Graphene supercapacitor capacity energy-saving control method based on intelligent algorithm

Abstract

The invention provides a graphene super-capacitor capacity energy-saving control method based on an intelligent algorithm, and the method comprises the steps: obtaining the harmonic content data of a power grid, carrying out the spectrum analysis of the harmonic content through a Fourier transform algorithm, and obtaining the change trend of the harmonic content; acquiring historical capacitor bank impedance characteristic data and a corresponding harmonic content change trend, and calculating impedance matching degrees of the graphene super capacitor under different load conditions; according to the impedance matching degree of the graphene super capacitor, generating a capacitor bank switching sequence by adopting a dynamic programming algorithm, and determining a target switching sequence of the graphene super capacitor; switching of the graphene super capacitor according to the target switching sequence is executed through an edge calculation unit, the access state of the graphene super capacitor is adjusted in real time, and harmonic wave changes of a power grid are dynamically matched; the method comprises the following steps: acquiring transient current data of a system, and decomposing the transient current data by adopting a wavelet transform algorithm to obtain a frequency component of transient current;

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a graphene supercapacitor capacity energy-saving control method based on an intelligent algorithm. Background Art

[0002] In the graphene supercapacitor capacity control system, the edge computing unit is used to realize the real-time switching control of the capacitor group. When the system detects a sudden change in the harmonic content of the power grid during operation, the edge computing unit needs to dynamically adjust the switching sequence according to the impedance characteristics of the capacitor to cope with the change in load conditions. However, this process faces a technical contradiction: in the case of a sudden change in the harmonic content of the power grid, the impedance characteristics of the capacitor group will change with the frequency, resulting in the edge computing unit being unable to directly use the original switching timing strategy. If the switching timing is not adjusted in time, the system may generate a large current shock during the transient process, affecting the stability of the power grid. Specifically, a sudden change in the harmonic content of the power grid will cause the impedance of the capacitor group at a specific frequency to be significantly reduced, thereby changing the response characteristics of the capacitor group to harmonic currents. The edge computing unit needs to collect the harmonic data of the power grid in real time, and recalculate the optimal switching timing in combination with the impedance and frequency characteristic curves of the capacitor group. However, due to the nonlinear characteristics of the impedance characteristics of the capacitor group and the large range of harmonic content changes, the edge computing unit may face the contradiction between computational complexity and real-time performance when dynamically adjusting the switching sequence. In addition, the adjustment of the switching timing also needs to consider the suppression of transient current shocks. If the switching interval is too short, the capacitor bank may be frequently switched, exacerbating the transient current fluctuations in the system; if the switching interval is too long, the system may not be able to respond to the sudden change of harmonics in time, causing the harmonics in the power grid to deteriorate further. Therefore, how to take into account the multiple requirements of capacitor impedance characteristics, switching timing optimization and transient current suppression in a dynamic environment with sudden changes in the harmonic content of the power grid to achieve efficient and stable operation of the system is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a graphene supercapacitor capacity energy-saving control method based on an intelligent algorithm, which mainly includes: Obtain the harmonic content data of the power grid, use the Fourier transform algorithm to perform spectrum analysis on the harmonic content, and obtain the changing trend of the harmonic content; According to the trend of harmonic content changes, the bands with abnormal harmonic content are identified. For the abnormal bands, the edge computing unit is triggered to perform real-time harmonic content evaluation, determine the degree of harmonic excess and the location of the harmonic source, generate corresponding harmonic suppression strategies, including switching capacitor groups and adjusting the working parameters of active filter devices, and send them to the execution unit; Obtain historical capacitor bank impedance characteristic data and corresponding harmonic content change trends, and calculate the impedance matching of graphene supercapacitors under different load conditions; According to the impedance matching of graphene supercapacitors, a dynamic programming algorithm is used to generate a capacitor bank switching sequence and determine the target switching sequence of graphene supercapacitors; The edge computing unit executes the switching of graphene supercapacitors according to the target switching sequence, and adjusts the access status of graphene supercapacitors in real time to dynamically match the harmonic changes of the power grid; Obtain system transient current data, use wavelet transform algorithm to decompose the transient current data, and obtain the frequency components of the transient current; According to the transient current frequency component, the pulse amplitude, duration and rising rate of the transient current are calculated. If the transient current impact exceeds the preset threshold, the protection mechanism is triggered, and different levels of protection measures are taken according to the severity of the impact. The capacitor bank switching optimization process is started and the capacitor bank switching sequence is adjusted again; An adaptive filtering algorithm is used to optimize the adjusted capacitor group switching sequence, including minimizing the switch inrush current and suppressing resonance during the switching process. The optimized switching sequence is sent to the execution unit to coordinate with other control links.

[0004] The technical solution provided by the embodiment of the present invention may have the following beneficial effects: The present invention discloses a graphene supercapacitor capacity energy-saving control method based on an intelligent algorithm. The method first performs spectrum analysis on the harmonic content of the power grid through Fourier transform to identify abnormal harmonic bands. For abnormal bands, the present invention uses an edge computing unit to perform real-time evaluation, determine the degree of harmonic excess and the source location, and generate a suppression strategy. Graphene supercapacitors are innovatively introduced, and a dynamic programming algorithm is used to generate an optimal switching sequence based on its impedance matching. The present invention also combines wavelet transform to analyze transient currents, triggers a protection mechanism when the impact exceeds the threshold, and optimizes the switching of capacitor groups. Finally, an adaptive filtering algorithm is used to further optimize the switching sequence, minimize switch inrush currents, and suppress resonance; the present invention realizes accurate identification, dynamic suppression, and adaptive optimization of power grid harmonics, significantly improving the stability and reliability of power grid operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0005] Figure 1 The present invention is a flow chart of a graphene supercapacitor capacity energy-saving control method based on an intelligent algorithm. DETAILED DESCRIPTION

[0006] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0007] like Figure 1 In this embodiment, a graphene supercapacitor capacity energy-saving control method based on an intelligent algorithm may specifically include: S101. Obtain the harmonic content data of the power grid, use the Fourier transform algorithm to perform spectrum analysis on the harmonic content, and obtain the changing trend of the harmonic content.

[0008] A dynamic window harmonic collector is used to read the harmonic signal data generated by the power grid equipment during operation according to the preset harmonic sampling frequency; after the harmonic signal data is subjected to wavelet denoising filtering, the amplitude, phase and frequency characteristics are extracted through the Fourier transform algorithm, and the harmonic content is divided into corresponding intervals using a density-based clustering method, and the values ​​of each harmonic distortion rate and the total distortion amount are obtained from the divided intervals; a harmonic propagation characteristic matrix is ​​constructed according to the values ​​of each harmonic distortion rate and the total distortion amount, and when the harmonic voltage fluctuation amplitude exceeds the preset operating threshold, a recursive neural network is used to quantitatively analyze the harmonic superposition effect, and the harmonic interference degree index is obtained from the harmonic propagation characteristic matrix; the harmonic content change trend analysis is performed according to the harmonic interference degree and the harmonic attenuation coefficient to obtain time series data.

[0009] Specifically, a dynamic window harmonic collector is used to read the harmonic signal data generated during the operation of the power grid equipment according to the preset harmonic sampling frequency, and the harmonic signal data is subjected to wavelet denoising filtering to obtain the first harmonic data sequence. According to the first harmonic data sequence, the harmonic amplitude data of multiple time points are calculated by the Fourier transform algorithm, and the amplitude, phase, and frequency characteristics are extracted from the harmonic amplitude data of the multiple time points to obtain the second harmonic feature sequence. For the second harmonic feature sequence, a density-based clustering method is used to divide the harmonic content into corresponding intervals, and the values ​​of each harmonic distortion rate and the total distortion are obtained from the divided intervals. A harmonic propagation feature matrix is ​​constructed based on the values ​​of each harmonic distortion rate and the total distortion, and the harmonic attenuation coefficient is calculated by the harmonic distribution density. For the harmonic attenuation coefficient and the harmonic propagation feature matrix, when the harmonic voltage fluctuation amplitude exceeds the preset operating threshold, a recursive neural network is used to quantitatively analyze the harmonic superposition effect to determine the degree of harmonic interference in the harmonic propagation path. According to the harmonic interference degree and the harmonic attenuation coefficient, a harmonic propagation evaluation index is constructed, and the harmonic content change trend analysis is performed on the second harmonic feature sequence to obtain time series data. In the operation of the power grid, the change trend analysis of the harmonic content is of great significance to the safe operation of the power grid. The dynamic window harmonic collector collects the harmonic signals of the power grid equipment based on the preset sampling frequency. The sampling frequency is usually set to 64 times the fundamental frequency, that is, the sampling frequency is 3200 Hz in a 50 Hz power grid, and the harmonic components within 31 times can be accurately collected. The collected harmonic signal data often contains random noise, and effective signals can be effectively extracted by wavelet denoising. The db4 wavelet basis function is selected for 5-layer decomposition, the soft threshold is set to 0.25, and the high-frequency coefficients are thresholded and reconstructed to obtain the first harmonic data sequence. The obtained first harmonic data sequence is subjected to Fourier transform analysis, 10 minutes is selected as the time window, and the window sliding interval is 1 minute to obtain the harmonic amplitude data at each time point. The amplitude, phase and frequency characteristics of each harmonic are extracted from it to form the second harmonic feature sequence. The amplitude feature reflects the content of each harmonic, the phase feature characterizes the phase relationship between harmonics, and the frequency feature reflects the frequency distribution of harmonics. The density-based clustering method is used to divide the harmonic content into intervals. The clustering radius is set to 0.15, the minimum number of samples is set to 5, and the harmonic data is divided into three intervals: high, medium, and low. The distortion rate of each harmonic is calculated. The distortion rate of the third harmonic is 2.8%, the distortion rate of the fifth harmonic is 2.1%, the distortion rate of the seventh harmonic is 1.6%, and the total harmonic distortion rate is 3.9%. When constructing the harmonic propagation feature matrix, the matrix dimension is the product of the number of each measurement point and the harmonic order, and the matrix elements are the amplitude and phase of each harmonic at the corresponding measurement point. According to the propagation attenuation law of harmonics between the measurement points, the attenuation coefficients of harmonics of different frequencies are calculated. The attenuation coefficient of low-order harmonics is about 0.92, and the attenuation coefficient of high-order harmonics is about 0.85.When the voltage fluctuation amplitude exceeds 2% of the rated voltage, a three-layer recursive neural network structure is used to analyze the harmonic superposition effect. The number of hidden layer neurons is 12, 8, and 4, respectively, and the hyperbolic tangent function is used as the activation function. The network input is the amplitude and phase data of each harmonic, and the output is the harmonic interference degree index. An interference degree index greater than 0.8 indicates that the harmonic interference is serious and suppression measures need to be taken. A comprehensive evaluation index is constructed based on the harmonic interference degree and attenuation coefficient, with weights of 0.6 and 0.4, respectively. When the evaluation index is greater than 0.75, it indicates that the harmonic propagation has a significant impact. Combined with the analysis of the second harmonic characteristic sequence, it is found that the harmonic content shows periodic fluctuation characteristics, and the harmonic content increases significantly during periods of heavy grid load.

[0010] S102. According to the changing trend of harmonic content, identify the bands with abnormal harmonic content. For the abnormal bands, trigger the edge computing unit to perform real-time harmonic content evaluation, determine the degree of harmonic excess and the location of the harmonic source, generate corresponding harmonic suppression strategies, including switching capacitor groups and adjusting the working parameters of the active filter device, and send them to the execution unit.

[0011] According to the time series data, an adaptive threshold detection algorithm based on statistical test is used to judge the abnormal harmonic frequency band, and the phase characteristics and amplitude characteristics are extracted from the abnormal harmonic frequency band to obtain a harmonic feature sequence; an edge multi-point distributed acquisition device is used to sample the abnormal harmonic frequency band, and the waveform distortion degree is calculated according to the harmonic feature sequence to obtain the harmonic source intensity value, and the attenuation rate is calculated by amplitude comparison between measuring points to obtain the harmonic propagation trajectory; a propagation distance matrix is ​​constructed according to the harmonic propagation trajectory, and the minimum spanning tree algorithm is used to locate the harmonic source node, and a harmonic source position description is generated; for the harmonic source position description, a capacitor group switching timing table is generated according to the harmonic compensation capacity of each measuring point, and a load balancing algorithm is used to perform task allocation on the capacitor group switching timing table to generate an edge control instruction set.

[0012] Specifically, the harmonic content reference value is calculated according to the time series data, and the abnormal harmonic frequency band is judged by the adaptive threshold detection algorithm based on statistical test within the reference value interval, and the phase characteristics and amplitude characteristics are extracted from the abnormal harmonic frequency band to obtain the first harmonic characteristic sequence. The edge multi-point distributed acquisition device is used to sample the abnormal harmonic frequency band at a sampling rate of 256 times the fundamental frequency, and the waveform distortion degree is calculated according to the first harmonic characteristic sequence to obtain the harmonic source intensity value, and the attenuation rate is calculated by comparing the amplitude between the measurement points to obtain the harmonic propagation trajectory. According to the harmonic propagation trajectory, a propagation distance matrix is ​​constructed, and the harmonic source node is located by the minimum spanning tree algorithm, and a harmonic source location description including the fault distance and propagation direction is generated. For the place where the harmonic source intensity value exceeds the standard limit, a capacitor group switching timing table is generated based on the harmonic compensation capacity of each measurement point, and the harmonic source location description is matched with the nearest compensation device parameters through a dynamic programming algorithm. A compensation strategy instruction is generated according to the capacitor group switching timing table, and the instruction execution priority is calculated based on the response time and compensation capacity of each compensation device to generate a hierarchical compensation control sequence. The load balancing algorithm is used to assign tasks to the hierarchical compensation control sequence, and the execution load is distributed according to the computing power of the edge node and the network delay to generate an edge control instruction set. In the power system, abnormal harmonic detection first needs to establish a benchmark value judgment standard for harmonic content. By statistically analyzing historical data, the mean and standard deviation of the harmonic content in different time periods are calculated, and data is collected with a time window of 10 minutes and a sliding interval of 1 minute. The adaptive threshold detection algorithm based on statistical inspection sets a dynamic warning threshold. When the harmonic content exceeds the mean plus 3 times the standard deviation, it is judged as an abnormal harmonic frequency band. The phase and amplitude characteristics of these abnormal frequency bands are extracted to obtain the first harmonic feature sequence as the basis for subsequent analysis. The edge acquisition device uses high-frequency sampling to conduct in-depth analysis of the abnormal harmonic frequency band. The sampling frequency is set to 256 times the fundamental frequency, that is, the sampling frequency is 12800 Hz in a 50 Hz power grid, and the sampling accuracy is 16 bits. The waveform distortion is calculated based on the sampled data. When the total harmonic distortion rate exceeds 4.5%, the attenuation rate is calculated by the amplitude ratio of adjacent measurement points. The amplitude ratio gradually decreases with the propagation distance, and the typical attenuation rate is 0.92 per kilometer. After obtaining the harmonic propagation trajectory, a matrix reflecting the propagation distance between each measurement point is constructed, and the matrix elements are the actual line distances between the measurement points. The minimum spanning tree algorithm is used to analyze the propagation distance matrix. By comparing the harmonic intensity and propagation direction of each node, the harmonic source node position is located, and the precise distance from the fault point to each measurement point and the propagation path direction angle are calculated. When it is found that the harmonic source intensity exceeds the standard limit, such as the third harmonic exceeds 2.4% and the fifth harmonic exceeds 1.8%, harmonic suppression needs to be carried out in time. Based on the harmonic characteristics of each measurement point, the compensation capacity demand is calculated, and a compensation table containing switching capacity and timing is generated.The selection of compensation equipment follows the principle of proximity, and the optimal compensation scheme is matched through the dynamic programming algorithm to ensure that the compensation capacity matches the strength of the harmonic source. When performing compensation control, the characteristics of different types of compensation equipment need to be considered. The response time of the passive filter is within 20 milliseconds, and the response time of the active filter can reach 5 milliseconds, but the compensation capacity is small. According to the response time and the size of the compensation capacity, the control instructions are prioritized. The equipment with a short response time has a higher priority, forming a hierarchical compensation control sequence. Finally, the control tasks are allocated through the load balancing algorithm. Considering the processing power of the edge node, the node with a processor frequency of more than 1.5 GHz can process 8 data streams at the same time, and the node with a main frequency between 800 MHz and 1.5 GHz can process 4 data streams. At the same time, the network delay is used as a constraint condition for task allocation. When the network delay exceeds 50 milliseconds, the number of allocated tasks is reduced to ensure that the control instructions can be executed in time.

[0013] S103, obtaining historical capacitor group impedance characteristic data and corresponding harmonic content change trends, and calculating the impedance matching degree of the graphene supercapacitor under different load conditions.

[0014] According to the impedance measurement values ​​recorded in the historical operation data of the capacitor group, the wavelet decomposition algorithm is used to perform time-frequency domain analysis on the impedance measurement values ​​to obtain an impedance characteristic sequence containing amplitude and phase information; for the impedance characteristic sequence and the operating parameters obtained by the distributed data acquisition device, the fused impedance parameters are obtained by the data fusion algorithm; according to the fused impedance parameters, an impedance matrix reflecting the load change is constructed, the row vector of the impedance matrix represents the load gear, and the column vector represents the time series, and the impedance offset is obtained by the least squares method; for the impedance offset and the load gear data, the recursive least squares method is used to calculate the harmonic distortion level, and the harmonic content change law is extracted from the harmonic distortion level to obtain the load response characteristic curve; based on the load response characteristic curve, the harmonic content change law is extracted, and the impedance matching degree between the graphene supercapacitor and the original capacitor group is calculated by support vector machine regression.

[0015] Specifically, according to the impedance measurement values ​​and conductivity parameters recorded in the historical operation data of the capacitor group, the impedance measurement values ​​are analyzed in the time and frequency domain by wavelet decomposition algorithm to obtain the first impedance characteristic sequence containing amplitude and phase information. A distributed data acquisition device is used to obtain the operating parameters of the graphene supercapacitor, including the charge and discharge rate, the internal resistance change rate and the conductivity value, and the operating parameters and the first impedance characteristic sequence are fused to obtain the second impedance characteristic sequence. According to the second impedance characteristic sequence, an impedance matrix reflecting the load change is constructed, the row vector of the matrix represents the load gear, the column vector represents the time series, and the impedance offset is calculated by the least squares method. For the impedance offset and load gear data, the recursive least squares method is used to calculate the harmonic distortion level under different load conditions to obtain the load response characteristic curve. Based on the load response characteristic curve, the change law of the harmonic content is extracted, and the impedance matching degree between the graphene supercapacitor and the original capacitor group is calculated by support vector machine regression. A load response range table is constructed according to the impedance matching degree, and the optimal impedance matching interval is obtained from the load response range table to generate an operating parameter group including a load limit and a matching threshold. In the power system, the parameters of the capacitor group change with the operation time. By analyzing the impedance measurement values ​​in the historical operation data, it is found that the impedance value shows periodic fluctuation characteristics. The wavelet decomposition algorithm is used to decompose the impedance measurement value into 5 layers, and the db4 wavelet basis function is selected to obtain a characteristic sequence containing amplitude and phase information, in which the impedance amplitude varies in the range of 0.1 to 0.5 ohms and the phase angle is offset in the range of -15 degrees to 15 degrees. The graphene supercapacitor has excellent conductivity, and the distributed data acquisition device records its operating parameters at a sampling frequency of 100 Hz. The charge and discharge rate varies in the range of 0 to 4.5 coulombs / second, the internal resistance change rate does not exceed 0.1% / hour, and the conductivity is maintained above 580 Siemens / meter. These parameters are combined with the impedance characteristic sequence through the data fusion algorithm to construct a data sequence reflecting real-time characteristics. During the load change process, the dimension of the impedance matrix is ​​the product of the number of load gears and the length of the time series. The typical load is divided into 5 gears, and the time series length is 1440 minutes. The impedance offset calculated by the least squares method shows that the impedance offset is within 0.05 ohms when lightly loaded and can reach 0.2 ohms when heavily loaded. The recursive least squares method is used to calculate the harmonic distortion level under various load conditions. Using a forgetting factor of 0.98, the calculation results show that the total harmonic distortion rate is below 2.5% when lightly loaded and can reach 4.8% when heavily loaded. The load response characteristic curve shows nonlinear characteristics, and the rate of increase of harmonic content gradually accelerates when the load increases. Based on the law of change of harmonic content, the support vector machine uses a radial basis kernel function with a kernel parameter of 0.1 to calculate the impedance matching degree between the graphene supercapacitor and the original capacitor group. The results show that in the range of 30% to 70% of the rated load, the impedance matching degree of the two capacitors is maintained above 0.85, and the harmonic suppression effect is optimal.The load response range table records the operating parameters at different load levels in detail, including the allowable load change rate, impedance deviation limit, and upper limit of harmonic content. When the load change rate exceeds 5% / minute, the impedance matching margin needs to be increased to maintain it above 0.9. Through comparative analysis, it is found that under conditions of large load fluctuations, the impedance characteristics of graphene supercapacitors are more stable and can continue to maintain good harmonic suppression effects.

[0016] S104, according to the impedance matching of the graphene supercapacitor, a capacitor group switching sequence is generated by using a dynamic programming algorithm to determine a target switching sequence of the graphene supercapacitor.

[0017] A switching state matrix is ​​constructed according to the impedance matching degree of the graphene supercapacitor and the working state data of the capacitor group, and a switching priority sequence is extracted from the switching state matrix; a capacity allocation ratio is calculated for the switching priority sequence in combination with the real-time power data of the capacitor group, and the capacity allocation ratio is adjusted through a recursive neural network to obtain a switching state matrix with an adjusted capacity allocation ratio; a piecewise linear regression is used to calculate the capacitor group loss value corresponding to the switching state matrix with the adjusted capacity allocation ratio to obtain a switching stability curve; a dynamic timing compensation algorithm is used to generate a control timing table for the switching stability curve and the correction coefficient table, and the switching time and capacity gear value are extracted from the control timing table to generate a target switching sequence of the graphene supercapacitor.

[0018] Specifically, according to the impedance matching degree of the graphene supercapacitor and the working state data of the capacitor group sampled in real time, a dynamic programming algorithm is used to construct a first switching state matrix containing the switching interval time and the compensation capacity value, and the switching priority sequence is extracted from the first switching state matrix. The capacity allocation ratio is calculated for the switching priority sequence in combination with the real-time power data of the capacitor group, and the capacity allocation ratio is dynamically adjusted through a recursive neural network to obtain a second switching state matrix. A correction coefficient table is constructed according to the second switching state matrix and the real-time load level value, and the loss value of the capacitor group is calculated by piecewise linear regression to obtain a switching stability curve. For the switching stability curve and the correction coefficient table, the least squares method is used to calculate the compensation capacity correction value of each group of capacitors to generate a basic switching instruction sequence. According to the basic switching instruction sequence and the response time limit of the capacitor group, a control timing table considering the response delay is generated by a dynamic timing compensation algorithm. According to the control timing table, a target switching sequence of the graphene supercapacitor is generated according to the capacity classification principle, and the switching time and capacity gear value are extracted from the target switching sequence. The switching control of graphene supercapacitors involves the coordination of multiple key parameters. The working status data of the capacitor group is obtained through high-speed sampling. The sampling frequency is set to 2000 Hz, and parameters such as voltage, current and power factor are recorded. The dynamic programming algorithm constructs a switching state matrix based on these data. The matrix dimension is the product of the number of capacitor groups and the length of the time window. The switching interval is not less than 20 milliseconds, and the compensation capacity is divided into 10 gears. The division of the switching priority level takes into account multiple characteristic parameters of the capacitor group. The capacitor group with larger capacity has a higher priority value, and the capacitor group with short response time has an increased priority. In actual operation, the priority of the capacitor group with a capacity of 500 kvar is 0.8, and the priority of the capacitor group with a capacity of 200 kvar is 0.6. When the power data shows that the power grid load fluctuates, the switching process of the capacitor group needs to maintain power balance, and the capacity allocation ratio is obtained by recursive calculation to obtain the optimal solution. Dynamic correction is performed on the switching state matrix, and the correction coefficient table reflects the impact of load changes on the switching process. When the load power fluctuation exceeds 15% of the rated power, the correction coefficient value varies in the range of 0.85 to 1.15. The loss value of the capacitor group is calculated using the piecewise linear regression method, and the loss rate is between 0.2% and 0.5% of the rated capacity. The correction value of the compensation capacity is calculated based on the switching stability curve, which shows that the compensation capacity needs to be appropriately increased at the moment of load mutation. The correction value calculated by the least squares method shows that the compensation capacity under light load conditions increases by 5%, and the compensation capacity under heavy load conditions increases by 8%. The generated basic switching instructions include two dimensions: switching time and capacity size. The dynamic timing compensation algorithm takes into account the response characteristics of different types of capacitors. The response time of ordinary capacitors is about 50 milliseconds, and the response time of graphene supercapacitors is only 5 milliseconds. Detailed execution time nodes are recorded in the control timing table to ensure that each group of capacitors is put into operation according to the optimal timing.The generated switching sequence is divided into three gears according to the capacity level: large, medium and small, corresponding to 100%, 60% and 30% of the rated capacity respectively. The selection of the switching time is based on the zero-crossing point of the grid voltage. The transient process generated by the switching of graphene supercapacitors near the zero-crossing point of the voltage is minimal. The measured data shows that with the optimized switching sequence, the voltage fluctuation amplitude of the capacitor switching process is reduced to 25% of the original, and the transient process of the system is significantly improved.

[0019] S105. Execute the switching of graphene supercapacitors according to the target switching sequence through the edge computing unit, and adjust the access status of the graphene supercapacitors in real time to dynamically match the harmonic changes of the power grid.

[0020] The harmonic content is sampled at a preset sampling frequency by an edge computing unit, and voltage, current and power parameters are extracted from the sampled data to obtain a first edge control instruction; a support vector machine regression algorithm is used to extract features according to the first edge control instruction, and a dynamic compensation increment is calculated in combination with the compensation capacity parameter to obtain a second edge control instruction; a response timing table is constructed for the second edge control instruction, and the edge node processing capacity load value is calculated by an optimal search algorithm to obtain an edge computing allocation scheme; a distributed computing method is used to generate an edge node task sequence according to the edge computing allocation scheme, and the compensation parameters are dynamically corrected by an adaptive filter to obtain a real-time compensation instruction, and the operating parameters of the graphene supercapacitor are updated according to the real-time compensation instruction.

[0021] Specifically, according to the target switching sequence and the capacitor access status data, the harmonic content is sampled according to the preset sampling frequency by the edge computing unit, and the voltage, current and power parameters are extracted from the sampled data to generate the first edge control instruction. The support vector machine regression algorithm is used to extract the features of the first edge control instruction, and the dynamic compensation increment is calculated in combination with the compensation capacity parameter to generate the second edge control instruction including the capacity gear and the adjustment direction. A response timing table is constructed for the second edge control instruction, and the edge node processing capacity load value is calculated by the optimal search algorithm to obtain the edge computing allocation scheme. According to the edge computing allocation scheme and the real-time status of the edge node, the edge node task sequence is generated by the distributed computing method to obtain a resource allocation table including computing tasks and execution timing. According to the resource allocation table, the compensation parameter correction amount is calculated according to the correction time interval, and the compensation parameter is dynamically corrected by the adaptive filter to generate a real-time compensation instruction. According to the real-time compensation instruction, the operating parameters of the graphene supercapacitor are updated, and the switching execution result is judged according to the operating status monitoring data, and a feedback data set including the execution status and the correction parameters is generated. The edge computing unit realizes real-time control of the graphene supercapacitor through high-speed data acquisition. The sampling frequency is set to 12800 Hz, and 384 data points are collected in each working cycle, and three parameters, voltage, current and power, are recorded at the same time. The voltage sampling accuracy is 0.1 volt, and the current sampling accuracy is 0.01 ampere. The sampled data is digitally filtered to generate edge control instructions. The support vector machine regression algorithm uses the radial basis kernel function, and the kernel parameter is set to 0.1 to extract features from the control instructions. The feature dimensions include voltage fluctuation rate, current distortion rate and power factor. The calculated dynamic compensation increment varies within the range of plus or minus 10% of the original compensation capacity. The compensation gear is divided into 5 gears, and the capacity difference of each gear is 20% of the rated capacity. The processing capacity of the edge node is related to its hardware configuration. The processor with a main frequency of 1.5 gigahertz processes 4096 data points per second and has a memory capacity of 4 gigabytes. The optimal search algorithm calculates the load distribution scheme based on the processing capacity. When the load of a single node exceeds 75%, the task is automatically distributed to the adjacent node. Distributed computing adopts a master-slave architecture, where the master node is responsible for task allocation and the slave node performs specific calculations. The resource allocation table contains 64 time slots, each of which is 5 milliseconds long. Computational tasks are assigned to time slots according to their priority, and high-priority tasks occupy continuous time slots. The compensation parameters are corrected using a basic interval of 50 milliseconds, and the parameters are corrected through an adaptive filter. The forgetting factor of the filter is 0.96, and the convergence time does not exceed 3 correction cycles. During the dynamic correction process, the rate of change of the compensation parameters is limited to less than 20% per second. The real-time compensation instruction contains the switching control word and the parameter update value. The control word is 16 bits long, of which the upper 8 bits represent the switching state and the lower 8 bits represent the compensation capacity.The operation status monitoring collects the voltage and current data of the capacitor port. The voltage fluctuation during the switching process does not exceed 2% of the rated value. The execution status feedback includes the operation result code and the execution timestamp. All monitoring data is saved in binary format, and the data frame length is 256 bytes. The operating parameters of the graphene supercapacitor include the charge and discharge status, capacity utilization and internal resistance change rate. The operation data shows that under load fluctuation conditions, the capacity utilization rate is maintained between 65% and 85%, and the internal resistance change rate is less than 0.1% / hour. Through real-time monitoring and dynamic adjustment, the graphene supercapacitor always works in the best matching state and the operating parameters remain stable.

[0022] S106, acquiring system transient current data, and using a wavelet transform algorithm to decompose the transient current data to obtain frequency components of the transient current.

[0023] A Butterworth second-order bandpass filter is used to preprocess transient current data to obtain a preprocessed transient data sequence; a five-layer decomposition and reconstruction is performed on the preprocessed transient data sequence through a db4 wavelet basis function to obtain a decomposed and reconstructed transient data sequence containing high-frequency components and low-frequency components; amplitude extraction is performed based on the high-frequency components in the decomposed and reconstructed transient data sequence, and an instantaneous phase value is calculated through a Hilbert transform to obtain a transient feature matrix containing an amplitude sequence and a phase sequence; support vector machine regression is used to calculate the time series characteristics of the frequency components of the transient feature matrix, and the main frequency components are identified through an adaptive clustering method to obtain a frequency component feature table.

[0024] Specifically, transient current data is obtained according to a preset transient sampling frequency, and the transient current data is preprocessed using a Butterworth second-order bandpass filter to obtain a first transient data sequence. The db4 wavelet basis function is selected for the first transient data sequence to perform five-layer decomposition, and a second transient data sequence containing high-frequency components and low-frequency components is obtained by inverse reconstruction. The amplitude of the high-frequency component in the second transient data sequence is extracted, and the instantaneous phase value is calculated using the Hilbert transform to generate a transient feature matrix containing an amplitude sequence and a phase sequence. A time-frequency domain mapping table is constructed according to the transient feature matrix, and the frequency component time series characteristics are calculated by support vector machine regression to obtain a frequency component change curve. The frequency component change curve is arranged in descending order according to the component energy size, and the main frequency components are identified using an adaptive clustering method to generate a frequency component feature table. The attenuation rate of each component is calculated according to the frequency component feature table, and a transient feature vector containing an amplitude change rate and a phase offset is obtained by piecewise linear fitting. In the transient analysis of power systems, the selection of sampling frequency directly affects the accuracy of signal processing. It is usually set to 256 times the fundamental frequency, that is, the sampling frequency is 12800 Hz in a 50 Hz power grid. The passband of the Butterworth second-order bandpass filter is set between 10 Hz and 6400 Hz, and the attenuation rate of the cutoff frequency is 12 dB per octave, which effectively filters out power frequency interference and high-frequency noise. The db4 wavelet basis function is selected for five-layer decomposition, which has good time-frequency aggregation. The first layer of decomposition obtains the detail coefficients of the frequency band from 6400 to 3200 Hz, and the second layer of decomposition obtains the detail coefficients of the frequency band from 3200 to 1600 Hz, and so on until the fifth layer. The time domain waveforms of each frequency band are restored by reverse reconstruction. The high-frequency component reflects the rapid change characteristics in the transient process, and the low-frequency component reflects the basic waveform characteristics. The amplitude of the reconstructed high-frequency component is extracted, and the instantaneous phase is calculated using Hilbert transform. At the beginning of the transient, the amplitude of the high-frequency component can reach 80% of the fundamental amplitude, and the phase shows a rapid change characteristic, with a phase change rate of more than 360 degrees per millisecond. The transient feature matrix records the amplitude and phase information of each frequency band at different times. Short-time Fourier transform is used for time-frequency domain mapping, the time window length is set to 2 milliseconds, and the window sliding interval is 0.2 milliseconds. The radial basis function is used as the kernel function of the support vector machine, and the kernel parameter is set to 0.1. The calculated frequency components show obvious time series characteristics. The high-frequency component accounts for a large proportion in the initial stage of the transient and then decays rapidly. The main frequency components are identified by a density-based clustering method, and the clustering radius is set to 0.15 times the standard deviation of the amplitude. The analysis results show that the transient process mainly contains three frequency components, high-frequency oscillations of about 3000 Hz, medium-frequency oscillations of about 1500 Hz, and low-frequency oscillations of about 500 Hz.The attenuation characteristics of each frequency component are calculated by piecewise linear fitting. The attenuation rate of the high-frequency component is about 90% per millisecond, the attenuation rate of the medium-frequency component is about 60% per millisecond, and the attenuation rate of the low-frequency component is about 30% per millisecond. The phase shift shows a frequency-related law. The higher the frequency, the faster the phase shift rate. The transient characteristic vector also records the initial amplitude ratio, attenuation time constant and phase change curve of each component, which fully describes the dynamic characteristics of the transient process.

[0025] S107. Calculate the pulse amplitude, duration and rising rate of the transient current according to the transient current frequency component. If the transient current impact exceeds the preset threshold, trigger the protection mechanism, take different levels of protection measures according to the severity of the impact, start the capacitor group switching optimization process, and adjust the capacitor group switching sequence again.

[0026] The current rise rate and duration are calculated by the least square method according to the transient current frequency component to obtain a first transient matrix including pulse amplitude and duration characteristics; the amplitude comparison is performed according to the standard limit for the first transient matrix, and the impact characteristic parameters are extracted from the amplitude comparison results to generate a second transient matrix including amplitude excess multiples and duration; support vector machine regression is used to extract features from the second transient matrix, and a protection parameter table including action levels and response times is generated according to the impact amplitude excess multiples; for transient impacts in the protection parameter table that exceed the preset protection threshold, a recursive optimization algorithm is used to calculate the fluctuation suppression parameters to generate an adjusted capacitor group switching sequence.

[0027] Specifically, a transient characteristic data set is constructed according to the transient current frequency component, and the current rise rate and duration are calculated by the least squares method to obtain a first transient matrix containing pulse amplitude and continuous characteristics. Amplitude comparison is performed according to the standard limit for the first transient matrix, and the impact characteristic parameters are extracted from the amplitude comparison results to generate a second transient matrix containing amplitude overlimit multiples and duration. Support vector machine regression is used to extract features from the second transient matrix, and the protection action is divided into four levels according to the impact amplitude overlimit multiples, and a protection parameter table containing action level and response time is generated. For transient shocks exceeding the preset protection threshold in the protection parameter table, the fluctuation suppression parameters are calculated by a recursive optimization algorithm to obtain response parameter groups of different protection levels. The capacitor group optimization control sequence is constructed according to the response parameter group, and the switching time interval is calculated by a hierarchical search method to generate a timing control table containing the switching sequence of each group of capacitors. For the timing control table combined with the capacitor group capacity parameters, the switching optimization instruction is generated according to the hierarchical protection principle, and the capacitor group switching sequence is adjusted online by an adaptive correction algorithm. In the transient protection of power system, the extraction of transient characteristics is crucial for protection action. The transient current data obtained by high-frequency sampling contains rich frequency component information. The current rise rate calculated by the least squares method is above 200 amperes / microseconds, and the transient duration is in the range of 0.5 to 2 milliseconds. The pulse amplitude can reach 5 to 10 times the rated current to form the first transient matrix. The severity of transient impact is evaluated by the amplitude over-limit multiples. The over-limit multiples relative to the standard limit are divided into four levels, 1.5 to 2 times for mild over-limit, 2 to 3 times for moderate over-limit, 3 to 4 times for severe over-limit, and more than 4 times for severe over-limit. The second transient matrix records the over-limit multiples and duration at different times. The protection action is triggered when the over-limit duration exceeds 1 millisecond. The support vector machine uses the radial basis kernel function for feature extraction, and the kernel parameter is set to 0.1. The calculated protection action level corresponds to the degree of over-limit. Mild over-limit corresponds to level 1 protection with a response time of 20 milliseconds; moderate over-limit corresponds to level 2 protection with a response time of 10 milliseconds; severe over-limit corresponds to level 3 protection with a response time of 5 milliseconds; and serious over-limit corresponds to level 4 protection with a response time of 2 milliseconds. The recursive optimization algorithm calculates the fluctuation suppression parameters based on the characteristics of transient shocks. The switching capacity of level 1 protection is 25% of the rated capacity, the switching capacity of level 2 protection is 50%, the switching capacity of level 3 protection is 75%, and the switching capacity of level 4 protection is all available. The response parameter groups of different protection levels contain detailed capacity regulation strategies. The generation of the optimized control sequence adopts a hierarchical search method, and the switching time interval is shortened step by step with the protection level. The minimum switching interval of level 1 protection is 100 milliseconds, that of level 2 protection is 50 milliseconds, that of level 3 protection is 20 milliseconds, and that of level 4 protection is 10 milliseconds.The timing control table records the specific switching time and capacity of each capacitor group. The adaptive correction algorithm dynamically adjusts the switching sequence. When the transient impact is detected to be weakened, the switching action gradually returns to normal according to the preset rules. When the transient amplitude drops below 1.2 times the standard limit, the switching interval is extended to the normal level and the capacity adjustment step is reduced to the standard value. By continuously monitoring the transient characteristics, the protection mechanism always maintains an effective response to transient impacts while avoiding excessive protection actions.

[0028] S108. Adopt an adaptive filtering algorithm to optimize the switching sequence of the adjusted capacitor group, including minimizing the switch inrush current and suppressing resonance during the switching process, and send the optimized switching sequence to the execution unit to coordinate with other control links.

[0029] An adaptive Kalman filter is used to perform online optimization on the switching sequence of the capacitor group to obtain a first optimization parameter group including switching control parameters; a neural network regression is used to calculate the resonance suppression coefficient based on the first optimization parameter group to obtain a second optimization parameter group including the switching timing and the suppression amplitude; a least squares optimizer is constructed for the second optimization parameter group to calculate the switching execution sequence to obtain a control instruction table including the execution priority and the response timing; a collaborative control matrix is ​​constructed based on the control instruction table, and the response time of each node is calculated by the recursive least squares method to obtain a collaborative control sequence including the response timing and the execution order.

[0030] Specifically, according to the switching sequence of the capacitor group and the switch inrush current data, an adaptive Kalman filter is used to optimize the switching sequence online, and a first optimization parameter group is obtained by resonance suppression calculation, and the switching control parameters are extracted from the first optimization parameter group. The switching control parameters are combined with the inrush current peak limit condition, and the resonance suppression coefficient is calculated by neural network regression to generate a second optimization parameter group including the switching timing and the suppression amplitude. According to the second optimization parameter group combined with the filter gain value, a least squares optimizer is constructed to calculate the switching execution sequence, and a control instruction table including the execution priority and the response timing is obtained. Distributed node allocation is performed for the control instruction table, and the execution load parameters are calculated according to the node processing capacity, and an execution allocation table including node identification and load distribution is generated. According to the execution allocation table, a collaborative control matrix is ​​constructed, and the response time of each node is calculated by recursive least squares method to obtain a collaborative control sequence including the response timing and execution order. According to the collaborative control sequence, an adjustment instruction group is generated according to the preset execution rules, and the adjustment instruction group is dynamically corrected by an adaptive optimizer, and an optimization control table including the suppression parameters and the switching time is extracted from the correction result. Inrush suppression and resonance control during the switching process of capacitor banks are the key to ensuring the safe operation of the power grid. The adaptive Kalman filter uses high-speed sampling to obtain switch inrush data. The sampling frequency is set to 12800 Hz and the measurement accuracy is 0.1%. The inrush peak usually reaches 10 to 15 times the rated current, and the duration is in the range of 1 to 2 milliseconds. The inrush peak is controlled within 5 times the rated current through filtering optimization. The neural network regression adopts a three-layer structure. The input layer contains three features: inrush amplitude, duration and phase angle. The number of hidden layer nodes is 8, and the output layer gives the resonance suppression coefficient. Batch normalization is used for network training, and the learning rate is set to 0.01. After 500 iterations, the prediction accuracy reaches 98%. The optimized switching timing requires that the switching interval of adjacent capacitor banks is not less than 20 milliseconds. The least squares optimizer calculates the switching execution sequence based on the filter gain value, and the gain value is dynamically adjusted in the range of 0.1 to 0.9. The execution priority is divided into four levels. The response time of high-priority tasks is less than 5 milliseconds, and the response time of low-priority tasks is within 20 milliseconds. The control instruction table records in detail the execution time and response requirements of each switching action. The load distribution of distributed nodes takes into account processor performance and network delay. Nodes with a main frequency of 1.5 GHz or above can process 8 data streams at the same time, and nodes with a main frequency between 800 MHz and 1.5 GHz can process 4 data streams. When the network delay exceeds 50 milliseconds, the number of assigned tasks is automatically reduced to ensure that the control instructions are executed in time. The collaborative control matrix reflects the timing relationship between each execution node, and the matrix dimension is the product of the number of nodes and the control cycle. The recursive least squares method uses a forgetting factor of 0.98 to calculate the node response time in real time, and triggers task reallocation when the response delay exceeds 20% of the preset value.The adaptive optimizer dynamically adjusts the control parameters and adjusts the switching timing according to the resonance suppression effect. The suppression parameters include fundamental compensation and harmonic compensation. The fundamental compensation adopts segmented adjustment, and the adjustment amplitude of each segment does not exceed 20% of the rated capacity. The harmonic compensation adopts continuous adjustment, and the adjustment rate is limited to 10% per second. The optimization control table records a detailed switching execution plan, including the switching time, compensation capacity and execution node identification, to provide precise guidance for the coordinated control of the capacitor group.

[0031] It should be noted that the above examples are only some specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and there are many variations. All variations that can be directly derived or associated with the content disclosed by a person skilled in the art should be considered as the protection scope of the present invention.

Claims

1. A graphene supercapacitor capacity energy-saving control method based on intelligent algorithm, characterized in that: The method comprises: Obtain the harmonic content data of the power grid, use the Fourier transform algorithm to perform spectrum analysis on the harmonic content, and obtain the changing trend of the harmonic content; According to the trend of harmonic content changes, the bands with abnormal harmonic content are identified. For the abnormal bands, the edge computing unit is triggered to perform real-time harmonic content evaluation, determine the degree of harmonic excess and the location of the harmonic source, generate corresponding harmonic suppression strategies, including switching capacitor groups and adjusting the working parameters of active filter devices, and send them to the execution unit; Obtain historical capacitor bank impedance characteristic data and corresponding harmonic content change trends, and calculate the impedance matching of graphene supercapacitors under different load conditions; According to the impedance matching of graphene supercapacitors, a dynamic programming algorithm is used to generate a capacitor bank switching sequence and determine the target switching sequence of graphene supercapacitors; The edge computing unit executes the switching of graphene supercapacitors according to the target switching sequence, and adjusts the access status of graphene supercapacitors in real time to dynamically match the harmonic changes of the power grid; Obtain system transient current data, use wavelet transform algorithm to decompose the transient current data, and obtain the frequency components of the transient current; According to the transient current frequency component, the pulse amplitude, duration and rising rate of the transient current are calculated. If the transient current impact exceeds the preset threshold, the protection mechanism is triggered, and different levels of protection measures are taken according to the severity of the impact. The capacitor bank switching optimization process is started and the capacitor bank switching sequence is adjusted again; An adaptive filtering algorithm is used to optimize the adjusted capacitor group switching sequence, including minimizing the switch inrush current and suppressing resonance during the switching process. The optimized switching sequence is sent to the execution unit to coordinate with other control links.

2. The method according to claim 1, characterized in that The method of obtaining the harmonic content data of the power grid and performing spectrum analysis on the harmonic content using a Fourier transform algorithm to obtain a variation trend of the harmonic content includes: Use a dynamic window harmonic collector to read the harmonic signal data generated by the power grid equipment during operation according to the preset harmonic sampling frequency; After the harmonic signal data is subjected to wavelet denoising filtering, the amplitude, phase and frequency characteristics are extracted by Fourier transform algorithm, the harmonic content is divided into corresponding intervals by density-based clustering method, and the distortion rate value and total distortion amount of each harmonic are obtained from the divided intervals; A harmonic propagation characteristic matrix is ​​constructed according to the values ​​of the distortion rates of each harmonic and the total amount of distortion; when the harmonic voltage fluctuation amplitude exceeds a preset operating threshold, a recursive neural network is used to quantitatively analyze the harmonic superposition effect, and a harmonic interference degree index is obtained from the harmonic propagation characteristic matrix; According to the harmonic interference degree and the harmonic attenuation coefficient, the harmonic content variation trend analysis is performed to obtain time series data.

3. The method according to claim 1, characterized in that According to the change trend of harmonic content, the band with abnormal harmonic content is identified, and for the abnormal band, the edge computing unit is triggered to perform real-time harmonic content evaluation, determine the degree of harmonic excess and the location of the harmonic source, generate corresponding harmonic suppression strategies, including switching capacitor groups and adjusting the working parameters of the active filter device, and send them to the execution unit, including: According to the time series data, an adaptive threshold detection algorithm based on statistical test is used to determine the abnormal harmonic frequency band, and phase characteristics and amplitude characteristics are extracted from the abnormal harmonic frequency band to obtain a harmonic feature sequence; The abnormal harmonic frequency band is sampled by using an edge multi-point distributed acquisition device, the waveform distortion is calculated according to the harmonic characteristic sequence to obtain the harmonic source intensity value, and the attenuation rate is calculated by comparing the amplitudes between the measurement points to obtain the harmonic propagation trajectory; Constructing a propagation distance matrix according to the harmonic propagation trajectory, locating the harmonic source node using a minimum spanning tree algorithm, and generating a harmonic source position description; In view of the harmonic source position description, a capacitor group switching schedule is generated according to the harmonic compensation capacity of each measuring point, a load balancing algorithm is used to perform task allocation on the capacitor group switching schedule, and an edge control instruction set is generated.

4. The method according to claim 1, characterized in that: The obtaining of historical capacitor group impedance characteristic data and corresponding harmonic content variation trend, and calculating the impedance matching of the graphene supercapacitor under different load conditions, includes: According to the impedance measurement values ​​recorded in the historical operation data of the capacitor bank, a wavelet decomposition algorithm is used to perform time-frequency domain analysis on the impedance measurement values ​​to obtain an impedance characteristic sequence containing amplitude and phase information; Obtaining fused impedance parameters through a data fusion algorithm for the impedance characteristic sequence and the operating parameters acquired by the distributed data acquisition device; An impedance matrix reflecting load changes is constructed according to the fused impedance parameters, wherein the row vector of the impedance matrix represents the load gear, and the column vector represents the time series, and the impedance offset is obtained by the least square method; The harmonic distortion level is calculated by using a recursive least square method for the impedance offset and the load gear data, and the variation law of the harmonic content is extracted from the harmonic distortion level to obtain a load response characteristic curve; The variation law of harmonic content is extracted based on the load response characteristic curve, and the impedance matching degree between the graphene supercapacitor and the original capacitor group is calculated by support vector machine regression.

5. The method according to claim 1, characterized in that The method of generating a capacitor bank switching sequence by using a dynamic programming algorithm according to the impedance matching of the graphene supercapacitor and determining a target switching sequence of the graphene supercapacitor comprises: Constructing a switching state matrix according to the graphene supercapacitor impedance matching degree and the capacitor bank working state data, and extracting a switching priority sequence from the switching state matrix; Calculating a capacity allocation ratio based on the switching priority sequence in combination with the real-time power data of the capacitor bank, adjusting the capacity allocation ratio through a recursive neural network, and obtaining a switching state matrix of the adjusted capacity allocation ratio; Using piecewise linear regression to calculate the capacitor bank loss value corresponding to the switching state matrix of the adjusted capacity allocation ratio, to obtain a switching stability curve; According to the switching stability curve and the correction coefficient table, a dynamic timing compensation algorithm is used to generate a control timing table, and the switching time and capacity gear value are extracted from the control timing table to generate a target switching sequence of the graphene supercapacitor.

6. The method according to claim 1, characterized in that The method executes the switching of the graphene supercapacitor according to the target switching sequence through the edge computing unit, and adjusts the access state of the graphene supercapacitor in real time to dynamically match the harmonic changes of the power grid, including: The harmonic content is sampled by the edge computing unit according to a preset sampling frequency, voltage, current and power parameters are extracted from the sampled data, and a first edge control instruction is obtained; A support vector machine regression algorithm is used to extract features according to the first edge control instruction, and a dynamic compensation increment is calculated in combination with a compensation capacity parameter to obtain a second edge control instruction; Constructing a response timing table for the second edge control instruction, calculating the edge node processing capacity load value through an optimal search algorithm, and obtaining an edge computing allocation plan; According to the edge computing allocation scheme, a distributed computing method is used to generate an edge node task sequence, and the compensation parameters are dynamically corrected through an adaptive filter to obtain real-time compensation instructions. According to the real-time compensation instructions, the operating parameters of the graphene supercapacitor are updated.

7. The method according to claim 1, characterized in that The method of acquiring system transient current data and decomposing the transient current data using a wavelet transform algorithm to obtain frequency components of the transient current includes: The transient current data is preprocessed by using a Butterworth second-order bandpass filter to obtain a preprocessed transient data sequence; Performing five-layer decomposition and reconstruction on the preprocessed transient data sequence by using db4 wavelet basis function to obtain a decomposed and reconstructed transient data sequence including high-frequency components and low-frequency components; Extracting amplitude from high-frequency components in the decomposed and reconstructed transient data sequence, calculating instantaneous phase values ​​through Hilbert transform, and obtaining a transient characteristic matrix including an amplitude sequence and a phase sequence; The support vector machine regression is used to calculate the time series characteristics of the frequency components for the transient characteristic matrix, and the main frequency components are identified by an adaptive clustering method to obtain a frequency component characteristic table.

8. The method according to claim 1, characterized in that The method calculates the pulse amplitude, duration and rising rate of the transient current according to the transient current frequency component, and triggers the protection mechanism if the transient current impact exceeds the preset threshold value, takes different levels of protection measures according to the severity of the impact, starts the capacitor bank switching optimization process, and adjusts the capacitor bank switching sequence again, including: The current rise rate and duration are calculated by using the least square method according to the transient current frequency component, so as to obtain a first transient matrix including pulse amplitude and duration characteristics; Performing amplitude comparison on the first transient matrix according to standard limits, extracting impulse characteristic parameters from the amplitude comparison results, and generating a second transient matrix including amplitude over-limit multiples and duration; Support vector machine regression is used to extract features from the second transient matrix, and a protection parameter table including action levels and response times is generated according to the multiples of the impulse amplitude exceeding the limit; For transient shocks exceeding a preset protection threshold in the protection parameter table, a recursive optimization algorithm is used to calculate fluctuation suppression parameters to generate an adjusted capacitor bank switching sequence.

9. The method according to claim 1, characterized in that: The adaptive filtering algorithm is used to optimize the adjusted capacitor bank switching sequence, including minimizing the switch inrush current and suppressing resonance during the switching process, and the optimized switching sequence is sent to the execution unit to coordinate with other control links, including: An adaptive Kalman filter is used to perform online optimization on the switching sequence of the capacitor bank to obtain a first optimization parameter group including switching control parameters; According to the first optimization parameter group, a resonance suppression coefficient is calculated by using a neural network regression to obtain a second optimization parameter group including a switching timing and a suppression amplitude; Constructing a least squares optimizer for calculating the switching execution sequence for the second optimization parameter group to obtain a control instruction table including execution priority and response timing; A collaborative control matrix is ​​constructed according to the control instruction table, and the response time of each node is calculated by recursive least square method to obtain a collaborative control sequence including a response timing and an execution order.

Images