Multi-objective optimization method and device for power module

Abstract

The invention discloses a power module multi-objective optimization method and device, and the method comprises the steps: determining a key structure parameter and a constraint range of a to-be-optimized power module, and constructing a multi-dimensional parameter space based on the key structure parameter and the constraint range; performing global uniform sampling on the multi-dimensional parameter space based on a Sobo sequence to generate a sample point set; performing electromagnetic-thermal coupling simulation based on the sample point set to obtain a target performance index, and constructing a training data set based on the target performance index and the sample point set; constructing an agent model of the MLP, and training the agent model based on the training data set to obtain a target optimization model; and constructing a multi-target optimization model based on the target performance index, solving the multi-target optimization model based on a tabu search algorithm, a fast non-dominated sorting genetic fusion algorithm and the target optimization model, and obtaining optimal key parameters of the to-be-optimized power module.

Description

Technical Field

[0001] This invention relates to the field of multi-objective optimization technology, and in particular to a method and apparatus for multi-objective optimization of power modules. Background Technology

[0002] Currently, power modules, as core components of power electronic systems, play a crucial role in power conversion and power regulation in fields such as new energy power generation and electric vehicle drive. With the widespread adoption of wide-bandgap semiconductor devices such as silicon carbide (SiC) and gallium nitride (GaN), chip operating frequencies have broken through the MHz level, and power density requirements have increased to 5kW / cm². 3 The above places more stringent requirements on packaging technology: under high-frequency operating conditions, the parasitic inductance of the commutation circuit must be controlled within 5nH to avoid voltage overshoot, while the chip junction temperature uniformity (temperature difference <3℃) and space compactness must be optimized in a coordinated manner.

[0003] Current mainstream packaging solutions, such as double-sided heat dissipation and three-dimensional stacking, improve performance by optimizing current and heat conduction paths, but still face multi-dimensional performance conflicts. For example, while double-sided heat dissipation structures can reduce interface thermal resistance to 0.25K / W, the multi-layer soldered interfaces increase package thickness by 30%, limiting power density to 3.5kW / cm². 3 Copper clip bonding technology can compress the main circuit inductance to below 3nH, but the processing accuracy error (±10μm) can cause the non-uniformity of parallel current of multiple chips to reach 8%-12%. These contradictions highlight the need for packaging design to achieve a fine balance between electrical performance, thermal management and structural reliability.

[0004] Existing multi-objective optimization methods for power modules rely on metaheuristic algorithms such as genetic algorithms (GA) to generate optimization schemes through Pareto optimization of objectives such as parasitic inductance and junction temperature. However, their parametric modeling is often based on simplification assumptions, such as simplifying the three-dimensional converter circuit into a rectangular combination model, ignoring eddy current effects and inter-chip thermal coupling, resulting in large simulation errors of parasitic inductance; and their optimization process depends on fixed design parameters, making it difficult to meet the requirements of dynamic design. Summary of the Invention

[0005] This invention provides a method and apparatus for multi-objective optimization of power modules, thereby improving the flexibility of multi-objective optimization and thus improving the performance of power modules.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a multi-objective optimization method for power modules, comprising:

[0007] The key structural parameters and constraint ranges of the power module to be optimized are determined, and a multi-dimensional parameter space is constructed based on the key structural parameters and constraint ranges.

[0008] Global uniform sampling is performed on the multidimensional parameter space based on the Sobol sequence to generate a sample point set;

[0009] Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain the target performance index, and a training dataset is constructed based on the target performance index and the sample point set.

[0010] Construct an agent model for the MLP, and train the agent model on the training dataset to obtain the target optimized model;

[0011] A multi-objective optimization model is constructed based on the target performance index, and the multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

[0012] This invention achieves end-to-end automation from parameter space to optimal solution by unifying the modeling of key structural parameters and their constraint ranges, combined with global uniform sampling of Sobol sequences, physical coupling simulation, MLP surrogate model, and TS-NSGA-II multi-objective optimization. While ensuring comprehensive coverage of the design space, it significantly reduces the number of time-consuming simulations and substantially shortens the design cycle. Simultaneously, it can simultaneously optimize multiple performance indicators such as parasitic inductance, junction temperature, and silver solder paste pressure, improving the overall performance and reliability of the power module.

[0013] Furthermore, determining the key structural parameters and constraint ranges of the power module to be optimized, and constructing a multi-dimensional parameter space based on the key structural parameters and constraint ranges, includes:

[0014] Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters;

[0015] The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

[0016] This invention pre-screens structural parameters with significant impact based on business experience, clarifies their constraints according to manufacturing processes and performance requirements, and then constructs a multidimensional parameter space using Cartesian product. This ensures that the optimization variables are both sufficiently engineering-specific and mathematically form a multidimensional space that can be systematically explored. It effectively avoids the computational overhead caused by redundant variables, provides a reliable foundation for subsequent sampling and simulation, and improves the controllability and stability of the entire optimization process.

[0017] Furthermore, the step of performing global uniform sampling of the multidimensional parameter space based on the Sobol sequence to generate a sample point set includes:

[0018] The dimension of the Sobol sequence is determined based on the multidimensional parameter space;

[0019] A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence;

[0020] The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

[0021] This invention uses Sobol sequences to generate low-discrepancy sampling vectors based on the parameter space dimension and a preset sample size, and maps them to the design variable range. This significantly reduces sample redundancy while ensuring global uniform coverage. Compared with traditional random sampling, it can more efficiently cover the multi-dimensional design space, improve the representativeness and generalization ability of the surrogate model training data, thereby accelerating the convergence speed of subsequent optimization and improving the quality of the solution.

[0022] Furthermore, the step of performing electromagnetic-thermal coupling simulation based on the sample point set to obtain target performance indicators, and constructing a training dataset based on the target performance indicators and the sample point set, includes:

[0023] Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure.

[0024] The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data;

[0025] The original dataset is constructed by pairing the normalized data and the sample point set.

[0026] The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

[0027] This invention accurately extracts three performance indicators—parasitic inductance, junction temperature, and nano-silver solder paste pressure—by sequentially performing electromagnetic-thermal coupling three-dimensional simulations on a sample point set. These indicators are then normalized and the dataset is partitioned, eliminating dimensional differences and ensuring a scientific division between the training and test sets. This significantly improves the prediction accuracy and robustness of the MLP surrogate model while avoiding repeated, time-consuming simulations during online optimization, greatly enhancing the efficiency of the entire optimization process.

[0028] Furthermore, the construction of a multi-objective optimization model based on the target performance index, and the solution of the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized, includes:

[0029] A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range.

[0030] The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set.

[0031] The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set.

[0032] This invention constructs a multi-objective optimization function by minimizing parasitic inductance, junction temperature, and silver solder paste pressure. It integrates tabu search and fast non-dominated sorting genetic algorithm (TS-NSGA-II) to balance global exploration and local refinement, which can efficiently obtain a Pareto optimal solution set that is evenly distributed and comprehensively covered. Subsequently, based on the Pareto front, the optimal parameter combination with comprehensive performance is selected, realizing the optimal trade-off of power module under multiple conflicting indicators, taking into account the diversity of results and the convenience of decision-making in practical applications.

[0033] Secondly, the present invention provides a power module multi-objective optimization device, comprising: a parameter construction module, a sampling module, a simulation module, a model construction module, and an optimization module;

[0034] The parameter construction module is used to determine the key structural parameters and constraint range of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint range.

[0035] The sampling module is used to perform global uniform sampling on the multidimensional parameter space based on the Sobol sequence to generate a sample point set;

[0036] The simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set.

[0037] The model building module is used to build an agent model for MLP, and to train the agent model based on the training dataset to obtain the target optimized model;

[0038] The optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

[0039] Furthermore, the parameter construction module is used to determine the key structural parameters and constraint ranges of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint ranges, including:

[0040] Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters;

[0041] The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

[0042] Furthermore, the sampling module is used to perform global uniform sampling on the multidimensional parameter space based on the Sobol sequence to generate a sample point set, including:

[0043] The dimension of the Sobol sequence is determined based on the multidimensional parameter space;

[0044] A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence;

[0045] The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

[0046] Furthermore, the simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set, including:

[0047] Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure.

[0048] The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data;

[0049] The original dataset is constructed by pairing the normalized data and the sample point set.

[0050] The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

[0051] Furthermore, the optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized, including:

[0052] A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range.

[0053] The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set.

[0054] The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set. Attached Figure Description

[0055] Figure 1 A flowchart illustrating a multi-objective optimization method for a power module provided in an embodiment of the present invention;

[0056] Figure 2 A flowchart of the TS-NSGA-II algorithm provided in this embodiment of the invention;

[0057] Figure 3 This is a schematic diagram of a power module to be optimized, provided in an embodiment of the present invention.

[0058] Figure 4 This is a schematic diagram of another structure of a power module to be optimized, provided as an embodiment of the present invention;

[0059] Figure 5 This is a performance comparison chart of TSNSGA-II and NSGA-II provided for an embodiment of the present invention. Detailed Implementation

[0060] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0061] The terms "first" and "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0062] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0063] Example 1

[0064] See Figure 1 , Figure 1 This is a flowchart illustrating a multi-objective optimization method for a power module according to an embodiment of the present invention. The embodiment of the present invention provides a multi-objective optimization method for a power module, including steps 101 to 105, as detailed below:

[0065] Step 101: Determine the key structural parameters and constraint range of the power module to be optimized, and construct a multi-dimensional parameter space based on the key structural parameters and constraint range;

[0066] In this embodiment, determining the key structural parameters and constraint ranges of the power module to be optimized, and constructing a multi-dimensional parameter space based on the key structural parameters and constraint ranges, includes:

[0067] Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters;

[0068] The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

[0069] In this embodiment, based on the electromagnetic performance and thermal management requirements of the power module, key structural parameters were determined comprehensively using industry experience and finite element simulation results. These parameters include semiconductor chip spacing, lead-out terminal length, lead-out terminal width, heat sink height, and substrate thickness. Simultaneously, based on manufacturing process limits (such as minimum stampable terminal width and chip assembly tolerances) and performance requirements (such as maximum junction temperature not exceeding 125°C and parasitic inductance not exceeding 5nH), reasonable value ranges were set for each of the above parameters (e.g., chip spacing 0.5mm–2.0mm, terminal length 3mm–10mm, terminal width 1mm–5mm, heat sink height 2mm–8mm, and substrate thickness 0.5mm–1.5mm). Finally, the discrete values ​​of each key structural parameter were combined using Cartesian products to generate a multidimensional parameter space encompassing all possible process and performance constraints, laying a comprehensive and orderly design benchmark for subsequent global sampling and optimization.

[0070] In this embodiment, structural parameters with significant impact are pre-selected based on business experience, and their constraints are defined according to manufacturing processes and performance requirements. A multidimensional parameter space is then constructed using Cartesian products to ensure that the optimization variables are both sufficiently engineering-specific and mathematically form a systematically explorable multidimensional space. This effectively avoids the computational overhead caused by redundant variables, provides a reliable foundation for subsequent sampling and simulation, and improves the controllability and stability of the entire optimization process.

[0071] Step 102: Perform global uniform sampling on the multidimensional parameter space based on the Sobol sequence to generate a sample point set;

[0072] In this embodiment, the step of globally uniformly sampling the multidimensional parameter space based on the Sobol sequence to generate a sample point set includes:

[0073] The dimension of the Sobol sequence is determined based on the multidimensional parameter space;

[0074] A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence;

[0075] The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

[0076] In this embodiment, for the five-dimensional parameter space consisting of five key structural parameters—semiconductor chip spacing, lead terminal length, lead terminal width, heat sink height, and substrate thickness—and their pre-defined value ranges, the dimension d=5 of this parameter space is first used as the dimension of the Sobol sequence, and the total number of samples N (e.g., N=1000) is preset according to experience or design scale. Then, the Sobol sequence is used to generate a set containing N low-dispersion, uniformly distributed sampling vectors in a unit hypercube. Finally, according to the linear mapping rule of the minimum and maximum values ​​of each parameter, each component in the Sobol vector is transformed to the actual value range of the corresponding parameter, thereby obtaining a sample point set that covers the entire multi-dimensional parameter space with minimal redundancy, providing high-quality data support for subsequent electromagnetic-thermal coupling simulation and surrogate model training.

[0077] In this embodiment, a Sobol sequence is used to generate low-discrepancy sampling vectors based on the parameter space dimension and a preset sample size, and these vectors are mapped to the design variable range. This significantly reduces sample redundancy while ensuring global uniform coverage. Compared with traditional random sampling, this method can more efficiently cover the multi-dimensional design space, improve the representativeness and generalization ability of the surrogate model training data, thereby accelerating the subsequent optimization convergence speed and improving the quality of the solution.

[0078] Step 103: Perform electromagnetic-thermal coupling simulation based on the sample point set to obtain the target performance index, and construct a training dataset based on the target performance index and the sample point set;

[0079] In this embodiment, the step of performing electromagnetic-thermal coupling simulation based on the sample point set to obtain target performance indicators, and constructing a training dataset based on the target performance indicators and the sample point set, includes:

[0080] Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure.

[0081] The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data;

[0082] The original dataset is constructed by pairing the normalized data and the sample point set.

[0083] The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

[0084] In this embodiment, the generated N sets of parameter sample points are used as inputs. Three-dimensional field simulations are performed on each parameter combination in the electromagnetic-thermal coupling modules of Ansys Q3DExtractor and COMSOL Multiphysics to accurately extract the corresponding parasitic inductance L, junction temperature T_j, and nano-silver solder paste pressure P. Subsequently, L, T_j, and P are linearly normalized according to their minimum and maximum values ​​in all samples to eliminate dimensional differences, resulting in normalized performance vectors. Each normalized performance vector is then paired with its corresponding parameter sample point according to a one-to-one correspondence principle to construct an original dataset containing input parameters and output performance indicators. Finally, the original dataset is divided into training and testing sets according to a pre-set ratio (e.g., 80%:20%) to ensure the scientific nature of the surrogate model training and validation. Based on the training and testing sets, a final training dataset for subsequent MLP surrogate model training is generated.

[0085] In this embodiment, by sequentially performing electromagnetic-thermal coupling three-dimensional simulations on the sample point set, three performance indicators—parasitic inductance, junction temperature, and nano-silver solder paste pressure—are accurately extracted. These indicators are then normalized and the dataset is partitioned, eliminating dimensional differences and ensuring a scientific division between the training and test sets. This significantly improves the prediction accuracy and robustness of the MLP surrogate model while avoiding repeated, time-consuming simulations during online optimization, greatly enhancing the efficiency of the entire optimization process.

[0086] Step 104: Construct an agent model for the MLP, and train the agent model based on the training dataset to obtain the target optimized model;

[0087] In this embodiment, a multilayer perceptron (MLP) surrogate model with the number of input nodes equal to the dimension of the key structural parameters is first constructed. Its input layer consists of structural parameters, and its output layer consists of normalized performance metrics. The hidden layers use the ReLU activation function, and the network depth is dynamically adjusted according to the data complexity (e.g., 2 to 4 layers).

[0088] In this embodiment, the model includes several hidden layers (e.g., three layers, with 128, 64, and 32 nodes per layer, respectively) all using the ReLU activation function. The output layer has three neurons, corresponding to the predicted values ​​of the normalized parasitic inductance L, junction temperature T_j, and nanosilver solder paste pressure P, respectively. Subsequently, the surrogate model is iteratively trained using the training set. During training, mean squared error (MSE) is used as the loss function, combined with the Adam optimizer for parameter updates, and an early stopping mechanism is introduced (the validation set loss is terminated early when there is no significant decrease after 20 consecutive epochs). Finally, R on the test set is obtained. 2 A high-precision target optimization model with an accuracy of ≥0.98; this model can replace time-consuming finite element simulation in real time and accurately in subsequent multi-objective optimization processes, significantly improving the algorithm's convergence efficiency.

[0089] Step 105: Construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

[0090] In this embodiment, the step of constructing a multi-objective optimization model based on the target performance index, and solving the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized includes:

[0091] A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range.

[0092] The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set.

[0093] The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set.

[0094] Please refer to Figure 2 , Figure 2 A flowchart of a TS-NSGA-II algorithm is provided for an embodiment of the present invention.

[0095] In this embodiment, based on a high-precision MLP surrogate model, the parasitic inductance L, junction temperature T_j, and nano-silver solder paste pressure P are simultaneously used as multi-objective optimization functions to be minimized. Five key structural parameters, including semiconductor chip spacing, lead-out terminal length and width, heat sink height, and substrate thickness, are used as optimization variables, and the preset value range of each parameter is used as a constraint condition to construct a formalized multi-objective optimization model. Subsequently, the surrogate model is called to quickly predict the performance of any combination of structural parameters. Then, the TS-NSGA-II (Tabu-Search-Enhanced NSGA-II) algorithm is adopted. That is, in the standard fast non-dominated sorting genetic algorithm (NSGA-II) process, a tabu search local refinement strategy based on the neighborhood of the current solution is introduced for each generation of the population to enhance the ability to escape local optima and improve the uniformity of solution distribution.

[0096] In this embodiment, a fast non-dominated sorting genetic algorithm with fusion tabu search (TS-NSGA-II) is used to solve the multi-objective optimization model. Specifically: First, the population is initialized and the fitness value of each individual is calculated. Then, fast non-dominated sorting and crowding distance calculation are performed on the population to construct the parent population. Next, a secondary selection is performed, using a dynamic crossover strategy and a mutation operation with fusion tabu search mechanism to generate a offspring population with local refinement capabilities, which is then merged with the parent population to perform a new round of fast non-dominated sorting and crowding distance update. Subsequently, a new parent population is generated according to the elitist strategy and enters the next generation of evolution. If the current generation number Gen has not reached the maximum number of iterations maxGen, the above process is repeated; otherwise, the evolution is terminated, and the final Pareto solution set is subjected to multi-attribute decision analysis using the entropy weight method-TOPSIS to select the optimal solution as the optimal combination of structural parameters for the power module. This process integrates global search and local escape capabilities, improves the uniformity of solution distribution and optimization convergence speed, and ensures the physical feasibility and optimal performance of the obtained solution.

[0097] In this embodiment, the above-mentioned multi-objective optimization model is solved and iterated until the population converges to obtain a Pareto optimal solution set containing multiple optimal compromise solutions. Finally, based on the decision-maker's application focus or further entropy weight-TOPSIS comprehensive evaluation, a single optimal combination of structural parameters that takes into account various performance indicators is selected from the Pareto optimal solution set as the final result of this power module optimization design.

[0098] In this embodiment, a multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and silver solder paste pressure. The tabu search and fast non-dominated sorting genetic algorithm (TS-NSGA-II) are combined to balance global exploration and local refinement, which can efficiently obtain a Pareto optimal solution set that is evenly distributed and comprehensively covered. Subsequently, the parameter combination with the best overall performance is selected based on the Pareto front, realizing the optimal trade-off of the power module under multiple conflicting indicators, taking into account the diversity of results and the convenience of decision-making in practical applications.

[0099] Please refer to Figure 3 and Figure 4 , Figure 3 This is a schematic diagram of a power module to be optimized according to an embodiment of the present invention. Figure 4 This is a schematic diagram of another structure of a power module to be optimized, provided as an embodiment of the present invention.

[0100] In this embodiment, the key structural parameters of the power module are determined as follows: h1 is the height of the gap in the lower half of the terminal; h2 is the height of the upper half of the terminal; h3 is the width of the upper half of the terminal; h4 is the thickness of the copper layer on the DBC substrate; h5 is the thickness of the ceramic layer on the DBC substrate; h6 is the thickness of the lower copper layer on the DBC substrate; d1 is the distance between chip Q1 and the upper busbar terminal; d2 is the distance between chip Q1 and chip Q2; d3 is the distance between chip Q1 and the left boundary; d4 is the distance between chip Q1 and the right boundary; and d5 is the distance between the busbar terminal and the left boundary.

[0101] In this embodiment, the packaging structure of the power module to be optimized is optimized using five key structural parameters: semiconductor chip spacing, lead terminal length and width, heat sink height, and substrate thickness. The value ranges are set according to manufacturing process limits and performance requirements. First, a global uniform sampling is performed in the five-dimensional parameter space using a Sobol sequence, generating 5000 parameter sample points. These are then exported as a parameter matrix, serving as input for the electromagnetic-thermal coupling simulation of Ansys Q3D and COMSOL Multiphysics. Subsequently, three-dimensional coupling simulations are performed on each sample point combination, extracting three performance indicators: parasitic inductance L, junction temperature T, and nano-silver solder paste pressure P. These indicators are then linearly normalized to construct a dataset containing the mapping relationship between structural dimensions and normalized performance. This dataset is then divided into training and test sets at an 80%:20% ratio for subsequent model validation. Next, a multilayer perceptron (MLP) surrogate model containing three hidden layers, a ReLU activation function, and an Adam optimizer is built based on this training set. Iterative training is performed using mean squared error as the loss function until the test set R... 2A high-precision performance prediction model was obtained with an accuracy ≥0.98. Furthermore, the L, T, and P objectives were minimized, and a multi-objective optimization problem was constructed by combining parameter value constraints. This problem was solved using a fast non-dominated sorting genetic algorithm (TS-NSGA-II) that incorporates tabu search.

[0102] Please refer to Figure 5 , Figure 5 This is a performance comparison chart of TSNSGA-II and NSGA-II provided for an embodiment of the present invention. Figure 5 (a) Hypervolume convergence comparison. Figure 5 (b) Pareto Front Comparison Figure 5 (c) Generation Distance Convergence Comparison Figure 5 (d) Spacing convergence comparison. Figure 5 It can be seen that the improved TSNSGA-II can generate solution sets faster, and the solution sets approach the optimal frontier more quickly. At the same time, the smaller the TSNSGA-II spacing convergence value, the more uniform the distribution of the solution sets.

[0103] As shown in the performance comparison with the traditional NSGA-II, TS-NSGA-II can generate functional solution sets faster and more uniformly approximate the Pareto front under the same steps. Finally, the objective weights of the three objectives are calculated using the entropy weight method, and the Pareto solution sets are sorted by ideal solution approximation using the TOPSIS method. The parameter combination with the best overall performance is selected as the final optimization result. Please refer to Table 1, which is a schematic table of optimization results provided by an embodiment of the present invention.

[0104] Table 1

[0105]

[0106] This invention also provides a power module multi-objective optimization device, including: a parameter construction module, a sampling module, a simulation module, a model construction module, and an optimization module;

[0107] The parameter construction module is used to determine the key structural parameters and constraint range of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint range.

[0108] The sampling module is used to perform global uniform sampling on the multidimensional parameter space based on the Sobol sequence to generate a sample point set;

[0109] The simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set.

[0110] The model building module is used to build an agent model for MLP, and to train the agent model based on the training dataset to obtain the target optimized model;

[0111] The optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

[0112] In this embodiment, the parameter construction module is used to determine the key structural parameters and constraint ranges of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint ranges, including:

[0113] Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters;

[0114] The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

[0115] In this embodiment, the sampling module is used to perform global uniform sampling of the multidimensional parameter space based on the Sobol sequence to generate a sample point set, including:

[0116] The dimension of the Sobol sequence is determined based on the multidimensional parameter space;

[0117] A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence;

[0118] The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

[0119] In this embodiment, the simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set, including:

[0120] Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure.

[0121] The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data;

[0122] The original dataset is constructed by pairing the normalized data and the sample point set.

[0123] The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

[0124] In this embodiment, the optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized, including:

[0125] A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range.

[0126] The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set.

[0127] The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set.

[0128] In this embodiment of the invention, a terminal device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described multi-device access platform processing method.

[0129] In this embodiment of the invention, a computer-readable storage medium is also provided, which includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the above-described multi-device access platform processing method when it is running.

[0130] For example, a computer program can be divided into one or more modules, one or more of which are stored in memory and executed by a processor to perform the present invention. The one or more modules can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in a terminal device.

[0131] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor, memory, and display. Those skilled in the art will understand that the above components are merely examples of terminal devices and do not constitute a limitation on the terminal device. It may include more or fewer components, or combinations of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, etc.

[0132] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device through various interfaces and lines.

[0133] Memory can be used to store computer programs and / or modules. The processor implements various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory, and by accessing data stored in the memory. Memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, application programs required for at least one function (such as sound playback, text conversion, etc.), etc.; the data storage area can store data created based on the use of the mobile phone (such as audio data, text message data, etc.). In addition, memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0134] In this invention, modules based on multi-objective optimization of power modules, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. Computer-readable media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. Those skilled in the art can understand and implement this invention without any inventive effort.

[0135] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A multi-objective optimization method for power modules, characterized in that, include: The key structural parameters and constraint ranges of the power module to be optimized are determined, and a multi-dimensional parameter space is constructed based on the key structural parameters and constraint ranges. Global uniform sampling is performed on the multidimensional parameter space based on the Sobol sequence to generate a sample point set; Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain the target performance index, and a training dataset is constructed based on the target performance index and the sample point set. Construct an agent model for the MLP, and train the agent model on the training dataset to obtain the target optimized model; A multi-objective optimization model is constructed based on the target performance index, and the multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

2. The multi-objective optimization method for power modules as described in claim 1, characterized in that, The process of determining the key structural parameters and constraint ranges of the power module to be optimized, and constructing a multi-dimensional parameter space based on the key structural parameters and constraint ranges, includes: Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters; The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

3. The multi-objective optimization method for power modules as described in claim 2, characterized in that, The step of performing global uniform sampling of the multidimensional parameter space based on the Sobol sequence to generate a sample point set includes: The dimension of the Sobol sequence is determined based on the multidimensional parameter space; A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence; The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

4. The power module multi-objective optimization method as described in claim 3, characterized in that, The step of performing electromagnetic-thermal coupling simulation based on the sample point set to obtain target performance indicators, and constructing a training dataset based on the target performance indicators and the sample point set, includes: Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure. The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data; The original dataset is constructed by pairing the normalized data and the sample point set. The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

5. The power module multi-objective optimization method as described in claim 4, characterized in that, The process involves constructing a multi-objective optimization model based on the target performance index, and solving the multi-objective optimization model using a tabu search algorithm, a fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized, including: A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range. The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set. The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set.

6. A power module multi-objective optimization device, characterized in that, include: The module includes a parameter building module, a sampling module, a simulation module, a model building module, and an optimization module. The parameter construction module is used to determine the key structural parameters and constraint range of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint range. The sampling module is used to perform global uniform sampling on the multidimensional parameter space based on the Sobol sequence to generate a sample point set; The simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set. The model building module is used to build an agent model for MLP, and to train the agent model based on the training dataset to obtain the target optimized model; The optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the target optimization model to obtain the optimal key parameters of the power module to be optimized.

7. The power module multi-objective optimization device as described in claim 6, characterized in that, The parameter construction module is used to determine the key structural parameters and constraint ranges of the power module to be optimized, and to construct a multi-dimensional parameter space based on the key structural parameters and constraint ranges, including: Based on business experience, the key structural parameters of the power module to be optimized are determined, and the constraint range is determined based on the key structural parameters; The multidimensional parameter space is constructed based on the key structural parameters and the constraint range using a Cartesian product approach.

8. The power module multi-objective optimization device as described in claim 7, characterized in that, The sampling module is used to perform global uniform sampling of the multidimensional parameter space based on the Sobol sequence to generate a sample point set, including: The dimension of the Sobol sequence is determined based on the multidimensional parameter space; A set of sample vectors is generated based on the dimension, the preset total number of samples, and the Sobol sequence; The constraint range shall set the sampling vectors in the sampling vector set to the multi-parameter space to generate a sample point set.

9. A power module multi-objective optimization device as described in claim 8, characterized in that, The simulation module is used to perform electromagnetic-thermal coupling simulation based on the sample point set, obtain target performance indicators, and construct a training dataset based on the target performance indicators and the sample point set, including: Electromagnetic-thermal coupling simulation is performed based on the sample point set to obtain target performance indicators; the target performance indicators include parasitic inductance, junction temperature, and nano-silver solder paste pressure. The parasitic inductance, junction temperature, and nano-silver solder paste pressure were normalized to obtain normalized data; The original dataset is constructed by pairing the normalized data and the sample point set. The original dataset is divided according to a preset ratio to obtain a training set and a test set, and a training dataset is constructed based on the training set and the test set.

10. A power module multi-objective optimization device as described in claim 9, characterized in that, The optimization module is used to construct a multi-objective optimization model based on the target performance index, and solve the multi-objective optimization model based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm, and the target optimization model to obtain the optimal key parameters of the power module to be optimized, including: A multi-objective optimization function is constructed by minimizing parasitic inductance, junction temperature, and nano-silver solder paste pressure; optimization variables are constructed using the joint structure parameters, and the value range of the optimization variables is constructed using the constraint range; a multi-objective optimization model is constructed based on the multi-objective optimization function, the optimization variables, and the value range. The multi-objective optimization model is solved based on the tabu search algorithm, the fast non-dominated sorting genetic fusion algorithm and the objective optimization model to obtain the Pareto optimal solution set. The optimal key parameters of the power module to be optimized are obtained based on the Pareto optimal solution set.

Images