Signal processor comprehensive evaluation and selection method based on single-machine multi-dimension data screening

Abstract

The application discloses a signal processor comprehensive evaluation and optimization method based on single-machine multi-dimension data screening, and belongs to the technical field of signal processors. The method comprises the following steps: acquiring test data of a plurality of first candidate signal processors in a plurality of evaluation dimensions under each test scene; for each test scene, performing clustering processing based on the test data of the first candidate signal processors in the plurality of evaluation dimensions to obtain a corresponding category result; determining a target category result based on the category results of all test scenes; and performing pairwise comparison on the first candidate signal processors based on the target category result and the test data of the first candidate signal processors in the plurality of evaluation dimensions under each test scene to determine a target signal processor. The method can comprehensively evaluate the performance of the signal processor in various scenes in multiple evaluation dimensions, and can realize the optimization selection of the signal processor.

Description

Technical Field

[0001] This invention belongs to the field of signal processing technology, specifically relating to a comprehensive evaluation and selection method for signal processing machines based on single-machine multi-dimensional data screening. Background Technology

[0002] The ship's signal processor (hereinafter referred to as the signal processor) is a core component of the ship's electronic system. It is responsible for acquiring, processing, and analyzing signals from various sensors such as radar, sonar, and communication systems. Its performance directly affects the ship's navigation accuracy, target recognition capability, and communication quality. With the increasing intelligence of ships, the types of signals that the signal processor needs to handle are becoming increasingly complex, placing higher demands on its performance indicators such as processing speed, power consumption, reliability, and environmental adaptability. Currently, the selection of signal processors mainly adopts a single-scenario testing and evaluation method, that is, testing candidate signal processors under specific operating conditions (such as a standard laboratory environment) and directly selecting the product with the best performance based on the test results. However, the ship's navigation environment is complex and variable. Signal processors may face various harsh operating conditions such as high temperature, high humidity, high salt spray, and strong vibration. The test results of a single scenario cannot comprehensively reflect the processor's overall performance in actual use. Under different operating conditions, the performance of various evaluation dimensions may vary significantly. For example, a processor may have excellent processing speed at room temperature, but its performance may drop sharply in a high-temperature environment. Therefore, how to comprehensively evaluate the performance of signal processors across multiple evaluation dimensions in various scenarios and select the best signal processor has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0003] This disclosure provides a method for comprehensive evaluation and selection of signal processors based on multi-dimensional data filtering of a single machine. This method aims to select the best signal processor by comprehensively evaluating its performance across multiple evaluation dimensions in various scenarios.

[0004] In some embodiments, a comprehensive evaluation and selection method for signal processors based on single-machine multi-dimensional data filtering includes: Test data for multiple first candidate signal processors in multiple evaluation dimensions were obtained for each test scenario. For each test scenario, clustering is performed based on the test data of each first candidate signal processor across multiple evaluation dimensions to obtain the corresponding category results; Based on the category results of all test scenarios, determine the target category result; Based on the target category results and the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios, the first candidate signal processors are compared pairwise to determine the target signal processor.

[0005] The beneficial effects of this invention are as follows: By acquiring test data for each candidate signal processor across all test scenarios and performing clustering on each scenario, a target category is determined. This allows for a comprehensive assessment of the signal processor's performance across various test scenarios, enabling clustering across all evaluation dimensions. Furthermore, based on the target category and test data from multiple candidate signal processors across multiple evaluation dimensions in each test scenario, pairwise comparisons are made to determine the optimal signal processor, i.e., the target signal processor. In this way, by clustering evaluation dimensions under different test scenarios, determining the target category based on the clustering results (i.e., category results), and combining test data from all test scenarios to determine the optimal processor through pairwise comparisons, the evaluation results comprehensively reflect the signal processor's performance under various operating conditions (i.e., test scenarios). This provides a more complete picture of the actual performance of each candidate signal processor, allowing for optimal selection by comprehensively evaluating the signal processor's performance across multiple evaluation dimensions in various scenarios. Attached Figure Description

[0006] One or more embodiments are illustrated by way of example with reference to the accompanying drawings. These illustrations and drawings do not constitute a limitation on the embodiments. Elements having the same reference numerals in the drawings are shown as similar elements. The drawings are not to be scaled. And wherein: Figure 1 This is a flowchart of a comprehensive evaluation and selection method for signal processors based on single-machine multi-dimensional data filtering, provided by the present invention. Detailed Implementation

[0007] To provide a more detailed understanding of the features and technical content of the embodiments of this disclosure, the implementation of the embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of this disclosure. In the following technical description, for ease of explanation, several details are used to provide a full understanding of the disclosed embodiments. However, one or more embodiments may still be implemented without these details. In other cases, well-known structures and devices may be simplified in their depiction to simplify the drawings.

[0008] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate for the embodiments of this disclosure described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion.

[0009] Unless otherwise stated, the term "multiple" means two or more.

[0010] In this embodiment of the disclosure, the character " / " indicates that the objects before and after it are in an "or" relationship. For example, A / B means: A or B.

[0011] The term "and / or" describes an association between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or A and B.

[0012] The term "correspondence" can refer to an association or binding relationship. The correspondence between A and B means that there is an association or binding relationship between A and B.

[0013] Combination Figure 1 As shown in the embodiments of this disclosure, a comprehensive evaluation and selection method for signal processors based on single-machine multi-dimensional data filtering is provided, including: Step S101: Obtain test data of multiple first candidate signal processors in multiple evaluation dimensions under each test scenario.

[0014] The test scenarios include normal navigation scenarios (i.e., the ship cruises under normal sea conditions, such as a temperature of 25°C, humidity of 50%, normal power supply, and no vibration), high temperature and high humidity scenarios (i.e., the ship sails in tropical waters, such as a temperature of 45°C, humidity of 90%, and air conditioning at full load), strong vibration scenarios (i.e., the ship sails in severe sea conditions, such as continuous vibration frequency of 10-50Hz and amplitude of ±2mm), and full load scenarios (i.e., all systems of the ship operate simultaneously, such as full power supply and all radars / sonars working simultaneously).

[0015] The evaluation dimensions include a variety of aspects such as completeness check of the whole machine, complete machine slot check, module type check, internal backplane test, external communication interface test, weight measurement, input voltage test, noise test, power consumption test, heat dissipation check, architecture design check, health management test, real-time test, indicator light check, fan speed test, power on / off and reset function test, chassis number setting function test, grounding requirement test, chassis size measurement, module size measurement, appearance and marking test, key design test, independent controllability check, homogenization check, compatibility test, environmental adaptability test, electromagnetic compatibility test, stress test, and comprehensive application verification.

[0016] Step S102: For each test scenario, clustering is performed based on the test data of each first candidate signal processor in multiple evaluation dimensions to obtain the corresponding category results.

[0017] The categorization results include multiple dimensional categories, and each dimensional category includes multiple evaluation dimensions.

[0018] Step S103: Determine the target category result based on the category results of all test scenarios.

[0019] Step S104: Based on the target category results and the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios, perform pairwise comparisons on each first candidate signal processor to determine the target signal processor.

[0020] The signal processor comprehensive evaluation and selection method based on single-machine multi-dimensional data screening provided in this disclosure involves acquiring test data for each first candidate signal processor in each evaluation dimension under various test scenarios, and then performing clustering processing on each test scenario to obtain the category result corresponding to each test scenario. Based on the category results of all test scenarios, a target category result is determined, thereby comprehensively assessing the performance of the signal processor under various test scenarios and achieving clustering and classification processing for each evaluation dimension. Finally, based on the target category result and the test data of multiple first candidate signal processors in multiple evaluation dimensions under each test scenario, pairwise comparisons are performed on each first candidate signal processor to determine the optimal signal processor, i.e., the target signal processor. In this way, by clustering the evaluation dimensions under different test scenarios, and then determining the target category result based on the clustering results (i.e., category results), and combining the test data under all test scenarios, the optimal processor is determined by pairwise comparison. This makes the evaluation result comprehensive, taking into account the performance of the signal processor under various operating conditions (i.e. test scenarios), so as to more comprehensively reflect the actual performance of each first candidate signal processor. That is, by comprehensively evaluating the performance of the signal processor under multiple evaluation dimensions in various scenarios, the optimal signal processor can be selected.

[0021] Preferably, based on the test data of each first candidate signal processor across multiple evaluation dimensions, clustering is performed to obtain the corresponding category results, including: The test data of all first-candidate signal processors in the same evaluation dimension are used together as the vector of that evaluation dimension; Based on the vectors of each evaluation dimension, the K-means clustering algorithm is used to cluster multiple evaluation dimensions to obtain the corresponding category results.

[0022] In this way, under this test scenario, the test data of all processors under the same evaluation dimension are formed into a vector, and the K-means clustering algorithm is used to cluster them. This allows for the automatic identification of evaluation dimensions with similar characteristics based on the distribution characteristics of the test data itself, thereby objectively and accurately obtaining K dimensional categories, i.e., obtaining the corresponding category results.

[0023] For example, the test data of all first candidate signal processors in the same evaluation dimension are used together as the vector of that evaluation dimension. For example, for each first candidate signal processor, its test data in evaluation dimension A under test scenario 1, test data in evaluation dimension A under test scenario 2, test data in evaluation dimension A under test scenario 3, and test data in evaluation dimension A under test scenario 4 are used together as the vector of that first candidate signal processor in evaluation dimension A.

[0024] The results for this category include K dimensional categories. The number of clusters K can be determined through expert experience, the elbow rule (i.e., calculating the clustering error (SSE, sum of squares within clusters) under different K values, and selecting the inflection point where the error decreases significantly to determine the number of clusters K), or principal component analysis (performing principal component analysis on the vectors of each evaluation dimension to determine the cumulative variance contribution rate, and determining the number of clusters K based on the cumulative variance contribution rate).

[0025] For example, the category results corresponding to the normal navigation scenario (K=5) are as follows: Dimension Category A (Performance and Computation): Processing speed, computational accuracy, parallel processing capability, and real-time performance testing; Dimension Category B (Power Consumption and Heat Dissipation): Operating power consumption, standby power consumption, noise testing, fan speed testing, and heat dissipation method inspection; Dimension Category C (Structural and Physical): Weight measurement, chassis size measurement, module size measurement, grounding requirement testing, and chassis number setting function testing; Dimension Category D (Interface Basics): External communication interface testing, internal backplane testing, compatibility testing, power on / off testing, reset function testing, and indicator light inspection; Dimension Category E (Comprehensive Assurance): Procurement cost, maintenance cost, mean time between failures (MTBF), health management testing, independent controllability inspection, homogenization inspection, appearance and labeling testing, overall system completeness inspection, environmental adaptability testing, electromagnetic compatibility testing, stress test, comprehensive application verification, overall system slot inspection, module type inspection, input voltage testing, key layout design testing, and architecture design inspection.

[0026] The category results for the high temperature and high humidity scenario (K=6) are as follows: Dimension Category A (Performance and Power Consumption Coupling): Processing speed, computational accuracy, parallel capability, operating power consumption, standby power consumption, real-time performance testing, noise testing, fan speed testing, and heat dissipation inspection; Dimension Category B (Structural Physics): Weight measurement, chassis size measurement, module size measurement, grounding requirement testing, and chassis number setting function testing; Dimension Category C (Interface Basics): External communication interface testing, internal backplane testing, compatibility testing, power-on / off testing, reset function testing, indicator light inspection, overall machine slot inspection, and module type inspection; Dimension Category D (Cost): Procurement cost and maintenance cost; Dimension Category E (Reliability): Mean Time Between Failures (MTBF) and health management testing; Dimension Category F (Environmental Compliance): Environmental adaptability testing, electromagnetic compatibility testing, stress testing, comprehensive application verification, independent controllability inspection, homogenization inspection, appearance and labeling testing, overall machine completeness inspection, input voltage testing, key layout design testing, and architecture design inspection.

[0027] The category results for the strong vibration scenario (K=6) are as follows: Dimension Category A (Performance and Computation): Processing speed, computational accuracy, parallel processing capability, and real-time performance testing; Dimension Category B (Power Consumption and Heat Dissipation): Operating power consumption, standby power consumption, noise testing, fan speed testing, and heat dissipation method inspection; Dimension Category C (Structural Strength): Weight measurement, chassis size measurement, module size measurement, grounding requirement testing, chassis number setting function testing, and architecture design inspection; Dimension Category D (Interface Basics): External communication interface testing, internal backplane testing, compatibility testing, power on / off testing, reset function testing, indicator light inspection, overall machine slot inspection, and module type inspection; Dimension Category E (Cost, Reliability, and Coupling): Procurement cost, maintenance cost, mean time between failures (MTBF), and health management testing; Dimension Category F (Environmental Compliance): Environmental adaptability testing, electromagnetic compatibility testing, stress testing, comprehensive application verification, independent controllability inspection, homogenization inspection, appearance and labeling testing, overall machine completeness inspection, input voltage testing, and key layout design testing.

[0028] The category results for the full-load scenario (K=5) are as follows: Dimension Category A (Performance, Power Consumption, Reliability Coupling): Processing speed, computational accuracy, parallel processing capability, operating power consumption, standby power consumption, mean time between failures (MTBF), health management testing, real-time performance testing, noise testing, fan speed testing, and heat dissipation inspection; Dimension Category B (Structural Physics): Weight measurement, chassis size measurement, module size measurement, grounding requirement testing, and chassis number setting function testing; Dimension Category C (Interface Basics): External communication interface testing, internal backplane testing, compatibility testing, power-on / off testing, reset function testing, indicator light inspection, overall machine slot inspection, and module type inspection; Dimension Category D (Cost Separation): Procurement cost and maintenance cost; Dimension Category E (Comprehensive Assurance): Environmental adaptability testing, electromagnetic compatibility testing, stress testing, comprehensive application verification, independent controllability inspection, homogenization inspection, appearance and labeling testing, overall machine completeness inspection, input voltage testing, key layout design testing, and architecture design inspection.

[0029] Preferably, based on the category results of all test scenarios, the target category result is determined, including: Based on the category results of all test scenarios, the number of times each evaluation dimension pair is classified into the same dimension category in each test scenario is counted to obtain the co-occurrence frequency of each evaluation dimension pair; where each evaluation dimension pair includes two evaluation dimensions. Evaluation dimensions with co-occurrence frequencies greater than or equal to a preset frequency threshold are marked as having a strong correlation. The target category is determined based on the strong correlations.

[0030] In this way, by statistically analyzing the number of times each evaluation dimension pair is classified into the same dimension category in different test scenarios, the clustering results of multiple test scenarios are transformed into a unified co-occurrence frequency matrix. This allows the target category result to comprehensively reflect the performance of the signal processor under various operating conditions (i.e., test scenarios). Furthermore, by setting co-occurrence frequencies, evaluation dimension pairs with co-occurrence frequencies greater than or equal to a preset frequency threshold are marked as having a strong correlation. This allows for the filtering out evaluation dimension pairs that cluster together in different test scenarios. These strongly correlated evaluation dimension pairs have a stable intrinsic correlation, and their correlation remains stable as the operating conditions (i.e., test scenarios) change, thereby improving the repeatability and objectivity of the classification results.

[0031] Preferably, the target category result is determined based on each strong correlation, including: The evaluation dimensions are mapped to nodes, and strong relationships are mapped to edges to construct an undirected graph of dimensions; Identify all connected components in a dimensional undirected graph; The evaluation dimensions corresponding to each node within the same connected component are determined to be of the same dimension category, while the evaluation dimensions corresponding to nodes not included in any connected component are determined to be of an independent dimension category, in order to obtain the target category result.

[0032] In this way, by mapping evaluation dimensions to nodes and strong relationships to edges, an undirected graph of dimensions is constructed, transforming abstract co-occurrence frequency data into an intuitive graph structure, making the relationships between dimensions clear and intuitive. Furthermore, by identifying connected components, interconnected dimension clusters can be quickly discovered. Evaluation dimensions corresponding to nodes within a dimension cluster are assigned to the same dimension category, while evaluation dimensions corresponding to nodes not included in any connected component are assigned to independent dimension categories to obtain the target category result. This ensures that strongly correlated evaluation dimensions are correctly classified while preserving the independence of evaluation dimensions that are unstable in different testing scenarios and unsuitable for forced classification with other evaluation dimensions, resulting in a more accurate classification result (target category result).

[0033] For example, there are 4 test scenarios, and the clustering results (i.e., category results) for each test scenario are as follows: Clustering results for test scenario 1 (normal navigation): Dimension category A: D1, D2, D3; Dimension category B: D4, D5; Dimension category C: D6, D7, D8.

[0034] Clustering results for test scenario 2 (high temperature and high humidity): Dimension category A: D1, D2, D3, D4, D5; Dimension category B: D6, D7; Dimension category C: D8.

[0035] Clustering results for test scenario 3 (strong vibration): Dimension category A: D1, D2, D3; Dimension category B: D4, D5; Dimension category C: D6, D7, D8.

[0036] Clustering results for test scenario 4 (full load): Dimension category A: D1, D2, D3, D4, D5, D8; Dimension category B: D6; Dimension category C: D7.

[0037] 1) Count the co-occurrence frequency of each evaluation dimension pair: Traverse all evaluation dimension pairs (a total of C(8,2)=28 pairs), and count whether they are classified into the same dimension category in each test scenario, as shown in the table below: 2) If the preset frequency threshold is 3, then the evaluation dimension pairs with a co-occurrence frequency greater than or equal to 3 include D1 and D2, D1 and D3, D2 and D3, D4 and D5, and D6 and D7.

[0038] 3) Construct an undirected graph of dimensions with evaluation dimensions as nodes and strong relationships as edges, and use this graph to determine the target category results: Dimension category A: D1, D2, D3 (interconnected, forming connected component 1); Dimension category B: D4 and D5 (forming connected component 2); Dimension category C: D6-D7 (forming connected component 3); Dimension category D: D8 isolated (no strong relationship edges, forming a separate category).

[0039] Preferably, based on the category results of all test scenarios, the target category result is determined, including: Based on the category results of all test scenarios, the similarity between the category results of each test scenario pair is calculated to obtain the similarity of all test scenario pairs; where a test scenario pair includes two test scenarios. Based on the similarity of all test scenario pairs, the average similarity of each test scenario with other test scenarios is calculated to obtain the average similarity of each test scenario. The category result of the test scenario with the highest average similarity is used as the target category result.

[0040] In this way, by calculating the average similarity of each test scenario with all other test scenarios, the test scenario most similar to the clustering results (i.e., category results) of the other test scenarios is determined, and the target category result is determined accordingly. This ensures that the category result corresponding to the test scenario can represent the common features of all test scenarios to the greatest extent. Furthermore, directly selecting the clustering result of the real test scenario as the target category result can better preserve the physical meaning and interpretability of the original clustering, and the computational load is also smaller.

[0041] For example, the similarity between test scenario pairs includes: the similarity between test scenario 1 and test scenario 2 is 0.52; the similarity between test scenario 1 and test scenario 3 is 0.82; the similarity between test scenario 1 and test scenario 4 is 0.38; the similarity between test scenario 2 and test scenario 3 is 0.48; the similarity between test scenario 2 and test scenario 4 is 0.45; and the similarity between test scenario 3 and test scenario 4 is 0.42.

[0042] The average similarity of each test scenario to other test scenarios includes: Average similarity of test scenario 1 = (sim(1,2)+sim(1,3)+sim(1,4)) / 3 = (0.52+0.85+0.38) / 3 = 0.583; Average similarity of test scenario 2 = (sim(2,1)+sim(2,3)+sim(2,4)) / 3 = (0.52+0.48+0.45) / 3 = 0.483; Average similarity of test scenario 3 = (sim(3,1)+sim(3,2)+sim(3,4)) / 3 = (0.85+0.48+0.42) / 3 = 0.583; Average similarity of test scenario 4 = (sim(4,1)+sim(4,2)+sim(4,3)) / 3 = (0.38+0.45+0.42) / 3 = 0.417.

[0043] The average similarity between test scenario 1 and test scenario 3 is the same. The category result of one of the test scenarios can be randomly selected as the target category result, or the category result of one of the test scenarios can be selected as the target category result through expert voting, or the category result of one of the test scenarios can be selected as the target category result by comparing the similarity distribution of test scenario 1 and test scenario 3 with other test scenarios.

[0044] Preferably, based on the target category results and the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios, pairwise comparisons are performed on each first candidate signal processor to determine the target signal processor, including: Based on the test data of multiple first candidate signal processors in various test scenarios and in multiple evaluation dimensions, the category scores of each first candidate signal processor in each target category are determined; where the target category represents the dimensional category in the target category result. For each first candidate signal processor pair, compare their category scores in each target category to determine the win / loss result of the first candidate signal processor pair in each target category; wherein, the first candidate signal processor pair includes two first candidate signal processors; The target signal processor is determined based on the win-loss results of all first candidate signal processor pairs in each target category.

[0045] In this way, based on the test data of each evaluation dimension under all test scenarios, the score (i.e. category score) of each first candidate signal processor in each target category is calculated. Then, a pairwise comparison method is used, with each target category as a game. By statistically analyzing the win and loss results of each game, the relative superiority or inferiority of the comprehensive performance of each first candidate signal processor is determined, so that the selection of the final target signal processor has higher stability.

[0046] For example, based on the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios, the category score of each first candidate signal processor in each target category is determined, including: standardizing the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios. For each target category of each first candidate signal processor, the average of the sum of the standardized test data of the first candidate signal processor in all test scenarios for that target category is calculated as the category score.

[0047] The category score of the first candidate signal processor P1 in dimension A is calculated as follows: The test data for each test scenario is standardized, and then the calculation is performed, for example: Test data (standardized) for each evaluation dimension within dimension category A in each test scenario: Test Scenario 1: 0.85 (Evaluation Dimension 1), 0.92 (Evaluation Dimension 2), 0.78 (Evaluation Dimension 3), 0.88 (Evaluation Dimension 4). Test Scenario 2: 0.82 (Evaluation Dimension 1), 0.89 (Evaluation Dimension 2), 0.75 (Evaluation Dimension 3), 0.85 (Evaluation Dimension 4). Test Scenario 3: 0.80 (Evaluation Dimension 1), 0.88 (Evaluation Dimension 2), 0.72 (Evaluation Dimension 3), 0.82 (Evaluation Dimension 4). Test Scenario 4: 0.75 (Evaluation Dimension 1), 0.85 (Evaluation Dimension 2), 0.70 (Evaluation Dimension 3), 0.80 (Evaluation Dimension 4).

[0048] Calculate the category score of P1 in dimension A: The score for category A in test scenario 1 is (0.85 + 0.92 + 0.78 + 0.88) / 4 = 0.8575; The score for category A in test scenario 2 is (0.82 + 0.89 + 0.75 + 0.85) / 4 = 0.8275; The score for category A in test scenario 3 is (0.80 + 0.88 + 0.72 + 0.82) / 4 = 0.8050; The score for category A in test scenario 4 is (0.75 + 0.85 + 0.70 + 0.80) / 4 = 0.7750; The category score of P1 in dimension A is (0.8575+0.8275+0.8050+0.7750) / 4=0.81625.

[0049] Similarly, calculate the category scores of each first candidate signal processor for all target categories in each test scenario.

[0050] Preferably, the target signal processor is determined based on the win / loss results of all first candidate signal processor pairs in each target category, including: Based on the win-loss results of all first candidate signal processors in each target category, determine the total number of wins for each first candidate signal processor. The first candidate signal processor with the most total wins is selected as the target signal processor.

[0051] In this way, by comparing each of the first candidate signal processors pairwise across all target categories, ensuring the comprehensiveness of the multi-dimensional evaluation, and then determining the total number of wins for each first candidate signal processor based on the win / loss results, a higher total number of wins indicates more victories in comparisons with all other first candidate signal processors, and thus stronger overall performance indicators and capabilities. The first candidate signal processor with the highest total number of wins is selected as the target signal processor, thus achieving a more comprehensive multi-dimensional selection process to obtain the target signal processor.

[0052] For example, the first candidate signal processors P1, P2, and P3 have four categories: Compared to P2, P1 has 2 wins and P2 has 2 wins; compared to P3, P1 has 1 win and P2 has 3 wins; compared to P3, P2 has 1 win and P3 has 2 wins; and P2 and P3 have a tie of 1 game. Therefore, P1 has a total of 3 wins, P2 has a total of 3 wins, and P3 has a total of 5 wins. Thus, P3 is selected as the target signal processor.

[0053] In this embodiment, if there are first candidate signal processors tied for first place in total wins, one of them can be determined as the target signal processor through expert voting, which is not limited here.

[0054] Preferably, the target signal processor is determined based on the win / loss results of all first candidate signal processor pairs in each target category, including: Based on the win-loss results of all first candidate signal processors in each target category, determine the total number of negative fields for each first candidate signal processor. The first candidate signal processor with the lowest total negative field count is selected as the target signal processor.

[0055] In this way, by comparing each candidate signal processor pairwise across all target categories, ensuring comprehensive multi-dimensional evaluation, the total number of negative tests for each candidate signal processor is determined based on the win / loss results. A higher total number of negative tests indicates more failures compared to other candidate signal processors, resulting in lower overall performance and capabilities. The candidate signal processor with the lowest total number of negative tests is selected as the target signal processor, thus achieving a more comprehensive multi-dimensional selection process to obtain the target signal processor.

[0056] Preferably, the target signal processor is determined based on the win / loss results of all first candidate signal processor pairs in each target category, including: Initialize the overall score of each first candidate signal processor; For each first candidate signal processor pair, the win-loss relationship and the strength of the win-loss are determined based on all the win-loss results between them; Based on the win-loss relationship and win-loss strength of each first candidate signal processor pair, the ELO algorithm is used to iteratively update the comprehensive score of each first candidate signal processor in multiple rounds until the comprehensive score of each first candidate signal processor converges. The first candidate signal processor with the highest converged composite score is selected as the target signal processor.

[0057] In this way, the ELO algorithm quantifies the performance gap between the first candidate signal processors by considering the strength of the opponents and the intensity of the victory or defeat. In multiple rounds of iteration, the scores of each first candidate signal processor converge, so that it can accurately reflect the stable performance value of each first candidate signal processor (i.e., the converged comprehensive score). This allows for a more intuitive and reasonable multi-dimensional comprehensive selection to obtain the target signal processor.

[0058] For example, based on all win-loss results between the two parties, the win-loss relationship and the strength of the win-loss relationship are determined, including: iterating through all win-loss results between the two parties and determining whether it is a draw. If it is not a draw, the party with more wins is identified as the winner, and the party with fewer wins is identified as the loser, thus determining the win-loss relationship as winner and loser, and the strength of the win-loss relationship is calculated as (number of wins by the winner - number of wins by the loser) / total number of target categories. If it is a draw, the win-loss relationship is a draw, and the strength of the win-loss relationship is 0.

[0059] Based on the win-loss relationship and strength of each first-candidate signal processor pair, the ELO algorithm is used to iteratively update the overall score of each first-candidate signal processor in multiple rounds. This includes: if the win-loss relationship is a draw, the scores for both players remain unchanged (but this does not affect their score updates relative to other first-candidate signal processors). If the win-loss relationship is between a winner and a loser, the scores are updated according to the following formulas: Winner's new score = Winner's old score (i.e., the overall score at the start of this round) + M × (1 - Winner's expected score) × S; Loser's new score = Loser's old score (i.e., the overall score at the start of this round) + M × (0 - Loser's expected score) × S. Where M represents the adjustment coefficient, controlling the overall magnitude of each score update, and S represents the strength of the win-loss relationship. A larger S indicates a more significant advantage and a larger score update magnitude; a smaller S indicates a weaker advantage and a smaller score update magnitude.

[0060] In this embodiment, the adjustment coefficient M can be determined based on expert experience, or the adjustment coefficient M can be gradually reduced with each iteration to accelerate fractional convergence.

[0061] Alternatively, in this embodiment, all win / loss results between the two are iterated. If it is determined whether it is a tie, the weight of each target category can be determined by retrieving a preset weight set (which includes the preset weights of each target category, and the sum of the preset weights is 1). The weight of each target category is then used as the weighted score of the winner in its dimension category (i.e., if the first candidate signal processor P1 wins in dimension category A, then P1 obtains the weight of dimension category A as the weighted score; if the first candidate signal processor P2 wins in dimension category A, then P2 obtains the weight of dimension category A as the weighted score). If there is no winner in the dimension category (i.e., P1 and P2 have the same category score in the dimension category), then each party obtains half of the category weight, thereby obtaining the final weight score of the two. If the weight scores are different, then the party with the larger weight score is the winner between the two, and the party with the smaller weight score is the loser between the two, and the win / loss relationship is determined as the winner and the loser. The win / loss intensity = (the weight score of the winner - the weight score of the loser). If the weighted scores are the same, the win-loss relationship is a draw, and the win-loss strength is 0.

[0062] Based on the win-loss relationship and strength of each first-candidate signal processor pair, the ELO algorithm is used to iteratively update the overall score of each first-candidate signal processor in multiple rounds. This includes: if the win-loss relationship is a draw, the scores for both players remain unchanged (but this does not affect their score updates relative to other first-candidate signal processors). If the win-loss relationship is between a winner and a loser, the scores are updated according to the following formulas: Winner's new score = Winner's old score (i.e., the overall score at the start of this round) + M × (1 - Winner's expected score) × S; Loser's new score = Loser's old score (i.e., the overall score at the start of this round) + M × (0 - Loser's expected score) × S. Where M represents the adjustment coefficient, controlling the overall magnitude of each score update, and S represents the strength of the win-loss relationship. A larger S indicates a more significant advantage and a larger score update magnitude; a smaller S indicates a weaker advantage and a smaller score update magnitude.

[0063] Preferably, test data for multiple first candidate signal processors in multiple evaluation dimensions are acquired under each test scenario, including: Test data for multiple second candidate signal processors in multiple evaluation dimensions were obtained for each test scenario. Iterate through the test data of each second candidate signal processor across all evaluation dimensions in all test scenarios; The second candidate signal processor whose test data for all evaluation dimensions in all test scenarios reaches the preset baseline data line is selected as the first candidate signal processor, so as to obtain the test data of multiple first candidate signal processors in multiple evaluation dimensions in each test scenario.

[0064] In this way, by pre-setting the basic data lines, it can be ensured that each first candidate signal processor can meet the minimum requirements in all test scenarios and all evaluation dimensions, thereby avoiding the situation where the selection result of the target signal processor is unreliable due to non-compliance in individual evaluation dimensions or individual operating conditions (i.e. test scenarios).

[0065] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made to it in form and detail without departing from the scope defined by the claims of the present invention.

Claims

1. A comprehensive evaluation and selection method for signal processors based on single-machine multi-dimensional data filtering, characterized in that, include: Test data for multiple first candidate signal processors in multiple evaluation dimensions were obtained for each test scenario. For each test scenario, clustering is performed based on the test data of each first candidate signal processor across multiple evaluation dimensions to obtain the corresponding category results; Based on the category results of all test scenarios, determine the target category result; Based on the target category results and the test data of multiple first candidate signal processors in multiple evaluation dimensions under various test scenarios, the first candidate signal processors are compared pairwise to determine the target signal processor.

2. The method according to claim 1, characterized in that, The clustering process, based on the test data of each first candidate signal processor across multiple evaluation dimensions, yields corresponding category results, including: The test data of all first-candidate signal processors in the same evaluation dimension are used together as the vector of that evaluation dimension; Based on the vectors of each evaluation dimension, the K-means clustering algorithm is used to cluster multiple evaluation dimensions to obtain the corresponding category results.

3. The method according to claim 1, characterized in that, The determination of the target category result based on the category results of all test scenarios includes: Based on the category results of all test scenarios, the number of times each evaluation dimension pair is classified into the same dimension category in each test scenario is counted to obtain the co-occurrence frequency of each evaluation dimension pair; where each evaluation dimension pair includes two evaluation dimensions. Evaluation dimensions with co-occurrence frequencies greater than or equal to a preset frequency threshold are marked as having a strong correlation. The target category is determined based on the strong correlations.

4. The method according to claim 3, characterized in that, The result of determining the target category based on each strong correlation includes: The evaluation dimensions are mapped to nodes, and strong relationships are mapped to edges to construct an undirected graph of dimensions; Identify all connected components in a dimensional undirected graph; The evaluation dimensions corresponding to each node within the same connected component are determined to be of the same dimension category, while the evaluation dimensions corresponding to nodes not included in any connected component are determined to be of an independent dimension category, in order to obtain the target category result.

5. The method according to claim 1, characterized in that, The determination of the target category result based on the category results of all test scenarios includes: Based on the category results of all test scenarios, the similarity between the category results of each test scenario pair is calculated to obtain the similarity of all test scenario pairs; where a test scenario pair includes two test scenarios. Based on the similarity of all test scenario pairs, the average similarity of each test scenario with other test scenarios is calculated to obtain the average similarity of each test scenario. The category result of the test scenario with the highest average similarity is used as the target category result.

6. The method according to any one of claims 1 to 5, characterized in that, The process of determining the target signal processor by comparing each candidate signal processor pairwise based on the target category results and test data from multiple first candidate signal processors across multiple evaluation dimensions in various test scenarios includes: Based on the test data of multiple first candidate signal processors in various test scenarios and in multiple evaluation dimensions, the category scores of each first candidate signal processor in each target category are determined; where the target category represents the dimensional category in the target category result. For each first candidate signal processor pair, compare their category scores in each target category to determine the win / loss result of the first candidate signal processor pair in each target category; wherein, the first candidate signal processor pair includes two first candidate signal processors; The target signal processor is determined based on the win-loss results of all first candidate signal processor pairs in each target category.

7. The method according to claim 6, characterized in that, The determination of the target signal processor based on the win-loss results of all first candidate signal processors in each target category includes: Based on the win-loss results of all first candidate signal processors in each target category, determine the total number of wins for each first candidate signal processor. The first candidate signal processor with the most total wins is selected as the target signal processor.

8. The method according to claim 6, characterized in that, The determination of the target signal processor based on the win-loss results of all first candidate signal processors in each target category includes: Initialize the overall score of each first candidate signal processor; For each first candidate signal processor pair, the win-loss relationship and the strength of the win-loss are determined based on all the win-loss results between them; Based on the win-loss relationship and win-loss strength of each first candidate signal processor pair, the ELO algorithm is used to iteratively update the comprehensive score of each first candidate signal processor in multiple rounds until the comprehensive score of each first candidate signal processor converges. The first candidate signal processor with the highest converged composite score is selected as the target signal processor.

9. The method according to claim 1, characterized in that, The step of acquiring test data for multiple first candidate signal processors in multiple evaluation dimensions under each test scenario includes: Test data for multiple second candidate signal processors in multiple evaluation dimensions were obtained for each test scenario. Iterate through the test data of each second candidate signal processor across all evaluation dimensions in all test scenarios; The second candidate signal processor whose test data for all evaluation dimensions in all test scenarios reaches the preset baseline data line is selected as the first candidate signal processor, so as to obtain the test data of multiple first candidate signal processors in multiple evaluation dimensions in each test scenario.

Images