Collision danger dynamic association identification method under complex meeting situation

Abstract

The invention discloses a collision risk dynamic association identification method under a complex encounter situation, which comprises the following steps of: constructing a collision risk degree calculation model based on domain intrusion parameters, and realizing collision risk calculation of a ship and other target ships under a complex scene based on space-time coupling characteristics; dangerous combinations are screened through a space and time dual dynamic threshold mechanism, and high-risk ships are screened; the optimal clustering hierarchy is determined by quantifying the dynamic balance relation between the intra-cluster compactness and the inter-cluster separation degree, the number of the optimal hierarchy is determined, and division of clusters with similar risk degrees is achieved; information entropy weight calculation of comprehensive collision danger indexes is carried out based on an entropy weight method, collision avoidance priority ranking is carried out on the ships subjected to cluster division, and operation of the ships according to collision avoidance priorities is realized based on an improved speed obstacle method; and on the basis of the re-navigation opportunity of the nearest meeting point, after the ship operates according to the collision avoidance priority, the ship returns to the original track to sail.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of maritime traffic risk analysis, and relates to a collision danger dynamic correlation identification method under a complex encounter situation. BACKGROUND

[0002] Autonomous ships replace manual operation with intelligent decision-making systems, which can systematically eliminate human factors that cause navigation accidents, and improve collision risk prevention and control accuracy to sub-second response level.

[0003] Ship collision risk analysis is the core functional module of the autonomous navigation system, and its technical features directly determine the safety, reliability and intelligence level of the autonomous navigation system. With the promotion of the International Maritime Organization (IMO) classification standard for ship autonomy (from auxiliary decision-making to fully autonomous navigation), collision risk analysis technology has evolved from single parameter judgment to multi-dimensional, multi-modal intelligent decision-making system.

[0004] Collision danger identification, as the core decision-making module of the autonomous navigation system, directly defines the safety benchmark and cognitive dimension of the intelligent navigation system. With the deep implementation of the International Maritime Organization (IMO) classification standard for autonomous ships, this field has broken through the traditional DCPA / TCPA single parameter threshold judgment, forming a three-dimensional technical architecture of multi-modal data fusion, dynamic risk field prediction and hybrid enhanced intelligent decision-making. By integrating hydrology and meteorology, ship maneuvering characteristics and complex navigation environment factors, a closed-loop intelligent system of "environment cognition-situation deduction-game decision" is constructed, realizing the leap from static geometric analysis to intelligent perception of navigation situation. This breakthrough in cognitive computing architecture not only reshapes the safety boundary of autonomous navigation, but also becomes the core engine driving the reconstruction of global shipping technology ecology, laying the key technical cornerstone for the intelligent shipping era.

[0005] The ship collision danger analysis technology system has formed a four-dimensional collaborative innovation framework: in the dimension of classical geometric modeling, a dynamic safety evaluation model is established with DCPA (closest point of approach) / TCPA (time to closest point of approach) as the core parameters, a hierarchical safety threshold is established by introducing the ship field theory, and an improved closest point of approach algorithm considering the ship maneuvering characteristics is developed; in the dimension of multi-source information fusion, the traditional sensor data barrier is broken through, multi-modal sensing data such as radar, AIS, camera and sonar are integrated, the extended Kalman filter algorithm is used to realize the spatio-temporal alignment and accuracy compensation of multi-source data, and the ship-shore sensing data is shared in real time through 5G+Internet of Things technology in the port channel scene; in the dimension of intelligent cognitive computing, the implicit risk characteristics in the historical navigation data are mined by using deep reinforcement learning, a collision danger quantitative evaluation function is constructed by combining fuzzy mathematics theory, and the continuity evaluation of risk level is realized; in the dimension of rule reasoning, a hybrid reasoning engine based on COLREGs rules is developed, the international collision avoidance rules are coded into a production rule base, a knowledge graph is constructed combining with expert experience, and a decision suggestion conforming to the practice of navigation is formed by analyzing the rule conflict scene through semantic network. The organic integration of the four-dimensional technology marks the transition of ship collision danger analysis from traditional threshold judgment to cognitive intelligent decision-making, and builds a three-dimensional safety protection system for intelligent navigation system.

[0006] The current ship collision danger analysis technology faces three technical bottlenecks in complex encounter scenarios, which seriously restricts the improvement of navigation safety level.

[0007] (1) The traditional analysis framework excessively focuses on the numerical calculation of collision probability, and has not established the mechanism of the effect of danger correlation on the optimization of decision-making model, resulting in exponential growth of the decision-making complexity of multi-ship cooperative avoidance;

[0008] (2) In the risk quantification level, although geometric probability indicators (DCPA / TCPA), ship field model and velocity obstacle method (VO) have realized single-ship risk assessment, the dynamic analysis capability of multi-ship conflict network topology is generally lacking, and the correlation discrimination criterion between conflict targets has not been established;

[0009] (3) The existing methods fail to establish a time sequence coordination mechanism for ship avoidance actions in complex encounter scenarios, making it difficult to realize the spatio-temporal decoupling of maneuvering response and conflict resolution of action sequence, resulting in a significant decline in the effectiveness of collision avoidance decision-making in complex encounter situations. SUMMARY

[0010] To solve the above problems, the technical scheme adopted by the present application is as follows: a collision danger dynamic correlation identification method under complex encounter situation, comprising the following steps:

[0011] Based on the DDV and TDV double parameters of the spatial intrusion characteristics in the ship field, through the three-dimensional evaluation framework of the spatial intrusion state, time dimension prediction and value dimension representation, combined with the asymmetric Gaussian distribution fitting algorithm, a collision risk calculation model based on field intrusion parameters is constructed to realize the collision risk calculation of the ship and other target ships in complex scenarios based on the space-time coupling characteristics;

[0012] The collision risk of the ship and other target ships in complex scenarios based on the space-time coupling characteristics evaluated by the collision risk evaluation model is screened through the spatial and temporal double dynamic threshold mechanism to screen high-risk ships.

[0013] For the screened high-risk ships, based on the improved hierarchical clustering algorithm, the time collision risk index is integrated into the improved contour coefficient, the dynamic balance relationship between the compactness within the cluster and the separation degree between the clusters is determined to determine the optimal clustering level and the optimal level number, and the division of the cluster with similar time collision risk is realized.

[0014] Based on the entropy weight method, the information entropy weight of the comprehensive collision risk index is calculated, and the ships after the cluster division are prioritized for collision avoidance,

[0015] Based on the improved speed obstacle method, the ship is operated according to the collision avoidance priority;

[0016] Based on the re-meeting point of the re-navigation opportunity, after the ship is operated according to the collision avoidance priority, the ship returns to the original trajectory for navigation.

[0017] Further, the collision risk calculation model based on the field intrusion parameters comprises the following:

[0018] Based on the circular ship field model, the intrusion degree DDV of the target ship to the ship safety field is derived, and the DDV calculation formula is as follows:

[0019] DDV = max(1-DCPA / D ij ,0) (1)

[0020] Wherein, D ij is the ship field radius;

[0021] The membership function r s is established based on the Gaussian equation as follows:

[0022]

[0023] Wherein, a represents the collision risk peak value, b represents the peak position, c represents the variance, and controls the width of the Gaussian curve;

[0024] For any two-ship encounter situation, the TDV calculation formula of the field intrusion remaining time is as follows:

[0025]

[0026] wherein V R represents the relative speed of ship j to ship i;

[0027] The time collision risk set is R u The membership function r u is:

[0028]

[0029] wherein, represents the time for two ships to reach the applicable distance of the situation.

[0030] Further: the process of screening the high-risk ship by screening the dangerous combination through the spatial and temporal double dynamic threshold mechanism is as follows:

[0031] When the spatial collision risk of the target ship to the ship satisfies r s ≥r0, r0 is a preset safety threshold, and the time collision risk reaches r t ≥r0+Δr, the ship pair (i,j) will be automatically marked as a high-risk combination;

[0032]

[0033] wherein Δr is a dynamic adjustment parameter, satisfying and the ship pair sequence set satisfying i≠j is P, satisfying and the ship pair sequence set satisfying i,j∈P,i≠j is P';

[0034] The time similarity function is used to quantify the dispersion degree of the meeting ship:

[0035]

[0036] wherein: is the time dimension collision risk membership function;

[0037] A meeting ship motion similarity quantization model is constructed based on the time dimension collision risk membership function;

[0038] A motion similarity quantization matrix of the meeting ship pair for judging the closeness of the time connection is constructed:

[0039]

[0040] wherein M simThe greater the value, the more similar the risk characteristics of the two ships in time urgency. Only when the ship pair (i, j) belongs to the high-risk set P', the time collision risk correlation is calculated, otherwise it is directly assigned as 0, indicating that there is no significant conflict correlation between the two ships.

[0041] Further: the improved hierarchical clustering algorithm includes the following processes:

[0042] Firstly, each ship in P' is regarded as a cluster (C1, C2, C3, …, C N ), and the initial cluster state C 0 is described as follows:

[0043] C 0 = [C1, C2, C3, …, C N ] (27)

[0044] The calculation of the distance between clusters is described as follows:

[0045]

[0046] Where C m and C n represent clusters m and n respectively; |C m | and |C n | represent the number of ships in clusters C m and C n , is the sum of the time collision risk degrees of the corresponding ships in the two clusters;

[0047] During the kth iteration, a global search is performed on the current risk matrix to determine the maximum inter-cluster distance:

[0048] [p, q] = {p, q | L avg (C p , C q ) = max L avg , p, q ∈ N k-1 , p ≠ q, N k-1 = N - 2 * (k - 1)} (29)

[0049] Where [p, q] is the index with the maximum inter-cluster distance, N k-1 is the number of clusters in the (k-1)th iteration, and the corresponding clusters are merged to form new clusters;

[0050]

[0051] Where, is the merged cluster;

[0052] The updated cluster at the kth iteration is represented as follows:

[0053]

[0054] synchronize and update the similarity matrix of the (k-1)th iteration;

[0055]

[0056] When the remaining object quantity meets the preset termination condition, the iteration process will be terminated, at this time, the remaining objects have met the condition of being uniquely attributed to the corresponding category or not needing further merging, ensuring the certainty of the classification result and the efficiency of the algorithm.

[0057] |N k |≤2 (33)Further: the improved contour coefficient calculation formula is as follows:

[0058]

[0059] wherein C m is a cluster containing ship i, C n is a cluster different from the cluster C m , a(i) is the average value of the time collision risk between ship i and other ships in the cluster C m , and b(i) is the minimum average time collision risk between ship i and other ships except the cluster C m .

[0060] For a cluster containing only a single ship (i.e., |C i | = 1), the contour coefficient is set to zero.

[0061]

[0062] For a candidate cluster set generated by hierarchical clustering Calculate the weighted average value of the improved S under this candidate hierarchy:

[0063]

[0064] After completing the weighted improved contour coefficient calculation of all candidate levels, the optimal level number is determined.

[0065]

[0066] Further: the comprehensive collision risk index expression is as follows:

[0067] r = w s · r s + w t · r t (40)

[0068] wherein r is the comprehensive collision risk index, w s , wt The corresponding spatial collision hazard level r s Time-based collision risk level r t The weight value;

[0069] Since the collision risk index is a negative indicator, it is processed using a negative indicator standardization method during the data standardization phase.

[0070]

[0071] Among them, w1 and w2 are dynamically determined by the entropy weight method;

[0072] Establish a proportional matrix based on standardized data;

[0073]

[0074] The expression for the information entropy of the indicator is as follows:

[0075]

[0076] The final weight expression for the indicator is obtained as follows:

[0077]

[0078] Furthermore: the process of enabling the vessel to operate according to collision avoidance priorities based on the improved speed barrier method is as follows:

[0079] The relative velocity vectors of the two ships, defined from point P O Along V OT The emitted rays are represented as:

[0080] λ(P O (t),V OT (t))={P O (t)+V OT (t)*t|t≥0} (27)

[0081] Where, φ O (t), P O (t), V O (t)}、φ O (t), P O (t) and V O (t) represent the heading, position, and speed of the ship's OS at time t, respectively. T (t), P T (t), V T (t)}、φ T (t), P T (t) and V T (t) represents the heading, position, and speed of the other ship at time t;

[0082] When falling into the area between the two tangent lines and , it is considered that the ship and the target ship have a collision risk, and the relative velocity set V OT that defines the collision of the ship and the target ship is defined as the relative collision zone RCC OT (Relative collision zone), RCC OT is expressed as:

[0083]

[0084] When there are multiple obstacles, it is necessary to convert the RCC of each obstacle into a unified speed space, and convert the relative speed space into the absolute speed space of the ship, so that the speed of the ship encountering the target ship collision obstacle is:

[0085]

[0086] where represents the Minkowski operation, which moves the relative collision cone RCC OT according to the speed vector of the ship, and the speed vector of the ship cannot be in the V OS|TS region;

[0087] Considering the actual maneuvering characteristics of the giving-way ship, according to the navigation practice, the range of the heading change of the giving-way ship taking evasive action is [-π / 3, π / 3], and if the ship is a giving-way ship, the actual allowable course range ACR of the ship is:

[0088]

[0089] where, and represent the upper and lower limits of the ship's heading angle in ACR, respectively;

[0090] The heading obstacle region of the ship relative to the other ship is defined as:

[0091]

[0092] where, and correspond to the left and right heading boundaries of the tangent lines and , respectively.

[0093] Similarly, compared with the COR OT , the safety heading region (Safety course region, SCR) of the ship relative to the other ship is expressed as:

[0094]

[0095] The navigable area of the ship is represented as:

[0096]

[0097] Further, the expression of the re-passing opportunity of the closest point of approach is as follows:

[0098] t r = TCPA + n * t s (34)

[0099] wherein t r is the re-passing opportunity of the ship, n is a positive integer set by the driver autonomously; t s is a time step.

[0100] A collision risk dynamic correlation identification method under a complex encounter situation, comprising:

[0101] A collision risk degree calculation model module based on domain intrusion parameters: for based on the DDV and TDV double parameters of the domain space intrusion characteristics of the ship, through the three-dimensional evaluation framework of the space intrusion state, time dimension prediction and value dimension representation, combined with the asymmetric Gaussian distribution fitting algorithm, a collision risk degree calculation model based on domain intrusion parameters is constructed, realizing the collision risk calculation of the ship and other target ships in a complex scene based on the space-time coupling characteristics;

[0102] A high-risk ship screening module: for the collision risk of the ship and other target ships in a complex scene based on the space-time coupling characteristics evaluated by the collision risk assessment model, the dangerous combination is screened through the space and time double dynamic threshold mechanism, and the high-risk ship is screened;

[0103] A danger degree cluster division module: for the high-risk ships screened, based on the improved hierarchical clustering algorithm, the time collision danger degree index is integrated into the improved contour coefficient, the dynamic balance relationship between the compactness in the cluster and the separation degree between the clusters is determined to determine the optimal clustering level and the optimal level number, and the division of the clusters with similar danger degrees is realized;

[0104] A collision avoidance priority sorting module: for calculating the information entropy weight of the comprehensive collision danger index based on the entropy weight method, and sorting the ships in the divided clusters according to the collision avoidance priority;

[0105] Operation module I: for realizing the operation of the ship according to the collision avoidance priority based on the improved speed obstacle method;

[0106] Operation module II: for realizing the return of the ship to the original trajectory after the operation of the ship according to the collision avoidance priority based on the re-passing opportunity of the closest point of approach.

[0107] A computer device comprises a processor and a memory, the memory storing program modules, the program modules being implemented when the processor runs, realizing any one of the methods.

[0108] The application provides a collision risk dynamic correlation identification method in a complex encounter situation, and the application is aimed at collision risk assessment, a key link in conflict analysis of the encounter situation, and the collision risk calculation model based on the domain intrusion parameter is constructed by comprehensively considering the domain intrusion degree and the domain intrusion residual time, fusing the four-phase theory of collision avoidance, the circular ship domain model and Gaussian fitting, and the collision risk identification precision is significantly improved.

[0109] Compared with the prior art, the application has the following beneficial effects:

[0110] The application focuses on the collision risk assessment problem in the multi-ship encounter conflict analysis, and constructs a real-time calculation model based on the domain intrusion parameter.

[0111] The adaptive clustering level judgment mechanism is constructed by fusing the average link algorithm and the contour coefficient method, the connection strength and the separation degree between clusters are quantified, the dynamic analysis of the conflict network topology and the adaptive judgment of the optimal level are realized, and the accuracy and the calculation efficiency of the target correlation identification are significantly improved.

[0112] The collision risk index weight calculation model is constructed based on the information entropy theory, the index information utility value is dynamically quantified by the entropy weight method, the conflict level and the motion situation data are combined, the limitations of the traditional subjective weighting method are effectively solved, and the objective division of the ship avoidance priority in the same cluster is realized. BRIEF DESCRIPTION OF DRAWINGS

[0113] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0114] Figure 1 is a ship collision avoidance parameter identifier;

[0115] Figure 2 is an average link level clustering;

[0116] Figure 3 is an avoidance action sequence comparison;

[0117] Figure 4 is a navigable area calculation model;

[0118] Figure 5 is a first round clustering family pedigree chart;

[0119] Figure 6 is an improved contour coefficient chart;

[0120] Figure 7 is a second round clustering family pedigree chart;

[0121] Figure 8 is the relative motion trajectory of each giving-way ship to his ship, (a) OS relative motion trajectory space-time characteristics, (b) TS1 relative motion trajectory space-time characteristics, (c) TS5 relative motion trajectory space-time characteristics, (d) TS4 relative motion trajectory space-time characteristics, (e) TS2 relative motion trajectory space-time characteristics, (f) TS7 relative motion trajectory space-time characteristics chart;

[0122] Figure 9 is a relative distance change curve, (a) OS relative distance time sequence evolution mode, (b) TS1 relative distance time sequence evolution mode, (c) TS5 relative distance time sequence evolution mode, (d) T4S relative distance time sequence evolution mode, (e) TS2 relative distance time sequence evolution mode, (f) TS7 relative distance time sequence evolution mode;

[0123] Figure 10 is a relative bearing change curve, (a) OS relative heading dynamic change law, (b) TS1 relative heading dynamic change law, (c) TS5 relative heading dynamic change law, (d) TS4 relative heading dynamic change law, (e) TS2 relative heading dynamic change law, (f) TS7 relative heading dynamic change law curve chart;

[0124] Figure 11is the avoidance effect of two dangerous identification methods, (a) 10 ship initial situation setting, (b) 8 ship encounter initial situation setting;

[0125] Figure 12 is the cluster group clustering and decision flow optimization, (a) 10 ship encounter scenario clustering result; (b) 8 ship encounter scenario clustering result. DETAILED DESCRIPTION

[0126] It should be noted that the embodiments in the present application and the features in the embodiments can be combined with each other without conflict, and the present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0127] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings of the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. The description of at least one exemplary embodiment is actually only illustrative, not as any limitation on the present application and its application or use. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0128] A collision danger dynamic correlation identification method under a complex encounter situation, comprising the following steps:

[0129] S1: Based on the DDV and TDV double parameters of the spatial intrusion characteristics of the ship field, through a three-dimensional evaluation framework of spatial intrusion state, time dimension prediction and value dimension representation, combined with an asymmetric Gaussian distribution fitting algorithm, a collision danger degree calculation model based on field intrusion parameters is constructed, and collision risk calculation based on space-time coupling characteristics of the ship and other target ships in a complex scene is realized;

[0130] S2: For the collision risk based on space-time coupling characteristics of the ship and other target ships in a complex scene evaluated by the collision danger evaluation model, dangerous combinations are screened through a spatial and temporal double dynamic threshold mechanism, and high-risk ships are screened;

[0131] S3: For the screened high-risk ships, based on an improved hierarchical clustering algorithm, the time collision danger degree index is integrated into the improved contour coefficient, the dynamic balance relationship between the compactness within the cluster and the separation degree between the clusters is determined to determine the optimal clustering level and the optimal level number, and the division of the cluster with similar danger degree is realized;

[0132] S4: Based on the entropy weight method, the information entropy weight of the comprehensive collision danger index is calculated, and the ships after cluster division are prioritized for collision avoidance,

[0133] S5: Based on the improved speed obstacle method, the ship is operated according to the collision avoidance priority;

[0134] S6: Based on the re-navigation opportunity of the closest point of approach, after the ship is operated according to the collision avoidance priority, the ship returns to the original track.

[0135] The steps S1 / S2 / S3 / S4 / S5 / S6 are sequentially executed;

[0136] Based on the dynamic evolution of ship collision risk, the traditional collision avoidance stage can be divided into four progressive stages, in which the legal effect of the collision avoidance responsibility starts from the second stage. At this time, the collision risk assessment system is changed from the traditional DCPA / TCPA parameter system to the DDV(Degree of domain violation) / TDV(time to domain violation) two-parameter evaluation model based on the domain space intrusion characteristics of the ship. Through the three-dimensional evaluation framework (space intrusion state, time dimension prediction and magnitude dimension representation), the model constructs a collision risk quantification system that is more in line with maritime practice. Combined with the asymmetric Gaussian distribution fitting algorithm, the spatio-temporal coupling characteristics of collision risk in complex scenarios are accurately captured, providing a more practical evaluation basis for intelligent collision avoidance decision systems.

[0137] The collision risk degree calculation model based on the domain intrusion parameter includes the following:

[0138] Based on the circular ship domain model, the intrusion degree DDV of the target ship to the safety domain of the ship is derived to reflect the interactive relationship between the real-time motion state of the ship and the surrounding environment, as shown in Figure 1 For any pair of meeting ships, the DDV calculation formula is as follows:

[0139] DDV=max(1-DCPA / D ij ,0) (1)

[0140] Wherein, D ij is the domain radius of the ship;

[0141] The membership function r s based on the Gaussian equation is:

[0142]

[0143] Wherein, a represents the peak value of the collision risk degree, b represents the peak position, c represents the variance, and controls the width of the Gaussian curve;

[0144] For any two-ship meeting situation, the TDV calculation formula of the domain intrusion remaining time is as follows:

[0145]

[0146] wherein V R denotes the relative speed of the ship j with respect to the ship i;

[0147] The time collision risk set is R u The membership function r u is:

[0148]

[0149] wherein, denotes the time for the two ships to reach the applicable distance of the situation.

[0150] The process of screening the high-risk ship through the spatial and temporal dual dynamic threshold mechanism is as follows:

[0151] In a complex multi-ship encounter scene, accurately identifying the high-risk collision ship combination is the key prerequisite for starting effective collision avoidance decision. Based on the collision risk assessment model, the dangerous combination is screened through the dual dynamic threshold mechanism. When the spatial collision risk degree of the target ship and the ship satisfies r s ≥r0(r0 is a preset safety threshold), and the time collision risk reaches r t ≥r0+Δr(Δr is a dynamic adjustment parameter), the system will automatically mark the ship pair as a high-risk combination. This method quantifies the spatio-temporal coupling characteristics of spatial collision risk, providing a basis for dynamic analysis of conflict network topology.

[0152] When the spatial collision risk degree of the target ship and the ship satisfies r s ≥r0, r0 is a preset safety threshold, and the time collision risk reaches r t ≥r0+Δr, the ship pair (i,j) will be automatically marked as a high-risk combination;

[0153]

[0154] wherein Δr is a dynamic adjustment parameter, satisfying and the ship pair sequence set P satisfying r t ij ≥r0+Δr and i,j∈P,i≠j is P'.

[0155] Based on the spatio-temporal urgency screening mechanism, only the ship pairs meeting the risk threshold enter the subsequent stage, effectively identifying key target ships and filtering safe interactions. Similarity measurement as the core of the clustering algorithm, the encounter ship dispersion degree is quantified by time similarity, and the time similarity function is:

[0156] d(i,j)=1-r t ij(i,j∈P) (46)

[0157] where: r t ij is the time dimension collision risk membership function;

[0158] A meeting ship motion similarity quantization model is constructed based on the time dimension collision risk membership function;

[0159] A motion similarity quantization matrix of the meeting ship pair of the time contact closeness degree discrimination is constructed:

[0160]

[0161] Wherein M sim The greater the value, the more similar the risk characteristics of the two ships in time urgency, only when the ship pair (i,j) belongs to the high-risk set P', the time collision risk correlation is calculated. Otherwise, it is directly assigned as 0, indicating that the two ships have no significant conflict association, which is convenient for subsequent clustering analysis.

[0162] For the screened high-risk ships, the time collision danger index is integrated into the improved contour coefficient based on the improved hierarchical clustering algorithm, and the dynamic balance relationship between the intra-cluster compactness and the inter-cluster separation degree is determined to determine the optimal clustering level and the optimal level number, and the process of dividing the cluster with similar danger degree is as follows:

[0163] The traditional ship collision risk assessment pays too much attention to the quantification of risk and ignores the relevance of multi-ship interaction. Although methods such as density-based spatial clustering (DBSCAN) and K-Means clustering are widely used, they have problems such as poor adaptability to density changes and sensitivity to initial parameters. The present application proposes an improved hierarchical clustering algorithm, as shown in Figure 2 The algorithm includes four key steps: dynamic average linkage reconstruction of inter-cluster distance, identification and merging of extreme ship combinations, updating of similarity matrix, and adaptive termination threshold setting.

[0164] The improved hierarchical clustering algorithm includes the following processes:

[0165] First, each ship in P' is regarded as a cluster (C1, C2, C3, …, C N ), and the initial cluster state C 0 is described as follows:

[0166] C 0 =[C1, C2, C3, …, C N ] (48)

[0167] The calculation of the inter-cluster distance is described as follows:

[0168]

[0169] where C m and C n represent clusters m and n respectively; |C m | and |C n | denote the number of ships in clusters C m and C n , is the sum of the time collision risk of corresponding ships in two clusters;

[0170] During the kth iteration, a global search is performed on the current risk matrix to determine the maximum inter-cluster distance:

[0171] [p,q] = {p,q | L avg (C p ,C q ) = maxL avg ,p,q ∈ N k-1 ,p ≠ q, N k-1 = N - 2*(k - 1)} (50)

[0172] where [p, q] is the index of the cluster with the maximum inter-cluster distance, N k-1 is the number of clusters in the (k - 1)th iteration, and the corresponding clusters are merged to form a new cluster;

[0173]

[0174] where, is the merged cluster;

[0175] Correspondingly, the updated clusters at the kth iteration are represented as follows:

[0176]

[0177] The similarity matrix of the (k - 1)th iteration is synchronized and updated;

[0178]

[0179] When the number of remaining objects meets the preset termination condition, the iteration process will terminate, at this time, the remaining objects have met the condition of being uniquely attributed to the corresponding category or not needing further merging, ensuring the certainty of the classification result and the efficiency of the algorithm.

[0180] |N k | ≤ 2 (54)

[0181] To optimize the ship cluster division and prevent over-merging, this study proposes an adaptive termination mechanism based on the improved silhouette coefficient. To address the problem of fragmented clusters or over-merging in traditional clustering algorithms in multi-ship encounter scenarios, this paper innovatively integrates the time collision risk index into the silhouette coefficient evaluation system to determine the optimal clustering level by quantifying the dynamic balance between the compactness within the cluster and the separation between clusters. The improved silhouette coefficient S(i) uses linear mapping to maintain monotonicity (0 is the worst, 1 is the best) for ship i. The improved silhouette coefficient calculation formula is as follows:

[0182]

[0183] where C m is the cluster containing ship i, C n is the cluster different from the cluster C m , a(i) is the average time collision risk between ship i and other ships in cluster C m , and b(i) is the minimum average time collision risk between ship i and other clusters except C m .

[0184] For clusters containing only a single ship (i.e., |C i | = 1), the silhouette coefficient is set to zero.

[0185]

[0186] For the candidate cluster set generated by hierarchical clustering Calculate the weighted average of the improved S under this candidate hierarchy:

[0187]

[0188] After completing the weighted improved silhouette coefficient calculation of all candidate levels, determine the optimal level number.

[0189]

[0190] In complex encounter situations, collision avoidance priority division is the core mechanism of ship cooperative collision avoidance. The key is not the single-ship intelligent optimization, but the collaborative transformation of group decision-making through a dynamic priority framework. Since the temporal and spatial urgency of each ship in different clustering clusters is different, their temporal and spatial danger function values are also different. Based on this, an information entropy weight calculation model of the comprehensive collision danger index is constructed based on the entropy weight method, providing theoretical support for priority division and ensuring the balance between safety and efficiency in complex encounter scenarios. The expression of the comprehensive collision danger index is as follows:

[0191] r = w s · r s + w t · r t(61)

[0192] Where r is the comprehensive collision hazard index, w s w t The corresponding spatial collision hazard level r s Time-based collision risk level r t The weight value.

[0193] Because the collision risk index is a negative indicator (i.e., the higher the index value, the higher the risk), it is detrimental to overall safety assessment. Therefore, a negative index standardization method is used during the data standardization phase.

[0194]

[0195] Among them, w1 and w2 are dynamically determined by the entropy weight method;

[0196] Establish a proportional matrix based on standardized data;

[0197]

[0198] The expression for the information entropy of the indicator is as follows:

[0199]

[0200] The final weight expression for the indicator is obtained as follows:

[0201]

[0202] To accurately calculate the decision-making timing and path options for ship collision avoidance and rerouting, this invention proposes a real-time optimization method based on discretized dynamic programming. For example... Figure 3 As shown, the entire decision-making process includes five basic steps: (1) multi-dimensional collision risk assessment; (2) adaptive hierarchical clustering based on average links; (3) priority action ranking based on entropy weight method; (4) avoidance scheme optimization based on improved speed barrier method; and (5) determination of rendezvous timing based on the nearest encounter point. The process is repeated iteratively until all ships can safely leave each other.

[0203] To improve the practical application effectiveness of ship collision avoidance decision models, this invention proposes a dynamic extraction strategy for navigable areas. This strategy integrates the Velocity Obstacle Method with good seamanship requirements, transforming the relative motion state of two ships into a velocity vector field through velocity space transformation, and constructing a navigable area boundary model that includes safety distance constraints, ship maneuverability constraints, and steering amplitude constraints.

[0204] like Figure 4 As shown, the process of enabling the vessel to operate according to collision avoidance priorities based on the improved speed barrier method is as follows:

[0205] Two-ship relative velocity vector, define a point from P O Along V OT The emitted ray is represented as:

[0206] λ(P O (t),V OT (t))={P O (t)+V OT (t)*t|t≥0} (27)

[0207] Where, φ O (t),P O (t),V O (t)}, φ O (t),P O (t) and V O (t) represent the heading, position and speed of the own ship OS at time t, φ T (t),P T (t),V T (t)}, φ T (t),P T (t) and V T (t) are the heading, position and speed of the target ship at time t;

[0208] When falling into the area between the two tangent lines and , it is considered that the own ship and the target ship have a collision danger, and the relative speed set V OT that makes the own ship collide with the target ship is defined as the relative collision zone RCC OT (Relative collision zone), RCC OT is expressed as:

[0209]

[0210] When there are multiple obstacles, it is necessary to convert the RCC of each obstacle into a unified speed space. For the convenience of practical application, it is necessary to convert the relative speed space into the absolute speed space of the ship, so the speed obstacle of the own ship that may collide with the target ship is:

[0211]

[0212] Where represents the Minkowski operation, which moves the relative collision cone RCC OT according to the speed vector of the target ship, and the speed vector of the own ship cannot be in the V OS|TS region;

[0213] Considering the actual maneuvering characteristics of the giving ship, according to the maritime practice, the range of the giving ship's changing course for taking evasive action is [-π / 3, π / 3], if the own ship is the giving ship, the actual allowable course region (ACR) of the ship is:

[0214]

[0215] where, and represent the upper and lower limits of the ship's course angle in the ACR, respectively.

[0216] The course barrier region of the own ship relative to the other ship is defined as:

[0217]

[0218] where, and correspond to the left and right course boundaries of the tangent and , respectively.

[0219] Similarly, compared with the COR OT , the safety course region (SCR) of the own ship relative to the other ship is expressed as:

[0220]

[0221] In order to enable the straight-ahead ship to quickly and accurately identify the intention of the giving ship to take evasive action, the change of the relative course angle between the two ships should be greater than π / 18, which can be considered to meet the spirit of the good seamanship requirements. Therefore, the ship's navigable region is expressed as:

[0222]

[0223] After the giving ship passes through the closest point of approach, the expression of the re-navigation time of the closest point of approach is as follows:

[0224] t r = TCPA + n * t s (34)

[0225] where, t r is the ship's re-navigation time, n is a positive integer set by the driver autonomously; t s is the time step.

[0226] Multi-ship collision avoidance decision essentially belongs to a multi-stage dynamic optimization problem. Its decision-making process can be divided into several related stages according to the time dimension. Each stage needs to make adaptive decisions based on the current situation to form a complete dynamic decision sequence. In order to evaluate the influence of collision risk correlation identification on decision effectiveness, this study constructs a complex simulation scenario including non-coordinated avoidance behavior and multi-ship population interaction, and focuses on quantitative analysis and visualization verification for multi-ship encounter situations. As shown in Table 1, the simulation configures the initial state parameters of 10 ships, including position coordinates, speed, heading and ship field radius, etc. Among them, OS and TS1, TS2 form an encounter situation with similar intersection angles;

[0227] Table 1

[0228]

[0229] At the 30th step of the simulation process, TS3 suddenly turns left, which may cause a chain reaction of cross encounters between TS2 and TS4-TS8, which are continuously approaching; At the same time, TS6 sailing from south to north may also produce a secondary cross encounter situation with OS, TS4 and TS5. By comparing the differences in decision effectiveness of different correlation analysis methods, the system outputs two types of simulation results: one is the quantitative evaluation module, including ship population relationship graph construction, improved contour coefficient calculation, avoidance priority sorting; The second is the visual analysis module, including relative motion trajectory spatiotemporal feature extraction, relative distance time series evolution pattern recognition and relative heading dynamic change rule analysis and other multi-dimensional situation awareness elements. This system builds a full-process closed loop covering collision risk analysis to decision effectiveness evaluation, realizes the systematic verification of the core value of collision risk correlation identification in complex encounter scenarios, and provides theoretical support for the improvement of real-time decision-making ability of intelligent navigation system.

[0230] In order to fully prove the feasibility and practicality of the proposed collision risk correlation identification method, first of all, the collision risk analysis of multi-ship encounter situation is carried out, and the collision risk correlation results are shown in Table 2

[0231] Table 2

[0232]

[0233] At the 35th time step, OS and TS1 meet the preset threshold conditions of cross encounter (rs>0.5, rt>0.5). By calculating the similarity matrix and inter-cluster distance, the family pedigree chart as shown in Figure 5 is constructed, which intuitively presents the ship clustering process and the hierarchical structure of the encounter situation. By calculating the contour coefficient, the ships are divided into two clusters, as shown in Figure 6The first cluster contains TS1, TS2, and TS3, and the second cluster contains TS4, TS5, and TS7. The cluster pedigree chart is shown in FIG. 6. As shown, the cluster pedigree chart shows the clustering process and the hierarchical relationship between the ships. Similarly, the second clustering result also contains two large clusters. Cluster 1 contains TS2 and TS3, and cluster 2 contains TS7 and TS8. Since each cluster contains only two ships, the action sequence is determined according to the International Regulations for Preventing Collisions at Sea. It should be noted that in cluster 1, TS2 is assigned the obligation to give way because TS3 takes non-coordinated action to disrupt the original crossing formation and trigger the automatic collision avoidance protocol. In cluster 2, TS7 has a higher priority according to the crossing encounter provisions. Figure 7 The first cluster contains TS1, TS2, and TS3, and the second cluster contains TS4, TS5, and TS7. The cluster pedigree chart is shown in FIG. 6. As shown, the cluster pedigree chart shows the clustering process and the hierarchical relationship between the ships. Similarly, the second clustering result also contains two large clusters. Cluster 1 contains TS2 and TS3, and cluster 2 contains TS7 and TS8. Since each cluster contains only two ships, the action sequence is determined according to the International Regulations for Preventing Collisions at Sea. It should be noted that in cluster 1, TS2 is assigned the obligation to give way because TS3 takes non-coordinated action to disrupt the original crossing formation and trigger the automatic collision avoidance protocol. In cluster 2, TS7 has a higher priority according to the crossing encounter provisions.

[0234] According to the collision analysis, the overall action plan of each ship is determined using the navigable area calculation model. The simulation results include the optimized trajectory, the relative motion trajectory, and the relative distance change curve, as shown in FIG. 7. Figures 8-10 The relative distance change curve is shown in FIG. 8, where (a) is the relative distance time sequence evolution pattern of OS, (b) is the relative distance time sequence evolution pattern of TS1, (c) is the relative distance time sequence evolution pattern of TS5, (d) is the relative distance time sequence evolution pattern of T4S, (e) is the relative distance time sequence evolution pattern of TS2, and (f) is the relative distance time sequence evolution pattern of TS7. Figure 8 The relative distance change curve is shown in FIG. 8, where (a) is the relative distance time sequence evolution pattern of OS, (b) is the relative distance time sequence evolution pattern of TS1, (c) is the relative distance time sequence evolution pattern of TS5, (d) is the relative distance time sequence evolution pattern of T4S, (e) is the relative distance time sequence evolution pattern of TS2, and (f) is the relative distance time sequence evolution pattern of TS7.

[0235] As can be seen, the giving-way ship's avoidance action includes avoidance and re-navigation throughout the entire avoidance process.

[0236] At the same time, Figure 9 The relative distance change curve is shown in FIG. 8, where (a) is the relative distance time sequence evolution pattern of OS, (b) is the relative distance time sequence evolution pattern of TS1, (c) is the relative distance time sequence evolution pattern of TS5, (d) is the relative distance time sequence evolution pattern of T4S, (e) is the relative distance time sequence evolution pattern of TS2, and (f) is the relative distance time sequence evolution pattern of TS7.

[0237] The relative distance change curve is shown in FIG. 8, where (a) is the relative distance time sequence evolution pattern of OS, (b) is the relative distance time sequence evolution pattern of TS1, (c) is the relative distance time sequence evolution pattern of TS5, (d) is the relative distance time sequence evolution pattern of T4S, (e) is the relative distance time sequence evolution pattern of TS2, and (f) is the relative distance time sequence evolution pattern of TS7. Figure 9 As can be seen from the relative distance change curve given in FIG. 8, the giving-way ship does not infringe on the ship's domain of the other ship during the entire avoidance process, i.e., the minimum encounter distance is greater than the corresponding ship's domain radius, meeting the safety requirements.

[0238] Figure 10are the relative position change curves, (a) OS relative heading dynamic change law, (b) TS1 relative heading dynamic change law, (c) TS5 relative heading dynamic change law, (d) TS4 relative heading dynamic change law, (e) TS2 relative heading dynamic change law, (f) TS7 relative heading dynamic change law curve diagram;

[0239] In addition, from Figure 10 The relative position change curves given by the present patent can be seen that the relative heading change of each giving way ship is greater than π / 18, which meets the requirements of large angle avoidance in the International Regulations for Preventing Collisions at Sea. Therefore, the decision-making method proposed by the present patent can quickly and accurately generate a cooperative collision avoidance decision-making scheme that meets the requirements of the International Regulations for Preventing Collisions at Sea.

[0240] To verify the influence of dangerous association recognition on decision-making performance, the avoidance effects of the two dangerous recognition methods were compared in the 8-ship and 10-ship encounter scenarios. Figure 11 ), Figure 11 are the avoidance effects of the two dangerous recognition methods, (a) 10-ship initial situation setting, (b) 8-ship encounter initial situation setting; the parameter configurations of the initial eight ships remain consistent. The first round of clustering results show that the dynamic clustering mechanism has a significant advantage in collision avoidance decision-making:

[0241] Figure 12 is the cluster group clustering and decision-making process optimization, (a) 10-ship encounter scenario clustering result; (b) 8-ship encounter scenario clustering result.

[0242] (1) The first cluster group only retains OS, TS1, and TS2. Thanks to the adaptive adjustment of the stopping threshold by the contour coefficient method, the decision-making complexity and avoidance time span are reduced, thereby reducing the probability of abnormal behavior optimization.

[0243] (2) The second cluster group maintains TS5, TS4, and TS7, but the action sequence is precisely sorted by the entropy weight method (TS5>TS4>TS7 instead of TS4>TS5>TS7). This method standardizes the collision risk index and gives it a reasonable weight, ensuring the accuracy of the action sequence within the cluster.

[0244] Decision-making performance improvement, as shown in Table 3:

[0245] Table 3

[0246]

[0247] (1)Timeliness: The average first-round action initiation time is reduced from 61.4 steps to 38.0 steps (increased by 38.2%), and small-scale cluster parallel decision eliminates action delay, and the response speed of key vessels (such as OS) is increased by 34%; The average sailing recovery time is reduced from 231.2 steps to 137.3 steps (41.0% reduction), effectively alleviating the interruption of the route and reducing fuel consumption.

[0248] (2)Accuracy: The dynamic danger re-evaluation mechanism monitors the state transition of independent clusters in real time. For example, TS3 is initially marked as an independent cluster, but is reorganized to a TS2-TS3 conflict cluster based on real-time risk assessment at step 81, reducing the collision avoidance time from 169.8 steps to 99.25 steps (41.5% reduction).

[0249] Algorithm architecture improvement;

[0250] The present application uses progressive load distribution to replace the traditional global optimization framework, reduces the first-round maximum cluster size from five ships to three ships, and avoids the computational bottleneck of large-scale cluster serial decision through a phased strategy, significantly improving the real-time performance and scalability of the system.

[0251] A collision danger dynamic correlation identification method under a complex encounter situation, comprising:

[0252] A collision danger degree calculation model module based on domain intrusion parameters: for the DDV and TDV double parameters based on the domain space intrusion characteristics of the ship, through the three-dimensional evaluation framework of space intrusion state, time dimension prediction and value dimension representation, combined with the asymmetric Gaussian distribution fitting algorithm, a collision danger degree calculation model based on domain intrusion parameters is constructed, realizing the collision risk calculation of the ship and other target ships in complex scenarios based on the space-time coupling characteristics;

[0253] A high-risk ship screening module: for the collision risk of the ship and other target ships in complex scenarios based on the space-time coupling characteristics evaluated by the collision danger degree evaluation model, the dangerous combination is screened through the space and time double dynamic threshold mechanism, and the high-risk ship is screened;

[0254] A danger degree cluster division module: for the high-risk ships screened, based on the improved hierarchical clustering algorithm, the time collision danger degree index is integrated into the improved contour coefficient, the dynamic balance relationship between the intra-cluster compactness and the inter-cluster separation degree is determined to determine the optimal clustering level, the optimal level number is determined, and the division of the cluster with similar danger degree is realized;

[0255] An avoidance priority sorting module: for calculating the information entropy weight of the comprehensive collision danger index based on the entropy weight method, and sorting the avoidance priority of the divided cluster ships;

[0256] The operation module I is used for realizing the operation of the ship according to the collision avoidance priority based on the improved speed barrier method.

[0257] The operation module II is used for realizing the return of the ship to the original track after the operation of the ship according to the collision avoidance priority based on the reposition time of the closest point of approach.

[0258] A computer device comprises a processor and a memory, wherein the memory stores program modules, and the program modules are executed by the processor to implement the method.

[0259] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A collision risk dynamic correlation identification method under complex encounter situation, characterized in that: The method comprises the following steps: Based on the DDV and TDV double parameters of the spatial intrusion characteristics of the ship field, a collision risk degree calculation model based on field intrusion parameters is constructed through a three-dimensional evaluation framework of spatial intrusion state, time dimension prediction and value dimension representation, combined with an asymmetric Gaussian distribution fitting algorithm, to realize collision risk calculation of the own ship and other target ships in a complex scene based on time and space coupling characteristics; The collision risk of the own ship and other target ships in a complex scene based on time and space coupling characteristics evaluated by the collision risk evaluation model is screened through a spatial and time dual dynamic threshold mechanism to screen high-risk ships; For the screened high-risk ships, an improved hierarchical clustering algorithm is used to integrate the time collision risk index into the improved contour coefficient, to determine the optimal clustering level and the optimal level number by quantifying the dynamic balance relationship between the compactness within the cluster and the separation degree between the clusters, so as to divide the ships with similar time collision risk degrees. The information entropy weight of the comprehensive collision risk index is calculated based on the entropy weight method, and the ships after division are prioritized for collision avoidance, The own ship is operated according to the collision avoidance priority based on the improved speed obstacle method. Based on the re-navigation time of the nearest encounter point, the own ship returns to the original trajectory after operation according to the collision avoidance priority.

2. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The collision risk degree calculation model based on the field intrusion parameters comprises the following: Based on the circular ship field model, the intrusion degree DDV of the target ship to the own ship safety field is quantified, and the DDV calculation formula is as follows: DDV = max(l - DCPA / D ij , 0) (1) wherein D ij is the radius in the field of ships; Based on the Gaussian equation to establish its membership function r s is: Wherein, a represents the peak value of the collision risk degree, b represents the peak position, and c represents the variance, controlling the width of the Gaussian curve. For any two-ship encounter situation, the TDV calculation formula of the field intrusion remaining time is as follows: where V R denotes the relative speed of ship j with respect to ship i; The time collision risk set is R u The membership function r u is: wherein, denotes the time at which the distance of applicability of the two-ship arrival situation is reached.

3. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The process of screening the dangerous combination and screening the high-risk ships through the spatial and time dual dynamic threshold mechanism is as follows: When the spatial collision risk of the target ship with the own ship satisfies r s ≥r0, r0 is a preset safety threshold, and the time collision risk reaches r t ≥r0+Δr, the ship pair (i, j) will be automatically marked as a high-risk combination. P' = {(i,j) | r t ij ≥ r0+ Δr, i,j e P, i≠ j} (4) wherein Δr is a dynamic adjustment parameter, satisfying and the sequence set of ship pairs with i≠j is P, satisfying r t ij and the sequence set of ship pairs with i≠j is P', satisfying r0+Δr The time similarity function is used to quantify the dispersion degree of the encounter ships, and the function is defined as: d(i,j) = 1 - r t ij (i,j∈P) (5) where: r t ij is the time dimension collision risk membership function; A motion similarity quantization model of encounter ships is constructed based on the time dimension collision risk membership function; A motion similarity quantization matrix of encounter ship pairs is constructed for time contact tightness discrimination: where M sim The larger the value, the more similar the risk profiles of the two ships in terms of time criticality. The time collision risk correlation is only calculated when the ship pair (i,j) belongs to the high-risk set P', otherwise it is directly assigned a value of 0, indicating that the two ships have no significant conflict correlation.

4. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The improved hierarchical clustering algorithm comprises the following process: First, each ship in P' is considered as a cluster (C1, C2, C3, …, C N ), the initial cluster state C 0 is described as follows: C 0 = [C1, C2, C3,..., C N ] (7) The calculation expression of the distance between clusters is as follows: where C m and C n represent clusters m and n, respectively; |C m | and |C n | denote the number of ships in clusters C m and C n , is the total time collision risk for corresponding ships in both clusters. During the kth iteration, the global search is performed on the current danger matrix to determine the maximum distance between clusters: [p,q] = {p,q | L avg (C p ,C q ) = maxL avg ,p,q∈N k-1 ,p≠q,N k-1 = N - 2 * (k - 1)} (9) where [p, q] is the index with the largest inter-cluster distance, N k-1 is the number of clusters at the (k - 1)th iteration, merging the corresponding clusters to form new clusters; wherein is a merged cluster; The cluster representation updated at the kth iteration is as follows: The similarity matrix of the (k-1)th iteration is updated synchronously; When the number of remaining objects meets the preset termination condition, the iteration process will be terminated, at this time, the remaining objects have met the condition of unique attribution to the corresponding category or do not need further merging, ensuring the certainty and efficiency of the classification result. |N k |≤2 (13) 5. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The calculation formula of the improved contour coefficient is as follows: where C m is a cluster containing vessel i, C n is a cluster different from the cluster C m , a(i) is the average of the time collision risk between vessel i and the vessels in the other clusters C m , b(i) is the minimum average time collision risk between vessel i and the vessels in the other clusters, except the cluster C m ; For clusters containing only a single vessel (i.e. |C i | = 1), the profile coefficient is set to zero; For hierarchical clustering generated candidate cluster set Compute the weighted average of S under this candidate hierarchy: After the weighted improved contour coefficient calculation of all candidate levels is completed, the optimal level number is determined.

6. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The expression of the comprehensive collision risk index is as follows: r = w s • r s + w t • r t (20) where r is the comprehensive collision risk index, w s , t are the weight values of the corresponding spatial collision risk r s , temporal collision risk r t . Since the collision risk index is a negative index, a negative index standardization method is used for processing in the data standardization stage: Wherein, w1 and w2 are the weights dynamically determined by the entropy weight method; A proportion matrix is established based on the standardized data; The index information entropy expression is as follows: The index weight expression is obtained:

7. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The process of realizing the operation of the ship according to the collision avoidance priority based on the improved speed obstacle method is as follows: The relative velocity vector of the two ships, defined in point P O Along V OT The emitted ray is represented as: λ(P O (t),V OT (t)) = {P O (t) + V OT (t) * t | t ≥ 0} (27) where φ O (t), P O (t), V O (t) and φ O (t) represent the heading, position and speed of the own ship at time t, respectively; φ O (t), P O (t), V T (t) and φ T (t) represent the heading, position and speed of the other ship at time t, respectively; and T (t) = φ T (t) - φ T (t) = P T (t) - P (t) = V (t) - V (t). When falling into the region between the two tangent lines and , it is considered that the own ship and the target ship have a collision danger possibility, and the relative velocity set V OT that makes the own ship collide with the target ship is defined. OT (Relative collision zone), RCC OT is expressed as: When there are multiple obstacles, it is necessary to convert the RCC of each obstacle into a unified speed space to express the relative speed space into the absolute speed space of the ship, so that the speed obstacle of the ship that may collide with the target ship is wherein represents the Sinekowski operation, which acts to move the relative collision cone RCC according to the own ship's velocity vector OT , the own ship's velocity vector cannot be in the region of V OS|TS . Considering the actual maneuvering characteristics of the giving-way ship, according to the navigation practice, the range of the heading change of the giving-way ship taking evasive action is [-π / 3, π / 3], and if the ship is a giving-way ship, the actual allowable heading range ACR of the ship is: wherein, and respectively represent the upper and lower limits of the ship's heading angle in ACR; The heading obstacle area of the ship relative to the other ship is defined as: wherein, and correspond to the left and right heading boundaries of the tangent and lines, respectively. Similarly, COR OT The safety course region (SCR) of the own ship relative to the other ship is expressed as The ship navigable area is expressed as:

8. The collision risk dynamic correlation identification method under complex encounter situation according to claim 1, characterized in that: The expression of the re-navigation opportunity of the closest point of approach is as follows: t r = TCPA + n * t s (34) Wherein, t r is the ship resumption time, n is a positive integer set by the driver autonomously; t s is the time step.

9. A collision risk dynamic correlation identification method under complex encounter situation, characterized in that, It comprises the following: The collision risk calculation model based on the domain intrusion parameter is constructed: the collision risk calculation model based on the domain intrusion parameter is constructed based on the domain space intrusion characteristics of the ship, the DDV distance intrusion amount and the TDV time intrusion amount are adopted, the three-dimensional evaluation framework of the space intrusion state, the time dimension prediction and the value dimension representation is adopted, the asymmetric Gaussian distribution fitting algorithm is adopted, the collision risk calculation model based on the domain intrusion parameter is constructed based on the time and space coupling characteristics, and the collision risk calculation of the ship and the target ship in a complex scene is realized; The high-risk ship screening module is used for screening the high-risk ships by screening the dangerous combinations through the space and time double dynamic threshold mechanism based on the collision risk evaluation model of the ship and other target ships in a complex scene based on the time and space coupling characteristics. The danger degree cluster division module is used for dividing the high-risk ships screened based on the improved hierarchical clustering algorithm, integrating the time collision danger degree index into the improved contour coefficient, determining the optimal clustering level by quantifying the dynamic balance relationship between the intra-cluster compactness and the inter-cluster separation degree, determining the optimal level number, and realizing the division of the clusters with similar danger degrees. The collision avoidance priority sorting module is used for calculating the information entropy weight of the comprehensive collision danger index based on the entropy weight method, and sorting the ships in the divided clusters according to the collision avoidance priority. The operation module I is used for realizing the operation of the ship according to the collision avoidance priority based on the improved speed obstacle method. The operation module II is used for realizing the return of the ship to the original track after the operation of the ship according to the collision avoidance priority based on the re-navigation opportunity of the closest point of approach.

10. A computer device comprising: The processor and the memory store program modules, and the program modules realize the method of any one of claims 1-8 when the processor runs.

Images