Adaptive power regulation method and system for power integrated circuits

Abstract

The application provides an adaptive power regulation method and system for a power integrated circuit, and relates to the technical field of power regulation. The method comprises the following steps: dividing the power integrated circuit and outputting multiple power domains; constructing multiple power feature vector sets according to power requirement data of each power domain, performing Gaussian mixture model clustering, and constructing a complementarity matrix based on multiple clustering results; establishing a pairing priority list according to the complementarity matrix, obtaining a complementary power regulator pair based on the pairing priority list; configuring an adaptive power regulator for the complementary power regulator pair, and performing adaptive power regulation on each complementary power regulator pair based on an adaptive power switching regulation logic model. The application can solve the technical problem of poor power regulation stability of the power integrated circuit in the prior art, and achieve the technical effect of improving the power regulation stability.

Description

Technical Field

[0001] This application relates to the field of power regulation technology, and in particular to an adaptive power regulation method and system for power integrated circuits. Background Technology

[0002] Currently, most existing power regulation technologies rely on fixed parameter control or adaptive strategies based on a single power model, lacking system analysis of the complementary characteristics between multiple power domains. This approach controls different loads through static power allocation tables or voltage tracking mechanisms; while simple in structure, it cannot dynamically respond to instantaneous demand changes in different power domains. Some improved schemes introduce feedback-based voltage regulation loops or dynamic frequency adjustment mechanisms, but these technologies primarily address single power paths, failing to adequately consider the temporal, spectral, and thermal coupling characteristics between power fluctuations across multiple domains.

[0003] In summary, existing technologies suffer from the technical problem that, due to the complementary modeling and dynamic adaptive control mechanism for multiple power domains, power integrated circuits cannot coordinate and schedule according to the time characteristics, spectral distribution, and thermal response differences of different power domains under complex load conditions, which further affects the overall energy utilization and operational stability of the system. Summary of the Invention

[0004] The purpose of this application is to provide an adaptive power regulation method and system for power integrated circuits, in order to solve the technical problem in the prior art that due to the complementary modeling and dynamic adaptive regulation mechanism for multiple power domains, power integrated circuits cannot coordinate and schedule according to the time characteristics, spectral distribution and thermal response differences of different power domains under complex load conditions, which further affects the overall energy utilization and operational stability of the system.

[0005] In view of the above problems, this application provides an adaptive power regulation method and system for power integrated circuits.

[0006] In a first aspect, this application provides an adaptive power regulation method for power integrated circuits, implemented through an adaptive power regulation system for power integrated circuits, comprising: dividing the power integrated circuit into multiple power domains; extracting power demand data for each power domain, and constructing multiple power feature vector sets based on the power demand data; performing Gaussian mixture model clustering on the multiple power feature vector sets to construct a complementarity matrix based on the multiple clustering results; establishing a pairing priority list according to the complementarity matrix, and obtaining complementary power regulator pairs based on the pairing priority list; configuring adaptive power regulators for the complementary power regulator pairs, wherein each adaptive power regulator includes an adaptive power switching regulation logic model constructed based on the redundancy switching logic between the complementary power regulator pairs, and performing adaptive power regulation on each complementary power regulator pair based on the adaptive power switching regulation logic model.

[0007] Preferably, the adaptive power regulation method for power integrated circuits further includes: performing Gaussian mixture model clustering on the plurality of power feature vector sets to obtain multiple clustering results; analyzing the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity between every two clustering results; calculating the comprehensive weight of the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity to obtain a comprehensive complementarity index; and constructing a complementarity matrix based on the comprehensive complementarity index, wherein each element in the complementarity matrix is ​​the comprehensive complementarity index value between the corresponding two clustering results.

[0008] Preferably, the adaptive power control method for power integrated circuits further includes: wherein each power feature vector set includes average power, peak power distribution, duty cycle, pulse amplitude distribution, thermal response sensitivity, and time-domain persistence distribution; defining initialization parameters for the Gaussian mixture model (GMM), including initialization mean, initialization covariance, initialization weights, and regularization term; performing expectation-maximization algorithm fitting on the initialized GMM according to the clustering quantity variable until a preset number of iterations is reached, and outputting the optimal solution of the clustering quantity variable; and using the optimal solution of the clustering quantity variable to perform probabilistic clustering on the multiple power feature vector sets to obtain multiple clustering results.

[0009] Preferably, the adaptive power regulation method for power integrated circuits further includes: outputting a list of paired comprehensive complementarity index values ​​according to the complementarity matrix; and outputting a pair of complementary power regulators after sorting the list of paired comprehensive complementarity index values.

[0010] Preferably, the adaptive power regulation method for power integrated circuits further includes: configuring an adaptive power regulator for the complementary power regulator pair, wherein the adaptive power regulator is a DC-DC power conversion circuit.

[0011] Preferably, the adaptive power regulation method for power integrated circuits further includes: the DC-DC power conversion circuit includes an output terminal and a redundant output terminal; wherein the output terminal is used to control the power domain of the original complementary power regulator, and the redundant output terminal is used to control the power domain of the complementary power regulator that triggers a fault.

[0012] Preferably, the adaptive power regulation method for power integrated circuits further includes: reading the sensor data pairs of each complementary power regulator for its respective power domain, including output voltage, output current and sensing temperature; performing fault detection on the sensor data pairs for its respective power domain to determine whether any sensor data in any power domain triggers a fault; if a fault is triggered, the adaptive power switching regulation logic model enters a fault transfer state to stop the operation of the complementary power regulator that triggered the fault and activate the redundant output terminal of the other complementary power regulator.

[0013] Preferably, the adaptive power regulation method for power integrated circuits further includes: constructing a multi-source state evaluation vector, inputting the multi-source state evaluation vector into a pre-trained fault classification model, wherein the fault classification model is a model based on a convolutional neural network and recurrent units learning from historical fault event samples; determining whether to trigger a fault based on the power anomaly level output by the pre-trained fault classification model, wherein a fault is triggered when the power anomaly level is greater than a preset level.

[0014] Preferably, the adaptive power regulation method for power integrated circuits further includes: reading the initial power control parameters of the complementary power regulator that triggered the fault; obtaining the real-time power control parameters of the complementary power regulator; using the initial power control parameters as the adaptive target and the real-time power control parameters as the input for redundancy regulation; obtaining the conversion power control parameters of the redundant output terminal, which are used to control the power domain of the complementary power regulator that triggered the fault.

[0015] Secondly, this application also provides an adaptive power regulation system for power integrated circuits, used to execute the adaptive power regulation method for power integrated circuits as described in the first aspect, comprising: a power domain output module for dividing the power integrated circuit and outputting multiple power domains; a power feature vector set construction module for extracting power demand data for each power domain in the multiple power domains and constructing multiple power feature vector sets based on the power demand data; a complementarity matrix construction module for performing Gaussian mixture model clustering on the multiple power feature vector sets and constructing a complementarity matrix based on the multiple clustering results; a complementary power regulator pair acquisition module for establishing a pairing priority list according to the complementarity matrix and acquiring complementary power regulator pairs based on the pairing priority list; and an adaptive power regulation module for configuring adaptive power regulators for the complementary power regulator pairs, each adaptive power regulator including an adaptive power switching regulation logic model constructed based on the redundancy switching logic between the complementary power regulator pairs, and performing adaptive power regulation on each complementary power regulator pair based on the adaptive power switching regulation logic model.

[0016] The technical solution provided in this application has at least the following technical effects or advantages: by realizing the technical goals of coordinated power allocation, adaptive redundancy control and rapid fault switching among multiple power domains, it achieves the technical effects of improving energy utilization efficiency, enhancing system reliability and realizing self-healing power control management.

[0017] The above description is merely an overview of the technical solution of this application. To enable a clearer understanding of the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the adaptive power regulation method for power integrated circuits used in this application.

[0020] Figure 2This is a schematic diagram of the adaptive power regulation system for power integrated circuits used in this application.

[0021] Figure labeling: Power domain output module 1, power feature vector set construction module 2, complementarity matrix construction module 3, complementary power regulator pair acquisition module 4, adaptive power regulation module 5. Detailed Implementation

[0022] This application provides an adaptive power regulation method and system for power integrated circuits, solving the technical problem in existing technologies where complementary modeling and dynamic adaptive regulation mechanisms for multiple power domains prevent power integrated circuits from coordinating and scheduling based on the time characteristics, spectral distribution, and thermal response differences of different power domains under complex load conditions, further affecting the overall energy utilization and operational stability of the system. The method achieves the technical goals of coordinated power allocation among multiple power domains, adaptive redundancy regulation, and rapid fault switching, thereby improving energy utilization efficiency, enhancing system reliability, and realizing self-healing power regulation management.

[0023] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0024] Example 1, please refer to the appendix. Figure 1 This application provides an adaptive power regulation method for power integrated circuits, applied to an adaptive power regulation system for power integrated circuits, specifically including the following steps:

[0025] S1: Divide the power integrated circuit to output multiple power domains.

[0026] Specifically, a power integrated circuit (IC) is a circuit structure that integrates power management, power conversion, and signal control on a single chip to achieve efficient scheduling and control of energy flow. Dividing a power IC involves dividing the complete IC into several independent but energy-coupled sub-regions, called power domains, based on its functional areas or power load characteristics. The partitioning process is based on the voltage level, load type, power consumption characteristics, or heat distribution of each module. For example, in a single chip, the processor module, memory module, and peripheral interface module can each constitute three power domains. This partitioning allows the power requirements of different modules to be independently identified and controlled, thereby improving the accuracy of energy efficiency management.

[0027] S2: Extract the power demand data of each power domain from the multiple power domains, and construct multiple sets of power feature vectors based on the power demand data.

[0028] Specifically, power demand data for each of the multiple power domains is extracted, and energy demand analysis is performed separately. This yields data on the electrical energy consumption exhibited during operation for each power domain. Power demand data refers to the real-time energy demand information of the circuit at different times and under different operating modes, including parameters such as average power, peak power, voltage fluctuation, current amplitude, and energy change rate.

[0029] The power demand data extracted from each power domain is mathematically characterized to form a set of vectors that reflect the power behavior characteristics of that power domain. A power feature vector is a data structure that exists in a multi-dimensional form, with each dimension representing a specific characteristic indicator, such as average power, duty cycle, thermal response rate, or time duration distribution. The set of feature vectors can be used to describe the power characteristic distribution of a power domain under different states.

[0030] S3: Perform Gaussian mixture model clustering on the multiple power feature vector sets to construct a complementarity matrix based on the multiple clustering results.

[0031] Furthermore, this application also includes: performing Gaussian mixture model clustering on the multiple power feature vector sets to obtain multiple clustering results; analyzing the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity between every two clustering results; calculating the comprehensive weight of the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity to obtain a comprehensive complementarity index; constructing a complementarity matrix based on the comprehensive complementarity index, wherein each element in the complementarity matrix is ​​the comprehensive complementarity index value between the corresponding two clustering results.

[0032] Furthermore, this application also includes: wherein each power feature vector set includes average power, peak power distribution, duty cycle, pulse amplitude distribution, thermal response sensitivity, and time-domain persistence distribution; defining initialization parameters for the Gaussian mixture model (GMM), including initialization mean, initialization covariance, initialization weights, and regularization term; performing expectation-maximization algorithm fitting on the initialized GMM according to the clustering quantity variable until a preset number of iterations is reached, and outputting the optimal solution of the clustering quantity variable; and using the optimal solution of the clustering quantity variable to perform probabilistic clustering on the multiple power feature vector sets to obtain multiple clustering results.

[0033] Specifically, each power feature vector set includes average power, peak power distribution, duty cycle, pulse amplitude distribution, thermal response sensitivity, and time-domain persistence distribution. Average power reflects the average power consumption level over a certain period and is a fundamental indicator for measuring energy stability. Peak power distribution describes the frequency and range at which power reaches its extreme values ​​within a short time, reflecting the transient load capacity of the circuit. Duty cycle refers to the proportion of time the power signal is in a high-power state within one cycle, used to measure the efficiency of power modulation or switching. Pulse amplitude distribution describes the probability of pulses of different amplitudes occurring in the power signal, reflecting the nonlinear characteristics of power fluctuations in the circuit. Thermal response sensitivity indicates the sensitivity of power changes to temperature rise, reflecting the chip's thermal coupling and heat dissipation capabilities. Time-domain persistence distribution describes the duration of power maintenance at a certain level, reflecting the temporal stability of load fluctuations.

[0034] The initialization parameters of the Gaussian Mixture Model (GMM) are defined, including the initial mean, initial covariance, initial weights, and regularization term. The initial mean refers to the initial position of each potential cluster centroid in the feature space, determining the initial distribution shape of the model. The initial covariance defines the degree of diffusion of each cluster across different feature dimensions, affecting the cluster shape and extent. The initial weights set the proportion of each Gaussian component in the overall distribution, thus influencing the model's initial focus on different patterns. The regularization term is a constraint parameter to prevent overfitting and maintain model stability during iteration.

[0035] The initial Gaussian Mixture Model (GMM) is fitted using the Expectation-Maximization (EM) algorithm based on the cluster size variable. This involves continuously optimizing the model parameters using the Expectation-Maximization algorithm to determine the optimal number of clusters until a preset number of iterations is reached. The optimal solution for the cluster size variable is then output, yielding the number of clusters best suited to describe the power characteristic data structure. The cluster size variable refers to the number of data clusters assumed by the GMM, which influences the accuracy and generalization ability of the clustering. The Expectation-Maximization algorithm alternates between the expectation step and the maximization step in each iteration, allowing the GMM to gradually approach the optimal distribution in a probabilistic sense. The preset number of iterations is set empirically or based on convergence conditions.

[0036] By utilizing the optimal solution of the clustering quantity variable to perform probabilistic clustering on multiple power feature vector sets and obtaining multiple clustering results, it is possible to re-perform clustering analysis on all power feature vector sets using the optimal clustering quantity, thereby obtaining clustering results that better reflect the data distribution. Probabilistic clustering assigns a probability of belonging to different categories to each sample, thus reflecting the fuzzy boundaries between power features.

[0037] Furthermore, the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity of each pair of clustering results are analyzed to represent a multidimensional analysis of the complementary characteristics between different power behavior patterns. Temporal complementarity refers to the degree to which two power patterns do not overlap or change in opposite directions on the time axis; for example, one module operates at high power while another is in a low-power state. Spectral complementarity reflects the degree of difference in the frequency domain characteristics of power signals; for example, one power domain is mainly concentrated in the low-frequency band, while the other is in the high-frequency band. Thermal complementarity indicates the synergy of different power domains in thermal response; if the heating peak of one power domain corresponds to the low-heat stage of another power domain, the complementarity is strong. Reliability complementarity measures the redundancy capability of two power domains in stable operation, i.e., whether the other domain can provide support when one domain fails. When two clustering patterns are staggered in time, complementary in spectrum, balanced in thermal distribution, and have high reliability, it indicates that the two clustering patterns have strong synergistic potential.

[0038] A comprehensive complementarity index is obtained by calculating the combined weights of temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity. This comprehensive weighting balances the overall evaluation results by allocating the importance proportions of different complementarity indices; for example, temporal complementarity accounts for 40% of the weight, thermal complementarity for 30%, and spectral complementarity and reliability each for 15%. A complementarity matrix is ​​constructed based on this comprehensive complementarity index. Each element in the complementarity matrix represents the comprehensive complementarity index value between two corresponding clustering results. That is, the complementarity matrix is ​​a two-dimensional data structure where rows and columns correspond to different clustering results, and each matrix element represents the comprehensive complementarity value between a pair of clusters. If there are five clustering results, the complementarity matrix is ​​a 5x5 square matrix, with each element value between 0 and 1; a larger value indicates stronger complementarity.

[0039] S4: Establish a pairing priority list according to the complementarity matrix, and obtain complementary power regulator pairs based on the pairing priority list.

[0040] Furthermore, this application also includes: outputting a list of paired comprehensive complementarity index values ​​according to the complementarity matrix; and outputting a pair of complementary power controllers after sorting the list of paired comprehensive complementarity index values.

[0041] Specifically, based on the complementarity matrix, the comprehensive complementarity index values ​​between each cluster result are extracted and reorganized into a list, outputting a list of paired comprehensive complementarity index values. The paired comprehensive complementarity index values ​​are a one-dimensional index set formed by extracting the comprehensive complementarity values ​​of each pair of cluster results, allowing for direct comparison of the synergistic potential of different cluster combinations.

[0042] The power regulator pairs are sorted in descending or ascending order based on their overall complementarity index values ​​to select the most complementary pairs. For example, a numerical comparison algorithm can be used to prioritize combinations with high overall complementarity. A complementary power regulator pair refers to two power domains or their corresponding control units with strong complementary characteristics, which can compensate for each other in power switching, load balancing, or thermal management. After sorting, target pairing results can be generated based on the high-complementarity combinations at the top of the list and output for use by the power management module.

[0043] S5: Configure an adaptive power regulator for the complementary power regulator pair. Each adaptive power regulator includes an adaptive power switching control logic model built based on the redundancy switching logic between the complementary power regulator pairs. Perform adaptive power control on each complementary power regulator pair based on the adaptive power switching control logic model.

[0044] Furthermore, this application also includes: configuring an adaptive power regulator for the complementary power regulator pair, wherein the adaptive power regulator is a DC-DC power conversion circuit.

[0045] Furthermore, this application also includes: the DC-DC power conversion circuit includes an output terminal and a redundant output terminal; wherein, the output terminal is used to control the power domain of the original complementary power regulator, and the redundant output terminal is used to control the power domain of the complementary power regulator that triggers a fault.

[0046] Furthermore, this application also includes: reading the sensor data pairs of each complementary power regulator for its respective power domain, including output voltage, output current and sensing temperature; performing fault detection on the sensor data pairs for its respective power domain to determine whether any sensor data in any power domain triggers a fault; if a fault is triggered, the adaptive power switching control logic model enters a fault transfer state to stop the operation of the complementary power regulator that triggered the fault and activate the redundant output terminal of the other complementary power regulator.

[0047] Furthermore, this application also includes: constructing a multi-source state evaluation vector, inputting the multi-source state evaluation vector into a pre-trained fault classification model, the fault classification model being a model based on a convolutional neural network and recurrent units learning from historical fault event samples; determining whether a fault is triggered based on the power anomaly level output by the pre-trained fault classification model, wherein a fault is triggered when the power anomaly level is greater than a preset level.

[0048] Furthermore, this application also includes: reading the initial power control parameters of the complementary power regulator that triggered the fault; obtaining the real-time power control parameters of the complementary power regulator; using the initial power control parameters as the adaptive target and the real-time power control parameters as the input for redundant regulation, to obtain the conversion power control parameters of the redundant output terminal, which are used to control the power domain of the complementary power regulator that triggered the fault.

[0049] Specifically, to improve the adaptability of complementary power regulators to dynamic load changes, an adaptive power control unit is added to each pair of complementary devices. An adaptive power regulator is a control device that can automatically adjust the output voltage or current according to changes in power domain demand. It employs a DC-DC power conversion circuit, a power electronic circuit structure that converts the input DC voltage into another form of DC voltage. The DC-DC power conversion circuit achieves efficient energy transfer and voltage regulation control by changing parameters such as duty cycle and switching frequency, maintaining a balanced and stable power output under different load conditions. Furthermore, the DC-DC power conversion circuit includes an output terminal and a redundant output terminal, indicating that the DC-DC power conversion circuit not only completes conventional energy output but also has redundancy design capabilities. The output terminal is used to control the power domain of the original complementary power regulator, meaning that during normal operation, the main output signal of the DC-DC power conversion circuit is transmitted to the original complementary device to control its corresponding power region. The redundant output is used to control the power domain of the complementary power regulator that triggered the fault in the event of a fault. This means that when a regulator stops working due to overload, short circuit or device failure, it can automatically switch to the redundant output, which will continue to provide stable energy to that power domain, preventing energy interruption or system collapse.

[0050] Furthermore, the sensor data pairs for each complementary power regulator's power domain are read, i.e., the operating status information of each power domain is acquired in real time. The sensor data pairs include output voltage, output current, and sensing temperature, which together reflect the operating status of the power domain. The output voltage represents the real-time change of potential, the output current reflects the load consumption, and the sensing temperature is used to monitor the thermal operating level of the circuit and devices.

[0051] Furthermore, constructing a multi-source state assessment vector refers to integrating data from different sensors or system modules into a multi-dimensional feature representation for comprehensive evaluation of the current operating status of the power system. For example, voltage, current, temperature, load change rate, frequency fluctuations, and equipment vibration signals collectively characterize the system's health status from multiple perspectives. The state assessment vector is a mathematical expression formed after monitoring data has been normalized, feature extracted, and combined.

[0052] The multi-source state evaluation vector is input into a pre-trained fault classification model to determine whether there is a potential fault risk. The fault classification model learns from historical fault event samples based on convolutional neural networks and recurrent units, combining spatial feature extraction and time series memory capabilities. The convolutional neural network is responsible for capturing the spatial distribution patterns of multi-dimensional features, such as power fluctuation patterns or temperature change trends, while the recurrent units are responsible for learning time dependencies, such as the dynamic evolution of data before and after the fault.

[0053] The determination of whether a fault is triggered based on the power anomaly level output by the pre-trained fault classification model means that the fault classification model generates a numerical value or category label representing the risk level of the current system state during the output phase. The power anomaly level reflects the degree to which the current power domain deviates from the normal state and is divided into multiple levels, such as level 0 representing normal, level 1 representing slight fluctuations, level 2 representing potential anomalies, and level 3 representing severe anomalies. The determination of whether a fault is triggered is made by comparing the power anomaly level with a preset level, where the preset level is a threshold determined by system safety standards or historical statistical data. When the power anomaly level exceeds the preset level, a fault is triggered, indicating that the current state has reached an unacceptable level of risk, requiring protective actions such as stopping output, initiating redundancy switching, or notifying the higher-level monitoring system.

[0054] When a sensor data triggers a fault in any power domain, the adaptive power switching control logic model enters a fault transfer state to stop the operation of the faulty complementary power regulator, preventing further damage or energy waste. Further, the initial power control parameters of the faulty complementary power regulator are read, extracting the original control parameters used by the regulator under normal operating conditions. These initial power control parameters refer to the standard operating settings recorded by the complementary power regulator during system configuration or normal operation, including output voltage setpoints, output current upper limit, switching frequency, duty cycle, and power feedback coefficient, reflecting the regulator's ideal operating mode. Then, the real-time power control parameters of the complementary power regulator are obtained, including current output power, voltage fluctuations, frequency adjustment status, and load feedback signals. Using the initial power control parameters as the adaptive target and the real-time power control parameters as input, redundant control is performed. The conversion power control parameters at the redundant output terminal, calculated by the control algorithm, are the final output settings calculated by the redundant control algorithm and will be executed on the backup channel to control the power domain of the faulty complementary power regulator, ensuring that this power domain continues to maintain normal power supply.

[0055] In summary, the adaptive power regulation method for power integrated circuits provided in this application has the following technical effects: by achieving the technical goals of coordinated power allocation among multiple power domains, adaptive redundancy regulation, and rapid fault switching, it achieves the technical effects of improving energy utilization efficiency, enhancing system reliability, and realizing self-healing power regulation management.

[0056] Example 2: Based on the same inventive concept as the adaptive power regulation method for power integrated circuits in the foregoing examples, this application also provides an adaptive power regulation system for power integrated circuits. Please refer to the appendix. Figure 2The system includes: a power domain output module 1, used to divide the power integrated circuit and output multiple power domains; a power feature vector set construction module 2, used to extract power demand data for each power domain and construct multiple power feature vector sets based on the power demand data; a complementarity matrix construction module 3, used to perform Gaussian mixture model clustering on the multiple power feature vector sets and construct a complementarity matrix based on the clustering results; a complementary power regulator pair acquisition module 4, used to establish a pairing priority list according to the complementarity matrix and acquire complementary power regulator pairs based on the pairing priority list; and an adaptive power regulation module 5, used to configure adaptive power regulators for the complementary power regulator pairs, each adaptive power regulator including an adaptive power switching regulation logic model constructed based on the redundancy switching logic between the complementary power regulator pairs, and performing adaptive power regulation on each complementary power regulator pair based on the adaptive power switching regulation logic model.

[0057] Furthermore, the adaptive power control system for power integrated circuits is also used for: performing Gaussian mixture model clustering on the multiple power feature vector sets to obtain multiple clustering results; analyzing the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity between every two clustering results; calculating the comprehensive weight of the temporal complementarity, spectral complementarity, thermal complementarity, and reliability complementarity to obtain a comprehensive complementarity index; and constructing a complementarity matrix based on the comprehensive complementarity index, where each element in the complementarity matrix is ​​the comprehensive complementarity index value between the corresponding two clustering results.

[0058] Furthermore, the adaptive power control system for power integrated circuits is also used for: wherein each power feature vector set includes average power, peak power distribution, duty cycle, pulse amplitude distribution, thermal response sensitivity, and time-domain persistence distribution; defining initialization parameters for the Gaussian mixture model (GMM), including initialization of mean, initialization of covariance, initialization of weights, and regularization term; fitting the initialized GMM to the cluster quantity variable using the expectation-maximization algorithm until a preset number of iterations is reached, and outputting the optimal solution for the cluster quantity variable; and using the optimal solution for the cluster quantity variable to perform probabilistic clustering on the multiple power feature vector sets to obtain multiple clustering results.

[0059] Furthermore, the adaptive power regulation system for power integrated circuits is also used to: output a list of paired comprehensive complementarity index values ​​according to the complementarity matrix; and output a pair of complementary power regulators after sorting the list of paired comprehensive complementarity index values.

[0060] Furthermore, the adaptive power regulation system for power integrated circuits is also used to: configure an adaptive power regulator for the complementary power regulator pair, wherein the adaptive power regulator is a DC-DC power conversion circuit.

[0061] Furthermore, the adaptive power regulation system for power integrated circuits is also used in the following ways: the DC-DC power conversion circuit includes an output terminal and a redundant output terminal; wherein the output terminal is used to control the power domain of the original complementary power regulator, and the redundant output terminal is used to control the power domain of the complementary power regulator that triggers a fault.

[0062] Furthermore, the adaptive power regulation system for power integrated circuits is also used to: read the sensor data pairs of each complementary power regulator for its respective power domain, including output voltage, output current and sensing temperature; perform fault detection on the sensor data pairs of the respective power domain to determine whether any sensor data in any power domain triggers a fault; if a fault is triggered, the adaptive power switching regulation logic model enters a fault transfer state to stop the operation of the complementary power regulator that triggered the fault and activate the redundant output terminal of the other complementary power regulator.

[0063] Furthermore, the adaptive power control system for power integrated circuits is also used for: constructing a multi-source state evaluation vector, inputting the multi-source state evaluation vector into a pre-trained fault classification model, wherein the fault classification model is a model that learns from historical fault event samples based on a convolutional neural network and recurrent units; determining whether to trigger a fault based on the power anomaly level output by the pre-trained fault classification model, wherein a fault is triggered when the power anomaly level is greater than a preset level.

[0064] Furthermore, the adaptive power regulation system for power integrated circuits is also used to: read the initial power control parameters of the complementary power regulator that triggered the fault; obtain the real-time power control parameters of the complementary power regulator; perform redundant regulation with the initial power control parameters as the adaptive target and the real-time power control parameters as the input to obtain the conversion power control parameters of the redundant output terminal, which are used to control the power domain of the complementary power regulator that triggered the fault.

[0065] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The adaptive power regulation method and specific examples for power integrated circuits in the foregoing Embodiment 1 are also applicable to the adaptive power regulation system for power integrated circuits in this embodiment. Through the foregoing detailed description of the adaptive power regulation method for power integrated circuits, those skilled in the art can clearly understand the adaptive power regulation system for power integrated circuits in this embodiment. Therefore, for the sake of brevity, it will not be described in detail here.

[0066] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0067] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for adaptive power regulation for a power integrated circuit, characterized by, The method comprises: dividing the power integrated circuit to output a plurality of power domains; extracting power demand data of each power domain in the plurality of power domains, and constructing a plurality of power feature vector sets according to the power demand data; performing Gaussian mixture model clustering on the plurality of power feature vector sets to construct a complementarity matrix based on a plurality of clustering results; establishing a pairing priority list according to the complementarity matrix, and obtaining a complementary power regulator pair based on the pairing priority list; configuring an adaptive power regulator for the complementary power regulator pair, each adaptive power regulator including an adaptive power switching control logic model constructed based on the redundancy switching logic between the complementary power regulator pair, and performing adaptive power control on each complementary power regulator pair based on the adaptive power switching control logic model; wherein each power feature vector set includes average power, peak power distribution, duty cycle, pulse amplitude distribution, thermal response sensitivity, and time domain duration distribution; defining the initialization parameters of the GMM Gaussian mixture model, including the initialization mean, the initialization covariance, the initialization weight, and the regularization term; performing expectation maximization algorithm fitting on the initialized GMM Gaussian mixture model according to the cluster number variable until a preset iteration number is reached, and outputting the optimal solution of the cluster number variable; performing probability clustering on the plurality of power feature vector sets using the optimal solution of the cluster number variable to obtain a plurality of clustering results.

2. The adaptive power regulation method for a power integrated circuit of claim 1, wherein, Constructing a complementarity matrix based on a plurality of clustering results, the method comprising: performing Gaussian mixture model clustering on the plurality of power feature vector sets to obtain a plurality of clustering results; analyzing the time complementarity, frequency spectrum complementarity, heat complementarity, and reliability complementarity between each two clustering results in the plurality of clustering results; comprehensive weight calculation is performed on the time complementarity, frequency spectrum complementarity, heat complementarity, and reliability complementarity to obtain a comprehensive complementarity index, and a complementarity matrix is constructed based on the comprehensive complementarity index, wherein each element in the complementarity matrix is a comprehensive complementarity index value between the corresponding two clustering results.

3. The adaptive power regulation method for a power integrated circuit of claim 1, wherein, Establishing a pairing priority list according to the complementarity matrix, and obtaining a complementary power regulator pair based on the pairing priority list, the method comprising: outputting a pairing comprehensive complementarity index value list according to the complementarity matrix; sorting the pairing comprehensive complementarity index value list to output a complementary power regulator pair.

4. The adaptive power regulation method for a power integrated circuit of claim 1, wherein, Configuring an adaptive power regulator for the complementary power regulator pair, the adaptive power regulator being a DC-DC power conversion circuit.

5. The adaptive power regulation method for a power integrated circuit of claim 4, wherein, The DC-DC power conversion circuit comprises an output end and a redundant output end; wherein the output end is used to control the power domain to which the original complementary power regulator belongs, and the redundant output end is used to control the power domain to which the complementary power regulator triggered by the fault belongs.

6. The adaptive power regulation method for a power integrated circuit of claim 5, wherein, Performing adaptive power control on each complementary power regulator pair based on the adaptive power switching control logic model, the method comprising: reading the sensor data pair of the power domain to which each complementary power regulator pair belongs, including output voltage, output current, and sensing temperature; The sensor data pair of the power domain is subjected to fault detection to determine whether any sensor data of the power domain triggers a fault; If a fault is triggered, the adaptive power switching control logic model enters a fault transfer state to stop the operation of the complementary power controller that triggers the fault and activate the redundant output end of another complementary power controller.

7. The adaptive power regulation method for a power integrated circuit of claim 6, wherein, The method for fault detection of the sensor data pair of the power domain comprises: A multi-source state evaluation vector is constructed, and the multi-source state evaluation vector is input into a pre-trained fault classification model, which is a model based on convolutional neural network and recurrent unit and learned from historical fault event samples; The power abnormality level output by the pre-trained fault classification model is used to determine whether a fault is triggered, wherein a fault is triggered when the power abnormality level is greater than a preset level.

8. The adaptive power governing method for a power integrated circuit of claim 6, wherein, After activating the redundant output end of another complementary power controller, the method comprises: Reading the initial power control parameters of the complementary power controller that triggers the fault; Obtaining real-time power control parameters of the complementary power controller, using the initial power control parameters as adaptive targets, and using the real-time power control parameters as inputs to perform redundant control to obtain conversion power control parameters of the redundant output end, which are used to control the power domain of the complementary power controller that triggers the fault.

9. An adaptive power regulation system for a power integrated circuit, characterized by, The steps for implementing the adaptive power control method for a power integrated circuit according to any one of claims 1 to 8 comprise: A power domain output module is configured to divide a power integrated circuit into multiple power domains and output the power domains; A power feature vector set construction module is configured to extract power demand data of each power domain in the multiple power domains and construct multiple power feature vector sets based on the power demand data; A complementarity matrix construction module is configured to perform Gaussian mixture model clustering on the multiple power feature vector sets and construct a complementarity matrix based on multiple clustering results; A complementary power controller pair acquisition module is configured to establish a pairing priority list according to the complementarity matrix and acquire a complementary power controller pair based on the pairing priority list; An adaptive power control module is configured to configure adaptive power controllers for the complementary power controller pair, each adaptive power controller comprising an adaptive power switching control logic model constructed based on the redundancy switching logic between the complementary power controller pair, and the adaptive power control module is configured to perform adaptive power control on each complementary power controller pair based on the adaptive power switching control logic model.

Images