A recall hard constraint double threshold screening and sorting method for permanent magnet synchronous motor parameter design

Abstract

The application provides a recall hard constraint double threshold screening and sorting method for permanent magnet synchronous motor parameter design, and belongs to the technical field of motor design and optimization. The method comprises the following steps: obtaining a sample set based on parameterized finite element modeling; generating a feasibility label according to motor engineering constraints; training a classifier, and adopting boundary sensitive weight and probability calibration; selecting a first threshold value by using a criterion on the premise of a recall hard constraint on a verification set; generating a second threshold value according to adaptive quantiles on the left side to form a high confidence straight-through area and a candidate pool; for the samples in the straight-through area or the candidate pool, a local preference pair is constructed based on normalized multi-objectives by using a second-stage pairwise sorting model, and then training is conducted, ranking scores are output, and a Top-k priority simulation list is given. The application has controllable missed judgment risk, adjustable candidate set scale and improved optimal hit rate, and is suitable for motor parameter design, batch screening parameters and rapid optimization parameters.

Description

Technical Field

[0001] This invention relates to a recall hard constraint dual threshold screening and sorting method for parameter design of permanent magnet synchronous motors, belonging to the field of motor design and optimization technology. Background Technology

[0002] Permanent magnet synchronous motors are widely used in electric vehicles, power tools, and home appliance compressors. The electromagnetic performance of these motors is influenced by a combination of multiple variables, including rotor and stator dimensions, slot-pole fit, magnet dimensions and arrangement, air gap, lamination thickness, and winding parameters. Performance indicators typically include average electromagnetic torque, torque ripple / coiling torque, total harmonic distortion (THD), efficiency / copper and iron losses, temperature rise, and manufacturability constraints. In engineering, parametric modeling combined with finite element method (e.g., Maxwell) simulation is often used to evaluate design points. However, single-point simulation is time-consuming, batch scanning is costly, and the proportion of feasible solutions under multiple constraints is often very low.

[0003] The typical process is: sampling within the design space based on the experimental design → electromagnetic simulation → establishing a surrogate model based on the data → using an optimizer to search for the optimal structure in the surrogate or simulation loop. Existing motor optimization processes lack a constraint judgment module, making it impossible to determine whether a sample is 'qualified / feasible'. The absence of this process presents the following common challenges:

[0004] The feasible region is sparse and the class is extremely imbalanced: Under multiple constraints and parallel decision-making, the proportion of samples that meet all indicators is often significantly less than 10%, or even lower. Directly training the classifier is prone to bias towards the "infeasible" class, resulting in low recall, missing high-quality design points, and limiting the space for subsequent optimization.

[0005] Sampling and computational resources cannot be allocated according to value: Existing technologies generally retain or discard samples in a "one-size-fits-all" manner under the same threshold, lacking the concept of a "candidate pool" and resource grading strategy. They cannot ensure high recall while guiding the limited simulation / experiment slots to be consumed preferentially on more valuable design points through the ranking of suboptimal samples.

[0006] Limitations of replacing "ordered optimality" with "direct optimality": Multi-objective optimization often uses a weighted or constrained approach to perform a global search. However, within the "feasible" sample, how to stably rank the samples based on comprehensive performance and perform Top-k selection? Existing solutions often use point estimation from single-point regression, which is susceptible to scale differences and noise. They lack a pairwise comparison learning and ranking mechanism, resulting in insufficient ranking robustness and hit rate. Existing improvement directions are extremely insufficient: Some studies combine "feasibility classification + multi-objective optimization," but lack a hierarchical strategy of "high recall hard constraints + candidate pool + ranking learning"; other works introduce Bayesian optimization to control the simulation budget, but it is prone to early convergence or insufficient exploration when the constraint feasibility is extremely sparse. Summary of the Invention

[0007] The purpose of this invention is to address the problems existing in the prior art, and to provide a recall-hard-constraint dual-threshold screening and ranking method for permanent magnet synchronous motor parameter design. This method, through a data-driven approach built around the motor design decision objectives, can explicitly apply high-recall hard constraints and provide transferable thresholds during the feasibility identification stage; form a "candidate pool" in resource-constrained scenarios and robustly rank feasible samples to improve Top-k hit rate; perform posterior calibration of probabilities to ensure stable decision-making across batches; and integrate with existing simulation / experiment closed loops to continuously update the model and thresholds. These requirements provide the practical background and technical motivation for this invention.

[0008] The objective of this invention is achieved through the following technical solution:

[0009] A recall hard-constraint dual-threshold screening and sorting method for permanent magnet synchronous motor parameter design includes the following steps:

[0010] Step 1: Data Acquisition and Standardization

[0011] Data samples are generated based on parametric motor models through experimental design or adaptive sampling. The performance indicators of permanent magnet synchronous motors were obtained using finite element simulation. ,right Perform numerical standardization, for Perform dimensionless transformation or interval scaling;

[0012] Step 2: Engineering Constraints and Label Generation

[0013] Based on the design red line, set the threshold values ​​for the permanent magnet synchronous motor performance indicators in step one, and generate the feasibility label y;

[0014] Step 3: Probability Estimation and Calibration

[0015] Choose a probabilistic classifier Using the threshold obtained in step two, combined with class imbalance and boundary-sensitive weights for training, the probabilistic classifier first outputs a score. The feasibility probability was obtained after isotonic or Platt calibration. , The feature vector of the motor scheme;

[0016] Step 4: Determining the threshold strategy:

[0017] Determine the high-confidence pass-through threshold on the independent validation set. With step three Candidate pool threshold ;

[0018] Step 5: Sample Triage

[0019] exist Left-side search for low threshold Based on the determination in step four and Threshold based on calibration probability The sample is divided into three domains: forming the three domains, S5A pass-through area: It is directly determined to be feasible; S5B candidate pool: Retain areas for review and sorting; S5C discard area: It does not proceed to subsequent expensive simulations;

[0020] Step Six: High-Confidence Learning Sort:

[0021] The sorting supervision signal is constructed in the pass-through region and candidate pool subdomain formed in step five:

[0022] Step 7: Closed-loop feedback and data transmission:

[0023] The newly added simulation or experimental results are fed back to the training dataset, and the calibration and retraining are performed periodically according to steps three through six; the threshold is reselected when the motor operating condition platform or red line changes. and ;

[0024] Step 8: Output the decision results.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] This invention obtains a sample set based on parametric finite element modeling; generates feasibility labels based on motor engineering constraints (lower limit of output torque, upper limit of cogging torque, upper limit of torque fluctuation, and upper limit of total harmonic distortion); in the training phase—a classifier—is trained using boundary-sensitive weights and probability calibration; and on the validation set, recall hard constraints are used as a prerequisite. The criterion selects the first threshold. ;according to The adaptive quantile on the left generates the second threshold. This method forms a "high-confidence direct pass zone" and a "candidate pool." For samples within the direct pass zone or candidate pool, a stage-two pairwise ranking model (RankSVM or equivalent model) is used to construct local preference pairs based on normalized multi-objectives (torque, THD, cogging, ripple) and train the model, outputting ranking scores and providing a Top-k priority simulation list. This invention significantly reduces simulation overhead while maintaining high recall and achieves stable priority ranking within high-confidence samples. Unlike existing single-stage classification or pure threshold / regression proxies, this invention integrates hard recall constraints, dual-threshold splitting, and local pairwise ranking into a unified process. The technical effects are: controllable missed detection risk, adjustable candidate set size, and improved optimal hit rate, making it suitable for motor parameter design, batch parameter screening, and rapid parameter optimization.

[0027] This invention can explicitly apply high recall hard constraints and provide transferable thresholds during the feasibility identification stage; form a "candidate pool" in resource-constrained scenarios and robustly rank feasible samples to improve Top-k hit rate; perform posterior calibration on probabilities to ensure stable decision-making across batches; and interface with existing simulation / experiment closed loops to continuously update the model and thresholds.

[0028] On typical datasets, this invention achieves a total selection recall of ≥0.98, a pass-through domain recall of ≥0.97, and a precision of ≥0.55; the pass-through domain Top-10 ≈ 0.40 and Top-20 ≈ 0.50. Compared to single-threshold or "regression proxy + empirical threshold," it improves Top-k hits by approximately 1.5 to 2.0 times under the same budget, reduces the number of simulation points by approximately 40% to 60%, and lowers R&D costs and shortens the iteration cycle without sacrificing feasible coverage.

[0029] This invention divides the overall sample into three regions using calibration probability and dual thresholds, and first determines these regions on the validation set using constraint optimization. ,make By reducing false negatives through mechanism, the actual measured recall in the direct pass domain is ≥0.95, and it can still guarantee "high recall" in extremely unbalanced scenarios; furthermore... Internal selection By controlling the candidate pool ratio to 20%–40%, forming three domains of direct access, candidate, and discard, and only reviewing and ranking the candidate pool, simulation or experimental costs are focused on value, saving approximately 30% of the second-stage computation and manual review workload. Simultaneously, samples from the direct access domain can directly enter the subsequent design process, significantly shortening the iteration cycle. A learning ranking model (with boundary weighting) is introduced within the candidate and direct access domains, learning "relative superiority" rather than a direct regression of absolute indicators, avoiding mismatches in "direct optimization" across operating conditions and aircraft models. The Top-k hit rate can reach 0.4–0.5 at k=10~20, prioritizing high-quality solutions with fewer review quotas. A corresponding closed-loop feedback system feeds new simulation / experiment results back into the retraining and reselection thresholds, ensuring the working point... It is auditable, reproducible, and improves with steady-state data, achieving triple benefits of high recall, value-based resource allocation, and stable ranking. Attached Figure Description

[0030] Figure 1 This is a flowchart of a recall hard constraint dual threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to the present invention.

[0031] Figure 2 This is a Precision–Recall curve for the test set of this invention.

[0032] Figure 3 This is the test set and the feasible probability histogram with threshold lines for this invention.

[0033] Figure 4 This is the confusion matrix diagram of the Pass∪Pool selection set in this invention.

[0034] Figure 5 This is a confusion matrix diagram of the "high confidence pass-through zone" of the present invention.

[0035] Figure 6 This is a confusion matrix diagram of the "candidate pool" of the present invention.

[0036] Figure 7 This is the Top-k hit rate curve of the present invention. Detailed Implementation

[0037] The present invention will be further described in detail below with reference to the accompanying drawings: This embodiment is implemented based on the technical solution of the present invention and provides detailed implementation methods, but the protection scope of the present invention is not limited to the following embodiments.

[0038] like Figure 1 As shown in this embodiment, a recall hard-constraint dual-threshold screening and sorting method for permanent magnet synchronous motor parameter design includes:

[0039] S1. Data Acquisition and Standardization: Data samples are generated based on parametric motor models (stator tooth width, magnet thickness, pole arc ratio, air gap, rotor inner diameter, winding parameters, etc.) through experimental design or adaptive sampling. Performance indicators obtained through finite element simulation , Represented as a characteristic matrix It is a set of real numbers. Matrix, where, The number of samples is the same as the number of rows. The characteristic number is the number of columns. ,in, For torque, For total harmonic distortion, For cogging torque, This is torque ripple. For Perform numerical standardization, for Perform dimensionless transformation or interval scaling.

[0040] S2. Engineering Constraints and Label Generation: Setting thresholds based on design red lines:

[0041]

[0042] A fixed threshold or a quantile-adaptive threshold can be used. Generate a feasibility label y, if...

[0043]

[0044] If y=1, then y=1; otherwise, y=0. Here, y=1 indicates feasibility, and y=0 indicates infeasibility.

[0045] S3. Probability Estimation and Calibration: Select a probabilistic classifier. (Gradient boosting, random forest, logistic regression, lightweight neural networks, etc., GBDT is selected in the example), combined with class imbalance and boundary-sensitive weight training, sample weights and feasible labels are based on the red line in S2, weights It can be constructed using the minimum margin:

[0046]

[0047] in, To amplify the boundary values, control the weight magnitude (actually measured to be 1–2). The value of the natural exponential function; Minimum normalization margin; It is a scale factor for torque, used to normalize different dimensions; The scale factor for total harmonic distortion; This is the scaling factor for cogging torque; This is the scaling factor for torque ripple;

[0048] When training a probability classifier, "more attention is paid to samples near the boundary / that are prone to errors." As a scale, A probabilistic classifier first outputs a score. The output scores are then calibrated using either isotonic or Platt to obtain reliable feasibility probabilities.

[0049]

[0050] in, This represents the feature vector of the motor design, i.e., the input of a sample. In this dataset, the feature vector of the motor design... , It is described as a set of independent variables such as motor geometry, materials, windings, or control settings. It always refers to the same set of motor structure parameter vectors.

[0051] S4. Threshold Strategy Determination: To reduce false positives and control candidate size while ensuring high recall, a high-confidence pass-through threshold is determined on the independent validation set. and Candidate pool threshold .

[0052] Set recall and precision lower bounds on the independent validation set. , On the validation set, select one threshold from all thresholds as the high threshold. It is required that it simultaneously meets the following conditions:

[0053]

[0054] Right now And within these threshold conditions, let maximum.

[0055] Comprehensive indicators adopt The scores are as follows:

[0056]

[0057] Note: Determined through comprehensive indicators Value, in the instance More emphasis on precision More emphasis on recalls; preferred .

[0058] Using validation sets The range is determined to "determine the lower boundary of the candidate pool". The idea is to keep the size of the candidate pool controllable and as small as possible without sacrificing quality.

[0059] in, For calibration probability The positive thresholds are all calculated on the independent validation set; express , To assess accuracy, how many of the solutions deemed feasible by the model are actually feasible? This is used to control false alarm rates and review costs. express , Recall rate is the percentage of all real-world feasible solutions that the model identifies, used to control for false negatives. To control recall weight.

[0060] S5. Sample Splitting (Pass-through / Candidate / Discard): In Left-side search for low threshold (Can be determined by the right quantile of the calibration probability or by the "selected proportion" constraint method), forming three domains: S5A through area: It is directly determined to be feasible; S5B candidate pool: Retain areas for review and sorting; S5C discard area: This avoids the need for subsequent expensive simulations. The determination satisfies the following: (1) the recall of the total selection set (direct and candidate) is not lower than the target; (2) the size of the candidate pool is within the target range (e.g., 20% to 40% of the total); (3) the accuracy of the total selection set is not lower than the preset lower limit, which can be adjusted within the range (0.55, 0.65) according to the project; if the cost of false alarms is high, the preset lower limit can be selected within the range (0.65, 0.75).

[0061] S6. High-confidence learning ranking: Construct ranking supervision signals in the pass-through and candidate subdomains: (1) Target fusion: The output Normalization defines a weighted overall advantage or disadvantage. (Or use Pareto dominance relation). (2) Paired samples: if or If it is dominant on most objectives, then To be positive; to enhance stability, sampling is only performed within the nearest neighbors, and boundary weights are assigned:

[0062]

[0063] (3) Training the sorter: The score function is obtained by training RankSVM, Pairwise-LR, LambdaMART or RankNet. (4) Reasoning and Top-k: Calculate the direct or direct ∪ candidate set. Output the top k as the priority simulation list.

[0064] S7. Closed-Loop Feedback and Data Backhaul: New simulation or experimental results are fed back to the training dataset, while the validation set is kept independently and used only for threshold selection and evaluation; periodic calibration and retraining are performed according to S3 to S6; when the motor operating condition platform or red line changes, the threshold is reselected to ensure portability and stability. The motor operating condition platform refers to significant changes in factors such as slot-pole ratio, material system, cooling structure, and load / speed conditions.

[0065] S8. Output Decision Results: The output includes: the operating point PR curve and its operating point coordinates. Probability Histogram Three-domain confusion matrix, pass-through domain sorted scatter plots, Top-k hit curve, and priority list. Among these, for That is, at a high threshold Precision at the location; for That is, at a high threshold Recall rate at each location.

[0066] In this invention, a recall hard constraint dual threshold screening and sorting system for permanent magnet synchronous motor parameter design includes: a data generation module, a label and constraint module, a stage-screening module, a dual threshold generation module, a stage-two sorting module, and a result output module.

[0067] The system operates in the following order: "Data → Label → Probability → Threshold → Triage → Sorting → Feedback → Output": First, the data generation module generates sample parameter vectors. The labeling and constraint module generates a feasibility label y based on the engineering boundary line; the phase one screening module trains a probability classifier and obtains the result using Platt / Isotonic calibration. The dual threshold generation module searches for thresholds on independent validation sets. : take first make Maximum and satisfy And then Internal selection The sample splitting module divides the samples into three regions based on the target range (e.g., 20%–40%) of the candidate pool: For direct access area, Enter the candidate pool, Discard; the Phase 2 sorting module ranks candidates within the pass zone and candidate pool according to normalized multi-indicator merits. Constructing nearest neighbor directed pairs The model is then trained by assigning boundary weights to obtain ranking scores. The output shows the Top-k hit rate; the closed-loop feedback and data feedback module will add simulation / experimental results to the training set, periodically repeating probability training and threshold updates; the results output module will uniformly generate PR curves and operating points. Probability histograms, three-domain confusion matrices, sorting scatter plots, Top-k curves, and priority lists are used to form an iterative and adaptive closed loop.

[0068] like Figure 2 The image shows the Precision-Recall (PR) curve (test set). Operating points are marked on the curve, corresponding to those satisfying... and of The selection criteria are given in the figure caption. With threshold.

[0069] like Figure 3The image shows a histogram of feasibility probabilities (test set). The probability axes are also labeled... and This demonstrates the division into three domains: among which, High-confidence direct access zone; Candidate pool; Discard area. (Explanation) It can be determined by quantile or candidate size constraints, and guarantees .

[0070] like Figure 4 The diagram shown is a confusion matrix of the "selection set (Pass∪Pool)". The four quadrants are labeled TN, FP, FN, and TP respectively; this matrix reflects the end-to-end effect (overall recall and overall precision) of the "recall hard constraint + candidate strategy".

[0071] like Figure 5 The diagram shown is a confusion matrix for the "high-confidence pass-through zone". It is used to illustrate the effect of pass-through determination under high recall (reduced FN) and controlled precision (tolerance for a certain FP), demonstrating the core technical benefits of hard constraint on recall.

[0072] like Figure 6 The image shown is a confusion matrix diagram for the "Pool-only" candidate pool. It is used to illustrate threshold proximity. At that time, there was an objective phenomenon that there were relatively few truly feasible samples in the candidate pool; as an example or comparative illustration, it is not included in the core effect comparison.

[0073] like Figure 7 The figure shows the Top-k hit rate curve. On the test set, the hit rate of the sorting module is approximately 0.40 when k=5, 10, and 15, and increases to approximately 0.50 when k=20. This indicates that the present invention can output a highly reliable priority simulation list, thereby reducing the number of experiments and simulations.

[0074] The above description is merely a preferred embodiment of the present invention. These specific embodiments are different implementations based on the overall concept of the present invention, and the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors, characterized in that, Includes the following steps: Step 1: Data Acquisition and Standardization Data samples are generated based on parametric motor models through experimental design or adaptive sampling. The performance indicators of permanent magnet synchronous motors were obtained using finite element simulation. ,right Perform numerical standardization, for Perform dimensionless transformation or interval scaling; Step 2: Engineering Constraints and Label Generation Based on the design red line, set the threshold values ​​for the permanent magnet synchronous motor performance indicators in step one, and generate the feasibility label y; Step 3: Probability Estimation and Calibration Choose a probabilistic classifier Using the threshold obtained in step two, combined with class imbalance and boundary-sensitive weights for training, the probabilistic classifier first outputs a score. The feasibility probability was obtained after isotonic or Platt calibration. , The feature vector of the motor scheme; Step 4: Determining the threshold strategy: Determining the high-confidence pass-through threshold on an independent validation set With step three Candidate pool threshold ; Step 5: Sample Triage exist Left-side search for low threshold Based on the determination in step four and Threshold based on feasible probability The sample is divided into three domains: forming a three-domain, direct access region. It is directly determined to be feasible; Candidate pool: Retain areas for review and sorting; discard areas: It does not proceed to subsequent expensive simulations; Step Six: High-Confidence Learning Sort: The sorting supervision signal is constructed in the pass-through region and candidate pool subdomain formed in step five: Step 7: Closed-loop feedback and data transmission: The newly added simulation or experimental results are fed back to the training dataset, and the calibration and retraining are performed periodically according to steps three through six; the threshold is reselected when the motor operating condition platform or red line changes. and ; Step 8: Output the decision results.

2. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 1, characterized in that, The data sample mentioned in step one The performance indicators of the permanent magnet synchronous motor ,in, For torque, For total harmonic distortion, For cogging torque, This refers to torque ripple.

3. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 2, characterized in that, The threshold settings for the performance indicators of the permanent magnet synchronous motor in step two adopt either a fixed threshold or a quantile adaptive threshold method. The thresholds for the performance indicators of the permanent magnet synchronous motor are as follows: Generate a feasibility label y, if If y=1, then y=1; otherwise, y=0. y=1 indicates that it is feasible, and y=0 indicates that it is not feasible.

4. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 3, characterized in that, Step three describes the training of weights by combining class imbalance and boundary-sensitive weights. Constructed using minimum margin, weights The formula is as follows: in, Increase the amount of data at the boundary and control the weighting magnitude; The value of the natural exponential function; Minimum normalization margin; It is a scale factor for torque, used to normalize different dimensions; The scale factor for total harmonic distortion; This is the scaling factor for cogging torque; This is the scaling factor for torque ripple; When training a probability classifier, more attention is paid to samples near the boundary / those prone to errors; among them, As a scale, .

5. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 1, characterized in that, In step four, determine the high-confidence pass-through threshold. The steps are as follows: Set recall and precision lower bounds on the independent validation set. , On the validation set, select one threshold from all thresholds as the high threshold. It is required that it simultaneously meets the following conditions: Comprehensive indicators adopt Fraction, The scores are as follows: Using validation sets The lower boundary of the candidate pool is determined within the specified interval. Where t is the feasible probability. The positive thresholds are all calculated on the independent validation set; express , For accuracy, express , For recall rate, To control recall weight.

6. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 1, characterized in that, The steps described in step five are as follows: Left-side search for low threshold The method of determining the right-hand quantile based on feasible probability or by using a selected proportion constraint method is employed; The determination satisfies the following conditions: (1) the recall of the direct access area and the candidate pool is not lower than the target; (2) the size of the candidate pool is within the target range; (3) the accuracy of the direct access area and the candidate pool is not lower than the preset lower limit.

7. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 1, characterized in that, The steps for constructing the sorting supervision signal in step six are as follows: Step 61, Target Fusion: The output will be fused. Normalization defines a weighted overall advantage or disadvantage. ; Step 62, Paired Samples: If the overall advantages and disadvantages obtained in Step 61 are... or If it is dominant on most objectives, then To enhance stability, sampling is only performed within the nearest neighbors, and boundary weights are assigned as follows: in, The first Feature vectors of each sample; These represent the feasible probabilities of the corresponding samples; These are pairwise weighting functions used to weight the sorting loss; To stabilize the term and prevent the denominator from being zero; Step 63: Train the sorter: Use RankSVM, Pairwise-LR, LambdaMART, or RankNet to train the score function. ; Step 64, Reasoning and Top-k: Calculate the score function from Step 63 for the direct or direct ∪ candidate set. Output the top k as the priority simulation list.

8. The recall hard-constraint dual-threshold screening and sorting method for parameter design of permanent magnet synchronous motors according to claim 1, characterized in that, The output decision results mentioned in step eight include: the operating point PR curve and its operating point coordinates. Labeling, probability histogram Three-domain confusion matrix, pass-through domain sorted scatter plots, Top-k hit curves, and priority list; among them, for That is, at a high threshold Precision at the location; for That is, at a high threshold Recall rate at each location.

Images