Intelligent navigation system collision avoidance function test method based on evaluation feedback mechanism

Abstract

The invention relates to an intelligent navigation system collision avoidance function test method based on an evaluation feedback mechanism, and belongs to the technical field of intelligent ships. The method comprises the following steps: performing simulation collision avoidance function tests on a plurality of collision avoidance algorithms for multiple times in a plurality of ship collision avoidance scenes, and calculating a comprehensive evaluation score of each collision avoidance scene and a first performance score of each collision avoidance algorithm based on each test result; constructing an optimization scene test set based on the comprehensive evaluation score; determining a plurality of preferred collision avoidance algorithms based on the first performance score; testing each preferred collision avoidance algorithm based on the optimized scene test set to obtain a second performance score of each preferred collision avoidance algorithm; and determining an optimal collision avoidance algorithm based on the second performance score to carry out a real ship collision avoidance experiment. According to the method, diversified ship collision avoidance scenes are constructed to verify the performance of the collision avoidance algorithm in different environmental conditions, the algorithm performance is comprehensively evaluated according to evaluation feedback, and the practicability and pertinence of testing are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent ships, and particularly relates to a method for testing the collision avoidance function of an intelligent navigation system based on an evaluation feedback mechanism. Background Art

[0002] With the continuous development of technology, the intelligence of ships is gradually being realized. It is very necessary to test the collision avoidance function of intelligent ships through simulation means. The current collision avoidance function tests generally have limitations, with insufficient diversity and complexity of test scenarios. During the test, the focus is mainly on a single collision avoidance algorithm, lacking diversified verification means. This test method is not only rather one-sided, but also often lacks clear goals and pertinence in the selection of test scenarios, appearing rather blind.

[0003] In order to improve the comprehensiveness and effectiveness of the test, it is necessary to broaden the test scope, cover multiple collision avoidance algorithms, comprehensively evaluate the performance and reliability of the collision avoidance function, and ensure that it can achieve the best collision avoidance effect in actual applications. Summary of the Invention

[0004] In view of the above analysis, the present invention aims to disclose a method for testing the collision avoidance function of an intelligent navigation system based on an evaluation feedback mechanism, which specifically includes the following steps:

[0005] Conduct multiple simulation collision avoidance function tests on multiple collision avoidance algorithms in multiple ship collision avoidance scenarios, and calculate the comprehensive evaluation score of each collision avoidance scenario and the first performance score of each collision avoidance algorithm based on the results of each test;

[0006] Construct an optimized scenario test set based on the comprehensive evaluation score; determine multiple preferred collision avoidance algorithms based on the first performance score;

[0007] Test each preferred collision avoidance algorithm based on the optimized scenario test set to obtain the second performance score of each preferred collision avoidance algorithm;

[0008] Determine the optimal collision avoidance algorithm based on the second performance score for a full-scale ship collision avoidance experiment.

[0009] Further, calculating the comprehensive evaluation score of each collision avoidance scenario and the first performance score of each collision avoidance algorithm based on the results of each test includes:

[0010] Calculate the comprehensive evaluation score of collision avoidance effectiveness for this test based on the normalized collision avoidance time value, the normalized cumulative change of the route, and the normalized cumulative change of the ship speed in the results of each test;

[0011] Calculate the comprehensive evaluation score of safety for this test based on the collision risk index CRI in the results of each test;

[0012] Calculate the evaluation score of the collision avoidance scenario used in this test and the performance score of the collision avoidance algorithm used in this test based on the comprehensive evaluation score of collision avoidance benefit and the comprehensive evaluation score of safety for each test.

[0013] Calculate the comprehensive evaluation score of each collision avoidance scenario by taking the average of all evaluation scores of each collision avoidance scenario; calculate the first performance score of each collision avoidance algorithm by taking the average of all performance scores of each collision avoidance algorithm.

[0014] Further, the construction of the optimized scenario test set based on the comprehensive evaluation score includes:

[0015] Determine multiple complex scenarios based on the comprehensive evaluation scores of all ship collision avoidance scenarios;

[0016] Generate an adjacent scenario set for each complex scenario;

[0017] Determine the optimized scenario test set based on all adjacent scenario sets.

[0018] Further, the determination of the optimal collision avoidance algorithm based on the second performance score includes:

[0019] Sort the second performance scores of each preferred collision avoidance algorithm from high to low;

[0020] If the difference in the second performance scores among the top m algorithms does not exceed the specified threshold, then select the algorithm with the smallest variance in the corresponding score distribution among the top m algorithms as the optimal collision avoidance algorithm through F-test; otherwise, select the algorithm with the highest second performance score as the optimal collision avoidance algorithm.

[0021] Further, the calculation formula for the comprehensive score of collision avoidance benefit is:

[0022]

[0023] where F is the comprehensive score of collision avoidance benefit, t norm 、 v norm are the normalized values of collision avoidance duration, cumulative change in course, and cumulative change in speed respectively, and w 1 、w 2 、w 3 are the corresponding weights of each parameter.

[0024] Further, the calculation formula for the comprehensive evaluation score of safety is:

[0025]

[0026] where S is the comprehensive evaluation score of safety, maxCRI and They are the maximum CRI value of this test, the average CRI value of this test, w 1 ', w 2 ' are the corresponding weights of each parameter respectively.

[0027] Furthermore, the calculation formula for the evaluation score of the collision avoidance scenario in the test result of this time is:

[0028] G = eF + sS;

[0029] where G is the evaluation score; e is the weight of the comprehensive score of collision avoidance effectiveness; s is the weight of the comprehensive safety score.

[0030] Furthermore, the calculation method for the performance score of the collision avoidance algorithm in the test result of this time is:

[0031] G' = 1 - (eF + sS);

[0032] where G' is the performance score.

[0033] Furthermore, the adjacent scenario set corresponding to each complex scenario generated based on each complex scenario includes:

[0034] For the complex scenario x i Set a restricted search area B i , where i is the complex scenario number;

[0035] Randomly sample scenario variables within B i to generate multiple sampled scenarios and store them as the sampled scenario set E i ;

[0036] Use the kd-tree to determine the adjacent scenario set E i ' based on the sampled scenario set E i '.

[0037] Furthermore, the use of the kd-tree to determine the adjacent scenario set E i ' based on the sampled scenario set E i includes:

[0038] Build a kd-tree based on the sampled scenario set E i ;

[0039] Calculate the L2 distance between each sampled scenario and the current complex scenario x i ;

[0040] Use the kd-tree search algorithm to find all sampled scenarios whose L2 distance from the current complex scenario x i is less than the threshold τ to construct the adjacent scenario set E i ';

[0041] Continue the search and check E iThe size of ' is stopped until the set threshold K is reached, and the adjacent scene set E is obtained i '.

[0042] The present invention can at least achieve one of the following beneficial effects:

[0043] By constructing diverse ship collision avoidance scenarios, verifying the performance of collision avoidance algorithms under different environmental conditions, the test scope is broadened. According to the evaluation feedback of the first round of tests, an optimized scenario test set is constructed and an optimal collision avoidance algorithm is selected. Based on the optimized scenario test set, multiple second-round simulation collision avoidance function tests are carried out on each optimal collision avoidance algorithm respectively. According to the evaluation feedback of the second round of tests, the algorithm performance is evaluated, the optimal collision avoidance algorithm is determined, the algorithm performance is comprehensively evaluated, and the practicality and pertinence of the test are enhanced.

[0044] Through the evaluation feedback mechanism, the complexity degree of the collision avoidance scenario and the performance excellence of the collision avoidance algorithm are determined respectively using the comprehensive evaluation scores of each collision avoidance scenario and the performance scores of each collision avoidance algorithm, effectively screening complex collision avoidance scenarios and high-performance collision avoidance algorithms, avoiding the blindness in the selection of test scenarios and collision avoidance algorithms, and enabling more targeted collision avoidance tests.

[0045] By relying on the results of the F-test, an algorithm with not only a high comprehensive score, but also a relatively stable score distribution and more consistent performance is selected as the optimal collision avoidance algorithm, further ensuring the scientific nature of the selection of the optimal collision avoidance algorithm. And through determining the optimal collision avoidance algorithm to conduct real ship collision avoidance experiments, economic waste is reduced, the navigation safety can be further improved, the adaptability and robustness of the algorithm are enhanced, the influence of human factors is reduced, and a scientific basis is provided for the study of collision avoidance.

[0046] Other features and advantages of the present invention will be described in the subsequent specification, and some advantages can be made obvious from the specification, or understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained from the content specifically pointed out in the specification, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The drawings are only for the purpose of showing specific embodiments and are not considered as a limitation of the present invention. Throughout the drawings, the same reference signs represent the same components.

[0048] Figure 1 is the flowchart of the method of the present invention;

[0049] Figure 2 is the schematic diagram of the composition of the collision avoidance scenario test cases of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] The preferred embodiments of the present invention will be specifically described below in conjunction with the accompanying drawings. The accompanying drawings form a part of this application and are used together with the embodiments of the present invention to explain the principles of the present invention, rather than to limit the scope of the present invention.

[0051] An embodiment of the present invention discloses a method for testing the collision avoidance function of an intelligent navigation system based on an evaluation feedback mechanism, which specifically includes the following steps:

[0052] Step S01: Conduct multiple rounds of simulation collision avoidance function tests on multiple collision avoidance algorithms respectively in multiple ship collision avoidance scenarios, and calculate the comprehensive evaluation scores of each collision avoidance scenario and the first performance scores of each collision avoidance algorithm based on the results of each round of tests;

[0053] Step S02: Construct an optimized scenario test set based on the comprehensive evaluation scores; determine multiple preferred collision avoidance algorithms based on the first performance scores;

[0054] Step S03: Conduct multiple rounds of simulation collision avoidance function tests on each preferred collision avoidance algorithm respectively based on the optimized scenario test set, and obtain the second performance scores of each preferred collision avoidance algorithm based on the results of each round of tests;

[0055] Step S04: Determine the optimal collision avoidance algorithm based on the second performance scores for in-ship collision avoidance experiments.

[0056] In this embodiment, by constructing diverse ship collision avoidance scenarios, the performance of collision avoidance algorithms under different environmental conditions is verified. An optimized scenario test set is constructed and preferred collision avoidance algorithms are selected based on the evaluation feedback of the first round of tests. Multiple rounds of simulation collision avoidance function tests are conducted on each preferred collision avoidance algorithm respectively based on the optimized scenario test set, and the algorithm performance is evaluated according to the evaluation feedback of the second round of tests to determine the optimal collision avoidance algorithm. Through two rounds of test evaluation feedback in this embodiment, the algorithm performance is comprehensively evaluated, enhancing the practicality and pertinence of the tests.

[0057] Specifically, in step SO1, multiple ship collision avoidance scenarios are generated based on the following steps:

[0058] s11: Decompose the collision avoidance scenario;

[0059] Specifically, analyze and disassemble complex collision avoidance scenarios to clarify the scenario elements in the scenarios, including environmental information, navigation information such as the relative course, relative speed, relative distance, and relative bearing between the own ship and the target ship, static obstacle information, dynamic obstacle information, etc. Further, the environmental information includes wind force, wave height, and flow velocity; the static obstacle information includes the relative bearing and relative distance between the own ship and the static obstacle and the number of static obstacles; the dynamic obstacle information includes the number of target ships in a head-on situation with the own ship, the number of target ships in a crossing situation with the own ship, and the number of target ships in an overtaking situation with the own ship;

[0060] S12. Generate multiple test cases corresponding to multiple ship collision avoidance scenarios;

[0061] Specifically, based on the deconstructed scenario elements, using the PICT pairwise combination testing method, by pairwise combining the scenario elements, specific test cases are designed to form a test set, including a head-on scenario test set, a crossing scenario test set, a overtaking scenario test set, and a static obstacle test set automatically generated according to the static obstacle information; each test case corresponds to each ship collision avoidance scenario.

[0062] Further, in step S01, when performing simulation tests based on the test cases corresponding to each collision avoidance scenario, the collision avoidance effects of each collision avoidance algorithm in the same collision avoidance scenario and the collision avoidance effects of the same collision avoidance algorithm in each collision avoidance scenario are tested.

[0063] Exemplarily, the collision avoidance algorithms include: a depth-determining policy gradient algorithm, an improved velocity obstacle model, an artificial potential field method, and a deep reinforcement learning algorithm, etc.

[0064] Further, the calculation of the comprehensive evaluation scores of each collision avoidance scenario and the first performance scores of each collision avoidance algorithm based on each test result includes:

[0065] Calculating the comprehensive collision avoidance benefit score of this test based on the normalized collision avoidance time value, the cumulative route change normalized value, and the cumulative speed change normalized value in each test result;

[0066] Calculating the comprehensive safety score of this test based on the collision risk index CRI in each test result;

[0067] Calculating the evaluation score of the collision avoidance scenario used in this test in this test and the performance score of the collision avoidance algorithm used in this test based on the comprehensive collision avoidance benefit score and the comprehensive safety score of each test;

[0068] Taking the mean of all evaluation scores of each collision avoidance scenario to obtain the comprehensive evaluation score of each collision avoidance scenario; taking the mean of all performance scores of each collision avoidance algorithm to obtain the first performance score of each collision avoidance algorithm.

[0069] Further, the calculation formula for the comprehensive collision avoidance benefit score is:

[0070]

[0071] where F is the comprehensive collision avoidance benefit score, t norm 、 v norm are the normalized collision avoidance duration value, the cumulative course change normalized value, and the cumulative speed change normalized value respectively, and w1 , w 2 , w 3 are the corresponding weights of each parameter respectively.

[0072] Further, the calculation method of the normalized value of the collision avoidance duration is:

[0073]

[0074] where, t act , mint, and maxt are the actual collision avoidance duration, the minimum value of the collision avoidance duration among all test results in this round of testing, and the maximum value of the collision avoidance duration among all test results in this round of testing respectively.

[0075] Further, the calculation method of the normalized value of the cumulative course change is:

[0076]

[0077] where, are the actual cumulative course change in this test, the minimum value of the course change among all test results in this round of testing, and the maximum value of the course change among all test results in this round of testing respectively.

[0078] Further, the calculation method of the normalized value of the cumulative speed change is:

[0079]

[0080] where, v act , minv, and maxv are the actual cumulative speed change in this test, the minimum value of the speed change among all test results in this round of testing, and the maximum value of the speed change among all test results in this round of testing respectively.

[0081] Further, the calculation formula of the comprehensive safety score is:

[0082]

[0083] where, S is the comprehensive safety score, maxCRI and are the maximum CRI value in this test and the average CRI value in this test respectively, w 1 ', w 2 ' are the corresponding weights of each parameter respectively.

[0084] Further, the calculation method of the collision risk index CRI can optionally use the calculation method based on DCPA and TCPA, the fuzzy mathematics calculation method, the grey relational analysis method, etc. Exemplarily, the present invention uses the following formula to calculate the CRI at time t:

[0085] CRI t = 0.36DCPAt +0.32 TCPA t +0.14 D t +0.10 B t +0.08 K t ;

[0086] Among them, the subscript t represents the time t, DPCA is the minimum distance of approach, TCPA is the time to closest point of approach, D is the relative distance, B is the relative bearing, and K is the ship speed ratio, all of which are the test data at time t of this simulation.

[0087] Furthermore, the calculation formula for the evaluation score of the collision avoidance scenario in the test result is:

[0088] G = eF + sS;

[0089] Among them, G is the evaluation score; e is the weight of the comprehensive score of collision avoidance effectiveness; s is the weight of the comprehensive safety score.

[0090] Specifically, for different collision avoidance scenarios, the higher the comprehensive score of collision avoidance effectiveness of the collision avoidance scenario, the longer the collision avoidance duration or the greater the cumulative change in course and speed, indicating that the scenario is more challenging for the effectiveness of the collision avoidance algorithm. If the comprehensive safety score of the collision avoidance scenario is higher, it indicates that the maximum CRI and average CRI of the scenario are greater, and the collision avoidance scenario is more dangerous. Therefore, for a collision avoidance scenario, the larger its evaluation score, the more complex and challenging the scenario is.

[0091] Furthermore, the calculation method for the performance score of the collision avoidance algorithm in the test result is:

[0092] G' = 1 - (eF + sS);

[0093] Among them, G' is the performance score.

[0094] Specifically, for different collision avoidance algorithms, the lower the comprehensive score of collision avoidance effectiveness, the shorter the collision avoidance duration or the smaller the cumulative change in course and speed in the collision avoidance scenario, and the better its effectiveness. If the comprehensive safety score is lower, it indicates that the collision avoidance during the collision avoidance scenario process is safer. For a collision avoidance algorithm, the higher the performance score, the better the performance of the collision avoidance algorithm, and it is the preferred collision avoidance algorithm in this test.

[0095] Furthermore, in step S01, the average value of all evaluation scores of each collision avoidance scenario is obtained to get the comprehensive evaluation score of each collision avoidance scenario; the average value of all performance scores of each collision avoidance algorithm is obtained to get the first performance score of each collision avoidance algorithm.

[0096] Specifically, in step S02, the construction of the optimized scenario test set based on the comprehensive evaluation score includes:

[0097] Determine multiple complex scenarios based on the comprehensive evaluation scores of all ship collision avoidance scenarios; specifically, use the collision avoidance scenarios with comprehensive evaluation scores higher than a set threshold as complex collision avoidance scenarios, and all complex collision avoidance scenarios form a set D = {x 1 , x 2 ,... x i}, i ∈ [1, n]; Exemplarily, the set threshold is 0.6;

[0098] Generate an adjacent scenario set for each complex scenario x i ;

[0099] Determine an optimized scenario test set based on all adjacent scenario sets.

[0100] Furthermore, generating an adjacent scenario set corresponding to each complex scenario includes:

[0101] Set a restricted search area B i for the complex scenario x i , where i is the complex scenario serial number; Exemplarily, for the relative speed V i in this complex scenario, the search area is set to [V i - v i , V i + v i , where v i is the search range value, and for the relative distance D i in this complex scenario, the search area is set to [D i - d i , D i + d i , where d i is the search range value;

[0102] Randomly sample scenario variables within B i to generate multiple sampled scenarios and store them as a sampled scenario set E i , specifically, when storing, store them in the form of an array;

[0103] Use a kd-tree to determine the adjacent scenario set E i ' based on the sampled scenario set E i .

[0104] Furthermore, using a kd-tree to determine the adjacent scenario set E i ' based on the sampled scenario set E i includes:

[0105] Build a kd-tree based on the sampled scenario set E i ;

[0106] Calculate the L2 distance between each sampling scenario and the current complex scenario x i ;

[0107] Use the kd-tree search algorithm to find all sampling scenarios whose L2 distance from the current complex scenario x i is less than the threshold τ to construct the adjacent scenario set E i ';

[0108] Keep searching and check the size of E i ' until the set threshold K is reached and then stop to obtain the adjacent scenario set E i '.

[0109] Furthermore, the calculation formula for the L2 distance is:

[0110]

[0111] where A and B respectively represent the current complex scenario x i and the sampling scenario, J represents the number of parameters in each scenario, a j and b j respectively represent the j-th scenario parameter values in the current complex scenario and the sampling scenario; ||·|| 2 represents the L2 norm.

[0112] Furthermore, for all complex scenarios D = {x 1 , x 2 ,... x i}, the corresponding optimized scenario test set E = {E 1 , E 2 ,... E i} is obtained.

[0113] Specifically, in step S02, determining multiple preferred collision avoidance algorithms based on the first performance score includes: selecting collision avoidance algorithms with a first performance score higher than a preset algorithm threshold as the preferred collision avoidance algorithms.

[0114] Specifically, in step S03, based on the optimized scenario test set E, a second round of multiple simulation collision avoidance function tests are respectively performed on each preferred forcing algorithm, and a second performance score of each preferred collision avoidance algorithm is obtained based on the results of each test in the second round; the calculation method of the second performance score is the same as that of the first performance score and will not be repeated here.

[0115] Specifically, in step S04, determining the optimal collision avoidance algorithm based on the second performance score includes:

[0116] Sort the second performance scores of each preferred collision avoidance algorithm from high to low;

[0117] If the difference in the second performance scores among the top m algorithms does not exceed a specified threshold, the algorithm with the smallest variance in the corresponding score distribution among the top m algorithms is selected as the optimal collision avoidance algorithm through the F-test; otherwise, the algorithm with the highest second performance score is selected as the optimal collision avoidance algorithm.

[0118] It should be noted that if the score distribution of an algorithm shows a large variance, it indicates that the performance of the algorithm under different test conditions may fluctuate significantly, that is, its stability is poor; on the contrary, if the variance of the score distribution is small, it means that the algorithm can maintain relatively consistent performance under various test conditions and has high stability. Therefore, with the help of the results of the F-test, the algorithm with not only a high comprehensive score but also a relatively stable score distribution and more consistent performance is selected as the optimal collision avoidance algorithm.

[0119] Furthermore, in step S04, an optimal collision avoidance algorithm is used to conduct a real ship collision avoidance experiment.

[0120] This embodiment discloses a method for testing the collision avoidance function of an intelligent navigation system based on an evaluation feedback mechanism, which broadens the scope of testing by constructing a variety of ship collision avoidance scenarios and verifying the performance of the collision avoidance algorithm under different environmental conditions. An optimized scenario test set is constructed and a preferred collision avoidance algorithm is selected based on the evaluation feedback from the first round of testing. A second round of multiple simulation collision avoidance function tests are performed on each preferred collision avoidance algorithm based on the optimized scenario test set. The algorithm performance is evaluated based on the evaluation feedback from the second round of testing, the optimal collision avoidance algorithm is determined, and the algorithm performance is comprehensively evaluated, thereby enhancing the practicality and pertinence of the test.

[0121] In the evaluation feedback, the complexity of the collision avoidance scenario and the performance of the collision avoidance algorithm are determined by using the comprehensive evaluation score of each collision avoidance scenario and the performance score of each collision avoidance algorithm respectively, so as to effectively screen complex collision avoidance scenarios and high-performance collision avoidance algorithms, avoid the blindness of test scenario and collision avoidance algorithm selection, and achieve more targeted collision avoidance tests.

[0122] By using the results of the F-test, we screened out the algorithms that not only had high comprehensive scores but also had relatively stable score distributions and more consistent performance as the optimal collision avoidance algorithms, which further ensured the scientific nature of the selection of the optimal collision avoidance algorithms. By determining the optimal collision avoidance algorithms, we conducted actual ship collision avoidance experiments to reduce economic waste, further improve navigation safety, enhance the adaptability and robustness of the algorithms, reduce the impact of human factors, and provide a scientific basis for collision avoidance research.

[0123] System Example

[0124] Another specific embodiment of the present invention discloses a collision avoidance function test system for an intelligent navigation system based on an evaluation feedback mechanism, comprising:

[0125] A scenario generation module for generating multiple ship collision avoidance scenarios for simulating and testing the collision avoidance function;

[0126] A simulation experiment module for conducting the first round of multiple simulation collision avoidance function tests on multiple collision avoidance algorithms respectively under multiple ship collision avoidance scenarios, and conducting the second round of multiple simulation collision avoidance function tests on each selected collision avoidance algorithm respectively based on the optimized scenario test set;

[0127] A function evaluation module for calculating the comprehensive evaluation scores of each collision avoidance scenario and the first performance scores of each collision avoidance algorithm based on the results of each test in the first round, and obtaining the second performance scores of each selected collision avoidance algorithm based on the results of each test in the second round;

[0128] An evaluation feedback module for constructing an optimized scenario test set based on the comprehensive evaluation scores; determining multiple selected collision avoidance algorithms based on the first performance scores; and determining the optimal collision avoidance algorithm based on the second performance scores.

[0129] Specifically, the scenario generation module includes a collision avoidance scenario deconstruction unit and a test case generation unit. The collision avoidance scenario deconstruction unit is used to deconstruct complex collision avoidance scenarios and clarify the scenario elements in the scenarios; the test case generation unit is used to generate multiple test cases based on the deconstructed scenario elements, and each test case corresponds to each collision avoidance scenario of the ship.

[0130] Specifically, the simulation experiment module includes a scenario parameter setting unit, a collision avoidance algorithm storage unit, and an experimental data output unit.

[0131] Further, the scenario parameter setting unit is used to set the ship collision avoidance scenario parameters of the test scenario. First, set the sizes of the environmental parameters, including wind force, wave height, and flow velocity. Then select the test scenario parameters from the test set generated by the scenario generation module according to the encounter types and the number of obstacles to be tested.

[0132] Further, the collision avoidance algorithm storage unit is used to store and manage the collision avoidance algorithms to be subjected to simulation experiments.

[0133] Further, the experimental data output unit is used to output the simulation test results.

[0134] Compared with the prior art, the beneficial effects of an intelligent navigation system collision avoidance function test system based on an evaluation feedback mechanism provided in this embodiment are basically the same as those provided in the method embodiment, and will not be elaborated here one by one.

[0135] It should be noted that the above embodiments are based on the same inventive concept, and for the parts not repeatedly described, reference can be made to each other.

[0136] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for testing the collision avoidance function of an intelligent navigation system based on an evaluation feedback mechanism, characterized in that: The steps include: A first round of multiple simulation collision avoidance function tests were conducted on multiple collision avoidance algorithms in multiple ship collision avoidance scenarios. Based on the test results of the first round, a comprehensive evaluation score of each collision avoidance scenario and a first performance score of each collision avoidance algorithm were calculated. Constructing an optimized scenario test set based on the comprehensive evaluation score; determining multiple preferred collision avoidance algorithms based on the first performance score; Based on the optimized scenario test set, a second round of multiple simulation collision avoidance function tests are performed on each preferred collision avoidance algorithm, and a second performance score of each preferred collision avoidance algorithm is obtained based on the second round of test results; Based on the second performance score, an optimal collision avoidance algorithm is determined to conduct a real ship collision avoidance experiment.

2. The testing method according to claim 1, characterized in that: The comprehensive evaluation score of each collision avoidance scenario and the first performance score of each collision avoidance algorithm calculated based on each test result include: The comprehensive score of the collision avoidance effectiveness of the test is calculated based on the normalized value of the collision avoidance time, the normalized value of the cumulative change of the route and the normalized value of the cumulative change of the speed in each test result; The comprehensive safety score of the test is calculated based on the collision risk index (CRI) in each test result; Based on the comprehensive score of collision avoidance effectiveness and comprehensive score of safety of each test, the evaluation score of the collision avoidance scenario used in the test and the performance score of the collision avoidance algorithm used in the test are calculated; The comprehensive evaluation score of each collision avoidance scenario is obtained by averaging all evaluation scores of each collision avoidance scenario; and the first performance score of each collision avoidance algorithm is obtained by averaging all performance scores of each collision avoidance algorithm.

3. The testing method according to claim 2, characterized in that: The constructing of the optimization scenario test set based on the comprehensive evaluation score comprises: Determining multiple complex scenarios based on the comprehensive evaluation scores of all ship collision avoidance scenarios; Generate a set of adjacent scenes of each complex scene based on the complex scene; The optimized scenario test set is determined based on all adjacent scenario sets.

4. The testing method according to claim 3, characterized in that: Determining the optimal collision avoidance algorithm based on the second performance score comprises: sorting the second performance scores of the preferred collision avoidance algorithms from high to low; If the difference in the second performance scores among the top m algorithms does not exceed a specified threshold, the algorithm with the smallest variance in the corresponding score distribution among the top m algorithms is selected as the optimal collision avoidance algorithm through the F-test; otherwise, the algorithm with the highest second performance score is selected as the optimal collision avoidance algorithm.

5. The testing method according to claim 2, characterized in that: The calculation formula of the collision avoidance effectiveness comprehensive score is: Among them, F is the comprehensive score of collision avoidance effectiveness, t norm , v norm are the normalized value of collision avoidance duration, the normalized value of cumulative heading change and the normalized value of cumulative speed change respectively. w1, w2 and w3 are the corresponding weights of each parameter respectively.

6. The testing method according to claim 5, characterized in that: The calculation formula of the comprehensive safety score is: Among them, S is the comprehensive safety score, maxCRI and are the maximum CRI value and the average CRI value of the test respectively, and w1' and w2' are the corresponding weights of each parameter respectively.

7. The testing method according to claim 6, characterized in that: The calculation formula for the evaluation score of the collision avoidance scenario in this test result is: G = eF + sS; Among them, G is the evaluation score; e is the weight of the comprehensive score of collision avoidance effectiveness; and s is the weight of the comprehensive safety score.

8. The testing method according to claim 7, characterized in that: The calculation method of the performance score of the collision avoidance algorithm in this test result is: G' = 1 - (eF + sS); Where G' is the performance score.

9. The testing method according to claim 3, characterized in that: Generating a set of adjacent scenes corresponding to each complex scene based on the complex scene includes: For complex scenes x i Set a restricted search area B i , where i is the complex scene number; In B i Randomly sample scene variables within, generate multiple sampling scenes and store them as sampling scene set E i ; Using kd-tree based on the sampled scene set E i Determine the adjacent scene set E i '.

10. The testing method according to claim 9, characterized in that: The kd-tree is used based on the sampled scene set E i Determine the adjacent scene set E i 'include: Based on the sampled scene set E i Construct a kd tree; Calculate the difference between each sampled scene and the current complex scene x i The L2 distance between Use the kd tree search algorithm to find all the complex scenes x i The sampled scenes whose L2 distance is less than the threshold τ are used to construct the adjacent scene set E i '; Keep searching, check E i ' until the threshold K is reached and the adjacent scene set E is obtained. i '.

Images