Intelligent control method and system for multi-mode switching of AC / DC module

Abstract

The invention relates to the technical field of AC / DC module control, and discloses an intelligent control method and system for multi-mode switching of an AC / DC module. According to the method, a dynamic topology mapping model is constructed according to an impedance spectrum matrix, and a physical connection relation is converted into a node admittance relation network; generating an energy distribution state sequence based on the power fluctuation vector, and analyzing migration constraint conditions among different operation modes by adopting an association rule mining algorithm; when an operation mode switching instruction is detected, activating a node admittance relation network in the dynamic topology mapping model, and inputting a current energy distribution state sequence to an association rule mining algorithm; according to a real-time matching result of the node admittance relation network and the migration constraint condition, generating a switching control instruction containing an impedance adjustment gradient and a power distribution weight; and the power semiconductor execution unit transmits the switching control instruction to the AC / DC module. According to the method, multi-mode intelligent switching control of the alternating-current and direct-current module can be realized.

Description

Technical Field

[0001] This invention relates to the field of AC / DC module control technology, specifically to an intelligent control method and system for multi-mode switching of AC / DC modules. Background Technology

[0002] In fields such as new energy power generation, energy storage systems, and microgrids, AC / DC modules, as core components for energy conversion and transmission, need to flexibly switch between multiple operating modes to adapt to different working conditions. For example, in distributed generation systems, modules may need to switch from AC grid-connected mode to DC off-grid mode, or from hybrid power supply mode to standby energy storage mode. These mode switching involves complex processes such as circuit topology reconfiguration, power flow adjustment, and impedance characteristic adaptation, placing extremely high demands on the dynamic response and adaptability of control methods.

[0003] In existing control methods, parameter acquisition primarily focuses on conventional electrical quantities such as voltage and current, failing to adequately capture the dynamic changes in impedance spectrum and the instantaneous characteristics of power fluctuations. This results in the raw parameter set failing to fully reflect the real-time operating status of the module, leaving subsequent control decisions without comprehensive data support. Regarding topology handling, traditional methods rely on pre-set static circuit models, solidifying physical connections into fixed matrix parameters. When internal components age, load changes abruptly, or external devices are connected, the physical connections dynamically change, and the static model cannot be updated in real time, easily leading to discrepancies between topology perception and actual conditions.

[0004] The analysis of energy distribution often relies on empirical thresholds, neglecting the inherent correlations between different operating modes. For example, when switching from DC boost mode to AC inverter mode, the amplitude, frequency, and duration of power fluctuations create specific constraints. Relying solely on thresholds makes it difficult to accurately analyze these constraints, leading to problems such as power surges and voltage oscillations during mode switching. Furthermore, insufficient coordination between control command generation and execution units, with command parameters often based on historical data, fails to dynamically match real-time node admittance relationships and energy distribution states. This results in a delay between the response of power semiconductor execution units and the actual needs of the module, affecting the smoothness of the switching process. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent control method for multi-mode switching of AC / DC modules to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent control method for multi-mode switching of AC / DC modules, the method comprising: Real-time acquisition of multi-dimensional electrical parameters of AC / DC modules in hybrid operation mode, generating an original parameter set including impedance spectrum matrix and power fluctuation vector; A dynamic topology mapping model is constructed based on the impedance spectrum matrix, and the physical connection relationship of AC / DC modules is transformed into a node admittance relationship network through this model; Energy distribution state sequences are generated based on power fluctuation vectors, and the migration constraints between different operating modes are analyzed using an association rule mining algorithm. When a running mode switching command is detected, the node admittance relationship network in the dynamic topology mapping model is activated, and the current energy distribution state sequence is input into the association rule mining algorithm. Based on the real-time matching results of the node admittance relationship network and migration constraints, a switching control command containing impedance adjustment gradient and power allocation weight is generated. The switching control command is transmitted to the power semiconductor execution unit of the AC / DC module.

[0007] Preferably, the multi-dimensional electrical parameters include the input voltage ripple coefficient, the output current harmonic component, the module temperature distribution value, and the heat dissipation efficiency factor; Set the input voltage ripple coefficient to the reference input value and the output current harmonic component to the target output value; The module temperature distribution value is calculated using an integral-differential hybrid algorithm: temperature sampling values ​​from multiple temperature measurement points inside the AC / DC module are obtained, spatial interpolation is performed on the temperature sampling values ​​to generate a three-dimensional temperature field, the standard deviation of the temperature gradient in each direction in the three-dimensional temperature field is calculated, and the ratio of the standard deviation to a preset temperature rise threshold is marked as the module temperature distribution value. The heat dissipation efficiency factor is expressed as the ratio of the power consumption of the heat dissipation device to the total power consumption of the AC / DC module.

[0008] Preferably, the construction process of the dynamic topology mapping model includes: A higher-order tensor decomposition is performed on the impedance spectrum matrix in the original parameter set to extract the frequency domain correlation features between the real and imaginary parts of the impedance. A node admittance relationship network is established based on frequency domain correlation features. The node admittance relationship network includes the admittance values ​​of each circuit node of the AC / DC module and the admittance weights of its connection paths. The power fluctuation vector is imported into the nodal admittance relation network for load sensitivity simulation, generating migration constraints that include voltage stability range and current margin.

[0009] Preferably, the calculation expression of the dynamic topology mapping model is: The predicted value of the output impedance adjustment gradient is obtained by convolving the admittance weight matrix in the nodal admittance relationship network with the input voltage ripple coefficient. The current margin in the migration constraint and the harmonic components of the output current are calculated using Euclidean distance to obtain the optimized value of the output power allocation weight. The predicted value of the impedance adjustment gradient and the optimized value of the power allocation weight constitute the core parameters of the switching control command.

[0010] Preferably, the implementation process of the association rule mining algorithm includes: A spatial gridding processing unit is set up to divide the operating area of ​​the AC / DC module into three-dimensional grids of equal size; Extract the spatiotemporal distribution data of power fluctuation vectors in a three-dimensional grid to form discrete segments of the energy distribution state sequence; For each discrete segment, the confidence of the association rule is evaluated to generate migration constraints between different operating modes.

[0011] Preferably, the specific operations for evaluating the confidence level of the association rule include: Tensor eigenvalue decomposition is performed on the energy distribution state sequence in the discrete segment to extract the energy ratio vector of the fundamental wave and harmonic waves. Calculate the autocorrelation function of the energy proportion vector within a preset time window to identify the periodic characteristics of the energy distribution state sequence; Establish a mapping relationship between periodic characteristics and AC / DC module operating modes, and determine the confidence score of migration constraints based on the matching degree of the mapping relationship.

[0012] Preferably, the calculation process for the confidence score of the migration constraint includes: The number of times the statistical energy distribution state sequence is triggered in historical operating mode switching events; Obtain the steady-state operating duration of the AC / DC module after each trigger; The ratio of the average duration of steady-state operation to the total duration of discrete segments is labeled as the confidence score of the migration constraint.

[0013] Preferably, the process of generating the switching control command includes: Get the trigger timestamp of the running mode switching command; Extract the node admittance relationship network parameters corresponding to the timestamp in the dynamic topology mapping model; Perform spatial vector projection calculations on the nodal admittance relationship network parameters and migration constraints; When the projection calculation result meets the preset impedance continuity condition, the predicted value of the output impedance adjustment gradient is obtained. When the projection calculation result meets the preset power compatibility conditions, the optimized value of the power allocation weight is output.

[0014] Preferably, the present invention also includes an intelligent control system for AC / DC module multi-mode switching, used to implement the above-mentioned intelligent control method for AC / DC module multi-mode switching, the system including a parameter acquisition module, a model construction module, a rule analysis module and an instruction generation module; The parameter acquisition module acquires multi-dimensional electrical parameters of the AC / DC module in real time, generates an original parameter set, and transmits it to the model building module. The model building module constructs a dynamic topology mapping model based on the impedance spectrum matrix, and forwards the power fluctuation vector to the rule analysis module. The rule analysis module generates migration constraints based on the power fluctuation vector and feeds the migration constraints back to the model building module. The instruction generation module generates switching control instructions based on the node admittance relationship network output by the model construction module and the migration constraints fed back by the rule analysis module.

[0015] Preferably, the system further includes a feedback calibration module, which performs the following operations: Collect actual impedance response data after the power semiconductor execution unit executes the switching control command; The residual is calculated by comparing the actual impedance response data with the predicted impedance adjustment gradient values ​​in the dynamic topology mapping model. The admittance weight matrix of the nodal admittance relationship network is updated using the residual calculation results; The updated admittance weight matrix is ​​synchronized to the dynamic topology mapping model in the model building module.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This intelligent control method acquires multi-dimensional electrical parameters of the AC / DC module in real time under hybrid operation mode, generating a raw parameter set including an impedance spectrum matrix and a power fluctuation vector. Compared to traditional methods that only acquire limited electrical quantities, this method can more comprehensively capture the module's operating status. The impedance spectrum matrix covers impedance characteristics at different frequencies, and the power fluctuation vector reflects the power variation characteristics over time. The raw parameter set formed by combining these two provides rich foundational information for subsequent topology mapping and constraint analysis.

[0017] The dynamic topology mapping model constructed based on the impedance spectrum matrix overcomes the limitations of traditional static topology models. This model can transform the physical connection relationships of AC / DC modules into a node admittance relationship network in real time. When the connection method of internal components changes or the external topology is adjusted, the node admittance relationship network can be dynamically updated accordingly, accurately reflecting the current circuit topology characteristics. This dynamic mapping capability ensures that the topology perception always remains consistent with the actual physical state, providing a reliable topology foundation for mode switching.

[0018] The energy distribution state sequence generated based on the power fluctuation vector fully records the distribution and change trajectory of energy in various parts of the module. An association rule mining algorithm is used to analyze this sequence, enabling the extraction of migration constraints between different operating modes from the patterns of energy fluctuations. These constraints are not based on empirical presets but rather stem from the inherent correlations in actual operating data, thus better reflecting the module's real operating characteristics and providing accurate basis for constraint judgments during mode switching.

[0019] When a mode switching command is detected, the node admittance relationship network in the dynamic topology mapping model is activated. Simultaneously, the current energy distribution state sequence is input into the association rule mining algorithm, forming a real-time linkage analysis of topological characteristics and energy state. The node admittance relationship network provides the foundation for the current circuit connections, while the energy distribution state sequence reflects the current state of energy flow. The combination of the two allows the switching analysis to have both structural support at the topological level and dynamic reference at the energy level. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent control method for multi-mode switching of AC / DC modules described in this invention. Figure 2 Design drawings for multi-dimensional electrical parameters; Figure 3 Design diagram for constructing a dynamic topology mapping model; Figure 4 A design diagram for evaluating the confidence of association rules. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Please see Figure 1 This invention provides an intelligent control method and system for multi-mode switching of AC / DC modules, the method comprising: The system acquires multi-dimensional electrical parameters of the AC / DC module in real time under hybrid operation mode, generating a raw parameter set including an impedance spectrum matrix and a power fluctuation vector. During this process, distributed sensing units synchronously sample the input, output, and key internal nodes of the AC / DC module. The sampling frequency is dynamically adjusted according to the module's rated power to ensure that the impedance spectrum matrix reflects the impedance characteristics over a wide frequency range and that the power fluctuation vector captures instantaneous power changes.

[0023] A dynamic topology mapping model is constructed based on the impedance spectrum matrix. This model transforms the physical connection relationships of AC / DC modules into a network of node admittance relationships. The dynamic topology mapping model needs to be based on the circuit structure characteristics of the modules, integrating the impedance parameters of components such as resistors, capacitors, and inductors into a matrix form, and then transforming them into admittance relationships between nodes through mathematical transformations. This allows the on / off states of physical connections and impedance changes to be reflected in the admittance values ​​and connection weights of network nodes.

[0024] An energy distribution state sequence is generated based on the power fluctuation vector, and an association rule mining algorithm is used to analyze the migration constraints between different operating modes. The energy distribution state sequence needs to record the power distribution in each sub-unit of the module according to the time axis. The association rule mining algorithm then extracts the voltage, current, power, and other constraint boundaries that need to be satisfied during mode migration by analyzing the energy distribution characteristics during historical mode switching.

[0025] When a mode switching command is detected, the node admittance relationship network in the dynamic topology mapping model is activated, and the current energy distribution state sequence is input into the association rule mining algorithm. The mode switching command can come from an external control signal or an automatic triggering mechanism within the module. After triggering, the node admittance relationship network enters a real-time calculation state, and the energy distribution state sequence serves as the input parameter for the current operating state.

[0026] Based on the real-time matching results between the node admittance relationship network and the migration constraints, a switching control command containing impedance adjustment gradients and power allocation weights is generated. The matching process requires multi-dimensional parameter comparison to ensure that the circuit topology reflected by the node admittance relationship network is consistent with the state range allowed by the migration constraints, thereby calculating the direction and magnitude of impedance adjustment, as well as the allocation ratio of each power unit.

[0027] The switching control command is transmitted to the power semiconductor execution unit of the AC / DC module. The transmission uses a high-speed communication protocol to ensure that the command is received by the execution unit within microseconds. The execution unit adjusts the gate drive signal according to the command to adjust parameters such as the conduction angle and switching frequency of the power semiconductor device, thereby completing the switching of the operating mode.

[0028] Example 1: See Figure 2 The multi-dimensional electrical parameters include the input voltage ripple coefficient, output current harmonic components, module temperature distribution value, and heat dissipation efficiency factor.

[0029] Obtaining the input voltage ripple coefficient requires multiple steps. First, a high-precision voltage sensor is installed at the input of the AC / DC module to continuously sample the input voltage. The sampling interval is determined based on the module's operating frequency, typically not exceeding one-tenth of the module's switching cycle. The acquired voltage signal contains a DC component and superimposed AC ripple. The DC component needs to be separated using a second-order Butterworth low-pass filter, and then the ripple component is obtained by subtracting the DC component from the original voltage signal. The effective value of the ripple component is calculated by integrating the square of the ripple signal over one cycle, dividing by the cycle length, and then taking the square root. Finally, the ratio of the effective value of the ripple component to the DC component is determined as the input voltage ripple coefficient. This coefficient serves as a reference input quantity for subsequent analysis of the stability of the module's input side.

[0030] The calculation of output current harmonic components begins with current sampling at the output terminal. A Hall current sensor is installed in the output circuit to collect the instantaneous value of the output current. The sampling frequency must be at least 20 times the module's operating frequency to ensure the capture of higher harmonics. The collected current signal is then subjected to a Fourier transform, decomposing it into the fundamental frequency and each harmonic component, obtaining the amplitude and phase information of each harmonic. The total effective value of all harmonic components is calculated by summing the squares of the effective values ​​of each harmonic and then taking the square root. This total effective value is compared with the effective value of the fundamental current; the ratio obtained is the output current harmonic component, which serves as the target output quantity to measure the quality of the module's output power.

[0031] The module temperature distribution values ​​are calculated using a hybrid integral-differential algorithm. Multiple temperature measurement points are arranged inside the AC / DC module, specifically at locations including the chip surface of the power transistors, the contact point between the heatsink and the power transistors, the core and windings of the transformer, the magnetic core of the inductor, and other major heat-generating components, as well as different areas of the module's outer casing. The number of measurement points is determined based on the module size, typically no fewer than 15. Pt100 platinum resistance sensors are used, with a sampling accuracy of ±0.1℃. Spatial interpolation is performed on the temperature samples from these measurement points using a three-dimensional linear interpolation method. Using the coordinates of each measurement point as known points, the temperature values ​​at locations within the module without measurement points are calculated using interpolation formulas, thus constructing a complete three-dimensional temperature field. The temperature gradient of this three-dimensional temperature field along the x, y, and z coordinate axes is calculated. The temperature gradient in each direction is obtained by dividing the temperature difference between two adjacent points by the distance. The standard deviation of the temperature gradient in each direction is then calculated, reflecting the degree of non-uniformity of temperature change in that direction. The average of the standard deviations of the temperature gradients in the three directions is compared with the preset temperature rise threshold, and the resulting ratio is the module temperature distribution value. The preset temperature rise threshold is determined based on the maximum allowable temperature rise of the least heat-resistant component in the module (usually a power semiconductor device).

[0032] Calculating the heat dissipation efficiency factor requires obtaining the power consumption of the heat dissipation device and the total power consumption of the AC / DC module separately. The heat dissipation device includes fans, water cooling systems, etc., and its power consumption is calculated by measuring the input voltage and current of the heat dissipation device, i.e., the product of voltage and current. For AC heat dissipation devices, the power factor also needs to be considered. The total power consumption of the AC / DC module is obtained by subtracting the output power from the input power. The input power is obtained by multiplying the average input voltage and the average input current, while the output power is the product of the average output voltage and the average output current. Dividing the power consumption of the heat dissipation device by the total power consumption of the AC / DC module yields the heat dissipation efficiency factor, which directly reflects the proportion of the heat dissipation system in the total energy consumption of the module. During the calculation process, all power parameter measurements are performed within the same time window, typically 10 times the module's operating cycle, to reduce errors caused by instantaneous fluctuations.

[0033] Example 2: See Figure 3 The construction of the dynamic topology mapping model begins with processing the impedance spectrum matrix in the original parameter set. The impedance spectrum matrix contains impedance data of the AC / DC module at different frequencies, covering multiple frequency bands from low to high frequencies, each corresponding to different real and imaginary impedance values. A high-order tensor decomposition is performed on this matrix, preserving the frequency dimension, real impedance dimension, and imaginary impedance dimension. The high-dimensional matrix is ​​decomposed into multiple low-dimensional factor matrices using a decomposition algorithm. These factor matrices reflect the intrinsic relationships between different dimensions. Frequency domain correlation features between the real and imaginary impedance parts are extracted from the decomposition results. These features include the ratio of the real to imaginary parts at specific frequency points, the consistency of trends with frequency, and the distribution of abrupt change points, reflecting the law of impedance characteristics changing with frequency.

[0034] A node admittance relationship network is established based on the extracted frequency domain correlation features. The nodes in the node admittance relationship network correspond to the key connection points in the AC / DC module circuit, including input nodes, output nodes, connection nodes of each power unit, and connection nodes of energy storage elements. The admittance value of each node is calculated based on the impedance at the corresponding location; that is, the admittance value is the reciprocal of the impedance value, while also considering the influence of parasitic parameters at the node (such as distributed capacitance and lead inductance) on the admittance. The admittance weight of the connection path is determined based on the component parameters (such as resistance, capacitance, and inductance) and the path length. The greater the obstruction of energy transfer by the components on the path, the smaller the admittance weight, and vice versa. The node admittance relationship network is represented in the form of an adjacency matrix, where the element values ​​correspond to the admittance weight of the connection path, thus clearly presenting the admittance correlation between each node.

[0035] Power fluctuation vectors are imported into the node admittance relation network for load sensitivity simulation. The power fluctuation vector contains the fluctuation values ​​of the input and output power of the AC / DC modules at different times, as well as the power exchange fluctuation values ​​between sub-modules. These fluctuation values ​​are used as inputs to perform simulation calculations through the node admittance relation network, simulating the voltage changes and current responses of each node under different power fluctuation conditions. The simulation results determine the voltage stability range, i.e., the maximum and minimum voltage values ​​at which each node voltage can remain within its normal operating range. When the voltage exceeds this range, the module may exhibit instability. Simultaneously, the current margin is calculated, which is the difference between the actual current and the maximum allowable current on each connection path. A positive current margin indicates that the path still has the capacity to carry additional current, while a negative value indicates potential overload. The voltage stability range and current margin together constitute migration constraints, used to limit the range of voltage and current changes during mode switching.

[0036] In the calculation of the dynamic topology mapping model, the admittance weight matrix and the input voltage ripple coefficient are convolved first. The admittance weight matrix is ​​the core parameter matrix of the node admittance relationship network, and the input voltage ripple coefficient is the real-time acquired reference input. The convolution operation is performed in the time dimension, using the admittance weight matrix as the convolution kernel and performing sliding calculations with the time series of the input voltage ripple coefficient. The calculation results reflect the influence of the admittance weight on the voltage ripple at different times, thereby obtaining the predicted value of the impedance adjustment gradient. This value represents the direction and rate at which the impedance needs to be adjusted to suppress voltage ripple.

[0037] Simultaneously, the Euclidean distance between the current margin and the output current harmonic components is calculated. The current margin is derived from the migration constraints, and the output current harmonic components are the target output quantity. Both are multi-dimensional vectors; the current margin vector contains the current margin values ​​for each path, and the output current harmonic component vector contains the component values ​​for each harmonic. The Euclidean distance is calculated using a formula to determine the spatial distance between the two vectors. The smaller the distance, the higher the degree of matching between the current margin and the output current harmonic components. Based on the magnitude of the Euclidean distance, an optimized value for the power allocation weight is obtained through a preset mapping rule. This value determines the proportion of each submodule in the total power to achieve the goal of reducing output current harmonics.

[0038] The predicted value of the impedance adjustment gradient and the optimized value of the power allocation weight together constitute the core parameters of the switching control command. These parameters directly determine the action amplitude and direction of the power semiconductor actuator. Through the synergistic effect of the parameters, the AC / DC module can be smoothly switched between different operating modes.

[0039] Example 3: See Figure 4The implementation of the association rule mining algorithm begins with spatial rasterization. A spatial rasterization unit is set up, which divides the operating area of ​​the AC / DC module into equal-sized three-dimensional grids based on the module's three-dimensional physical structural parameters, including length, width, and height. The grid size setting needs to comprehensively consider the distribution density of components within the module. In areas with dense components such as power semiconductor devices and transformers, the grid side length is set to a smaller value, typically not exceeding 5mm; in areas with sparse components, the grid side length can be appropriately increased, but not exceeding 20mm, to control the data volume while ensuring spatial resolution. Each three-dimensional grid is uniquely identified by three-dimensional coordinates (i,j,k), where i, j, and k correspond to the grid indices in the x, y, and z axes, respectively, ensuring that each grid accurately corresponds to a specific spatial location within the module.

[0040] After spatial rasterization, the spatiotemporal distribution data of power fluctuation vectors within the 3D raster are extracted. The power fluctuation vectors contain the instantaneous power values ​​of key nodes within the module at different times. These instantaneous values ​​are matched with the 3D coordinates of the 3D raster using timestamps to determine the power value within each grid at each time point. This data is arranged chronologically to form a power change sequence for each grid at consecutive time points. The sequences of all grids are then integrated to form discrete segments of the energy distribution state sequence. Each discrete segment contains spatial energy distribution information within a fixed time window. The length of the time window is determined based on the module's dynamic response characteristics, typically 5-10 switching cycles, to ensure that the complete power fluctuation cycle is captured.

[0041] For each discrete segment, the confidence level of the association rule is evaluated to generate migration constraints between different operating modes. During the evaluation process, the operating modes of the modules corresponding to different discrete segments are first compared to identify recurring energy distribution characteristics during mode switching. These characteristics include the power range of a specific grid and the power ratio between different grids. These characteristics are then transformed into quantitative constraints, such as the power of a grid not exceeding 80% of its rated power, or the power difference between two adjacent grids not exceeding a certain threshold.

[0042] The specific operation of association rule confidence assessment includes the following steps: Tensor feature decomposition is performed on the energy distribution state sequence in the discrete segment. The CP decomposition algorithm is used to decompose the sequence into the sum of multiple rank tensors, each rank tensor corresponding to a feature component. From these, the fundamental and harmonic energy proportion vectors are extracted. The fundamental energy proportion vector reflects the main component in energy transmission, while the harmonic energy proportion vector reflects the energy loss and interference components caused by nonlinear elements.

[0043] Calculate the autocorrelation function of the energy proportion vector within a preset time window. The autocorrelation function measures the similarity of sequences at different time points, and its calculation formula is as follows: ; in, This represents the value of the autocorrelation function. Indicates time delay. This indicates the number of sampling points within the time window. This represents the energy percentage vector value at a given time point. This represents the energy percentage vector value at a given time point. By calculating the autocorrelation function values ​​under different time delays, the periodicity of the energy distribution state sequence can be identified. When the autocorrelation function value shows a significant peak, the corresponding time delay is the period of the sequence, and the magnitude of the peak reflects the strength of the periodicity.

[0044] A mapping relationship is established between periodic characteristics and the operating modes of AC / DC modules. Each operating mode (e.g., rectification mode, inverter mode, standby mode, etc.) corresponds to a set of periodic characteristic parameters, including period length, peak intensity, and harmonic component ratio. The matching degree is calculated by comparing the periodic characteristics of the current energy distribution state sequence with the characteristic parameter sets of each operating mode. The matching degree calculation considers the differences of multiple characteristic parameters; the smaller the difference, the higher the matching degree. The confidence score of the migration constraint is determined based on the matching degree; a higher score indicates a better fit between the migration constraint and the current operating mode switching requirements.

[0045] Throughout the evaluation process, data preprocessing is required, including outlier removal (such as jump values ​​caused by sensor malfunctions) and smoothing (using moving averages to reduce high-frequency noise) to ensure the accuracy of the energy distribution state sequence. Simultaneously, the time window length can be dynamically adjusted according to the module's operating status. The window length is shortened during periods of severe power fluctuations to improve responsiveness to rapid changes; the window length is extended during stable operation to reduce computational load and improve the stability of feature extraction. Through these steps, the association rule mining algorithm can accurately analyze the migration constraints between different operating modes, providing a reliable decision-making basis for the multi-mode switching of AC / DC modules.

[0046] Example 4: The calculation process for the confidence score of migration constraints begins with the collection and organization of historical data. All records related to mode switching are extracted from the AC / DC module's operation logs, covering different operating scenarios, including light load, full load, overload, and operating data under different ambient temperatures (e.g., -10℃, 25℃, 40℃). These records contain the trigger time for each mode switch, the operating mode before and after the switch, the energy distribution state sequence, and the changes in operating parameters after the switch.

[0047] The number of times an energy distribution state sequence is triggered in historical operating mode switching events is recorded. For each energy distribution state sequence, historical switching events are examined one by one, and the number of times the sequence occurred and directly caused a mode transition is recorded. For example, if a certain energy distribution state sequence occurred 28 times in the past 100 mode switches, then its trigger count is 28. During the statistics, it is necessary to distinguish between active triggers (such as switching triggered by external commands) and passive triggers (such as switching after the module automatically detects an anomaly) to ensure the comprehensiveness of trigger scenarios. At the same time, atypical switching events caused by sudden failures (such as short circuits or component damage) are excluded to avoid these special cases interfering with the statistical results.

[0048] Acquire the steady-state operating duration of the AC / DC module after each trigger. After each mode switch, monitor key parameters such as the module's output voltage, output current, and module temperature in real time. Timing begins when these parameters are all within the preset allowable fluctuation range (e.g., output voltage fluctuation does not exceed ±2% of the rated value, output current harmonic component does not exceed 5%). Timing stops when any parameter exceeds the allowable range, and the duration recorded is the steady-state operating duration. For example, if the module's output voltage, current, and other parameters remain stable for 120 minutes after a trigger, then the steady-state operating duration for that trigger is 120 minutes. For cases where continuous stable operation exceeds a preset upper limit (e.g., 24 hours), the upper limit value is recorded to avoid excessive influence of individual excessively long stabilization times on the average value calculation.

[0049] The ratio of the average steady-state duration to the total duration of discrete segments is used as the confidence score for the migration constraint. First, the arithmetic mean of the steady-state durations across all triggered events is calculated, i.e., the sum divided by the number of triggers. For example, if the total steady-state duration of 28 triggers is 2520 minutes, then the average is 2520 ÷ 28 = 90 minutes. The total duration of discrete segments is the time length corresponding to the energy distribution state sequence, determined by the sampling frequency and the number of sampling points. If the sampling frequency is 10 Hz and the number of sampling points is 600, then the total duration of discrete segments is 600 ÷ 10 = 60 minutes. In this case, the confidence score is 90 ÷ 60 = 1.5.

[0050] The following is a data table related to a certain energy distribution state sequence: During the calculation process, historical data needs to be stratified. The number of triggers and the duration of steady-state operation are counted separately according to the operating scenario (e.g., load level, ambient temperature), and then a weighted average is calculated. The weights are determined based on the frequency of occurrence of different scenarios; the higher the frequency, the greater the weight, making the average more reflective of the actual operating situation. Simultaneously, a data update mechanism is established. After a certain number of new mode-switching events (e.g., 50 times), the confidence score is recalculated to adapt to changes in the module's operating state. For energy distribution state sequences with short operating times and insufficient historical data (e.g., fewer than 5 triggers), the confidence score is not calculated temporarily until a sufficient amount of data has accumulated. Through these steps, it is ensured that the confidence score accurately reflects the performance of migration constraints in practical applications, providing a reliable reference for AC / DC module mode switching.

[0051] Example 5: The generation process of the switching control command begins with acquiring the trigger timestamp of the operating mode switching command. The trigger timestamp is recorded by a high-precision real-time clock inside the module, which is synchronized with the clocks of the parameter acquisition module and the model building module to ensure time accuracy at the microsecond level. When an external control signal is input or the module detects conditions requiring a switching of the operating mode (such as a sudden load change or input power switching), the timestamp of that moment is immediately recorded as the time reference for all subsequent parameter extraction and calculation. The trigger timestamp is stored in digital form, containing year, month, day, hour, minute, second, and microsecond information, facilitating comparison and traceability with historical data.

[0052] Extract the node admittance relationship network parameters corresponding to the given timestamp in the dynamic topology mapping model. These parameters are stored in a real-time database, which records parameter values ​​at different times in chronological order, including the admittance values ​​of each node, the admittance weights of the connection paths between nodes, and the adjacency matrix of the network topology. Based on the trigger timestamp, the parameter set for the corresponding time moment is precisely retrieved from the database. If the trigger timestamp falls between two recorded times, linear interpolation is used to calculate the parameter values ​​for that time moment, ensuring that the parameters accurately reflect the module circuit state at the moment of triggering. During the extraction process, the parameters are validated for completeness, checking for missing or outlier values. If a problem is found, the parameters from the previous time moment are automatically retrieved as backups, and the status of the extraction result is marked for subsequent analysis.

[0053] The nodal admittance network parameters and migration constraints are subjected to spatial vector projection. The nodal admittance network parameters are transformed into vectors in a high-dimensional space, where the dimension of the vector equals the number of parameters. Each dimension's value corresponds to a standardized value of a specific parameter (standardized by dividing the parameter value by its maximum value). Similarly, the migration constraints are transformed into multiple constraint planes in this high-dimensional space. Each constraint plane is defined by an inequality equation, representing the allowed range of values ​​for a certain parameter. The spatial vector projection operation determines the relative positional relationship between the vector and the constraint plane by calculating the projection components of the nodal admittance vector onto each constraint plane. A positive projection component indicates that the parameter satisfies the constraint condition, while a negative component indicates that it does not.

[0054] When the projection calculation result meets the preset impedance continuity condition, the predicted value of the impedance adjustment gradient is output. The impedance continuity condition requires that all projection components of impedance-related parameters are positive, and the rate of change of impedance parameters between adjacent time points does not exceed a set range. Specifically, the difference between the nodal admittance value at the current time and the previous time point needs to be calculated, and then divided by the time interval to obtain the rate of change. This rate of change must be less than the preset maximum allowable rate of change. After these conditions are met, the predicted value of the impedance adjustment gradient is determined based on the magnitude of the projection components. The larger the projection component, the farther the parameter is from the constraint boundary, and the absolute value of the adjustment gradient can be appropriately increased; when the projection component is close to zero, the absolute value of the adjustment gradient decreases to avoid exceeding the constraint range. The direction of the impedance adjustment gradient is determined based on the deviation direction between the nodal admittance vector and the ideal state vector, and the deviation direction is obtained by calculating the vector difference.

[0055] When the projection calculation result meets the preset power compatibility conditions, the optimized value of the power allocation weight is output. The power compatibility conditions involve the projected components of power-related parameters, requiring that the power carrying capacity parameters of each submodule, the power transmission path loss parameters, etc., all meet the constraints. Specifically, this includes ensuring that the ratio of the current power to the rated power of each submodule does not exceed a set threshold, and that the projected components of the path loss parameters are positive. The optimized value of the power allocation weight is obtained by solving a multi-objective optimization problem. Objectives include minimizing the power allocation deviation of each submodule (the deviation being the difference between the actual allocated power and the ideal allocated power) and minimizing the total power loss. During the optimization process, the weights of each objective are adjusted according to the magnitude of the projected components. Parameters with larger projected components have higher objective weights to prioritize ensuring the satisfaction of those parameters.

[0056] In practice, both impedance continuity and power compatibility conditions must be met simultaneously. If either condition is not met, the parameter adjustment phase begins. The adjustment process involves fine-tuning the admittance relationship network parameters (e.g., slightly modifying the admittance weights of non-critical paths) and recalculating the space vector projection until both conditions are met. There is an upper limit to the number of adjustments (usually 5). If the conditions are still not met after exceeding this limit, the generation of the switching command is paused, an alarm signal is output, and the module is prompted to enter protection mode. After the switching control command is generated, a checksum is attached. The checksum is generated using a hash algorithm based on the command content. After receiving the command, the power semiconductor execution unit must first verify the correctness of the checksum before executing the command to prevent errors during command transmission.

[0057] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0058] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent control method for multi-mode switching of AC / DC modules, comprising the following steps: Real-time acquisition of multi-dimensional electrical parameters of AC / DC modules in hybrid operation mode, generating an original parameter set including impedance spectrum matrix and power fluctuation vector; A dynamic topology mapping model is constructed based on the impedance spectrum matrix, and the physical connection relationship of AC / DC modules is transformed into a node admittance relationship network through this model; Energy distribution state sequences are generated based on power fluctuation vectors, and the migration constraints between different operating modes are analyzed using an association rule mining algorithm. When a running mode switching command is detected, the node admittance relationship network in the dynamic topology mapping model is activated, and the current energy distribution state sequence is input into the association rule mining algorithm. Based on the real-time matching results of the node admittance relationship network and migration constraints, a switching control command containing impedance adjustment gradient and power allocation weight is generated. The switching control command is transmitted to the power semiconductor execution unit of the AC / DC module.

2. The intelligent control method for multi-mode switching of AC / DC modules according to claim 1, characterized in that: The multi-dimensional electrical parameters include input voltage ripple coefficient, output current harmonic components, module temperature distribution value, and heat dissipation efficiency factor. Set the input voltage ripple coefficient to the reference input value and the output current harmonic component to the target output value; The module temperature distribution value is calculated using an integral-differential hybrid algorithm: temperature sampling values ​​from multiple temperature measurement points inside the AC / DC module are obtained, spatial interpolation is performed on the temperature sampling values ​​to generate a three-dimensional temperature field, the standard deviation of the temperature gradient in each direction in the three-dimensional temperature field is calculated, and the ratio of the standard deviation to a preset temperature rise threshold is marked as the module temperature distribution value. The heat dissipation efficiency factor is expressed as the ratio of the power consumption of the heat dissipation device to the total power consumption of the AC / DC module.

3. The intelligent control method for multi-mode switching of AC / DC modules according to claim 1, characterized in that, The construction process of the dynamic topology mapping model includes: A higher-order tensor decomposition is performed on the impedance spectrum matrix in the original parameter set to extract the frequency domain correlation features between the real and imaginary parts of the impedance. A node admittance relationship network is established based on frequency domain correlation features. The node admittance relationship network includes the admittance values ​​of each circuit node of the AC / DC module and the admittance weights of its connection paths. The power fluctuation vector is imported into the nodal admittance relation network for load sensitivity simulation, generating migration constraints that include voltage stability range and current margin.

4. The intelligent control method for multi-mode switching of AC / DC modules according to claim 3, characterized in that, The calculation expression for the dynamic topology mapping model is: The predicted value of the output impedance adjustment gradient is obtained by convolving the admittance weight matrix in the nodal admittance relationship network with the input voltage ripple coefficient. The current margin in the migration constraint and the harmonic components of the output current are calculated using Euclidean distance to obtain the optimized value of the output power allocation weight. The predicted value of the impedance adjustment gradient and the optimized value of the power allocation weight constitute the core parameters of the switching control command.

5. The intelligent control method for multi-mode switching of AC / DC modules according to claim 1, characterized in that, The implementation process of the association rule mining algorithm includes: A spatial gridding processing unit is set up to divide the operating area of ​​the AC / DC module into three-dimensional grids of equal size; Extract the spatiotemporal distribution data of power fluctuation vectors in a three-dimensional grid to form discrete segments of the energy distribution state sequence; For each discrete segment, the confidence of the association rule is evaluated to generate migration constraints between different operating modes.

6. The intelligent control method for multi-mode switching of AC / DC modules according to claim 5, characterized in that, The specific operations for evaluating the confidence level of the association rules include: Tensor eigenvalue decomposition is performed on the energy distribution state sequence in the discrete segment to extract the energy ratio vector of the fundamental wave and harmonic waves. Calculate the autocorrelation function of the energy proportion vector within a preset time window to identify the periodic characteristics of the energy distribution state sequence; Establish a mapping relationship between periodic characteristics and AC / DC module operating modes, and determine the confidence score of migration constraints based on the matching degree of the mapping relationship.

7. The intelligent control method for multi-mode switching of AC / DC modules according to claim 6, characterized in that, The calculation process for the confidence score of the migration constraint includes: The number of times the statistical energy distribution state sequence is triggered in historical operating mode switching events; Obtain the steady-state operating duration of the AC / DC module after each trigger; The ratio of the average duration of steady-state operation to the total duration of discrete segments is labeled as the confidence score of the migration constraint.

8. The intelligent control method for multi-mode switching of AC / DC modules according to claim 1, characterized in that, The process of generating the switching control command includes: Get the trigger timestamp of the running mode switching command; Extract the node admittance relationship network parameters corresponding to the timestamp in the dynamic topology mapping model; Perform spatial vector projection calculations on the nodal admittance relationship network parameters and migration constraints; When the projection calculation result meets the preset impedance continuity condition, the predicted value of the output impedance adjustment gradient is obtained. When the projection calculation result meets the preset power compatibility conditions, the optimized value of the power allocation weight is output.

9. An intelligent control system for multi-mode switching of AC / DC modules, used to implement the intelligent control method for multi-mode switching of AC / DC modules as described in claim 1, comprising a parameter acquisition module, a model construction module, a rule analysis module, and an instruction generation module; The parameter acquisition module acquires multi-dimensional electrical parameters of the AC / DC module in real time, generates an original parameter set, and transmits it to the model building module. The model building module constructs a dynamic topology mapping model based on the impedance spectrum matrix, and forwards the power fluctuation vector to the rule analysis module. The rule analysis module generates migration constraints based on the power fluctuation vector and feeds the migration constraints back to the model building module. The instruction generation module generates switching control instructions based on the node admittance relationship network output by the model construction module and the migration constraints fed back by the rule analysis module.

10. The intelligent control system for multi-mode switching of AC / DC modules according to claim 9, characterized in that: The system also includes a feedback calibration module, which performs the following operations: Collect actual impedance response data after the power semiconductor execution unit executes the switching control command; The residual is calculated by comparing the actual impedance response data with the predicted impedance adjustment gradient values ​​in the dynamic topology mapping model. The admittance weight matrix of the nodal admittance relationship network is updated using the residual calculation results; The updated admittance weight matrix is ​​synchronized to the dynamic topology mapping model in the model building module.

Images