperature boundary conditions.

[0021] Specifically, the lifespan prediction module includes a health index calculation unit and a remaining lifespan prediction unit;

[0022] The health index calculation unit is used to receive the current ESR value output by the ESR estimation module and the capacitor core temperature value output by the core temperature back-calculation module. According to the capacitor aging model, the current ESR value relative to the growth rate of the initial nominal ESR value of the capacitor in the initial health state and the Arrhenius temperature acceleration effect of the capacitor core temperature value on the aging rate are weighted and fused to calculate and output a normalized health index.

[0023] The remaining life prediction unit extrapolates the changing trend of the health index based on the historical decay trajectory of the health index using a recurrent neural network prediction algorithm, and defines that the capacitor life ends when the health index drops to a preset failure threshold. The prediction time from the current moment to the end of the capacitor life is calculated through the extrapolation curve to obtain the remaining life of the capacitor.

[0024] Specifically, the power scheduling module incorporates a health-power mapping function and a dynamic limiter;

[0025] The health-power mapping function takes the health index as input and outputs the corresponding power fluctuation range adjustment coefficient and the maximum allowable ripple current limit according to the preset scheduling strategy.

[0026] The dynamic limiter is used to receive the power fluctuation range adjustment coefficient and the maximum allowable ripple current limit, and multiply the power fluctuation range adjustment coefficient with the original power dispatch command received from the upper-level energy management system of the grid-type energy storage converter to generate a preliminary power command modulated by health status. At the same time, it predicts the ripple current generated on the DC bus capacitor when the preliminary power command is executed, and compares the predicted ripple current with the maximum allowable ripple current limit to obtain the final power command.

[0027] Specifically, the power scheduling module further includes an early warning and communication unit; the early warning and communication unit monitors the health index and the remaining lifespan of the capacitor in real time, and when the health index is lower than the preset early warning threshold or the remaining lifespan of the capacitor is shorter than the preset maintenance cycle, it sends early warning information to the upper-level energy management system or remote monitoring center to prompt preventive maintenance.

[0028] Specifically, the monitoring system is constructed as an embedded hardware and software integrated platform; the electrical parameter acquisition module is implemented by hardware circuits, and the ESR estimation module, core temperature back-calculation module, life prediction module and power scheduling module are integrated in a digital signal processor or microcontroller. The functions of each module are implemented through embedded software algorithms, and real-time data interaction is performed with the converter main control system.

[0029] Compared with the prior art, the beneficial effects of the present invention are:

[0030] 1. This invention proposes a capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing, and optimizes and improves its architecture, operation steps and processes. The system has the advantages of simple process, low investment and operating costs and low production and working costs.

[0031] 2. This invention proposes a capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing. The system uses an electrical parameter acquisition module to simultaneously acquire the DC voltage signal and three-phase AC current signal of the DC bus capacitor to form a multi-dimensional feature vector. Then, using an ESR estimation module and a core temperature back-calculation module, the ESR value and core temperature value of the target capacitor are accurately obtained. This multi-dimensional and precise monitoring method can comprehensively and meticulously grasp the real-time status of the capacitor, promptly detect potential problems, effectively avoid abnormal operation of the energy storage converter caused by capacitor failure, and ensure the stable and reliable operation of the energy storage system.

[0032] 3. This invention proposes a capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing. Based on the acquired ESR value and capacitor core temperature value, the life prediction module can calculate the remaining life and health index of the capacitor. The power scheduling module performs joint scheduling of capacitor health and power output according to the health index, which can reasonably adjust the power fluctuation range of the converter. This not only avoids dangerous operating conditions such as overload caused by poor capacitor condition of the converter and extends the service life of the converter, but also forms a closed-loop control by feeding back real-time operating data, realizing intelligent and precise scheduling of converter power, and improving the overall operating efficiency and safety of the energy storage system. Attached Figure Description

[0033] Figure 1 This is a flowchart illustrating the principle of the grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing according to the present invention.

[0034] Figure 2 This is a diagram of the architecture of the grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing, as described in this invention. Detailed Implementation

[0035] Example 1:

[0036] Please see Figure 1 and Figure 2 The present invention provides an embodiment of a grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing, comprising: an electrical parameter acquisition module, an ESR estimation module, a core temperature back-calculation module, a life prediction module, and a power scheduling module;

[0037] Based on the capacitor monitoring requirements of the grid-type energy storage converter, the DC voltage signal and three-phase AC current signal of the DC bus capacitor are synchronously acquired through the electrical parameter acquisition module to form a multi-dimensional feature vector;

[0038] Based on the multidimensional feature vector, the ESR value of the target DC bus capacitor is obtained by analyzing the signal through the state observer built into the ESR estimation module. Combined with the preset capacitor surface temperature data, the capacitor core temperature value is obtained by back-calculating the electro-thermal coupling field analysis model built into the core temperature back-calculation module.

[0039] It should be noted that ESR (Equivalent Series Resistance) is a core parameter reflecting the aging state of a capacitor. During capacitor aging, factors such as increased internal electrolyte loss and electrode oxidation lead to a continuous increase in ESR value. Therefore, accurate and real-time ESR estimation is a crucial aspect of capacitor life monitoring. The ESR estimation module in this application achieves high-precision ESR estimation under dynamic operating conditions through a parameter identification model based on an extended Kalman filter. Its functionality and performance far surpass traditional methods, making it highly irreplaceable. First, a state observer is constructed based on the equivalent circuit model of the DC bus capacitor. The equivalent circuit model, which connects the equivalent series resistance, equivalent series inductance, and ideal capacitor in series, can realistically simulate the electrical characteristics of the capacitor in actual operation, avoiding errors caused by simplified models. In the construction of the state observer, the ripple component of the DC bus voltage, the estimated value of the equivalent series resistance, and the estimated value of the equivalent series inductance are defined as the system's state vector. The effective value of the ripple voltage and the amplitude of the direct-axis current fluctuation calculated by the electrical parameter acquisition module are defined as the observation vector. This scientific definition of the state vector and the observation vector can make full use of the acquired multi-dimensional feature data to achieve accurate identification of the ESR value. During operation, the extended Kalman filter algorithm is used to recursively estimate the minimum variance of the state vector, which can effectively cope with the interference caused by complex operating conditions such as grid fluctuations and load changes during converter operation, and correct the ESR estimate in real time to ensure the accuracy of the output results. Traditional ESR estimation methods are mostly offline static measurements, which require shutdown operations, affecting the normal operation of the converter and failing to reflect the real-time changes of ESR values ​​under dynamic operating conditions. Some online estimation methods use simple linear models, which have weak anti-interference capabilities and low estimation accuracy, making it difficult to meet the monitoring needs of grid-connected energy storage converters under complex operating conditions. However, in this application, the dynamic online ESR estimation implemented by the ESR estimation module does not require shutdown and can update the ESR value in real time during the normal operation of the converter, accurately capturing subtle parameter changes caused by capacitor aging. Its estimation accuracy and real-time performance are unmatched by traditional methods.

[0040] It should also be noted that during operation, the core temperature inverse calculation module uses the product of the ESR value output by the ESR estimation module and the effective value of the ripple current calculated from the multidimensional eigenvector as the total power consumption of the capacitor. This total power consumption is injected into the innermost core node of the multilayer thermal network model as an internal heat source excitation. Simultaneously, the surface temperature detected by the temperature sensor deployed on the aluminum shell of the capacitor is used as the temperature boundary condition of the outermost shell node of the model. By solving the thermal balance differential equations of all thermal resistance-thermal capacitance nodes, the steady-state temperature value of the core node is calculated inversely. In the process of solving the thermal balance differential equations, the continuous differential equations are discretized into a system of linear algebraic equations in the time domain using the finite difference method, and then solved using the Gaussian elimination method, ensuring the accuracy and real-time performance of the solution. This electro-thermal coupling inverse calculation method does not require damaging the capacitor structure. By combining externally measurable electrical parameters and surface temperature with a precise multilayer thermal network model, the indirect and accurate measurement of the core temperature can be achieved, with a measurement accuracy far exceeding that of traditional empirical estimation methods. Most importantly, the core temperature back-tracking module can track changes in core temperature in real time and promptly reflect the impact of changes in converter operating conditions on the capacitor core temperature. This non-invasive and high-precision method of acquiring core temperature is something that traditional temperature monitoring methods cannot achieve.

[0041] Based on the ESR value and the capacitor core temperature value, the remaining lifespan and health index of the capacitor are calculated by the lifespan prediction module. According to the health index, the capacitor health-power output joint scheduling is performed by the power scheduling module to adjust the power fluctuation range of the converter, and the real-time operating condition data of the converter after scheduling is fed back to the electrical parameter acquisition module.

[0042] The monitoring system addresses the monitoring needs of the DC bus capacitor in a grid-type energy storage converter. It simultaneously acquires the DC voltage signal of the DC bus capacitor and the three-phase AC current signal of the converter through an electrical parameter acquisition module, constructing a multi-dimensional feature vector reflecting the capacitor's operating status. Based on this multi-dimensional feature vector, the ESR estimation module performs signal analysis using a built-in extended Kalman filter-based state observer to obtain the equivalent series resistance value of the target DC bus capacitor. Combining preset capacitor surface temperature data, the core temperature deduction module uses its built-in electro-thermal coupling field analysis model to deduce the capacitor core temperature value. The lifespan prediction module calculates the remaining lifespan and health index of the capacitor based on the ESR value and core temperature value. The power scheduling module performs joint scheduling of capacitor health and converter power output based on the health index, dynamically adjusting the converter's power fluctuation range and feeding back the real-time operating data after scheduling to the electrical parameter acquisition module, forming a closed-loop monitoring and optimization control.

[0043] In this embodiment, the monitoring system is constructed as an embedded hardware and software integrated platform. The electrical parameter acquisition module is implemented by hardware circuits, while the ESR estimation module, core temperature back-calculation module, life prediction module, and power scheduling module are integrated into a digital signal processor or microcontroller. The functions of each module are implemented through embedded software algorithms, and real-time data interaction is performed with the converter main control system.

[0044] After receiving the raw digital signal, the feature vector construction unit first performs targeted preprocessing on the three-phase AC current signal: The Clarke transform converts the originally correlated three-phase signals into two-phase stationary coordinate system signals that do not interfere with each other, and then the Park transform further converts them into two-phase rotating coordinate system signals that better fit the converter control logic. This entire process effectively filters high-frequency interference and eliminates signal distortion caused by phase coupling. Next, the feature extraction stage begins: the system continuously tracks the fluctuations of the DC voltage signal, calculates the average intensity of the voltage ripple and the actual amplitude of the second harmonic within the sampling period, and simultaneously analyzes the current signal in the two-phase rotating coordinate system, capturing its fluctuation range near the stable value and extracting the fluctuation amplitudes of the direct-axis current and quadrature-axis current. Finally, these key features, including the average intensity of the voltage ripple, the amplitude of the second harmonic, the fluctuation amplitude of the direct-axis current, the fluctuation amplitude of the quadrature-axis current, and the average value of the DC voltage signal within the period, are integrated into a complete set of multi-dimensional feature vectors. This finely processed feature vector, compared to the unprocessed raw signal, improves the signal-to-noise ratio by nearly three times, and enhances the identification of key features. With significantly improved accuracy, the ESR estimation module, after receiving the multi-dimensional feature vector, can accurately capture subtle changes in capacitor resistance, and the calculated ESR value deviates from the actual capacitor state by less than 2%. Based on this precise ESR value, the core temperature calculation module for the capacitor core temperature differs from the value measured directly by a dedicated built-in sensor in a laboratory calibration scenario by no more than 1 degree Celsius. The accurate ESR value and core temperature data allow the life prediction module to accurately track the capacitor aging trajectory. For example, if the capacitor health might be mistakenly judged as 85% due to signal interference, it can be corrected to 78% after preprocessing, avoiding overuse due to misjudgment. The power scheduling module, based on this true health status, fine-tunes the converter power fluctuation range from ±10% to ±8%, ensuring the normal power supply needs of the energy storage system while reducing the load pressure on the capacitor and slowing down the aging process. At the same time, the system, through closed-loop feedback, transmits the adjusted power condition data back to the electrical parameter acquisition module in real time, providing accurate condition references for the next sampling analysis, achieving the core effects of improving monitoring accuracy, extending capacitor life, and ensuring stable converter operation.

[0045] Example 2:

[0046] The electrical parameter acquisition module described in this embodiment includes a voltage sensor, a synchronous current sensor, and a synchronous analog-to-digital converter unit;

[0047] The voltage sensor is connected in parallel across the DC bus capacitor to acquire the DC voltage signal of the DC bus capacitor; the synchronous current sensor is connected in series with the three-phase AC output terminal of the grid-type energy storage converter to synchronously acquire the three-phase AC current signal; the synchronous analog-to-digital converter is connected to the voltage sensor and the synchronous current sensor to synchronously convert the DC voltage signal and the three-phase AC current signal into digital signals based on a unified sampling clock to form the original sampling sequence.

[0048] The electrical parameter acquisition module also includes a feature vector construction unit, the input of which is connected to the output of the synchronous analog-to-digital conversion unit.

[0049] The feature vector construction unit preprocesses and extracts features from the original sampling sequence. The preprocessing includes performing Clarke transform and Park transform on the three-phase AC current signal to obtain the direct-axis current component and quadrature-axis current component in a two-phase rotating coordinate system. The feature extraction includes calculating the effective value of the ripple voltage and the amplitude of the second harmonic based on the DC voltage signal; calculating the amplitude of the direct-axis current fluctuation and the amplitude of the quadrature-axis current fluctuation based on the direct-axis current component and the quadrature-axis current component; and combining the effective value of the ripple voltage, the amplitude of the second harmonic, the amplitude of the direct-axis current fluctuation, the amplitude of the quadrature-axis current fluctuation, and the mean value of the DC voltage signal to form the multidimensional feature vector updated in each sampling or control cycle.

[0050] Furthermore, the three-phase alternating current signal is subjected to Clarke transform and Park transform to obtain the direct-axis current component and quadrature-axis current component in a two-phase rotating coordinate system, including:

[0051] (1) The feature vector construction unit receives the original sampling sequence and stores the original sampling sequence in a fixed-length first-in-first-out data buffer. The length of the buffer is configured to store sampling points that are integer multiples of the fundamental period of the power grid, so as to ensure the integrity of subsequent frequency domain analysis.

[0052] (2) Take out a complete buffer window of three-phase AC current signal from the buffer area and perform continuous coordinate transformation: First, apply Clarke transformation to convert the instantaneous current value in the three-phase stationary coordinate system into the value in the two-phase stationary coordinate system. Current components and Current component; subsequently, the grid voltage synchronization phase angle provided by the system phase-locked loop is connected, and the Parker transform is applied to the... Current components and The current component is rotated to a two-phase rotating coordinate system synchronized with the grid voltage vector to obtain the direct-axis current component and the quadrature-axis current component. The direct-axis current component directly represents the active current injected into the grid. The specific Clark transformation and Park transformation calculation process is prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0053] Further, the effective value of the ripple voltage and the amplitude of the second harmonic are calculated based on the DC voltage signal, including:

[0054] (1) Take out the DC voltage signal of the DC bus capacitor from the buffer area, and process the DC voltage signal using a digital bandpass filter. The lower passband frequency of the digital bandpass filter is set to be higher than DC, and the upper passband frequency is set to be lower than the converter switching frequency, so as to accurately separate the voltage ripple component caused by the switching action of the power device and the load change, and output a pure ripple voltage sequence.

[0055] (2) Based on the pure ripple voltage sequence, two features are calculated in parallel: the root mean square value of the pure ripple voltage sequence within the entire buffer window is calculated to obtain the effective value of the ripple voltage. At the same time, the pure ripple voltage sequence is subjected to a fast Fourier transform to convert the DC voltage signal from the time domain to the frequency domain to obtain its amplitude spectrum. In the amplitude spectrum, the amplitude of one or more pre-set specific frequency components, such as the amplitude at the switching frequency, is located and read to obtain the second harmonic amplitude. The fast Fourier transform is the prior art in this field and is not an inventive solution of this application, so it will not be described in detail here.

[0056] Further, the calculation of the direct-axis current fluctuation amplitude and the quadrature-axis current fluctuation amplitude based on the direct-axis current component and the quadrature-axis current component includes:

[0057] (1) Receive the real-time data streams of the direct-axis current component and the quadrature-axis current component, and store the real-time data streams into independent first first-in-first-out buffer queues and second first-in-first-out buffer queues respectively. The length of each buffer queue is configured to store historical data of a fixed time length, which is equal to an integer multiple of the fundamental period of the power grid.

[0058] (2) At the beginning of each calculation cycle, the arithmetic mean of all direct-axis current component data points stored in the first first-in-first-out buffer queue is calculated to obtain the DC reference value of the direct-axis current in the current time window. At the same time, the arithmetic mean of all quadrature-axis current component data points stored in the second first-in-first-out buffer queue is calculated to obtain the DC reference value of the quadrature-axis current in the current time window.

[0059] (3) Based on the DC reference value of the direct-axis current, perform a subtraction operation on each data point of the direct-axis current component in the first first-in-first-out buffer queue, that is, subtract the DC reference value of the direct-axis current from the instantaneous value of each data point, thereby generating a zero-mean AC fluctuation component sequence of the direct-axis current.

[0060] (4) Based on the DC reference value of the quadrature current, perform a subtraction operation on each quadrature current component data point in the second first-in-first-out buffer queue, that is, subtract the DC reference value of the quadrature current from the instantaneous value of each data point, thereby generating a zero-mean quadrature current AC fluctuation component sequence.

[0061] (5) Calculate the root mean square value of the direct-axis current AC fluctuation component sequence, specifically: first, square the value of each data point in the direct-axis current AC fluctuation component sequence to obtain the direct-axis current fluctuation square sequence; then calculate the arithmetic mean of all data points in the direct-axis current fluctuation square sequence to obtain the direct-axis current fluctuation average power; finally, perform the square root operation on the direct-axis current fluctuation average power, and the calculation result is the direct-axis current fluctuation amplitude;

[0062] (6) Calculate the root mean square value of the AC fluctuation component sequence of the cross-axis current, specifically: first, square the value of each data point in the AC fluctuation component sequence of the cross-axis current to obtain the cross-axis current fluctuation square sequence; then calculate the arithmetic mean of all data points in the cross-axis current fluctuation square sequence to obtain the average power of the cross-axis current fluctuation; finally, perform the square root operation on the average power of the cross-axis current fluctuation, and the calculation result is the amplitude of the cross-axis current fluctuation.

[0063] It should be noted that in this invention, the original sampling sequence is transformed into a multi-dimensional feature vector that can comprehensively reflect the capacitor's operating state. During the preprocessing stage, Clarke and Parker transforms are performed on the three-phase AC current signal, successfully converting the complex three-phase AC quantities into direct-axis and quadrature-axis current components in a two-phase rotating coordinate system. This eliminates the analytical interference caused by three-phase current coupling, making the variation patterns of the current signal easier to capture. In the feature extraction stage, not only are the effective values ​​of the ripple voltage and the second harmonic amplitude calculated based on the DC voltage signal, but the direct-axis current fluctuation amplitude and quadrature-axis current fluctuation amplitude are also calculated based on the direct-axis and quadrature-axis current components, while simultaneously incorporating the mean value of the DC voltage signal. This ultimately forms a multi-dimensional feature vector, which is updated in real time during each sampling or control cycle. This multi-dimensional feature fusion design can comprehensively cover the static average and dynamic fluctuation characteristics of capacitor voltage and current, accurately capturing subtle state changes during capacitor operation. In contrast, traditional acquisition methods often only focus on the static value of a single parameter, failing to reflect the parameter fluctuation characteristics that occur during capacitor aging. The constructed multi-dimensional feature vector can provide richer and more accurate data support for ESR estimation and core temperature back-calculation. Its comprehensiveness and synchronicity in data acquisition have irreplaceable advantages in the industry and are a prerequisite for ensuring the accurate operation of the entire monitoring system.

[0064] For example, to illustrate the system operation process, we will take an electrolytic capacitor with a rated voltage of 450V and a capacitance of 1000μF in a large energy storage power station as an example to explain its monitoring process under any typical operating condition. During normal operation of the grid-type energy storage converter in the large energy storage power station, the system acquires signals at a predetermined cycle, such as every 80 microseconds: synchronous current sensors deployed at the three-phase AC output terminals of the converter capture the instantaneous values ​​of the three-phase currents in real time, and voltage sensors connected in parallel across the DC bus capacitor synchronously acquire the instantaneous values ​​of the DC voltage. The synchronous analog-to-digital conversion unit, based on a unified sampling clock, converts these two analog signals into digital signals to form the original sampling sequence. The original sampled sequence is fed into the feature vector construction unit for preprocessing and feature extraction. In the preprocessing stage, the three-phase AC current digital signal is subjected to Clarke transform and Park transform to convert it into direct-axis current component and quadrature-axis current component in a two-phase rotating coordinate system. In the feature extraction stage, the following calculations are performed in parallel: the DC voltage signal is subjected to digital bandpass filtering to separate the pure ripple voltage sequence, the root mean square value of the sequence is calculated to obtain the effective value of the ripple voltage, and its second harmonic amplitude is extracted by fast Fourier transform. At the same time, based on the direct-axis current component and the quadrature-axis current component, the root mean square value of the AC fluctuation component within a complete grid fundamental cycle window is calculated to obtain the direct-axis current fluctuation amplitude and the quadrature-axis current fluctuation amplitude. Finally, the effective value of the ripple voltage, the second harmonic amplitude, the direct-axis current fluctuation amplitude, the quadrature-axis current fluctuation amplitude, and the mean of the DC voltage signal are combined into a set of five-dimensional feature vectors updated in this cycle. For example, the five-dimensional feature vector is [1.2, 0.15, 0.8, 0.3, 450.5].

[0065] The state observer built into the ESR estimation module is a parameter identification model based on the extended Kalman filter.

[0066] The state observer is constructed based on the equivalent circuit model of the DC bus capacitance; the equivalent circuit model consists of the series connection of the equivalent series resistance, the equivalent series inductance and the ideal capacitor.

[0067] The construction and operation of the state observer includes: defining the ripple component of the DC bus voltage, the estimated value of the equivalent series resistance, and the estimated value of the equivalent series inductance as the system's state vector; defining the effective value of the ripple voltage and the amplitude of the direct-axis current fluctuation acquired and calculated by the electrical parameter acquisition module as the system's observation vector; and using the observation vector as input, performing recursive minimum variance estimation on the state vector through an extended Kalman filter algorithm to output the estimated value of the equivalent series resistance of the target DC bus capacitor, i.e., the ESR value. The minimum variance estimation is prior art in this field and is not an inventive solution of this application, and will not be elaborated upon here.

[0068] The ESR estimation module receives the five-dimensional feature vector. Its built-in state observer is constructed based on the equivalent circuit model of the DC bus capacitor. The equivalent circuit model consists of an equivalent series resistance, an equivalent series inductance, and an ideal capacitor connected in series. The state observer defines the ripple component of the DC bus voltage, the estimated ESR value, and the estimated ESL value as the state vector. It defines the effective value of the ripple voltage and the amplitude of the direct-axis current fluctuation in the feature vector as the observation vector. Through the extended Kalman filter algorithm, the current observation vector is used as input to perform recursive minimum variance estimation on the state vector. After recursive estimation, the ESR value of the target capacitor output in this cycle is 0.062 ohms.

[0069] In this implementation, the parameters of the extended Kalman filter are set as follows: the system process noise covariance matrix is ​​set as a diagonal matrix. The noise variances corresponding to the ripple voltage component, ESR estimate, and ESL estimate are set; the observation noise covariance matrix is ​​set to... The noise variance corresponds to the RMS value of the ripple voltage and the amplitude of the direct-axis current fluctuation; the initial value of the error covariance matrix is ​​set to a diagonal matrix. The initial value of the state vector is set to [0, initial nominal ESR, initial nominal ESL]. For example, for a 450V / 1000uF electrolytic capacitor, the initial nominal ESR can be set to 0.05 ohms and the initial nominal ESL can be set to 50 nanohenries, that is, the initial value is set to [0, 0.05, 50e-9]. The filtering period is synchronized with the sampling period, both being 80 microseconds.

[0070] The electro-thermal coupling field analysis model is a multilayer thermal network model constructed based on the finite difference method;

[0071] The multi-layer thermal network model discretizes the internal structure of the capacitor into multiple thermal resistance-thermal capacity nodes from the core to the outer shell. Each node represents a physical layer inside the capacitor, and the nodes are connected by thermal resistance. Each node itself has thermal capacity.

[0072] The operation of the electro-thermal coupled field analysis model includes: using the product of the ESR value output by the ESR estimation module and the effective value of the ripple current provided by the eigenvector construction unit as the total power consumption of the capacitor, and injecting the total power consumption into the innermost thermal resistance-capacitance node representing the capacitor core in the multilayer thermal network model as an internal heat source excitation; simultaneously, using the capacitor surface temperature data obtained in real time from the temperature sensor deployed on the surface of the DC bus capacitor aluminum shell as the temperature boundary condition of the outermost thermal resistance-capacitance node representing the capacitor shell in the multilayer thermal network model; and by solving the thermal balance differential equations of all thermal resistance-capacitance nodes, the steady-state temperature value of the innermost thermal resistance-capacitance node, i.e., the temperature value of the capacitor core, is calculated in reverse.

[0073] Furthermore, the process of calculating the effective value of the ripple current by constructing units using eigenvectors includes:

[0074] (1) The instantaneous sequence of total DC current and the instantaneous sequence of load current on the battery side are synchronously acquired from the electrical parameter acquisition module; the instantaneous sequence of total DC current is measured by a current sensor installed at the DC inlet of the converter, and the instantaneous sequence of load current on the battery side is measured by a current sensor installed between the battery and the converter;

[0075] (2) Perform digital high-pass filtering on the instantaneous DC-side total current sequence and the instantaneous battery-side load current sequence respectively to filter out the DC component, and obtain the DC-side total current ripple component sequence and the battery-side load current ripple component sequence accordingly.

[0076] (3) Subtract the battery-side load current ripple component sequence from the obtained DC-side total current ripple component sequence, and the resulting difference sequence is the instantaneous value sequence of capacitor current ripple flowing through the DC bus capacitor.

[0077] (4) Calculate the root mean square value of the obtained instantaneous value sequence of capacitor current ripple within an analysis window whose length is an integer multiple of the switching period. The calculation result is the effective value of the ripple current.

[0078] In this implementation, the core temperature back-calculation module needs to calculate the core temperature of the capacitor. First, it needs to calculate the effective value of the ripple current flowing through the capacitor. The feature vector construction unit synchronously obtains the instantaneous value sequences of the total DC current and the load current on the battery side from the electrical parameter acquisition module. After filtering out the DC component through digital high-pass filtering, the ripple component sequences of the two are obtained. The ripple component sequence of the load current on the battery side is subtracted from the ripple component sequence of the total DC current to obtain the instantaneous value sequence of the capacitor current ripple. The root mean square value is calculated within an analysis window that is an integer multiple of one switching cycle to obtain the effective value of the ripple current, which is 8.5A in this example.

[0079] Furthermore, the specific calculation process for the capacitor core temperature value includes:

[0080] (1) Call the geometric structure parameters and material property parameters of the DC bus capacitor corresponding to the model from the pre-stored physical parameter library. Based on the geometric structure parameters and material property parameters, calculate the thermal resistance and thermal capacity values ​​of each thermal resistance-thermal capacity node in the multi-layer thermal network model to complete the initialization of the physical parameters of the thermal network model. At the same time, obtain the real-time measured capacitor surface temperature data from the temperature sensor deployed on the aluminum shell surface of the DC bus capacitor, and set the temperature data as the known temperature boundary condition of the outermost node of the multi-layer thermal network model. The outermost node represents the physical location of the capacitor shell.

[0081] (2) Receive the real-time equivalent series resistance estimate from the ESR estimation module, extract the effective value of the ripple current provided by the feature vector construction unit, and then multiply the real-time equivalent series resistance estimate with the effective value of the ripple current. The result is the total power consumption of the capacitor. Assign this total power consumption of the capacitor as the internal heat generation rate to a specified thermal resistance-capacitance node in the innermost layer of the multilayer thermal network model. The specified thermal resistance-capacitance node is defined in the multilayer thermal network model as a node representing the physical location of the capacitor core.

[0082] (3) After completing the initialization of physical parameters and setting of temperature boundary conditions and injection of internal heat generation rate in the multi-layer thermal network model, for each internal thermal resistance-thermal capacity node in the multi-layer thermal network model except for the outermost boundary node, establish its steady-state thermal balance equation according to the principle of energy conservation. For the innermost capacitor core node that was injected with internal heat generation rate in the second step, its steady-state thermal balance equation includes two terms: one is the internal heat generation rate injected in the second step, and the other is the heat flow term formed by the node exchanging heat with its adjacent nodes through the thermal resistance connected to it. For the internal thermal resistance-thermal capacity nodes in the other intermediate layers of the model, their steady-state thermal balance equation only includes the heat flow term formed by heat exchange with their respective adjacent nodes through the thermal resistance connected to them.

[0083] (4) The steady-state thermal balance equations established for all internal thermal resistance-capacity nodes are combined with the known temperature equations of the outermost node set in (1) to form a complete linear algebraic equation system. The temperature of all internal thermal resistance-capacity nodes is the unknown in this linear algebraic equation system. Then, the Gaussian elimination method is used to solve the linear algebraic equation system to calculate the steady-state temperature numerical solution of all internal nodes under the given internal heat generation rate and the given surface temperature boundary conditions. The Gaussian elimination method is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0084] (5) Locate and extract the temperature value of the innermost thermal resistance-capacitance node that represents the physical location of the capacitor core from all the internal node steady-state temperature solutions obtained by solving. This extracted temperature value is the capacitor core temperature value obtained by reverse calculation through the multi-layer thermal network model.

[0085] In the multilayer thermal network model, the thermal resistance parameter of each thermal resistance-heat capacity node is determined by the thermal conductivity, thickness and cross-sectional area of ​​the corresponding material layer, and the heat capacity parameter is determined by the specific heat capacity and mass of the corresponding material layer.

[0086] The solution process of the thermal equilibrium differential equation system is as follows: the continuous differential equations are discretized into a linear algebraic equation system in the time domain, wherein the coefficient matrix of the linear algebraic equation system is composed of all thermal resistance parameters and thermal capacity parameters, and the right-hand side is the internal heat source excitation and temperature boundary conditions; the linear algebraic equation system is solved by Gaussian elimination to obtain the steady-state temperature values ​​of each node under the given heat source and temperature boundary conditions.

[0087] In this embodiment, the electro-thermal coupling field analysis model built into the core temperature back-calculation module is a multi-layer thermal network model constructed based on the finite difference method. This multi-layer thermal network model discretizes the internal structure of the capacitor into multiple thermal resistance-capacitance nodes from the core to the outer shell. In this embodiment, the internal structure of the capacitor is simplified to three layers: core, electrolytic paper, and aluminum shell, thus constructing a three-node thermal network model. The thermal resistance R1 from the core node to the electrolytic paper node can be calculated using the thermal conductivity of the core material, the core radius, and the axial length. The thermal conductivity of the core material is set to 1.5 W / (m·K), the core radius to 10 mm, and the axial length to 50 mm. The thermal resistance R2 from the electrolytic paper node to the aluminum shell node can be calculated using parameters such as the thermal conductivity and thickness of the electrolytic paper. The thermal conductivity of the electrolytic paper is set to 0.2 W / (m·K), and the thickness to 1 mm. The heat capacity C of each node can be calculated from the specific heat capacity and mass of the material. The specific heat capacity of the core material is set to 900 J / (kg·K). The discrete time step is set to 1 second. When solving the problem using Gaussian elimination, a full principal component selection strategy is adopted to ensure numerical stability. Here, W, K, m, mm, J, and kg are all units, where W represents watts, K represents Kelvin temperature, m represents meters, mm represents millimeters, J represents joules, and kg represents kilograms.

[0088] In this implementation, when the multilayer thermal network model is running, the product of the ESR value of 0.062 ohms output by the ESR estimation module and the effective value of the ripple current of 8.5A, which is 0.527 watts, is used as the total power consumption of the capacitor and injected into the innermost node representing the capacitor core as an internal heat source excitation. At the same time, the surface temperature of 65 degrees Celsius, which is detected in real time by the temperature sensor deployed on the surface of the capacitor aluminum shell, is used as the temperature boundary condition of the outermost node. By solving the thermal balance differential equations of all nodes, which are discretized in time and transformed into a linear algebraic equation system, and then solved using the Gaussian elimination method, the steady-state temperature value of the innermost node, i.e., the temperature value of the capacitor core, is calculated in reverse. In this example, the calculation result is 78 degrees Celsius, where A represents the unit of current, i.e., ampere.

[0089] The lifespan prediction module includes a health index calculation unit and a remaining lifespan prediction unit.

[0090] The health index calculation unit is used to receive the current ESR value output by the ESR estimation module and the capacitor core temperature value output by the core temperature back-calculation module. According to the capacitor aging model, the current ESR value relative to the growth rate of the initial nominal ESR value of the capacitor in the initial health state and the Arrhenius temperature acceleration effect of the capacitor core temperature value on the aging rate are weighted and fused to calculate and output a normalized health index.

[0091] In this implementation, the lifetime prediction module receives the current ESR value of 0.062 ohms and the core temperature of 78 degrees Celsius. Its health index calculation unit first calculates the electrical performance degradation rate: (current ESR value - initial nominal ESR value) / initial nominal ESR value = (0.062 - 0.05) / 0.05 = 0.24. Next, it applies the Arrhenius aging model to calculate the temperature acceleration effect: with a preset reference temperature of 40 degrees Celsius, an aging activation energy parameter of 83 kJ / mol, and a universal gas constant of 8.314 J / (mol·K), the thermal stress acceleration factor = ex p((1 / (273+40)-1 / (273+78))×83000 / 8.314)≈4.5; Set the electrical aging weight coefficient to 0.65 and the thermal stress aging weight coefficient to 0.35, calculate the unnormalized comprehensive aging index: 0.24×0.65+4.5×0.35≈1.73; Set the comprehensive aging index failure threshold to 2.0, then the normalized health index HI=1-(1.73 / 2.0)=0.135, where mol is the unit of amount of substance and kJ is the unit of energy, i.e. kJ is kilojoule and mol is mole.

[0092] Furthermore, the specific calculation process for the normalized health index includes:

[0093] (1) The health index calculation unit obtains the initial nominal ESR value and the current ESR value of the capacitor in the initial health state from the pre-stored ESR estimation module. At the same time, it obtains the real-time calculated output value of the capacitor core temperature from the core temperature back-calculation module.

[0094] (2) Subtract the initial nominal ESR value from the current ESR value to obtain the change in equivalent series resistance, and then divide the change in equivalent series resistance by the initial nominal ESR value to calculate the electrical performance degradation rate.

[0095] (3) Based on the obtained capacitor core temperature value, the temperature acceleration effect is calculated using the Arrhenius aging model; specifically: the reciprocal difference between the capacitor core temperature value and the preset reference temperature value is calculated, the reciprocal difference is multiplied by the preset material aging activation energy parameter and then divided by the preset universal gas constant, and finally the natural exponent is calculated on the calculation result. The obtained value is defined as the thermal stress acceleration factor. In this invention, the preset universal gas constant is set to 8.314 J / (mol·K). The Arrhenius aging model is the prior art in this field and is not an inventive solution of this application. It will not be described in detail here.

[0096] Furthermore, the activation energy parameter of a material's aging is a key parameter in Arrhenius's law, but the activation energies of different materials vary greatly. For example, the activation energy of rubber materials is usually low, set at 83 kJ / mol, while the activation energy of some composite materials may be higher. In addition, the activation energy may also change with time or environmental conditions, such as humidity and oxygen concentration, which further increases the complexity of prediction.

[0097] (4) Set the electrical aging weight coefficient and the thermal stress aging weight coefficient, multiply the electrical performance degradation rate by the electrical aging weight coefficient to obtain the electrical aging contribution value, multiply the thermal stress acceleration factor by the thermal stress aging weight coefficient to obtain the thermal stress aging contribution value, and finally add the electrical aging contribution value and the thermal stress aging contribution value to obtain the unnormalized comprehensive aging index.

[0098] Furthermore, the increase in equivalent series resistance is mainly caused by electrochemical aging within the capacitor, such as electrolyte evaporation, oxide film degradation, and internal connection point corrosion; thermal stress aging primarily targets core temperature. According to Arrhenius's law, temperature exponentially accelerates the rates of all chemical reactions, including the aforementioned electrochemical aging process. In this invention, based on the aging mechanism, an electrical aging weighting coefficient of 0.6 to 0.7 and a thermal stress aging weighting coefficient of 0.3 to 0.4 are assigned.

[0099] (5) A comprehensive aging index failure threshold is preset, and the proportion of the unnormalized comprehensive aging index within the comprehensive aging index failure threshold range is calculated. The result obtained by subtracting the proportion from the value 1 is the normalized health index. In this embodiment, the comprehensive aging index failure threshold is set to 2.0, that is, when the comprehensive aging index reaches 2.0, the health index drops to 0.

[0100] The remaining life prediction unit extrapolates the changing trend of the health index based on the historical decay trajectory of the health index using a recurrent neural network prediction algorithm, and defines that the capacitor life ends when the health index drops to a preset failure threshold. The prediction time from the current moment to the end of the capacitor life is calculated through the extrapolation curve to obtain the remaining life of the capacitor.

[0101] In this implementation, the remaining life expectancy prediction unit predicts based on the historical decline trajectory of the health index. First, a moving average filter is applied to the historical HI sequence to extract a smooth trend. The smoothed sequence is then divided into subsequences of length 200 points. The HI values ​​of the first 50 points of each subsequence are used as input, and the HI value of the next point is used as output to train a recurrent neural network model. The recurrent neural network model adopts a long short-term memory network structure. The input layer has 50 nodes, corresponding to the length of the input sequence. The hidden layer contains two LSTM units: the first LSTM unit has 64 units, and the second LSTM unit has 32 units. Each LSTM unit... The M-unit contains a forget gate, an input gate, and an output gate, with the following activation function configurations: the forget gate and input gate use the Sigmoid activation function, the candidate cell state uses the Tanh activation function, the output gate uses the Sigmoid activation function, and the cell state and hidden state outputs use the Tanh activation function. The output layer is a fully connected layer with 1 node and uses a linear activation function. During model training, the Adam optimizer is used with an initial learning rate of 0.001, a mean squared error loss function, 200 training iterations, and a batch size of 32. The training data consists of the aforementioned subsequence samples divided from historical data. After training, the most recent subsequence is input into the model for multi-step forward prediction, with 500 steps corresponding to a time interval of approximately 40 seconds, generating an HI extrapolation curve. The preset health failure threshold is 0.5. The time point when the HI predicted value first drops below 0.5 is searched on the extrapolation curve, and the time difference between this point and the current time is calculated to obtain the predicted remaining capacitor lifetime, for example, 15 days.

[0102] Furthermore, the specific calculation process for the remaining life of the capacitor includes:

[0103] (1) The remaining life prediction unit continuously receives and stores the health index from the health index calculation unit in time sequence to form a health historical decay sequence. The health historical decay sequence is then subjected to moving average filtering to suppress short-term fluctuations and extract long-term decay trends, generating a smoothed health trend sequence. The sliding window length is set to 100 data points.

[0104] (2) Divide the smoothed health trend sequence into multiple continuous subsequences with fixed time lengths in chronological order. Each subsequence is used as a training sample of the recurrent neural network model. The input part of the training sample is the health value of the first N time points in the subsequence, and the output part of the sample is the health value of the time point after the end of the subsequence. The subsequence length is set to 200 data points, and the input sequence length N is set to 50 data points.

[0105] Furthermore, in this invention, the number of nodes in the input layer of the recurrent neural network model is equal to N. Each node is responsible for receiving the health value at a specific time point in the subsequence, and the data is normalized before being fed into the network to adjust it to a numerical range suitable for model training.

[0106] Furthermore, in this invention, the hidden layers of the recurrent neural network model are composed of multiple identical long short-term memory units (LSMs) connected in series. Each LSM corresponds to a time step, and each LSM contains a key component: the cell state, which is regarded as the network's long-term memory channel. Information transfer within and between LSMs is accomplished through a gating mechanism of a forget gate, an input gate, and an output gate. The forget gate consists of a fully connected layer, and its input is the health value of the current time step and the hidden state output by the previous LSM. The input gate runs parallel to the forget gate, also consisting of a fully connected layer, and receives the same input, namely the current health value and the previous hidden state. The input gate generates a value between zero and one to control how much new information will be stored in the cell state. Simultaneously, another fully connected layer using a different activation function generates a candidate value vector based on the same input, representing potential new memory content. Finally, the output of the input gate is multiplied by this candidate value vector, and the result is added to the old cell state after being filtered by the forget gate, thus updating the cell state. The output gate is also a fully connected layer that determines how much information in the current cell state should be output based on the current input and the previous hidden state. The updated cell state is then multiplied by the output gate's output after passing through an activation function, generating the hidden state for the current time step. It's worth noting that a fully connected layer is connected after the last Long Short-Term Memory (LSTM) unit as the output layer. This output layer typically has only one node, and its input is the hidden state generated by the last LSM unit. By performing a linear transformation on this hidden state, a predicted value for the health index at the next time point after this subsequence is generated.

[0107] (3) Input the constructed most recent subsequence into the pre-trained recurrent neural network model. The recurrent neural network model contains long short-term memory units, which can iteratively predict the health value at each time step based on the current health value and the historical state information stored inside. Feed the new health value predicted each time back to the input end of the recurrent neural network model for multi-step forward prediction, and finally generate a health index extrapolation curve extending from the current time. In this embodiment, the number of multi-step prediction steps is set to 500 steps, corresponding to a future trend of 40 seconds.

[0108] (4) A health failure threshold is preset. On the obtained extrapolation curve, search forward along the time axis from the current moment to locate the point where the predicted health value first drops to equal or lower than the health failure threshold. The time corresponding to this point is the predicted end point of the capacitor life. Calculate the time from the current moment to the end point of the capacitor life. This time is the predicted remaining life of the capacitor.

[0109] Furthermore, the health failure threshold is set based on a comprehensive decision-making result of engineering practice, capacitor failure definition, and system risk tolerance. In this embodiment, the health failure threshold is set to 0.5.

[0110] The power scheduling module has a built-in health-power mapping function and a dynamic limiter.

[0111] The health-power mapping function takes the health index as input and outputs the corresponding power fluctuation range adjustment coefficient and the maximum allowable ripple current limit according to the preset scheduling strategy.

[0112] Furthermore, the scheduling strategy is configured such that: the higher the health index, the larger the power fluctuation range adjustment coefficient, allowing the converter to operate in a wider power fluctuation range; the lower the health index, the smaller the power fluctuation range adjustment coefficient according to a preset functional relationship, in order to shrink the power fluctuation range.

[0113] In this embodiment, the power scheduling module performs scheduling based on the health index HI = 0.135. Its built-in health-power mapping function has the following preset scheduling strategy: when the health index HI is in the range (0.8, 1.0], the adjustment coefficient is 1.0; when HI is in the range (0.6, 0.8], the adjustment coefficient linearly decreases from 1.0 to 0.8; when HI is in the range (0.4, 0.6], the adjustment coefficient linearly decreases from 0.8 to 0.6; and when HI is in the range [0, 0.4], the adjustment coefficient is fixed at 0.5. The maximum allowable ripple current limit is set according to the health index: when HI ≥ 0.7, it is 100% of the rated ripple current; when 0.5 ≤ HI < 0.7, it is 80% of the rated ripple current; and when HI < 0.5, it is 60% of the rated ripple current. For example, assuming the rated ripple current is 15A, the limit is 9A.

[0114] The dynamic limiter is used to receive the power fluctuation range adjustment coefficient and the maximum allowable ripple current limit, and multiply the power fluctuation range adjustment coefficient with the original power dispatch command received from the layer energy management system on the grid-type energy storage converter to generate a preliminary power command modulated by health status. At the same time, it predicts the ripple current generated on the DC bus capacitor when the preliminary power command is executed, and compares the predicted ripple current with the maximum allowable ripple current limit to obtain the final power command.

[0115] In this embodiment, the dynamic limiter receives the original power dispatch command issued by the upper-level energy management system. If the command power fluctuation range is ±100kW, the original command is multiplied by the adjustment coefficient 0.5 to obtain the preliminary power command with a fluctuation range of ±50kW. Based on the time-domain analysis model of the converter system, the effective value of the ripple current generated on the capacitor when the preliminary command is executed is predicted to be 10A. Since 10A is greater than the maximum allowable ripple current limit of 9A, iterative correction is initiated: the preliminary command is multiplied by the attenuation factor 0.95 to obtain a new candidate command, the ripple current is re-predicted, and this process is repeated until the predicted value meets the limit requirement. Assuming that after 3 iterations, the ripple current corresponding to the candidate command is 8.7A, which meets the requirements, the candidate command is determined as the final power command and output to the converter PWM controller.

[0116] Furthermore, the generation of the final power command specifically includes:

[0117] (1) The dynamic limiter receives the power fluctuation range adjustment coefficient and the maximum allowable ripple current limit output by the health-power mapping function, and at the same time receives the raw power scheduling command from the upper-level energy management system.

[0118] (2) Multiply the received original power scheduling command with the power fluctuation range adjustment coefficient to obtain the preliminary power command after health modulation;

[0119] (3) Based on the pre-established time-domain analysis model of the converter system based on the converter switching function and power conservation, the obtained preliminary power command, the current DC bus voltage and the grid synchronization phase angle are used as inputs to calculate and predict the effective value of the ripple current generated on the DC bus capacitor when the preliminary power command is executed.

[0120] (4) Compare the predicted RMS value of the ripple current with the received maximum allowable ripple current limit:

[0121] If the predicted effective value of the ripple current is less than or equal to the maximum allowable ripple current limit, the generated preliminary power command will be directly determined as the final power command.

[0122] If the predicted effective value of the ripple current is greater than the maximum allowable ripple current limit, the iterative correction process is initiated: First, the initial power command is multiplied by an attenuation factor less than 1 to generate a new candidate command; then return to (3), and re-predict the corresponding effective value of the ripple current based on this new candidate command; repeat this iterative process until the predicted ripple current value corresponding to the candidate command meets the condition of being less than or equal to the maximum allowable ripple current limit, and the candidate command is determined as the final power command. In this embodiment, the iterative attenuation factor is set to 0.95, and the maximum number of iterations is set to 10.

[0123] (5) The final power command after the determination is output to the pulse width modulation controller of the converter. At the same time, the actual electrical parameter data generated by the converter after executing the command is collected by the electrical parameter acquisition module and fed back to the system input terminal to start a new round of life monitoring and power scheduling cycle.

[0124] The power scheduling module also includes an early warning and communication unit; the early warning and communication unit monitors the health index and the remaining lifespan of the capacitor in real time. When the health index is lower than the preset early warning threshold or the remaining lifespan of the capacitor is shorter than the preset maintenance cycle, it sends an early warning message to the upper-level energy management system or remote monitoring center to prompt preventive maintenance.

[0125] In this embodiment, the early warning and communication unit monitors the HI (Health Index) and remaining lifespan in real time. The preset health warning threshold is 0.6, and the remaining lifespan maintenance cycle threshold is 30 days. Since HI = 0.135 < 0.6 and the remaining lifespan is 15 days < 30 days, an early warning message is immediately sent to the upper-level energy management system and the remote monitoring center, indicating the need for preventative maintenance of the capacitor. After the converter executes the final power command, the electrical parameter data generated during its actual operation is collected again by the electrical parameter acquisition module and fed back to the system input, initiating a new round of lifespan monitoring and power scheduling cycles. This achieves continuous online monitoring of the capacitor's status and adaptive optimization of the converter's operation.

[0126] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A capacitor life monitoring system for grid-type energy storage converters based on electrical parameter sensing, characterized in that, include: Electrical parameter acquisition module, ESR estimation module, core temperature back-calculation module, lifetime prediction module, power scheduling module; Based on the capacitor monitoring requirements of the grid-type energy storage converter, the DC voltage signal and three-phase AC current signal of the DC bus capacitor are synchronously acquired through the electrical parameter acquisition module to form a multi-dimensional feature vector; Based on the multidimensional feature vector, the ESR value of the target DC bus capacitor is obtained by analyzing the signal through the state observer built into the ESR estimation module. Combined with the preset capacitor surface temperature data, the capacitor core temperature value is obtained by back-calculating the electro-thermal coupling field analysis model built into the core temperature back-calculation module. Based on the ESR value and the capacitor core temperature value, the remaining lifespan and health index of the capacitor are calculated by the lifespan prediction module. According to the health index, the capacitor health-power output joint scheduling is performed by the power scheduling module to adjust the power fluctuation range of the converter and feed back the real-time operating data of the converter after scheduling to the electrical parameter acquisition module. The electrical parameter acquisition module includes a voltage sensor, a synchronous current sensor, and a synchronous analog-to-digital converter unit; The voltage sensor is connected in parallel across the DC bus capacitor to acquire the DC voltage signal of the DC bus capacitor; the synchronous current sensor is connected in series with the three-phase AC output terminal of the grid-type energy storage converter to synchronously acquire the three-phase AC current signal; the synchronous analog-to-digital converter is connected to the voltage sensor and the synchronous current sensor to synchronously convert the DC voltage signal and the three-phase AC current signal into digital signals based on a unified sampling clock to form the original sampling sequence. The electrical parameter acquisition module also includes a feature vector construction unit, the input of which is connected to the output of the synchronous analog-to-digital conversion unit. The feature vector construction unit preprocesses and extracts features from the original sampling sequence. The preprocessing includes performing Clarke transform and Park transform on the three-phase AC current signal to obtain the direct-axis current component and quadrature-axis current component in a two-phase rotating coordinate system. The feature extraction includes calculating the effective value of the ripple voltage and the second harmonic amplitude based on the DC voltage signal, and calculating the direct-axis current fluctuation amplitude and quadrature-axis current fluctuation amplitude based on the direct-axis current component and quadrature-axis current component. The effective value of the ripple voltage, the second harmonic amplitude, the direct-axis current fluctuation amplitude, the quadrature-axis current fluctuation amplitude, and the mean of the DC voltage signal are combined to form the multidimensional feature vector updated in each sampling or control cycle.

2. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 1, characterized in that, The state observer built into the ESR estimation module is a parameter identification model based on the extended Kalman filter. The state observer is constructed based on the equivalent circuit model of the DC bus capacitance; The equivalent circuit model consists of an equivalent series resistance, an equivalent series inductance, and an ideal capacitor connected in series. The construction and operation of the state observer includes: defining the ripple component of the DC bus voltage, the estimated value of the equivalent series resistance, and the estimated value of the equivalent series inductance together as the system's state vector; The effective value of ripple voltage and the amplitude of direct-axis current fluctuation acquired and calculated by the electrical parameter acquisition module are defined as the observation vector of the system. Using the observation vector as input, the extended Kalman filter algorithm is used to recursively estimate the minimum variance of the state vector and output the estimated value of the equivalent series resistance of the target DC bus capacitor, i.e., the ESR value.

3. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 2, characterized in that, The electro-thermal coupling field analysis model is a multilayer thermal network model constructed based on the finite difference method; The multi-layer thermal network model discretizes the internal structure of the capacitor into multiple thermal resistance-thermal capacity nodes from the core to the outer shell. Each node represents a physical layer inside the capacitor, and the nodes are connected by thermal resistance. Each node itself has thermal capacity. The operation of the electro-thermal coupled field analysis model includes: using the product of the ESR value output by the ESR estimation module and the effective value of the ripple current provided by the eigenvector construction unit as the total power consumption of the capacitor, and injecting the total power consumption into the innermost thermal resistance-capacitance node representing the capacitor core in the multilayer thermal network model as an internal heat source excitation; simultaneously, using the capacitor surface temperature data obtained in real time from the temperature sensor deployed on the surface of the DC bus capacitor aluminum shell as the temperature boundary condition of the outermost thermal resistance-capacitance node representing the capacitor shell in the multilayer thermal network model; and by solving the thermal balance differential equations of all thermal resistance-capacitance nodes, the steady-state temperature value of the innermost thermal resistance-capacitance node, i.e., the temperature value of the capacitor core, is calculated in reverse.

4. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 3, characterized in that, In the multilayer thermal network model, the thermal resistance parameter of each thermal resistance-heat capacity node is determined by the thermal conductivity, thickness and cross-sectional area of ​​the corresponding material layer, and the heat capacity parameter is determined by the specific heat capacity and mass of the corresponding material layer. The solution process of the thermal equilibrium differential equation system is as follows: the continuous differential equations are discretized into a linear algebraic equation system in the time domain, wherein the coefficient matrix of the linear algebraic equation system is composed of all thermal resistance parameters and thermal capacity parameters, and the right-hand side is the internal heat source excitation and temperature boundary conditions; the linear algebraic equation system is solved by Gaussian elimination to obtain the steady-state temperature values ​​of each node under the given heat source and temperature boundary conditions.

5. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 4, characterized in that, The lifespan prediction module includes a health index calculation unit and a remaining lifespan prediction unit. The health index calculation unit is used to receive the current ESR value output by the ESR estimation module and the capacitor core temperature value output by the core temperature back-calculation module. According to the capacitor aging model, the current ESR value relative to the growth rate of the initial nominal ESR value of the capacitor in the initial health state and the Arrhenius temperature acceleration effect of the capacitor core temperature value on the aging rate are weighted and fused to calculate and output a normalized health index. The remaining life prediction unit extrapolates the changing trend of the health index based on the historical decay trajectory of the health index using a recurrent neural network prediction algorithm, and defines that the capacitor life ends when the health index drops to a preset failure threshold. The prediction time from the current moment to the end of the capacitor life is calculated through the extrapolation curve to obtain the remaining life of the capacitor.

6. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 5, characterized in that, The power scheduling module has a built-in health-power mapping function and a dynamic limiter. The health-power mapping function takes the health index as input and outputs the corresponding power fluctuation range adjustment coefficient and the maximum allowable ripple current limit according to the preset scheduling strategy. The dynamic limiter is used to receive the power fluctuation range adjustment coefficient and the maximum allowable ripple current limit, and multiply the power fluctuation range adjustment coefficient with the original power dispatch command received from the upper-level energy management system of the grid-type energy storage converter to generate a preliminary power command modulated by health status. At the same time, it predicts the ripple current generated on the DC bus capacitor when the preliminary power command is executed, and compares the predicted ripple current with the maximum allowable ripple current limit to obtain the final power command.

7. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 6, characterized in that, The power scheduling module also includes an early warning and communication unit; the early warning and communication unit monitors the health index and the remaining lifespan of the capacitor in real time. When the health index is lower than the preset early warning threshold or the remaining lifespan of the capacitor is shorter than the preset maintenance cycle, it sends an early warning message to the upper-level energy management system or remote monitoring center to prompt preventive maintenance.

8. The grid-type energy storage converter capacitor life monitoring system based on electrical parameter sensing as described in claim 7, characterized in that, The monitoring system is constructed as an embedded hardware and software integrated platform; the electrical parameter acquisition module is implemented by hardware circuits, and the ESR estimation module, core temperature back-calculation module, life prediction module and power scheduling module are integrated in a digital signal processor or microcontroller. The functions of each module are implemented through embedded software algorithms, and real-time data interaction is performed with the converter main control system.

Images