Optimization method of highly parallelized resistance-capacitance network

Abstract

The invention relates to the technical field of resistance-capacitance networks, in particular to an optimization method of a highly parallelized resistance-capacitance network. Comprising the following steps: numbering nodes of an original resistance-capacitance network, constructing an admittance matrix, and carrying out preliminary cluster division on the nodes; distributing the divided node clusters to a processor core for parallel reduction; the coupling coefficient of the nodes and the cluster is calculated through the inter-node admittance matrix so as to identify a high-coupling node cluster; collecting network temperature and node current in real time, calculating a node load fluctuation coefficient based on a current change rate, and determining temperature deviation; according to the coupling coefficient, the load fluctuation coefficient and the temperature deviation, the task allocation weight is dynamically adjusted, and the node cluster is allocated to the optimal processor core; calculating an error correction factor based on the task allocation weight and the dynamic change of the load and the temperature, and optimizing the reduced equivalent resistance-capacitance parameter by using the error correction factor; according to the invention, the influence of multiple physical fields and dynamic loads can be comprehensively considered.

Description

Technical Field

[0001] This invention relates to the field of resistor-capacitor network technology, and more specifically to an optimization method for highly parallelized resistor-capacitor networks. Background Technology

[0002] In the field of parallel optimization of large-scale RC networks, existing technologies often rely solely on simple grouping based on network topology when handling node cluster partitioning and task allocation. They fail to adequately consider the synergistic effects of electrical coupling characteristics, dynamic load changes, and temperature drift on RC parameters. This limitation has become a core bottleneck restricting optimization accuracy and efficiency. Specifically, existing methods only focus on physical connections when partitioning node clusters, ignoring the electrical coupling strength between nodes as reflected by the admittance matrix. This leads to highly coupled nodes being split across different processor cores, significantly increasing cross-core data interaction overhead and disrupting the locality of parallel processing. In the task allocation phase, the real-time fluctuations in node load are not considered. Fixed allocation patterns prevent nodes with drastically changing loads from receiving priority processing resources, easily causing processor core overload or idleness, resulting in low resource utilization. Furthermore, the impact of temperature drift on RC component characteristics is not included in parameter calculations, leading to significant deviations between the equivalent RC values ​​and the actual characteristics of the original network under different temperature environments, making it difficult to guarantee simulation accuracy.

[0003] The aforementioned problems make it difficult for existing technologies to achieve a balance between accuracy and efficiency when dealing with ultra-large-scale resistive-capacitive networks, failing to meet the actual needs of high-precision and high-efficiency parallel simulation. There is an urgent need for an optimization scheme that can comprehensively consider the effects of multiple physics fields and dynamic loads. Summary of the Invention

[0004] The purpose of this invention is to provide an optimization method for highly parallelized RC networks, comprising: The original RC network is numbered and the admittance matrix is ​​constructed. The nodes are then initially clustered. The partitioned node clusters are allocated to processor cores for parallel scaling. Also includes: The coupling coefficient between nodes and the cluster is calculated using the inter-node admittance matrix to identify highly coupled node clusters. Real-time acquisition of network temperature and node current; calculation of node load fluctuation coefficient based on current change rate and determination of temperature deviation. The task allocation weights are dynamically adjusted based on the coupling coefficient, load fluctuation coefficient, and temperature deviation, and the node cluster is allocated to the optimal processor core. The error correction factor is calculated based on the task allocation weight and the dynamic changes of load and temperature. The reduced equivalent resistance and capacitance parameters are then optimized using this error correction factor. The optimized parameters output by each processor core are combined to generate the final reduced network.

[0005] Preferably, the calculation of the coupling coefficient between the node and the cluster through the inter-node admittance matrix includes: An improved spectral clustering algorithm is used to decompose the admittance matrix of the original RC network into eigenvalues, extract eigenvectors, and pre-cluster the nodes using the K-means algorithm. The sum of the admittance values ​​of each node and other nodes in its cluster is calculated, and the ratio of the sum to the total admittance value of the node is used as the coupling coefficient between the node and the cluster. Nodes with significant coupling relationships are identified as highly coupled nodes and assigned to the same cluster.

[0006] Further preferably, the real-time acquisition of network temperature and node current includes: deploying distributed temperature sensors and current sampling circuits at key nodes of the RC network; setting dynamic adjustment rules for the sampling frequency, increasing the sampling frequency when the load fluctuation coefficient is at a high level, decreasing the sampling frequency when the load fluctuation coefficient is at a low level, and keeping the sampling frequency stable when the load fluctuation coefficient is at a medium level.

[0007] Further preferably, the step of dynamically adjusting the task allocation weight based on the coupling coefficient, load fluctuation coefficient, and temperature deviation includes: constructing a processor core load-efficiency model, monitoring the current load status of each core in real time, and marking cores with low load status as idle cores; prioritizing the allocation of node clusters with high task allocation weights to idle cores, and when there are no idle cores, allocating node clusters to non-idle cores with the lowest load, while ensuring that the core is within a reasonable load range after receiving new tasks.

[0008] More preferably, the task allocation weight is calculated using the following formula: ; in, For nodes With cluster The coupling coefficient, For real-time temperature, The reference temperature is 25℃. Temperature sensitivity coefficient, unit: °C -1 , For nodes The load fluctuation coefficient, This is the coupling-load balancing factor. For clusters The total number of nodes included.

[0009] More preferably, the error correction factor is calculated using the following formula: ; in, The time derivative of the load fluctuation coefficient. Temperature sensitivity of the coupling coefficient To dynamically adjust the weights, and , This is the basic error term.

[0010] More preferably, the equivalent resistance-capacitance parameters are calculated using the following formula: ; in, The set of nodes to be reduced. For nodes The original resistance value, For nodes The original capacitance value, The temperature coefficient of the capacitor. For nodes With nodes Task allocation weights For nodes With nodes Error correction factor.

[0011] Further preferably, the optimized parameters for merging the outputs of each processor core include: constructing parameter fusion rules; when the deviation of the same equivalent resistance parameter output by different cores is small, the arithmetic mean is taken as the final result; when the deviation is large, a secondary verification mechanism is initiated to recalculate the task allocation weight and error correction factor corresponding to the parameter, and the final equivalent resistance parameter is determined based on the new calculation result; for the equivalent capacitance parameter, the same fusion rules are used to process it to ensure that the parameters of the merged RC network remain consistent.

[0012] More preferably, the calculation of node load fluctuation coefficient based on current change rate includes: recording the current value of the node in multiple consecutive sampling periods, calculating the ratio of the current change in adjacent periods to the time interval, and obtaining multiple current change rates; and taking the ratio of the maximum value of the current change rate to the maximum operating current allowed by the network design as the node load fluctuation coefficient.

[0013] More preferably, the dynamic adjustment of task allocation weights further includes: adjusting the coupling-load balancing factor every preset period. Perform a calibration; when the average coupling coefficient of the node cluster is at a high level, increase the [adjustment / adjustment]. When the average load fluctuation coefficient is at a high level, reduce it. ; The adjustments are always kept within a reasonable range to ensure that the coupling coefficient and load fluctuation coefficient maintain a balanced weight ratio in task allocation.

[0014] Compared with the prior art, the present invention has the following advantages: This invention identifies highly coupled node clusters by calculating the coupling coefficient between nodes and the cluster, solving the problem of unreasonable cluster partitioning caused by neglecting electrical coupling in existing technologies. By combining real-time temperature and current dynamic adjustment of task allocation weights and calculating error correction factors, it addresses the issues of low resource utilization and insufficient simulation accuracy under load fluctuations and temperature drift. Its innovation lies in the synergistic consideration of multiphysics and dynamic loads, achieving a balance between accuracy and efficiency in the parallel optimization of large-scale resistive-capacitive networks. Attached Figure Description

[0015] To more clearly illustrate the embodiments of the present invention or the technical solutions in 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, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0016] Figure 1 This is a flowchart of an optimization method for a highly parallelized RC network according to the present invention. Figure 2 This is a flowchart illustrating the calculation of the coupling coefficient between a node and the cluster using the inter-node admittance matrix, as described in this invention. Figure 3 This is a flowchart illustrating the real-time acquisition of network temperature and node current in this invention. Figure 4 This is a flowchart illustrating the process of dynamically adjusting task allocation weights based on coupling coefficient, load fluctuation coefficient, and temperature deviation, as described in this invention. Figure 5 This is a flowchart illustrating the optimized parameters for merging the outputs of various processor cores in this invention. Figure 6 This is a flowchart of the calculation of node load fluctuation coefficient based on current change rate in this invention. Detailed Implementation

[0017] 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.

[0018] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Traditional highly parallelized RC network optimization methods suffer from problems such as node cluster partitioning not considering the quantization of coupling relationships, task allocation not taking into account the dynamic load and temperature effects, and equivalent parameter reduction ignoring multi-physics interference. These problems lead to large errors when scaling up large-scale networks, uneven utilization of processor core resources, and difficulty in adapting to temperature fluctuations and load changes, thus affecting simulation accuracy and efficiency.

[0020] Based on this, please refer to Figure 1 This embodiment provides an optimization method for highly parallelized RC networks, including: S1: Number the nodes of the original RC network and construct the admittance matrix, and perform preliminary clustering of the nodes; S2: Distribute the partitioned node cluster to processor cores for parallel reduction; S3: Calculate the coupling coefficient between nodes and the cluster using the inter-node admittance matrix to identify highly coupled node clusters; S4: Real-time acquisition of network temperature and node current, calculation of node load fluctuation coefficient based on current change rate and determination of temperature deviation; S5: Dynamically adjusts task allocation weights based on coupling coefficient, load fluctuation coefficient, and temperature deviation, allocating node clusters to the optimal processor cores; S6: Calculate the error correction factor based on the task allocation weight and the dynamic changes of load and temperature, and use the error correction factor to optimize the reduced equivalent resistance and capacitance parameters. S7: Combines the optimized parameters output by each processor core to generate the final reduced network.

[0021] This solution addresses the issues of insufficient accuracy and resource waste caused by traditional methods that rely solely on topology for parallel optimization by incorporating considerations of node coupling coefficients, dynamic load fluctuations, and temperature deviations. It achieves both accuracy and efficiency in reducing RC networks under multi-physics and dynamic load scenarios.

[0022] It is worth mentioning that the node coupling coefficient is quantified by the admittance matrix to reflect the electrical correlation strength between nodes; the load fluctuation coefficient is calculated based on the current change rate to capture the dynamic impact of the load in real time; the temperature deviation combines the reference temperature and the real-time temperature to correct the impact of temperature drift on the resistance and capacitance parameters. The three are optimized in a closed loop through task allocation weights and error correction factors to ensure that the task allocation and parameter reduction of each processor core are always adapted to the real-time network status.

[0023] The technical effects achieved by the above scheme include: accurate identification of highly coupled node clusters reduces cross-cluster signal interference and improves the locality of parallel reduction; real-time acquisition of dynamic load and temperature provides a real-time basis for task allocation, avoiding resource idleness or overload in fixed allocation mode; the introduction of error correction factors directly compensates for parameter deviations caused by multi-physics interference, making the equivalent RC parameters closer to the original network characteristics. The synergistic effect of these steps achieves a balance between accuracy and efficiency in large-scale RC networks, providing a more reliable reduction model for subsequent simulations.

[0024] Traditional methods for partitioning node clusters in RC networks often rely on simple grouping based on topological connections without quantifying the electrical coupling strength between nodes. This results in highly coupled nodes being split into different clusters, increasing the overhead of cross-cluster data interaction. Meanwhile, low-coupling nodes are forcibly grouped into the same cluster, reducing the independence of parallel processing and affecting overall reduction efficiency and accuracy.

[0025] Based on this, please refer to Figure 2 The calculation of the coupling coefficient between a node and the cluster using the inter-node admittance matrix includes: S31: An improved spectral clustering algorithm is used to perform eigenvalue decomposition on the admittance matrix of the original RC network; S32: Extract feature vectors and perform pre-clustering of nodes using the K-means algorithm; S33: Calculate the sum of the admittance values ​​of each node and other nodes in its cluster, and use the ratio of the sum to the total admittance value of the node as the coupling coefficient between the node and the cluster. S34: Nodes with significant coupling relationships are identified as highly coupled nodes and assigned to the same cluster.

[0026] This scheme quantifies the coupling strength between nodes using the admittance matrix, solving the problem of fragmented coupling relationships caused by traditional cluster partitioning relying solely on topology, and achieving accurate aggregation of highly coupled nodes. Notably, the improved spectral clustering algorithm captures global coupling features in the admittance matrix through eigenvalue decomposition, while the K-means algorithm performs preliminary grouping based on eigenvectors, ensuring the rationality of the pre-clustering results. The coupling coefficient, calculated as the ratio of the sum of admittance values ​​to the total admittance value, directly reflects the electrical dependence of a node on the cluster. When this coefficient reaches a significant level, the node is identified as a highly coupled node, ensuring its collaborative processing with other nodes within the cluster.

[0027] The technical effects achieved by the above embodiments include: the aggregation of highly coupled nodes reduces cross-cluster communication in parallel processing and reduces data interaction latency; the reasonable splitting of low-coupled nodes improves the independence of each processor core task and avoids resource competition; the quantification of coupling coefficient provides an electrical characteristic basis for subsequent task allocation, making the cluster partition not only conform to the topology but also adapt to the electrical behavior of the network, fundamentally improving the coordination and accuracy of parallel reduction.

[0028] In traditional RC network parameter acquisition methods, the sampling frequency is mostly a fixed value and cannot be dynamically adjusted according to the load status. When the load fluctuates drastically, a fixed low frequency will cause the loss of key current change information, affecting the accuracy of load fluctuation coefficient calculation. When the load is stable, a fixed high frequency will cause data redundancy, increase storage and processing overhead. At the same time, temperature acquisition is not linked with the load status, making it difficult to reflect the synergistic effect of the two.

[0029] Based on this, please refer to Figure 3 The real-time acquisition of network temperature and node current includes: S411: Deploy distributed temperature sensors and current sampling circuits at key nodes of the RC network; S412: Set the dynamic adjustment rules for the sampling frequency, and increase the sampling frequency when the load fluctuation coefficient is at a high level; S413: Reduce the sampling frequency when the load fluctuation coefficient is at a low level, and keep the sampling frequency stable when the load fluctuation coefficient is at a medium level.

[0030] This solution addresses the issues of information loss and resource waste caused by fixed sampling frequencies in scenarios with dynamically changing loads, achieving a balance between acquisition accuracy and efficiency. Notably, distributed temperature sensors and current sampling circuits are deployed at key nodes to ensure comprehensive monitoring of the network's core status. The sampling frequency adjustment directly responds to changes in the load fluctuation coefficient; when load fluctuations are severe, the frequency is increased to capture instantaneous changes; when the load is stable, the frequency is decreased to reduce redundant data; and during moderate fluctuations, the frequency remains stable to balance both requirements.

[0031] The technical effects achieved by the above embodiments include: high-frequency sampling during periods of severe load fluctuation ensures the accuracy of current change rate calculation and provides reliable data support for load fluctuation coefficient; low-frequency sampling during periods of stable load reduces the amount of data and lowers the acquisition and processing burden on the processor; and synchronous acquisition of temperature and current provides a basis for subsequent collaborative analysis of temperature deviation and load fluctuation, making the correlation of multi-physics parameters closer and improving the ability to capture network dynamic characteristics.

[0032] In traditional RC network parallel task allocation, polling or random allocation methods are often used, without considering the real-time load status and efficiency characteristics of the processor cores. This results in high-load cores continuously receiving tasks and becoming overloaded, while low-load cores are idle, leading to uneven resource utilization. At the same time, task allocation is not associated with the priority of the node cluster, and important clusters may be allocated to inefficient cores, affecting the overall processing speed.

[0033] Based on this, please refer to Figure 4 The dynamic adjustment of task allocation weights based on coupling coefficient, load fluctuation coefficient, and temperature deviation includes: S51: Build a processor core load-efficiency model, monitor the current load status of each core in real time, and mark cores with low load status as idle cores; S52: Prioritize assigning node clusters with high task allocation weights to idle cores. When there are no idle cores, assign node clusters to non-idle cores with the lowest load, and ensure that the core is within a reasonable load range after receiving new tasks.

[0034] This solution addresses the resource waste and load imbalance issues inherent in traditional allocation methods by employing a load-efficiency model and a priority allocation mechanism, thereby improving processor core utilization efficiency. Notably, the processor core load-efficiency model quantifies processing efficiency under different loads by monitoring core operating status in real time, providing a basis for task allocation decisions. Idle cores are marked based on low-load conditions, ensuring high-priority tasks receive sufficient resources. The allocation of non-idle cores is based on minimizing load and limiting the load ceiling after receiving new tasks, preventing processing delays caused by overload.

[0035] The technical effects achieved by the above embodiments include: idle cores prioritize processing high-weight tasks, reducing the processing waiting time of critical clusters and improving overall parallel efficiency; the selection of the least idle core and the control of the load limit balance the workload of each core and avoid local overload; the introduction of the load-efficiency model makes task allocation more in line with hardware characteristics, ensuring that each cluster can be processed on the core with the best efficiency, thereby improving the speed and stability of resistor-capacitor network reduction from the hardware level.

[0036] In traditional RC network task allocation weight calculation, only the number of nodes or topological distance are considered, without taking into account the influence of node coupling strength, dynamic load changes and temperature drift. This leads to a disconnect between weight allocation and the actual network state. Highly coupled node clusters may be assigned to different cores, nodes with drastic dynamic loads may not receive sufficient processing resources, and the weights of temperature-sensitive nodes may not be adjusted with temperature, affecting the coordination and accuracy of parallel processing.

[0037] Based on this, the task allocation weight is calculated using the following formula: ; in, For nodes With cluster The coupling coefficient, For real-time temperature, The reference temperature is 25℃. Temperature sensitivity coefficient, unit: °C -1 , For nodes The load fluctuation coefficient, This is the coupling-load balancing factor. For clusters The total number of nodes included.

[0038] This formula is used to quantize nodes. In the cluster The parallel processing priority in the formula is the core quantitative basis for achieving dynamic task allocation. The formula as a whole adopts a normalized structure, with the numerator as a node. The weighted score, with the denominator being the cluster. The weighted sum of scores for all nodes within the node ensures that the weight values ​​are within a reasonable range and that the sum can be used for priority sorting.

[0039] In the molecule, As a coupling-load balancing factor, its value is dynamically adjusted within a range, such as from 0.3 to 0.7, to adapt to changes in network conditions: when the nodes within the cluster are tightly coupled, Increase the value to enhance the effect of the coupling coefficient; when the load fluctuates drastically. Adjust it to highlight the role of the load fluctuation coefficient.

[0040] , for nodes With cluster The coupling coefficient, calculated through the admittance matrix, directly reflects the electrical connection strength between the node and the cluster. The larger the value, the more significant the electrical dependence of the node on the cluster, and the more likely it is to be assigned to the same processor core to reduce cross-core communication.

[0041] It is a temperature correction item, in which The temperature sensitivity coefficient is determined by the material properties of the resistive and capacitive components. and The difference (25℃) reflects the degree of temperature drift. The physical meaning of this exponential term is: when the temperature deviates from the reference value, the actual influence of the coupling coefficient will decrease or increase with the degree of temperature sensitivity, for example, the temperature-sensitive element... The larger the value, the more significant the correction effect of this item on the coupling coefficient when the temperature changes, ensuring that the weight calculation adapts to the impact of temperature changes on electrical characteristics.

[0042] middle, This is the load fluctuation coefficient, calculated based on the rate of change of node current, reflecting the degree of dynamic impact on the node load. When the node current changes rapidly, The value increases, through The weight allocation allows nodes with high load fluctuations to receive higher priority in task allocation, preventing processor cores from being overloaded due to concentrated load.

[0043] The summation operation in the denominator normalizes the weights, making... The value range is uniform, which facilitates horizontal comparison of priorities between different clusters and nodes, ensuring that the processor core can select high-priority node clusters according to a unified standard when allocating tasks, thereby improving resource utilization.

[0044] This formula overcomes the limitations of traditional weight calculations that rely on a single factor by using a synergistic model of coupling coefficient, load fluctuation coefficient, and temperature deviation, thus achieving precise matching between task allocation and the dynamic characteristics of the network's multi-physics field.

[0045] It is worth mentioning that the coupling coefficient Reflects the electrical connection between nodes and the cluster, ensuring that highly coupled nodes receive similar weights; temperature term Correct the impact of temperature drift on weights, so that the weights of temperature-sensitive nodes are dynamically adjusted with temperature; load fluctuation coefficient. This reflects the real-time load status of the nodes, with nodes under heavy load receiving higher priority; balancing factor Coordinate the weighting of coupling and load to adapt to different network conditions.

[0046] The technical effects achieved by the above embodiments include: the introduction of coupling coefficients ensures that highly coupled node clusters are allocated to nearby cores, reducing cross-core communication; the temperature correction term adapts weights to temperature changes, improving the processing accuracy of temperature-sensitive networks; the consideration of load fluctuation coefficients ensures that dynamically loaded nodes are processed in a timely manner, avoiding resource allocation lag; and the balance factor... The dynamic adjustment enables weight calculation to adapt to different network scenarios, thereby improving the rationality of task allocation and the collaborative efficiency of parallel processing.

[0047] Traditional error correction methods in RC network reduction often employ fixed coefficients or single factors, such as considering only temperature or load for correction. They fail to correlate task allocation weights, the dynamic changes in load over time, and the impact of temperature on the coupling coefficient, resulting in correction factors that cannot adapt to the real-time dynamic characteristics of the network. When the load fluctuates drastically or the temperature changes rapidly, fixed correction methods struggle to compensate for instantaneous errors, while ignoring the impact of task allocation weights leads to insufficient error correction for high-priority nodes, ultimately resulting in a significant deviation between the equivalent parameters and the original network characteristics.

[0048] Based on this, the error correction factor is calculated using the following formula: ; in, The time derivative of the load fluctuation coefficient. Temperature sensitivity of the coupling coefficient To dynamically adjust the weights, and , This is the basic error term.

[0049] This formula is used to quantify the error compensation amount during the reduction process of the RC network, and is a key correction term to ensure the accuracy of the equivalent parameters.

[0050] Its core logic is to incorporate task allocation weights, dynamic load changes, and the impact of temperature on coupling relationships into a unified correction framework to achieve closed-loop error control under multi-physics disturbances. Assign weights to tasks as the base coefficients for correction factors to ensure that high-priority nodes, such as highly coupled and high-load-fluctuation nodes, receive more accurate error compensation and avoid parameter deviations caused by insufficient correction of critical nodes.

[0051] This design reflects the principle that higher priority leads to more refined adjustments, which is consistent with the resource allocation logic of parallel processing.

[0052] It is a load dynamic correction item, in which This is the time derivative of the load fluctuation coefficient, which directly reflects the rate of load change. For example, this value increases significantly during a pulse current impact.

[0053] As a dynamically adjusted weight, it is related to negative correlation of squares The physical meaning is that load fluctuations on high-priority nodes have a more significant impact on errors and require intervention. Dynamic adjustments enhance the correction strength, for example when When it approaches 1, When the value is close to 0, the weight of dynamic load correction is reduced to avoid stability issues caused by excessive correction of high-priority nodes.

[0054] It is a temperature coupling correction term. The temperature sensitivity of the coupling coefficient reflects the rate of change of the coupling relationship with temperature, and is related to the temperature deviation. The product of ) quantifies the degree of influence of temperature change on the coupling relationship. Weight allocation and This complementarity ensures that the effect of temperature on the coupling relationship can be fully corrected when the load fluctuation is gradual. For example, in low-temperature environments, the temperature coefficient of resistance of some materials increases sharply, and this item can specifically compensate for the coupling error caused by this.

[0055] The basic error term is obtained by training based on historical simulation data. It is used to compensate for the inherent characteristics of resistors and capacitors, such as systematic errors caused by manufacturing tolerances, and to ensure that the correction factor can maintain basic accuracy even without significant load fluctuations and temperature changes.

[0056] This scheme addresses the limitations of traditional, static error correction methods by employing multi-factor collaborative modeling, achieving precise matching between error correction and the network's dynamic state. Notably, it also incorporates task weight allocation. As a base coefficient for correction factors, it ensures that high-priority nodes receive more accurate corrections; the time derivative of load fluctuations. The rate of load change is captured to compensate for errors caused by transient shocks; the temperature sensitivity of the coupling coefficient. The product of temperature deviation and temperature variation quantifies the impact of temperature change on the coupling relationship, correcting for the resulting error; the weights are dynamically adjusted. Follow Dynamic adjustments are made to adapt the load and temperature correction ratios to node priorities.

[0057] The technical effects achieved by the above embodiments include: when the load fluctuates drastically, the time derivative term enhances the correction strength, compensating for errors caused by instantaneous changes; in temperature-sensitive scenarios, the temperature sensitivity term ensures that changes in coupling relationships are accurately corrected; the introduction of task allocation weights allows for focused compensation of errors in high-priority nodes, avoiding insufficient accuracy in critical nodes; and the basic error term... This provides a stable baseline correction for the system, reducing the impact of random errors. Overall, this correction factor comprehensively covers the sources of error in the equivalent parameters, improving the accuracy and reliability of the RC network reduction.

[0058] In traditional calculations of equivalent parameters for RC networks, the original resistance and capacitance values ​​are often simply superimposed or allocated according to a fixed ratio. The task allocation weights and error correction factors are not incorporated into the calculation, and the temperature-dependent characteristics of capacitance are ignored, leading to inaccurate calculations of the equivalent resistance (…). ) and capacitor ( It cannot accurately reflect the overall characteristics of the reduced set of nodes. When there is complex coupling or temperature fluctuation between nodes, the fixed calculation mode will amplify the error, causing the reduced network to deviate significantly from the original network in terms of DC and AC characteristics.

[0059] Based on this, the equivalent resistance and capacitance parameters are calculated using the following formula: ; in, The set of nodes to be reduced. For nodes The original resistance value, For nodes The original capacitance value, The temperature coefficient of the capacitor. For nodes With nodes Task allocation weights For nodes With nodes Error correction factor.

[0060] This set of formulas is used to calculate the set of nodes that have been reduced. equivalent resistance and capacitor The core innovation is to incorporate task allocation weights, error correction factors, and component physical characteristics, such as the temperature coefficient of capacitors, into the calculation to ensure that the equivalent parameters are adapted to both DC and AC characteristics.

[0061] for The calculation, This represents the total original resistance of the nodes that were reduced. For nodes With nodes The task allocation weight is used to allocate the contribution ratio of each node resistance in the equivalent value, reflecting the principle that the resistance of high-weight nodes contributes more.

[0062] It is a resistance error correction term, when When positive, this term reduces the contribution of the corresponding resistance, compensating for resistance calculation errors caused by coupling or load fluctuations; when... When it is negative, this term increases the contribution of resistance and corrects the underestimation error.

[0063] This design ensures that the equivalent resistance reflects the overall electrical characteristics of the resistance in the original network, especially when there is strong coupling between nodes, thus avoiding resistance value distortion caused by simple superposition.

[0064] for The core difference between the calculation method and the resistance formula lies in the introduction of... Item, of which The temperature coefficient of a capacitor reflects the rate at which the capacitance changes with temperature.

[0065] The physical meaning of this additional item is that capacitors are generally more sensitive to temperature than resistors, and the effect of temperature changes on capacitors needs to be compensated separately. For example, ceramic capacitors have a large temperature coefficient, and their capacitance value may decrease significantly with increasing temperature. This item can be used to convert temperature deviations into a correction amount for the equivalent capacitance value, ensuring... It can still reflect the capacitance characteristics of the original node set over a wide temperature range.

[0066] Both sets of formulas aggregate the characteristics of the reduced nodes through summation operations, and the correction logic is deeply related to the task allocation weight and error correction factor, forming a closed loop of allocation, correction to equivalence. This enables the reduced RC network to accurately match the electrical behavior of the original network in both static (DC) and dynamic (AC) scenarios, providing a reliable model for subsequent simulations.

[0067] This scheme solves the problem of neglecting dynamic factors and component characteristics in traditional calculation methods by combining the synergistic effect of weights and correction factors and introducing the temperature characteristics of capacitors, thus achieving high-precision matching between equivalent parameters and the original network.

[0068] It is worth mentioning that in the resistance equivalent calculation, task weighting is performed. The contribution ratio of the resistance at each node is determined by the error correction factor. pass A compensation bias is introduced to ensure the accuracy of resistance superposition; in the equivalent capacitance calculation, in addition to weight allocation, an additional factor is introduced. This feature is specifically designed to compensate for the effects of temperature changes on capacitors, enabling the equivalent capacitance value to adapt to temperature fluctuation scenarios.

[0069] The technical effects achieved by the above embodiments include: the resistance equivalent calculation accurately aggregates the resistance characteristics of the reduced nodes through the synergy of weights and correction factors, avoiding resistance deviations caused by differences in node coupling; the capacitance equivalent calculation considers the temperature coefficient, so that the capacitance value can still reflect the capacitance characteristics of the original node when the temperature changes, solving the limitation of the capacitance value being fixed in traditional calculations; the overall equivalent parameters adapt to both DC and AC characteristics, ensuring that the reduced network can remain consistent with the original network under different operating conditions, providing a reliable model foundation for subsequent simulations.

[0070] Traditional parallel scaling down of RC networks often involves merging parameters using simple arithmetic averaging or directly selecting the result from a single core. This approach fails to consider the degree of deviation in output parameters across different processor cores. When the deviation is significant, forcibly averaging leads to a decrease in the accuracy of the final parameters. Furthermore, the lack of a secondary verification mechanism makes it impossible to trace and correct the cause of the deviation, affecting the consistency and reliability of the merged network. Especially in large-scale networks, differences in data interaction between cores can cause parameter deviations, which traditional merging methods struggle to eliminate.

[0071] Based on this, please refer to Figure 5 The optimized parameters for merging the outputs of each processor core include: S71: Construct parameter fusion rules. When the deviation of the same equivalent resistance parameter output by different cores is small, take the arithmetic mean as the final result. S72: When the deviation is large, the secondary verification mechanism is activated to recalculate the task allocation weight and error correction factor corresponding to the parameter, and the final equivalent resistance parameter is determined based on the new calculation results. S73: For equivalent capacitance parameters, the same fusion rules are used to ensure that the parameters of the merged RC network remain consistent.

[0072] This solution addresses the problems of traditional merging methods being insensitive to deviations and lacking correction capabilities through differentiated fusion and secondary verification, thereby improving the accuracy and consistency of the merged parameters.

[0073] It is worth mentioning that the parameter fusion rules dynamically adjust the strategy according to the magnitude of the deviation. When the deviation is small, the arithmetic mean balances efficiency and accuracy, avoiding over-computation. When the deviation is large, the secondary verification mechanism recalculates the task allocation weights and error correction factors to trace the source of the deviation, such as the calculation deviation caused by load fluctuations or temperature changes. Based on the corrected weights and factors, the final parameters are determined to ensure that the deviation is effectively eliminated. The capacitor parameters adopt the same rules to make the merging logic of resistors and capacitors consistent, avoiding network characteristic imbalance caused by different processing methods.

[0074] The technical effects achieved by the above embodiments include: rapid merging when the deviation is small improves the overall processing efficiency and avoids unnecessary computational overhead; secondary verification when the deviation is large ensures the accuracy of parameters and eliminates errors caused by data differences between cores; consistent processing logic for resistor and capacitor parameters ensures that the merged RC network maintains synergy in electrical characteristics and avoids simulation deviations caused by parameter mismatch; the overall merging process ensures the consistency of the final network parameters, providing a reliable data foundation for subsequent timing analysis, power consumption simulation, etc.

[0075] Traditional calculations of load fluctuation coefficients in RC networks often rely on the current change rate over a single sampling period, neglecting the dynamic trends across multiple consecutive periods. This results in coefficients that fail to accurately reflect the continuous fluctuation characteristics of the load. Particularly in scenarios involving instantaneous load surges or periodic fluctuations, single-period data may capture outliers, distorting the coefficients and affecting the accuracy of subsequent task weight allocation. Furthermore, the lack of a clear correlation between the current change rate and the network's maximum allowable current leads to a lack of a unified benchmark for the coefficients, making it difficult to compare the load states of different nodes horizontally.

[0076] Based on this, please refer to Figure 6 The calculation of the node load fluctuation coefficient based on the rate of change of current includes: S421: Record the current value of the node in multiple consecutive sampling periods, calculate the ratio of the current change in adjacent periods to the time interval, and obtain multiple current change rates. S422: The ratio of the maximum value of the current change rate to the maximum operating current allowed by the network design is used as the node load fluctuation coefficient.

[0077] This scheme overcomes the limitations of traditional single-cycle calculation by using multi-cycle sampling and maximum value selection, and achieves accurate quantification of the load fluctuation coefficient.

[0078] It is worth mentioning that the current recording of multiple consecutive sampling periods captures the dynamic changes of the load over a period of time, avoiding the random errors of a single period; the ratio of the current change in adjacent periods to the time interval, i.e., the current change rate, directly reflects the rate of load change, and the calculation of multiple change rates ensures comprehensive coverage of the fluctuation trend; selecting the maximum value as a representative highlights the most severe load fluctuation, which meets the priority consideration of high-load nodes in task allocation; the ratio to the maximum allowable current of the network provides a normalization benchmark for the coefficient, making the load fluctuation status of different nodes directly comparable. For example, a coefficient of 0.3 for node A and a coefficient of 0.5 for node B clearly shows that the load fluctuation of B is more severe.

[0079] The technical effects achieved by the above embodiments include: multi-cycle sampling ensures comprehensive capture of load fluctuation trends and avoids the one-sidedness of single-cycle data; the calculation of multiple current change rates and the selection of the maximum value accurately reflect the most severe load fluctuation state, providing a key basis for task allocation; normalization processing makes the load fluctuation coefficients of different nodes comparable, facilitating the processor core to prioritize based on the coefficient size; the overall calculation process enables the load fluctuation coefficient to truly reflect the load state of the node, improving the accuracy of subsequent task allocation weights and ensuring that high-fluctuation nodes receive sufficient processing resources.

[0080] In traditional RC network task allocation, the coupling-load balancing factor α is often a fixed value, which cannot be dynamically adjusted according to the real-time coupling status and load fluctuation level of the node cluster. As a result, in high coupling scenarios, α is not increased, resulting in insufficient weight of the coupling coefficient. Cluster partitioning and task allocation rely excessively on load fluctuation and ignore the electrical connection between nodes. In high load fluctuation scenarios, α is not decreased, resulting in insufficient weight of the load fluctuation coefficient. Task allocation cannot respond to load changes in a timely manner. Both of these will cause the task allocation weight to become disconnected from the actual network state, affecting the parallel processing efficiency.

[0081] Based on this, the dynamic adjustment of task allocation weights also includes: calibrating the coupling-load balancing factor α once every preset period. When the average coupling coefficient of the node cluster is at a high level, α is increased; when the average load fluctuation coefficient is at a high level, α is decreased. The adjustment of α is always kept within a reasonable range to ensure that the coupling coefficient and the load fluctuation coefficient maintain a balanced weight ratio in task allocation.

[0082] This scheme solves the problem that a fixed α cannot adapt to changes in network state by periodically and dynamically calibrating α, and achieves an adaptive balance between coupling and load weight ratio.

[0083] It is worth mentioning that the preset periodic calibration mechanism enables α to respond periodically to changes in network status, avoiding adjustment lag; when the average coupling coefficient is high, α is increased to enhance the proportion of the coupling coefficient in the weight calculation, ensuring that highly coupled node clusters are prioritized for aggregation; when the average load fluctuation coefficient is high, α is decreased to increase the weight proportion of the load fluctuation coefficient, making task allocation more sensitive to load changes; the reasonable range restriction prevents α from being overly biased towards a certain factor, ensuring that the weight proportion of coupling and load is always in a balanced state, neither over-relying on coupling and ignoring load, nor over-focusing on load and severing coupling.

[0084] The technical effects achieved by the above embodiments include: increasing α in highly coupled scenarios makes task allocation more closely match the electrical connections between nodes and reduces cross-cluster communication; decreasing α in high load fluctuation scenarios enables task allocation to quickly adapt to load changes and avoid resource allocation lag; periodic calibration ensures real-time matching of α with network status, improving the dynamic adaptability of task allocation weights; and limiting the reasonable range ensures that the weight ratio of coupling and load is always balanced, taking into account both the electrical characteristics of the network and dynamic load requirements, thereby improving the overall collaborative efficiency and accuracy of parallel processing.

[0085] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art can make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0086] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. An optimization method for highly parallelized RC networks, comprising: The original RC network is numbered and the admittance matrix is ​​constructed. The nodes are then initially clustered. The partitioned node clusters are allocated to processor cores for parallel scaling. Its characteristic is that it further includes: The coupling coefficient between nodes and the cluster is calculated using the inter-node admittance matrix to identify highly coupled node clusters. Real-time acquisition of network temperature and node current; calculation of node load fluctuation coefficient based on current change rate and determination of temperature deviation. The task allocation weights are dynamically adjusted based on the coupling coefficient, load fluctuation coefficient, and temperature deviation, and the node cluster is allocated to the optimal processor core. The error correction factor is calculated based on the task allocation weight and the dynamic changes of load and temperature. The reduced equivalent resistance and capacitance parameters are then optimized using this error correction factor. The optimized parameters output by each processor core are combined to generate the final reduced network.

2. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The calculation of the coupling coefficient between a node and the cluster using the inter-node admittance matrix includes: An improved spectral clustering algorithm is used to decompose the admittance matrix of the original RC network into eigenvalues, extract eigenvectors, and pre-cluster the nodes using the K-means algorithm. The sum of the admittance values ​​of each node and other nodes in its cluster is calculated, and the ratio of the sum to the total admittance value of the node is used as the coupling coefficient between the node and the cluster. Nodes with significant coupling relationships are identified as highly coupled nodes and assigned to the same cluster.

3. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The real-time acquisition of network temperature and node current includes: deploying distributed temperature sensors and current sampling circuits at key nodes of the RC network; setting dynamic adjustment rules for the sampling frequency, increasing the sampling frequency when the load fluctuation coefficient is at a high level, decreasing the sampling frequency when the load fluctuation coefficient is at a low level, and keeping the sampling frequency stable when the load fluctuation coefficient is at a medium level.

4. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The dynamic adjustment of task allocation weights based on coupling coefficient, load fluctuation coefficient, and temperature deviation includes: constructing a processor core load-efficiency model, monitoring the current load status of each core in real time, and marking cores with low load status as idle cores; prioritizing the allocation of node clusters with high task allocation weights to idle cores, and when there are no idle cores, allocating node clusters to non-idle cores with the lowest load, while ensuring that the core is within a reasonable load range after receiving new tasks.

5. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The task allocation weights are calculated using the following formula: ； in, For nodes With cluster The coupling coefficient, For real-time temperature, The reference temperature is 25℃. Temperature sensitivity coefficient, unit: °C -1 , For nodes The load fluctuation coefficient, This is the coupling-load balancing factor. For cluster The total number of nodes included.

6. The optimization method for a highly parallelized RC network according to claim 5, characterized in that, The error correction factor is calculated using the following formula: ； in, The time derivative of the load fluctuation coefficient. Temperature sensitivity of the coupling coefficient To dynamically adjust the weights, and , This is the basic error term.

7. The optimization method for a highly parallelized RC network according to claim 6, characterized in that, The equivalent resistance and capacitance parameters are calculated using the following formula: ； in, The set of nodes to be reduced. For nodes The original resistance value, For nodes The original capacitance value, The temperature coefficient of the capacitor. For nodes With nodes Task allocation weights For nodes With nodes Error correction factor.

8. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The optimized parameters for merging the outputs of each processor core include: constructing parameter fusion rules; when the deviation of the same equivalent resistance parameter output by different cores is small, the arithmetic mean is taken as the final result; when the deviation is large, a secondary verification mechanism is initiated to recalculate the task allocation weight and error correction factor corresponding to the parameter, and the final equivalent resistance parameter is determined based on the new calculation result; for the equivalent capacitance parameter, the same fusion rules are used to process it to ensure that the parameters of the merged RC network remain consistent.

9. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The method for calculating the node load fluctuation coefficient based on the current change rate includes: recording the current value of the node in multiple consecutive sampling periods, calculating the ratio of the current change in adjacent periods to the time interval, and obtaining multiple current change rates; and using the ratio of the maximum value of the current change rate to the maximum operating current allowed by the network design as the node load fluctuation coefficient.

10. The optimization method for a highly parallelized RC network according to claim 1, characterized in that, The dynamic adjustment of task allocation weights also includes: adjusting the coupling-load balancing factor every preset period. Perform a calibration; when the average coupling coefficient of the node cluster is at a high level, increase the [adjustment / adjustment]. When the average load fluctuation coefficient is at a high level, reduce it. ; The adjustments are always kept within a reasonable range to ensure that the coupling coefficient and load fluctuation coefficient maintain a balanced weight ratio in task allocation.

Images