An optimization method for key parameters of impedance characteristics of an electrically-driven filter

Abstract

The application relates to the technical field of electromagnetic compatibility (EMC) key parameter optimization, in particular to an optimization method for key parameters of impedance characteristics of an electric drive filter, which improves a genetic algorithm by adding early stop mechanism, dynamic convergence control, Gaussian variation, range clipping and mixed crossover strategies in the iteration process of the genetic algorithm, and optimizes a key parameter combination affecting the impedance characteristics of the electric drive filter by using the improved genetic algorithm, so that subjective deviation of traditional manual parameter adjustment can be avoided, a global optimal solution can be quickly located in a continuous hundred-billion-level parameter space, and repeated manual operation and calculation resource consumption in the research and development process can be effectively reduced.

Description

Technical Field

[0001] This invention relates to the field of electromagnetic compatibility (EMC) key parameter optimization technology, specifically to a method for optimizing key parameters of impedance characteristics of an electrically driven filter. Background Technology

[0002] Traditional methods for optimizing the key parameters of impedance characteristics in automotive electric drive filters employ general optimization algorithms such as gradient descent and basic genetic algorithms, and iteratively solve for the combination of key impedance parameters based on high-frequency electromagnetic field simulation models. However, in long-term use, those skilled in the art have discovered that this method has many technical shortcomings, such as:

[0003] ① Significant computational resource consumption: When performing electromagnetic compatibility (EMC) simulations, a high-frequency electromagnetic field model needs to be constructed based on key impedance characteristics. To accurately characterize the filter's impedance characteristics and meet simulation accuracy requirements, this high-frequency electromagnetic field model must employ high-resolution mesh generation. This operation directly leads to an exponential increase in the computational load of the model with the parameter dimension. Furthermore, optimization algorithms (such as genetic algorithms and gradient descent) need to iteratively optimize based on this high-frequency model, further adding to the computational load and significantly increasing the occupation and consumption of computational resources, resulting in a significantly low efficiency in the parameter optimization process.

[0004] ② Insufficient adaptability of optimization algorithms: Multiple key impedance characteristic parameters of a filter synergistically affect the actual performance of the impedance characteristics. Traditional optimization algorithms (such as gradient descent) are easily constrained by the initial parameter values ​​and fall into local optima. They can only search in a local region around the initial parameters and cannot fully cover the multi-dimensional continuous parameter space formed by the key impedance characteristic parameters. As a result, the parameters found may only fit the measured impedance characteristic values ​​in a certain frequency range, and cannot achieve an accurate match between the measured impedance characteristic values ​​and the full frequency range. Furthermore, such algorithms cannot automatically comply with the specific constraints of the key impedance characteristic parameters. It is necessary to rely on technicians with experience in electromagnetic field theory and EMC simulation to manually verify and correct the parameters (such as adjusting out-of-range inductance values ​​and optimizing combinations to improve the fitting degree), which makes it difficult to guarantee the engineering practicality of the optimization results.

[0005] ③ Poor parameter adaptability: Existing optimization algorithms can only optimize continuous parameters and cannot effectively adapt and optimize discrete parameters or non-ideal component characteristics (such as parasitic inductance and capacitance). Furthermore, simulation models often ignore nonlinear factors in actual application scenarios and fail to fully reflect real working conditions, resulting in large deviations between simulation results and measured data. This makes it impossible to provide accurate reference for parameter optimization, further leading to low efficiency and quality in the research and design of electrically driven filters, and making it difficult to support efficient research and development and intelligent upgrading in the field of electromagnetic compatibility (EMC) simulation.

[0006] ④ Low degree of process automation: The traditional method for optimizing key impedance characteristics involves two independent modules for the processing of measured data and parameter tuning. It has not built an automated collaborative link based on unified data transmission and verification between the processing and conversion of measured data and the parameter configuration of the tuning module. All core operations need to be manually completed by those skilled in the art, which makes the optimization process for key impedance characteristics cumbersome and redundant. It is also prone to errors due to manual operation, which reduces the overall optimization efficiency.

[0007] ⑤ Manual parameter adjustment relies on experience and is inefficient: Since the candidate parameters obtained by the general optimization algorithm do not fit well, those skilled in the art need to make subsequent manual adjustments based on electromagnetic field theory and past parameter adjustment experience. This process requires highly professional skills from the technicians and often requires a lot of time for repeated calibration and trial and error, which increases the cost of manual operation and extends the overall cycle of parameter optimization, seriously restricting the efficiency of project implementation.

[0008] In summary, the traditional methods for optimizing the key parameters of the impedance characteristics of automotive electric drive filters suffer from a combination of technical defects. These defects result in significant shortcomings in computational efficiency, optimization accuracy, engineering adaptability, automation level, and technology popularization. This not only greatly increases the R&D cycle and computational resource costs but also hinders the intelligent upgrade of the EMC simulation field. Summary of the Invention

[0009] The purpose of this invention is to address the shortcomings of existing technologies by providing a method for optimizing key parameters of the impedance characteristics of electrically driven filters. By incorporating strategies such as early stopping, dynamic convergence control, Gaussian mutation, range pruning, and hybrid crossover into the iterative process of the genetic algorithm, the genetic algorithm is improved. The improved genetic algorithm is then used to optimize the combination of key parameters affecting the impedance characteristics of the electrically driven filter. This method avoids the subjective bias of traditional manual parameter tuning and can quickly locate the global optimal solution in a continuous parameter space of hundreds of billions, effectively reducing repetitive manual operations and computational resource consumption during the research and development process.

[0010] The objective of this invention is achieved through the following approach:

[0011] A method for optimizing key parameters of the impedance characteristics of an electrically driven filter includes the following steps:

[0012] 1) Improve the genetic algorithm to obtain an improved genetic algorithm that can be used to optimize the key parameters of impedance characteristics of electrically driven filters;

[0013] 2) Collect the measured impedance characteristics of the electrically driven filter and calculate the complex impedance series based on the measured impedance characteristics;

[0014] 3) Based on the optimization requirements of key impedance characteristics, determine the configuration parameters of the improved genetic algorithm, including population size, crossover probability, mutation probability, maximum number of iterations, and encoding rules;

[0015] 4) Determine the value range of each key impedance characteristic parameter, construct the parameter space of the key impedance characteristic parameters, and uniformly sample within the parameter space to construct the initial population;

[0016] 5) Using an improved genetic algorithm, combined with a complex impedance sequence, the initial population is iteratively optimized to obtain the optimal combination of key impedance characteristics parameters.

[0017] This invention improves upon traditional genetic algorithms by combining complex impedance sequences calculated from measured data to iteratively optimize a population of key parameters for the impedance characteristics of electrically driven filters. This allows for rapid localization of the global optimum in the relevant parameter space, effectively avoiding the subjective biases of traditional manual parameter tuning and the shortcomings of general optimization algorithms. It also reduces repetitive manual operations and computational resource consumption during the R&D process, providing an efficient solution for optimizing key parameters of the impedance characteristics of electrically driven filters and promoting the intelligent development of the electromagnetic compatibility simulation field.

[0018] Preferably, in step 1), the improvement of the genetic algorithm includes:

[0019] 1-1) Add an early stopping mechanism to the traditional genetic algorithm. If the difference between the minimum fitness obtained in the current iteration and the minimum fitness obtained in the previous multiple iterations is ≤1e-3, the iteration is stopped. This is used to terminate redundant iterations and reduce computational power consumption.

[0020] 1-2) Dynamic convergence control is added to the traditional genetic algorithm so that the mutation probability decreases non-linearly with the number of iterations, which is used to balance the global exploration and local development capabilities of the genetic algorithm.

[0021] 1-3) Incorporate Gaussian mutation into the traditional genetic algorithm. Gaussian mutation is used to apply Gaussian perturbation to the key parameters of the impedance characteristics of individuals, thereby enhancing the local search capability of the genetic algorithm.

[0022] 1-4) Add range pruning to the traditional genetic algorithm to prune and correct parameters that exceed the preset range, so as to ensure the feasibility of individuals and avoid population degradation;

[0023] 1-5) By incorporating hybrid crossover into the traditional genetic algorithm, the parameters of the offspring are calculated by linear interpolation based on the parameters of the parent generation, thereby improving the efficiency of the offspring in inheriting the superior genes of the parent generation and the optimization potential of the offspring, resulting in an improved genetic algorithm.

[0024] This invention improves traditional genetic algorithms from three perspectives: iteration efficiency, search capability, and population quality by incorporating early stopping mechanisms, dynamic convergence control, Gaussian mutation, range pruning, and hybrid crossover strategies into the genetic algorithm. (Early stopping mechanisms and dynamic convergence control improve iteration efficiency; Gaussian mutation and hybrid crossover enhance search capability; and range pruning improves population quality.) This effectively addresses the core pain points of traditional genetic algorithms in optimizing key parameters of the impedance characteristics of electrically driven filters, such as iterative redundancy, local optimum traps, and population degradation, providing targeted technical support for efficient parameter optimization in this scenario.

[0025] Preferably, in step 2), the complex impedance series is calculated as follows:

[0026] 2-1) Collect the measured impedance characteristics of the same electric drive filter multiple times to obtain several sets of measured impedance characteristics. Each set of measured impedance characteristics includes the measured frequency, the measured impedance, and the measured impedance phase.

[0027] 2-2) Convert the measured impedance phase values ​​collected in step 2-1) from degrees to radians respectively;

[0028] 2-3) Based on the measured impedance phase value in each group of impedance characteristic measurements, convert the measured impedance value in the group of impedance characteristic measurements into the complex real part and complex imaginary part in rectangular coordinate form according to the following formula, and combine the complex real part and complex imaginary part into the form of "complex real part + j × imaginary part" as the complex impedance:

[0029] ;

[0030] ;

[0031] In the formula, r is the measured value of the impedance phase. The measured impedance phase value is in radians. `imag_part` is the real part of the complex number, and `imag_part` is the imaginary part of the complex number.

[0032] 2-4) Based on the measured frequency values, arrange and combine the complex impedances obtained in step 2-3) to form a complex impedance series in rectangular coordinate form.

[0033] This invention calculates a complex impedance series based on the measured values ​​of key impedance characteristics (i.e., measured impedance values), fundamentally eliminating the possibility of ineffective optimization based on incomplete data and accurately mapping the actual operating state of the filter. Furthermore, this complex impedance series can be directly used in the iterative algorithm optimization process (fitness calculation), avoiding the decrease in optimization efficiency or accuracy loss caused by incompatible data formats, ensuring that the improved genetic algorithm can conduct accurate iterations based on high-quality data sources.

[0034] Preferably, in step 4), the initial population is constructed according to the following formula:

[0035] 4-1) Based on the requirements of the electric drive filter and the physical characteristics of the components, determine the value range of each key impedance characteristic parameter, and construct the parameter space based on the value range of each key impedance characteristic parameter.

[0036] 4-2) Randomly and uniformly collect multiple sample points within the parameter space, the number of sample points being equal to the population size;

[0037] 4-3) According to the coding rules, the sample points collected in step 4-2) are coded as individuals, and all individuals are combined into an initial population.

[0038] Preferably, in step 5), the optimal combination of key impedance characteristics is determined in the following manner:

[0039] 5-1) Based on the value range set in step 4-1), perform a compliance test on the parameter range of each individual in the population in the following manner:

[0040] If all key impedance characteristic parameters of an individual are within the range set in step 4-1), the individual passes the qualification test, and four significant figures are retained for the key impedance characteristic parameters.

[0041] If any one of the key impedance characteristic parameters of an individual is outside the range set in step 4-1), then skip the fitness calculation process in step 5-2) and directly set the fitness of the individual to infinity.

[0042] 5-2) Calculate the fitness of each individual in the population based on the impedance calculation formula and the complex impedance series:

[0043] 5-3) Using a tournament selection algorithm, select 50% of the individuals from the population to serve as parent individuals;

[0044] 5-4) Based on the crossover probability, the selected parent individuals are randomly paired in pairs, and linear interpolation is used to generate offspring individuals;

[0045] 5-5) Based on the mutation probability, apply a Gaussian perturbation to each gene locus of each offspring individual generated in step 5-4) to obtain mutated offspring individuals and unmutated offspring individuals;

[0046] 5-6) Perform range pruning on each mutated and non-mutated offspring individual respectively;

[0047] 5-7) Use the offspring individuals obtained in step 5-6) as a new population, and increase the iteration count by 1;

[0048] 5-8) Based on the number of iterations and the early stopping mechanism, determine whether to continue iteration in the following ways:

[0049] If the number of iterations equals the maximum number of iterations, then stop iterating, calculate the fitness of each individual in the population, and select the individual with the lowest fitness as the optimal combination of key parameters for impedance characteristics.

[0050] If the number of iterations is less than the maximum number of iterations and the early stopping mechanism is satisfied, then stop the iteration, calculate the fitness of each individual in the population, and take the individual with the lowest fitness as the optimal combination of key parameters of impedance characteristics.

[0051] If the number of iterations is less than the maximum number of iterations, continue iterating, using the new population as the initial population, and repeat steps 5-1)-5-8).

[0052] This invention first verifies the value range of individuals in the population, removing invalid parameters from the source. Then, by calculating the fitness of individuals in the population, it selects parent individuals for subsequent crossover and mutation. Under the influence of crossover and mutation probabilities, the selected parent individuals generate new offspring individuals. The range of the new offspring individuals is then pruned again to ensure that the new parameters generated in the genetic operation still meet the physical characteristics of the filter components (avoiding mutations that cause parameters to go out of bounds and lead to population degradation). Finally, the fitness of the new offspring individuals determines whether to exit the iteration, stably outputting the optimal impedance characteristic key parameter combination that conforms to the physical characteristics of the electrically driven filter components.

[0053] Preferably, in step 5-2), the fitness is calculated according to the following formula:

[0054] S1) Calculate the theoretical complex impedance of an individual circuit according to the circuit impedance calculation formula;

[0055] S2) Calculate the fitness of an individual based on the theoretical complex impedance and the complex impedance series, using the following formula:

[0056]

[0057] In the formula, For theoretical complex impedance, For complex impedance series, For individual fitness.

[0058] Preferably, the attenuation formula for the nonlinear attenuation is:

[0059] ;

[0060] In the formula, This represents the current iteration number. The maximum number of iterations, This represents the probability of mutation.

[0061] Preferably, the perturbation formula for the Gaussian perturbation is:

[0062] ;

[0063] In the formula, These are the mutated parameter values. The length of the interval in which the parameter can take values. These are the original parameter values ​​before the mutation. This is a Gaussian distributed random perturbation term.

[0064] Preferably, the formula for the range trimming is:

[0065] ;

[0066] In the formula, These are the parameter values ​​after range clipping. These are the parameter values ​​after Gaussian mutation. The lower limit of the preset value range for the parameter. The upper limit of the preset value range for the parameter.

[0067] The beneficial effects of this invention include the following:

[0068] ① This invention introduces early stopping mechanism, dynamic convergence control and hybrid crossover strategy into traditional genetic algorithms. The early stopping mechanism automatically terminates the iteration when the difference between the results of multiple consecutive iterations is less than 1e-3, which can avoid the invalid computational consumption before the algorithm reaches the maximum number of iterations.

[0069] By dynamically adjusting the mutation probability, the iteration rhythm is optimized. In the early stage of iteration, the parameter space is explored with a high probability, and in the later stage of iteration, fine-tuning is performed with a low probability, thereby reducing unnecessary iteration load.

[0070] By employing efficient strategies such as hybrid cross-referencing, the quality of finding high-quality parameters in a single iteration is improved, and the global search cycle is shortened. This invention, through multi-dimensional optimization strategies, increases the automation rate of the impedance characteristic key parameter optimization process from 20% to 80% and improves the verification efficiency by 3 times. It can quickly lock high-quality parameter combinations in a continuous, multi-billion parameter space, significantly reducing the computational resource consumption caused by repeated iterations of high-frequency electromagnetic field models.

[0071] ② This invention incorporates Gaussian mutation, hybrid crossover strategy, and range pruning into the genetic algorithm. Gaussian mutation enhances local search capabilities and works in conjunction with the hybrid crossover strategy to balance global exploration and local development, avoiding getting trapped in local optima around the initial parameters. This can completely cover the multi-dimensional continuous space of key impedance characteristic parameters.

[0072] By utilizing a range pruning mechanism, it ensures that all parameter combinations generated in the iterations are always within the preset compliance range. It can adapt to discrete parameters and non-ideal component characteristics (such as parasitic inductance and parasitic capacitance) without manual intervention. This not only improves the accuracy of parameter optimization by 20%, but also effectively reduces the number of test rounds and enhances the adaptability of the algorithm to actual engineering scenarios.

[0073] ③ This invention calculates a complex impedance series based on the measured impedance characteristics of the electrically driven filter, and converts the impedance characteristic parameters into individual gene sequences through real number encoding. Then, based on the gene sequences and the complex impedance series, it directly optimizes the combination of impedance characteristic parameters. It completely eliminates the need to construct a traditional high-frequency simulation model, and can fundamentally avoid the defects caused by the simulation model ignoring nonlinear factors, which leads to significant deviations between the simulation results and the measured data and fails to match the real working conditions.

[0074] Meanwhile, this invention performs four-digit significant number formatting on the parameters of individuals in the population (unifying numerical precision), which can avoid the cumulative deviation of floating-point numbers in calculations (such as precision loss during parameter iteration). While ensuring the accuracy of impedance characteristic calculation, it balances the computational efficiency and numerical stability of the optimization process, effectively reducing the deviation between theoretical calculations and measured data (the fitting degree can be improved to over 90%). This provides a precise and reliable reference for optimizing key impedance characteristic parameters, supporting efficient research and development and intelligent advancement in the field of EMC (electromagnetic compatibility) simulation.

[0075] ④ To address the problems of cumbersome and redundant processes and frequent human error caused by the traditional separation of data processing and parameter tuning, this invention transforms the experience and judgment of R&D personnel into specific indicators of the optimization algorithm (such as early stopping threshold and mutation probability decay rule), and constructs a fully automated link from measured data to the optimal parameter combination (including automatic completion of measured data conversion, parameter combination generation and fitness calculation, etc.), reducing the number of human intervention nodes, which can not only avoid the errors introduced by human operation, but also simplify the cumbersome process and greatly improve the overall optimization efficiency.

[0076] ⑤ This invention deeply embeds cross-disciplinary expertise such as electromagnetic field theory and optimization algorithm principles into the core processing logic of the optimization algorithm, and combines it with various efficient optimization strategies (such as early stopping, dynamic convergence, Gaussian mutation, etc.) to form an integrated optimization scheme for key impedance characteristic parameters. Business personnel do not need to have in-depth knowledge of complex theories to directly call the algorithm to carry out parameter optimization work, which greatly reduces the technical threshold and job learning costs.

[0077] Definitions:

[0078] Early stopping mechanism: In the process of machine learning or model training, in order to prevent model overfitting, the training process is terminated in advance when the performance of the validation set no longer improves (or meets the preset conditions).

[0079] Dynamic convergence control is a control method that dynamically adjusts algorithm parameters (such as learning rate, crossover probability, and variable asynchronous length) based on the population evolution state or the changing trend of the objective function during the optimization algorithm iteration process.

[0080] Gaussian mutation: is a mutation operation in evolutionary algorithms. It is a strategy to generate new individuals by adding random perturbations that follow a Gaussian (normal) distribution to the gene values ​​of an individual.

[0081] Range trimming: In this invention, it refers to the method of forcibly adjusting a variable (or individual gene value) to within (or near) the boundary of the feasible region when it exceeds the preset feasible region after crossover, mutation and other operations during the optimization process.

[0082] Hybrid crossover: In this invention, it refers to an evolutionary algorithm operation that combines two or more crossover strategies (such as single-point crossover, two-point crossover, arithmetic crossover, etc.).

[0083] Tournament selection algorithm: It is a selection strategy in evolutionary algorithms. Its core is to select high-quality individuals from the population to enter the next generation or participate in crossover and mutation through "random sampling + competitive selection".

[0084] Complex Impedance Series: In this invention, the "complex impedance series" is a complex set of numbers in the form of "real part + imaginary part × j" formed by converting the impedance (amplitude) and phase in the measured data of the electrically driven filter through unit conversion (converting the impedance and phase from angle to radian) and polar coordinate to rectangular coordinate transformation. Attached Figure Description

[0085] Fig. 1 This is a flowchart of the present invention;

[0086] Fig. 2 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0087] like Figs. 1-2 As shown, a method for optimizing key parameters of the impedance characteristics of an electrically driven filter includes the following steps:

[0088] 1) Improve the genetic algorithm to obtain an improved genetic algorithm that can be used to optimize the key parameters of impedance characteristics of electrically driven filters;

[0089] 2) Collect the measured impedance characteristics of the electrically driven filter and calculate the complex impedance series based on the measured impedance characteristics;

[0090] 3) Based on the optimization requirements of key impedance characteristics, determine the configuration parameters of the improved genetic algorithm, including population size, crossover probability, mutation probability, maximum number of iterations, and encoding rules;

[0091] 4) Determine the value range of each key impedance characteristic parameter, construct the parameter space of the key impedance characteristic parameters, and uniformly sample within the parameter space to construct the initial population;

[0092] 5) Using an improved genetic algorithm, combined with a complex impedance sequence, the initial population is iteratively optimized to obtain the optimal combination of key impedance characteristics parameters.

[0093] Based on the above method, the following is an example:

[0094] 1) Improve the genetic algorithm to obtain an improved genetic algorithm that can be used to optimize the key parameters of the impedance characteristics of electrically driven filters, specifically including:

[0095] 1-1) Addressing the problem of ineffective computation caused by continued iteration after convergence in traditional genetic algorithms. This embodiment adds an early stopping mechanism to the traditional genetic algorithm. If the difference between the minimum fitness obtained in the current iteration and the minimum fitness obtained in the previous 30 iterations is ≤1e-3 (at which point the genetic algorithm has approached the global optimum, and continuing iteration cannot significantly improve the quality of the solution), the iteration stops, and the current optimal result is locked, avoiding meaningless waste of computational power.

[0096] 1-2) Traditional genetic algorithms use a fixed mutation probability, which leads to a contradiction: insufficient diversity in the early stages and premature convergence in the later stages. This embodiment adds dynamic convergence control to the traditional genetic algorithm, causing the mutation probability to decrease non-linearly with the number of iterations. Maintaining a high mutation probability in the early stages of iteration ensures population diversity, covering a wider search space (strengthening global search); gradually reducing the mutation probability in the later stages of iteration reduces disturbance to regions that are close to the optimal solution (strengthening local refinement), achieving dynamic adaptation between early exploration and later convergence. This resolves the contradiction of insufficient diversity in the early stages and premature convergence in the later stages of traditional genetic algorithms.

[0097] The attenuation formula for the nonlinear attenuation is:

[0098] ;

[0099] In the formula, This represents the current iteration number. The maximum number of iterations, The mutation probability;

[0100] 1-3) In traditional genetic algorithms, mutation causes the magnitude (step size) of gene modification to be uniformly distributed and remain unchanged throughout the evolutionary process, which is disconnected from the characteristics of the parameters, resulting in low optimization efficiency. In this embodiment, Gaussian mutation is added to the traditional genetic algorithm, applying Gaussian perturbation (mean of 0, standard deviation of 10% of the parameter range) to the impedance characteristic parameters of individuals with a 40% probability. This establishes a dynamic relationship between the variable step size and the value range of the parameter to be optimized (making the step size adapt to the numerical range and sensitivity characteristics of the parameter itself, avoiding indiscriminate uniform perturbation). This achieves accurate exploration of the neighborhood of the optimal solution and avoids the blind optimization caused by the traditional fixed step size, effectively improving the local optimization accuracy and overall optimization efficiency of the algorithm.

[0101] The perturbation formula for the Gaussian perturbation is:

[0102] ;

[0103] In the formula, These are the mutated parameter values. The length of the interval in which the parameter can take values. These are the original parameter values ​​before the mutation. This is a Gaussian distributed random disturbance term (the whole represents a normally distributed random number with a mean of 0 and a standard deviation of 0.1∆g).

[0104] 1-4) Traditional genetic algorithms are prone to parameter out-of-bounds errors (exceeding physical / business constraints) during mutation operations, leading to invalid solutions. This embodiment addresses this by incorporating range pruning into the traditional genetic algorithm. If a parameter exceeds a preset range after mutation, it is directly pruned and corrected according to boundary values ​​(e.g., if the lower limit is 'a' and the upper limit is 'b', a mutated value < 'a' is forcibly set to 'a', and a value > 'b' is forcibly set to 'b'). This filters out invalid solutions at the source, ensuring that all individuals in the population satisfy the parameter constraints and preventing invalid solutions from participating in subsequent iterations and causing the algorithm's optimization direction to deviate.

[0105] The formula for the range clipping is:

[0106] ;

[0107] In the formula, These are the valid parameter values ​​after range clipping. These are the parameter values ​​after Gaussian mutation. The lower limit (minimum valid value) of the preset value range for the parameter. The upper limit of the preset value range (maximum valid value) for the parameter;

[0108] Hybrid crossover is a crossover operator specifically designed for real number encoding, and its implementation process is as follows:

[0109] a. For each pair of parent individuals in the mating pool, independent crossover is performed based on gene loci;

[0110] b. After generating the offspring genes, immediately perform range pruning (see the third-order truncation mechanism in mutation operations).

[0111] c. The crossover probability is fixed at Pc=0.7, meaning that 70% of the individual pairs will undergo crossover.

[0112] In this embodiment, hybrid crossover is used to replace traditional single-point crossover, two-point crossover, and other crossover strategies. This solves the problem that traditional genetic algorithms are prone to losing core and effective information from the parent generation. Specifically, as follows:

[0113] 1-5) Adding a 0.1 hybrid crossover operator to the traditional genetic algorithm, that is, using the crossover logic of linear interpolation weighted fusion: based on the parent parameters, the offspring parameters are calculated by linear interpolation (e.g., offspring parameter = α × parent 1 parameter + (1-α) × parent 2 parameter, α∈[0,1], that is, the generated offspring parameters are a linear combination of the parent and offspring). This not only preserves the core characteristics of the parent parameters, but also introduces appropriate diversity through the randomness of the interpolation coefficients. Its essence is the weighted inheritance and fusion of the high-quality genes of the parent, so that the offspring can inherit high-quality genes while having new optimization possibilities, which can improve the overall optimization efficiency of the population.

[0114] It is worth noting that in this embodiment, the setting of the hybrid crossover operator α=0.1 is based on the following considerations:

[0115] When there are significant differences in parental genes (such as...) , The search range has been expanded to [

[11] , covering three orders of magnitude;

[0116] When parental genes are similar (e.g.) , The search range is expanded to [4.95, 5.15], enabling fine-grained local search;

[0117] In other words, this embodiment, by employing a hybrid crossover operator of 0.1, can expand the search range around the parent gene value to generate a continuous offspring gene, effectively balancing global exploration and local development.

[0118] This embodiment incorporates strategies such as early stopping, dynamic convergence control, Gaussian mutation, range pruning, and hybrid crossover into the traditional genetic algorithm, resulting in an improved genetic algorithm that returns the minimum fitness and the corresponding parameter combination, exhibiting high optimization efficiency.

[0119] 2) Collect multiple sets of measured impedance characteristic values ​​for the same electrically driven filter, and calculate the complex impedance series based on the measured impedance characteristic parameters, specifically including:

[0120] 2-1) Under laboratory conditions (such as an electromagnetic shielding laboratory), the impedance characteristics of the same electrically driven filter were collected multiple times using equipment such as a vector network analyzer, resulting in several sets of impedance characteristic measured values ​​(more than 400 sets). Each set of impedance characteristic measured values ​​includes the frequency measured value, the impedance measured value, and the impedance phase measured value.

[0121] 2-2) Convert the measured impedance phase values ​​from degrees to radians to meet the requirements of trigonometric function calculations:

[0122] ;

[0123] In the formula, The measured impedance phase value is in angular format. The measured impedance phase value is in radians.

[0124] 2-3) Based on the polar coordinate to rectangular coordinate conversion formula, and according to the measured impedance phase values ​​after conversion in step 2-2), convert the measured impedance values ​​into a complex impedance series in rectangular coordinate form:

[0125] Calculate the real part (real_part) and imaginary part (imag_part) of the complex number separately:

[0126] ;

[0127] ;

[0128] In the formula, r is the measured value of the impedance phase. The measured impedance phase value is in radians. `imag_part` is the real part of the complex number, and `imag_part` is the imaginary part of the complex number.

[0129] By combining the real and imaginary parts into a complex number and adding it to the measured data variables, we obtain the complex impedance:

[0130]

[0131] In the formula, `imag_part` is the real part of the complex number, and `imag_part` is the imaginary part of the complex number.

[0132] 2-4) Based on the measured frequency values, arrange and combine the complex impedances obtained in step 2-3) to form a complex impedance series in rectangular coordinate form.

[0133] 3) Based on the optimization requirements of impedance characteristic parameters, determine the configuration parameters of the improved genetic algorithm, including a population size of 3000, a crossover probability of 0.7, a mutation probability of 0.3, a maximum number of iterations of 2000, and a real-coded representation strategy as the encoding rule;

[0134] The real-number encoding rule directly maps circuit parameters to individual gene sequences, with each gene position corresponding to a physical quantity, forming a multi-dimensional real-number vector:

[0135] Each gene position directly corresponds to a circuit element parameter, avoiding conversion errors in binary encoding;

[0136] Differential properties in the real number space support gradient-guided search, enhancing local exploitation capabilities;

[0137] Genetic operations are performed directly in the real number space, ensuring the continuity of parameter changes.

[0138] 4) Determine the value range of each key impedance characteristic parameter, construct the parameter space of the key impedance characteristic parameters, and uniformly sample within the parameter space to construct the initial population. The specific method is as follows:

[0139] It is worth noting that different electrically driven filters have different structures, resulting in different key impedance characteristic parameters, parameter combinations, and parameter spaces. The impedance characteristic parameters involved in this embodiment specifically include inductors LX10, LX11, LX12, LX13, LX14, LX1, LX2, LX3, LX4, LX5, LX6, LX7, and LX8.

[0140] 4-1) Based on the requirements of the electric drive filter and the physical characteristics of the components, determine the value range of each key impedance characteristic parameter, and construct the parameter space based on the value range of each key impedance characteristic parameter.

[0141] In this embodiment, the value ranges of each impedance characteristic parameter are as follows:

[0142] The values ​​of inductors LX1, LX2, LX3, LX4, LX5, LX6, LX7, and LX8 all have the same range, which is 1×10⁻. 9 Henry (H) to 5 × 10⁻ 8 The lower limit is 1 nanohenry (nH), and the upper limit is 50 nanohenry (nH).

[0143] The inductance parameters LX9, LX10, LX11, LX12, LX13, and LX14 all have the same value range, which is 1×10⁻ 9 Henry (H) to 1×10⁻ 7The lower limit is 1 nanohenry (nH), and the upper limit is 100 nanohenry (nH).

[0144] 4-2) Randomly and uniformly collect multiple sample points within the parameter space, the number of which equals the population size. It is worth noting that the collected sample points are essentially combinations of impedance characteristic parameters. The specific form of this combination of key impedance characteristic parameters is as follows:

[0145] {

[0146] "LX9": "64.526n",

[0147] "LX10": "1.585n",

[0148] "LX11": "5.336n",

[0149] "LX12": "5.553n",

[0150] "LX13": "12.011n",

[0151] "LX14": "4.530n",

[0152] "LX1": "1.030n",

[0153] "LX2": "8.126n",

[0154] "LX6": "8.508n",

[0155] "LX5": "3.719n",

[0156] "LX3": "2.455n",

[0157] "LX4": "4.985n",

[0158] "LX7": "1.638n",

[0159] "LX8": "8.495n",

[0160] "LX15": "16.454n",

[0161] "LX16": "29.794n"}

[0162] 4-3) According to the coding rules, the sample points collected in step 4-2) are coded as individuals, and all individuals are combined into an initial population.

[0163] In this embodiment, the constructed population is actually composed of multiple individuals (i.e., a set of parameter combinations, such as {"LX9": "64.526n", "LX10": "1.585n", "LX11": "5.336n",...}), and each individual represents a candidate solution to the key parameter optimization problem.

[0164] The tournament selection algorithm is a selection strategy based on a competition mechanism, and its specific implementation steps are as follows:

[0165] Initialize the mating pool: Set the size of the mating pool to 50% of the original population (i.e., Nmate = 1500 when the original population N = 3000);

[0166] For the tournament rounds, repeat the following operations until the mating pool is full:

[0167] Step 1: Randomly select 3 individuals from the current population (tournament size = 3);

[0168] Step 2: Compare the fitness values of these three individuals. The calculation formula is: If Ei < Ej and Ei < Ek, then select individual i, where E is the fitness value, that is, the evaluation index;

[0169] Step 3: Add the winning individual to the mating pool;

[0170] In this embodiment, in the improved genetic algorithm, the tournament selection algorithm is introduced, and the specific method is as follows:

[0171] 5) Use the improved genetic algorithm, combined with the complex impedance column, to iteratively optimize the initial population to obtain the optimal impedance characteristic parameter combination, specifically as follows:

[0172] 5-1) According to the value range set in step 4-1), perform a qualification test on the parameter range of each individual in the population in the following manner:

[0173] If all the key parameters of the impedance characteristics of an individual are within the value range set in step 4-1), the individual passes the qualification test, and the key parameters of the impedance characteristics are reserved to four significant figures;

[0174] If, among the multiple key parameters of the impedance characteristics of an individual, there is any key parameter whose parameter value is not within the value range set in step 4-1), skip the fitness calculation process in step 5-2), and directly set the fitness of the individual to infinity ( ) and mark it as an invalid individual;

[0175] 5-2) According to the impedance calculation formula (i.e., the circuit model formula) and the complex impedance column, calculate the fitness of each individual in the population respectively, specifically including:

[0176] S1) Calculate the theoretical complex impedance of an individual circuit according to the circuit impedance calculation formula;

[0177] S2) Calculate the fitness of an individual based on the theoretical complex impedance and the complex impedance series, using the following formula (mean square error formula):

[0178]

[0179] In the formula, For theoretical complex impedance, For complex impedance series, For individual fitness.

[0180] In this embodiment, the mean square error formula is used to evaluate the degree of fit between the parameter combination and the measured data. The smaller the fitting error, the lower the fitness; conversely, the larger the fitness, the higher the fitness.

[0181] 5-3) Using tournament selection, 50% of the individuals in the population are selected as parent individuals;

[0182] It is worth noting that in this embodiment, a tournament selection algorithm is used to simulate the "tournament" process to select high-quality individuals (as parent individuals) to enter the mating pool and generate new offspring individuals. This effectively balances selection pressure and population diversity. The computational complexity of the entire process is low (only O(n)) and has the following advantages:

[0183] (1) Controllable selection pressure: When the tournament size is 3, the selection probability distribution is: Pselect=1−(1−1N)3. When N=3000, the probability of high-quality individuals being selected increases to about 0.001%. This can avoid the low screening efficiency caused by the low selection probability, and also prevent the rapid loss of population diversity caused by the high selection probability, thus achieving a balance between screening effectiveness and population diversity.

[0184] (2) Effectively avoid premature convergence: By randomly selecting individuals to participate in the competition, rather than directly locking the individuals with the highest fitness, the genetic differences within the population can be preserved to the greatest extent, preventing a few individuals with the best fitness from monopolizing the mating pool and causing the population to become genetically homogenized, thereby maintaining population diversity and providing a guarantee for the global search capability of subsequent iterations, and avoiding the algorithm from getting stuck in a local optimum.

[0185] 5-4) Based on the crossover probability, the selected parent individuals are randomly paired in pairs, and linear interpolation is used to generate offspring individuals;

[0186] 5-5) Based on the mutation probability, apply a Gaussian perturbation to each gene locus of each offspring individual generated in step 5-4) to obtain mutated offspring individuals and unmutated offspring individuals;

[0187] 5-6) Perform range pruning on each mutated and non-mutated offspring individual respectively;

[0188] 5-7) Use the offspring individuals obtained in step 5-6) as a new population, and increase the iteration count by 1;

[0189] 5-8) Based on the number of iterations and the early stopping mechanism, determine whether to continue iteration in the following ways:

[0190] If the number of iterations equals the maximum number of iterations, then stop iterating, calculate the fitness of each individual in the population, and select the individual with the lowest fitness as the optimal combination of impedance characteristic parameters.

[0191] It is worth noting that the impedance characteristics of an electrically driven filter are determined by multiple component parameters (inductance, capacitance, etc.). Therefore, the optimization process in this embodiment revolves around the overall adaptability of the "parameter combination", rather than optimizing a single isolated parameter to obtain the optimal impedance characteristic key parameter combination. It seeks the global optimal solution in a multi-dimensional parameter space of billions.

[0192] If the number of iterations is less than the maximum number of iterations and the early stopping mechanism is satisfied, then stop the iteration, calculate the fitness of each individual in the population, and take the individual with the lowest fitness as the optimal combination of key parameters of impedance characteristics.

[0193] If the number of iterations is less than the maximum number of iterations, continue iterating, using the new population as the initial population, and repeat steps 5-1)-5-8).

[0194] In summary, this invention addresses the pain points of traditional methods for optimizing key parameters of impedance characteristics in electrically driven filters, such as high computational resource consumption, susceptibility to local optima, low automation, and poor adaptability. It constructs a highly efficient and accurate improved genetic algorithm by innovatively incorporating five core strategies into the traditional genetic algorithm: early stopping mechanism, dynamic convergence control, Gaussian mutation, range pruning, and hybrid crossover. Using this improved genetic algorithm, iterative optimization is performed in the parameter space constructed from the key parameters of the impedance characteristics of the electrically driven filter. This enables rapid locking of the global optimum in a multi-dimensional parameter space of billions of parameters. It not only fundamentally avoids the subjective bias of traditional manual parameter tuning and the redundant calculations of high-frequency simulation models, but also increases the automation rate of optimization to 80%, improves the verification efficiency by 3 times, achieves a parameter fitting degree of over 90%, and adapts to discrete parameters and non-ideal component characteristics.

[0195] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications made to the present invention by those skilled in the art without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for optimizing key parameters of the impedance characteristics of an electrically driven filter, characterized in that, Includes the following steps: 1) Improve the genetic algorithm to obtain an improved genetic algorithm that can be used to optimize the key parameters of impedance characteristics of electrically driven filters; 2) Collect the measured impedance characteristics of the electrically driven filter and calculate the complex impedance series based on the measured impedance characteristics; 3) Based on the optimization requirements of key impedance characteristics, determine the configuration parameters of the improved genetic algorithm, including population size, crossover probability, mutation probability, maximum number of iterations, and encoding rules; 4) Determine the value range of each key impedance characteristic parameter, construct the parameter space of the key impedance characteristic parameters, and uniformly sample within the parameter space to construct the initial population; 5) Using an improved genetic algorithm, combined with a complex impedance sequence, the initial population is iteratively optimized to obtain the optimal combination of key impedance characteristics parameters.

2. The optimization method according to claim 1, characterized in that: In step 1), the ways to improve the genetic algorithm include: 1-1) Add an early stopping mechanism to the traditional genetic algorithm. If the difference between the minimum fitness obtained in the current iteration and the minimum fitness obtained in the previous multiple iterations is ≤1e-3, the iteration is stopped. This is used to terminate redundant iterations and reduce computational power consumption. 1-2) Dynamic convergence control is added to the traditional genetic algorithm so that the mutation probability decreases non-linearly with the number of iterations, which is used to balance the global exploration and local development capabilities of the genetic algorithm. 1-3) Incorporate Gaussian mutation into the traditional genetic algorithm. Gaussian mutation is used to apply Gaussian perturbation to the key parameters of the impedance characteristics of individuals, thereby enhancing the local search capability of the genetic algorithm. 1-4) Add range pruning to the traditional genetic algorithm to prune and correct parameters that exceed the preset range, so as to ensure the feasibility of individuals and avoid population degradation; 1-5) By incorporating hybrid crossover into the traditional genetic algorithm, the parameters of the offspring are calculated by linear interpolation based on the parameters of the parent generation, thereby improving the efficiency of the offspring in inheriting the superior genes of the parent generation and the optimization potential of the offspring, resulting in an improved genetic algorithm.

3. The optimization method according to claim 1, characterized in that: In step 2), the complex impedance series is calculated as follows: 2-1) Collect the measured impedance characteristics of the same electric drive filter multiple times to obtain several sets of measured impedance characteristics. Each set of measured impedance characteristics includes the measured frequency, the measured impedance, and the measured impedance phase. 2-2) Convert the measured impedance phase values ​​collected in step 2-1) from degrees to radians respectively; 2-3) Based on the measured impedance phase value in each group of impedance characteristic measurements, convert the measured impedance value in the group of impedance characteristic measurements into the complex real part and complex imaginary part in rectangular coordinate form according to the following formula, and combine the complex real part and complex imaginary part into the form of "complex real part + j × imaginary part" as the complex impedance: ； ； In the formula, r is the measured value of the impedance phase. The measured impedance phase value is in radians. `imag_part` is the real part of the complex number, and `imag_part` is the imaginary part of the complex number. 2-4) Based on the measured frequency values, arrange and combine the complex impedances obtained in step 2-3) to form a complex impedance series in rectangular coordinate form.

4. The optimization method according to claim 1, characterized in that: In step 4), the initial population is constructed according to the following formula: 4-1) Based on the requirements of the electric drive filter and the physical characteristics of the components, determine the value range of each key impedance characteristic parameter, and construct the parameter space based on the value range of each key impedance characteristic parameter. 4-2) Randomly and uniformly collect multiple sample points within the parameter space, the number of sample points being equal to the population size; 4-3) According to the coding rules, the sample points collected in step 4-2) are coded as individuals, and all individuals are combined into an initial population.

5. The optimization method according to claim 1, characterized in that: In step 5), the optimal combination of key impedance characteristics is confirmed in the following manner: 5-1) Based on the value range set in step 4-1), perform a compliance test on the parameter range of each individual in the population in the following manner: If all key impedance characteristic parameters of an individual are within the range set in step 4-1), the individual passes the qualification test, and four significant figures are retained for the key impedance characteristic parameters. If any one of the key impedance characteristic parameters of an individual is outside the range set in step 4-1), then skip the fitness calculation process in step 5-2) and directly set the fitness of the individual to infinity. 5-2) Calculate the fitness of each individual in the population based on the impedance calculation formula and the complex impedance series: 5-3) Using a tournament selection algorithm, select 50% of the individuals from the population to serve as parent individuals; 5-4) Based on the crossover probability, the selected parent individuals are randomly paired in pairs, and linear interpolation is used to generate offspring individuals; 5-5) Based on the mutation probability, apply a Gaussian perturbation to each gene locus of each offspring individual generated in step 5-4) to obtain mutated offspring individuals and unmutated offspring individuals; 5-6) Perform range pruning on each mutated and non-mutated offspring individual respectively; 5-7) Use the offspring individuals obtained in step 5-6) as a new population, and increase the iteration count by 1; 5-8) Based on the number of iterations and the early stopping mechanism, determine whether to continue iteration in the following ways: If the number of iterations equals the maximum number of iterations, then stop iterating, calculate the fitness of each individual in the population, and select the individual with the lowest fitness as the optimal combination of key parameters for impedance characteristics. If the number of iterations is less than the maximum number of iterations and the early stopping mechanism is satisfied, then stop the iteration, calculate the fitness of each individual in the population, and take the individual with the lowest fitness as the optimal combination of key parameters of impedance characteristics. If the number of iterations is less than the maximum number of iterations, continue iterating, using the new population as the initial population, and repeat steps 5-1)-5-8).

6. The optimization method according to claim 5, characterized in that: In step 5-2), the fitness is calculated according to the following formula: S1) Calculate the theoretical complex impedance of an individual circuit according to the circuit impedance calculation formula; S2) Calculate the fitness of an individual based on the theoretical complex impedance and the complex impedance series, using the following formula: ； In the formula, For theoretical complex impedance, For complex impedance series, For individual fitness.

7. The optimization method according to claim 2, characterized in that: The attenuation formula for the nonlinear attenuation is: ； In the formula, This represents the current iteration number. The maximum number of iterations, This represents the probability of mutation.

8. The optimization method according to claim 1, characterized in that: The perturbation formula for the Gaussian perturbation is: ； In the formula, These are the mutated parameter values. The length of the interval in which the parameter can take values. These are the original parameter values ​​before the mutation. This is a Gaussian distributed random perturbation term.

9. The optimization method according to claim 1, characterized in that: The formula for the range clipping is: ； In the formula, These are the parameter values ​​after range clipping. These are the parameter values ​​after Gaussian mutation. The lower limit of the preset value range for the parameter. The upper limit of the preset value range for the parameter.

Images