Ship collision avoidance optimization method under condition of uncertain obstacle ship motion information

Abstract

The invention discloses a ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information, and the method comprises the steps: constructing a membership function for measuring the ship collision risk degree, so as to construct a ship collision risk degree mathematical model, and obtaining the current ship collision risk degree; based on international maritime collision avoidance rules, according to a preset ship collision risk degree range, ship collision avoidance danger stages and ship encounter situations and way-giving relations corresponding to the ship collision avoidance danger stages are divided and obtained, and support is provided for formulating an autonomous collision avoidance strategy for the intelligent ship under different collision avoidance situations by combining a DWA collision avoidance algorithm and a VO collision avoidance algorithm; aiming at the condition that the ship navigation state has discontinuous output, a CNN-GRU ship navigation state prediction network is provided, the ship navigation state and the uncertainty of the state are predicted based on AIS historical data, and the uncertain distribution of the position of the ship navigation state is obtained, so that under the condition that a moving obstacle cannot be continuously perceived, the ship navigation state can be accurately predicted. It can still be ensured that the intelligent decision-making system provides effective collision avoidance decision support for the intelligent ship, the uncertainty of the ship position is fully considered, and the ship field is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent ship collision avoidance, and in particular to a ship collision avoidance optimization method under the condition of uncertain motion information of an obstacle ship. Background Art

[0002] During autonomous navigation, smart ships can rely on a variety of sensors to accurately locate themselves and detect navigation obstacles. However, they often encounter the problem of difficulty in continuously perceiving obstacle signals during collision avoidance, which makes it difficult for smart ships to accurately and in real time understand the navigation environment and encounter situation. This perception limitation poses a challenge to the effectiveness of collision avoidance algorithms based on observable obstacle motion states. Summary of the invention

[0003] The present invention provides a ship collision avoidance optimization method under the condition of uncertain motion information of an obstacle ship, so as to overcome the above technical problems.

[0004] In order to achieve the above object, the technical solution of the present invention is:

[0005] A ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information specifically comprises the following steps:

[0006] S1: Detect and obtain the status information of the obstacle ship and the target smart ship in the sensing area of ​​the navigable waters of the target smart ship;

[0007] The state information includes at least the ship position, ship speed and ship attitude;

[0008] S2: According to the status information of the obstacle ship and the target intelligent ship, a membership function for measuring the ship collision risk is constructed, and a ship collision risk mathematical model is constructed based on the membership function to obtain the current ship collision risk;

[0009] S3: Based on the International Maritime Collision Avoidance Regulations, the preset ship collision risk range is divided and the ship collision avoidance risk stage and the ship encounter situation and give way relationship corresponding to the ship collision avoidance risk stage are obtained;

[0010] The ship collision avoidance danger stage includes the safety stage, the collision danger stage, the emergency situation stage and the imminent danger stage;

[0011] S4: Based on the ship collision avoidance danger stage, confirm whether the target smart ship and the obstacle ship are in a safe stage according to the current ship collision risk;

[0012] If yes, the target smart ship is controlled to sail according to the originally set route;

[0013] Otherwise, execute step S5;

[0014] S5: confirm whether the obstacle ship cannot be stably tracked;

[0015] If so, the trained CNN+GRU network model based on the combination of the convolutional neural network (CNN) and the gated recurrent neural network (GRU) is used to obtain the status information of the obstacle ship that is difficult for the smart ship to obtain in real time due to the loss of detection signals;

[0016] And according to the state information of the obstacle ship obtained in real time, the real-time ship collision risk with the target smart ship is obtained, and step S6 is continued;

[0017] If not, directly execute step S6;

[0018] S6: Based on the real-time ship collision risk level / current ship collision risk level, determine whether the target smart ship and the obstacle ship are in an urgent danger stage;

[0019] If yes, the current achievable speed space of the target smart ship is obtained based on the DWA algorithm; and the optimal speed pair in the current achievable speed space is selected based on the objective function of the DWA algorithm;

[0020] If not, the DWA algorithm is improved based on the VO algorithm to obtain the current achievable speed space of the target smart ship, and the objective function of the local trajectory planning of the smart ship is constructed to select the optimal speed pair in the current achievable speed space;

[0021] S7: Based on the selected optimal speed pair in the current achievable speed space, ship collision avoidance is achieved under the condition of uncertain motion information of the obstacle ship.

[0022] Furthermore, the membership function constructed in S2 includes a membership function of the closest encounter distance of a ship, a membership function of the closest encounter time, a membership function of the distance of a dangerous encounter area, and a membership function of a relative true azimuth angle and a membership function of a relative ship speed ratio that affect the navigation safety of the smart ship;

[0023] The membership function D(DCPA) of the closest encounter distance of ships is:

[0024]

[0025] Where: d1 represents the distance of the ship area that can perform collision avoidance maneuvers, d2 represents the distance of the area where the collision avoidance maneuvers are performed at the latest, and d2 = nd1 (n = 1, 2, ..., k);

[0026] The closest encounter time membership function D(TCPA) is:

[0027]

[0028] Where: t1 and t2 represent the lower and upper bounds of the closest encounter time TCPA respectively; represents the relative velocity modulus of the obstacle ship relative to the own ship, and

[0029]

[0030] The dangerous encounter area distance membership function D(d) is:

[0031]

[0032] The relative true azimuth membership function D(R c ) and the relative ship speed ratio membership function D(K) is:

[0033]

[0034] Where: R c represents the relative true position of the obstacle ship and the target smart ship; K represents the relative speed ratio of the obstacle ship and the target smart ship; Q c Indicates that the obstacle ship is located at the true position of the target smart ship;

[0035] The mathematical model of ship collision risk constructed based on the membership function is:

[0036] CRI=ω1D(DCPA)+ω2D(TCPA)+ω3D(d)+ω4D(R c )+ω5D(K)

[0037] Where: CRI represents the risk of ship collision; ω1~ω5 represent the preset weight values.

[0038] Furthermore, based on the international maritime collision avoidance regulations described in S3, the ship collision avoidance risk stage Θ is divided and obtained according to the preset ship collision risk range, and its expression is:

[0039]

[0040] The relationship between the ship encounter situation and the give-way in the dangerous stage of ship collision avoidance is:

[0041]

[0042] Furthermore, the CNN+GRU network model constructed based on the convolutional neural network CNN and the gated recurrent neural network GRU in S5 has the following training process:

[0043] S51: Obtain historical ship time-series trajectory sequence data, perform normalization processing on it to obtain a standard sample set, and input the standard sample set into the CNN+GRU network model for training;

[0044] And the CNN+GRU network model includes an encoder unit and a decoder unit;

[0045] The encoder unit includes a convolutional neural network layer, a double-layer GRU network layer, and a positional Dropout network layer connected in sequence;

[0046] The convolutional neural network layer is used to extract data features from the ship time-series trajectory sequence data in the standard sample set;

[0047] The data features include ship position, ship speed and ship attitude;

[0048] The double-layer GRU network layer is used to further extract the data features of the ship time series trajectory sequence from the output of the convolutional neural network layer, and perform feature encoding on it to obtain an intermediate vector;

[0049] The position Dropout network layer is used to ignore some hidden neurons in proportion during the training process of the CNN+GRU network model to obtain an uncertainty estimation network model for the obstacle ship motion information;

[0050] The decoder unit is specifically a cyclic network structure composed of a single-layer GRU and a fully connected layer;

[0051] The single-layer GRU is used to perform a feature decoding operation on the intermediate vector;

[0052] The fully connected layer is used to extract the output of the single-layer GRU to obtain the state prediction information of the obstacle ship at the next moment;

[0053] S52: The mean square error function is used as a loss function to determine whether the output of the trained CNN+GRU network model converges, and the trained CNN+GRU network model with converged output is used as the final CNN+GRU network model.

[0054] Furthermore, the DWA algorithm described in S6 is improved based on the VO algorithm to obtain the current achievable speed space of the target smart ship, specifically:

[0055] S61: Define the position of the obstacle ship at the current time t as P t , whose expression is

[0056]

[0057] Where: N represents the acquisition period and N = (t-t0) / Δt; Δt represents the duration of the intelligent ship control period; t0 represents the start time of obstacle ship prediction; P TS Indicates the predicted starting position of the obstacle ship;

[0058] The speed interval v where the target smart ship OS collides with the obstacle ship TS isOS , expressed as

[0059]

[0060] Where: D(P,r sum / (t-t0)) represents the overlapping area between the target smart ship and the obstacle ship. Represents the Minkowski vector operator symbol; r sum represents the sum of the ship areas of the target smart ship and the obstacle ship; P OS (t0) represents the position of the target smart ship at time t0; P TS (t0) represents the position of the obstacle ship at time t0;

[0061] S62: The next moment’s ship speed v of the obstacle ship predicted by the CNN+GRU network model TS It obeys Gaussian distribution, that is

[0062] Then the position distribution of the obstacle ship is [P TS (t)-(v e -1.96σ x )Δt,P TS (t)+(v e +1.96σ x )Δt]

[0063] Among them, v e represents the mean of the navigation speed obtained from multiple prediction results; σ represents the standard deviation of the navigation speed;

[0064] According to the position distribution of the obstacle ship, determine t n The constraints satisfied by the target smart ship position and the obstacle ship position at the moment are:

[0065] S63: Based on the constraint conditions, the target smart ship OS and the obstacle ship TS collide according to the speed interval v OS , obtain the speed optional interval VNLHO considering the uncertainty of the dynamic motion information of the obstacle ship, and use it as the current achievable speed space of the target smart ship, and its expression is:

[0066]

[0067] Furthermore, the objective function J of the local trajectory planning of the intelligent ship is constructed in S6, and its expression is

[0068] minJ=ω v J v +ω ψ J ψ

[0069] J v =|Δv OS |=|v OS (t+Δt)-v OS (t)|

[0070] J ψ =|Δψ OS |=|ψ OS (t+Δt)-ψ OS (t)|

[0071] Where: J v ,J ψ They represent the defined speed change and heading change respectively; v represents the achievable linear velocity; ψ represents the feasible angular velocity; ω v ,ω ψ represents the weight parameter; Δv OS Indicates the change in achievable linear velocity; Δψ OS Represents the possible angular velocity change.

[0072] Beneficial effects: The present invention provides a ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information. By constructing a membership function for measuring the ship collision risk, a ship collision risk mathematical model is constructed to obtain the current ship collision risk; and based on the International Maritime Collision Avoidance Rules, the ship collision risk stage and the ship encounter situation and give way relationship corresponding to the ship collision risk stage are divided and obtained according to the preset ship collision risk range. By combining the DWA collision avoidance algorithm and the VO collision avoidance algorithm, support is provided for the formulation of autonomous collision avoidance strategies for smart ships under different collision avoidance situations. In addition, in view of the situation where there is a discontinuous output of the ship's navigation status, a CNN-GRU ship navigation status prediction network is proposed. The ship's navigation status and the uncertainty of the status are predicted based on the AIS historical data to obtain the uncertainty distribution of its position, so that when the moving obstacle cannot be continuously perceived, the intelligent decision-making system can still provide effective collision avoidance decision support for the smart ship, fully consider the uncertainty of the ship's position, and optimize the ship field. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0074] Figure 1 It is a flow chart of the ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information of the present invention;

[0075] Figure 2 This is a technical flow chart of the ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information in this embodiment;

[0076] Figure 3 This is a schematic diagram of the DWA algorithm search space in this embodiment;

[0077] Figure 4 This is a schematic diagram of a dynamic ship field model in this embodiment;

[0078] Figure 5 This is a simplified schematic diagram of the collision avoidance waypoints in this embodiment;

[0079] Figure 6 This is a schematic diagram of the CNN-GRU prediction network structure of the ship motion state considering trajectory uncertainty in this embodiment;

[0080] Figure 7 This is a schematic diagram of the uncertainty of ship trajectory prediction and motion state estimation process in this embodiment;

[0081] Figure 8 This is a schematic diagram of the position distribution of obstacle ships in this embodiment;

[0082] Fig. 9 Schematic diagram of the intelligent ship collision avoidance algorithm in the case of uncertain motion information of the obstacle ship in this embodiment;

[0083] Fig.10 This is a schematic diagram of the ship field considering the uncertainty expansion of the obstacle ship in this embodiment;

[0084] Fig.11 This is a schematic diagram of the encounter situation when the navigation status information of the blue ship is lost in this embodiment;

[0085] Fig.12 This is a schematic diagram of the encounter situation when the navigation status information of the red ship is lost in this embodiment;

[0086] Fig.13 This is a schematic diagram of the ship avoiding the ship in the blue area in this embodiment;

[0087] Fig.14 This is a schematic diagram of the ship passing through the clear blue area in this embodiment;

[0088] Fig.15 This is a schematic diagram of the ship avoiding the ship in the red area in this embodiment;

[0089] Fig.16 This is a schematic diagram of the ship passing through the red area and letting the ship clear in this embodiment. DETAILED DESCRIPTION

[0090] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0091] This embodiment provides a ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information, such as Figure 1 to Figure 2 As shown, the specific steps include:

[0092] S1: Detect and obtain status information of obstacle ships and target smart ships in the sensing area of ​​the navigable waters of the target smart ship; the status information includes at least the ship position, ship speed and ship attitude;

[0093] S2: According to the status information of the obstacle ship and the target intelligent ship, a membership function for measuring the ship collision risk is constructed, and a ship collision risk mathematical model is constructed based on the membership function to obtain the current ship collision risk;

[0094] Specifically, although the DWA algorithm is widely used in robot path planning tasks, it still has certain limitations. The DWA algorithm mainly generates speed commands based on dynamic windows, and calculates the optimal motion strategy by considering the dynamic constraints of the smart ship itself and the obstacles in the environment. It usually has a great advantage in quickly responding to dynamic environments. However, unlike land robots, smart ships sailing at sea have a long collision avoidance process and should follow the International Maritime Organization. (2002). International Regulations for Preventing Collisions at Sea (COLREGs). International Maritime Organization. This has certain limitations for the DWA algorithm that performs path planning based on dynamic constraints.

[0095] Therefore, in order to solve the above problems, the DWA algorithm is integrated with the VO algorithm in this embodiment. The VO and DWA algorithms are used in the stage when the smart ship follows the international maritime collision avoidance regulations. In the "urgent danger zone" where there is a serious collision risk between the two ships and all available actions need to be taken, the VO constraint is abandoned and only the DWA algorithm that can quickly respond to environmental dynamics is used. To achieve the above goals, the collision risk should be re-divided first, and the application time of each sub-algorithm should be clarified (such as Figure 2 shown).

[0096] First, the collision risk is constructed. The closest approach distance DCPA between two ships is used as one of the factors to measure the collision risk, and the closest approach distance membership function is:

[0097]

[0098] Where: d1 represents the range distance where collision avoidance maneuvers can be performed, d2 represents the range distance where collision avoidance maneuvers can be performed at the latest, and d2 = nd1 (n = 1, 2, ..., k). The setting of this range fully considers factors such as the size, speed, maneuverability and encounter situation of the ship;

[0099] In this embodiment, only the dynamic changes of the collision avoidance dimensions in the bow and stern directions of the obstacle ship or the target intelligent ship are considered, and it is simplified into a ship field model consisting of a semi-ellipse and a semi-circle:

[0100]

[0101] Where: R f ,R s The ellipse parameter radius and the circle parameter radius representing the ship domain model represents the coordinates of the siege ellipse; Representation and Respectively represent the horizontal and vertical coordinates of the ellipse; Q represents the set {R f ,R s};like Figure 4 As shown in the figure, sgn(x) represents the sign function. The ellipse and circle radii fully consider the ship size, the relative speed between the own ship and the target ship, the own ship speed, the encounter situation and other parameters. The dynamic radius can be described as

[0102]

[0103] Where: AD represents the ship's advance distance; V OS represents the speed of the ship; L represents the length of the ship; S(i) represents the encounter situation piecewise function, and

[0104]

[0105] Where: R c Indicates relative true direction; V r Indicates the relative speed of the obstacle ship relative to own ship;

[0106] In addition, TCPA is another important indicator that affects the collision risk. The closest encounter time membership function is:

[0107]

[0108] Define t1 and t2 as the lower and upper bounds of the closest encounter time TCPA, respectively. is the relative velocity modulus of the obstacle ship relative to the own ship, and its mathematical expression is

[0109]

[0110] Define the distance between the obstacle ship and the target smart ship as d, and d2 as the distance for the latest collision avoidance operation. If the distance between the two is within the dangerous encounter distance, it is judged as absolutely dangerous. If it exceeds d2, it is judged as no risk. That is, the membership function of the dangerous encounter distance is:

[0111]

[0112] In addition, the relative true bearing R c The ship speed ratio K also affects the navigation safety of smart ships. c For the obstacle ship to be located at the true azimuth of the target intelligent ship, the membership function of the relative true azimuth angle and the membership function of the relative ship speed ratio are designed as follows:

[0113]

[0114] The membership functions of all the above indicators are D∈[0,1], D(DCPA), D(TCPA), D(d), D(R c ) and D(K) are assigned weights ω1 to ω5 respectively, and the collision risk can be expressed mathematically as:

[0115] CRI=ω1D(DCPA)+ω2D(TCPA)+ω3D(d)+ω4D(R c )+ω5D(K)

[0116] S3: Based on the International Maritime Collision Avoidance Regulations, the preset ship collision risk range is divided and the ship collision avoidance risk stage and the ship encounter situation and give way relationship corresponding to the ship collision avoidance risk stage are obtained;

[0117] The ship collision avoidance danger stage includes the safety stage, the collision danger stage, the emergency situation stage and the imminent danger stage;

[0118] Specifically, after obtaining the collision risk index CRI, according to the international maritime collision avoidance regulations, this study divides the different stages of collision avoidance risk into the following stages, as shown in Table 1;

[0119] Table 1. Basis for classification of different stages of collision risk

[0120]

[0121] The expressions for the different stages of the collision hazard are:

[0122]

[0123] The judgment of the specific encounter situation and the relationship of giving way, that is, the ship collision avoidance rules corresponding to the dangerous stage of ship collision avoidance are as follows:

[0124]

[0125] Others = {Q c , R c Not included in the above cases} OS: Go straight

[0126] Based on the ship collision avoidance rules, the specific turning point position can be calculated according to the obstacle ship and the dynamic ship field. Specifically, for the convenience of calculation, the waypoints are selected such as Figure 5 As shown in the figure, when the own ship passes the target ship on the port side or starboard side, the waypoints that the own ship should select are wt1 and wt2 respectively; s is the radius of the passing point, N is the next target point, P is the vector from the obstacle ship position coordinate P2 to the own ship (target smart ship) position coordinate P1, and the position calculation formulas of the passing points wt1 and wt2 are as follows;

[0127]

[0128] S4: Based on the ship collision avoidance danger stage, confirm whether the target smart ship and the obstacle ship are in a safe stage according to the current ship collision risk, such as Fig. 9 As shown;

[0129] If yes, the target smart ship is controlled to sail according to the originally set route;

[0130] Otherwise, execute step S5;

[0131] S5: confirm whether the obstacle ship cannot be stably tracked;

[0132] If so, a CNN+GRU network model based on the trained convolutional neural network CNN and the gated recurrent neural network GRU is used to obtain the motion state information of the obstacle ship in the future time that is difficult for the smart ship to obtain in real time due to the loss of detection signals;

[0133] And according to the state information of the obstacle ship obtained in real time, the real-time ship collision risk with the target smart ship is obtained, and step S6 is continued;

[0134] If not, directly execute step S6;

[0135] In a specific embodiment, the CNN+GRU network model constructed based on the combination of the convolutional neural network CNN and the gated recurrent neural network GRU has a training process as follows:

[0136] S51: Obtain historical ship time-series trajectory sequence data, normalize it to obtain a standard sample set, and input the standard sample set into the CNN+GRU network model for unsupervised learning training;

[0137] And the CNN+GRU network model includes an encoder unit and a decoder unit;

[0138] The encoder unit includes a convolutional neural network layer, a double-layer GRU network layer, and a positional Dropout network layer connected in sequence;

[0139] The convolutional neural network layer is used to extract data features from the ship time-series trajectory sequence data in the standard sample set;

[0140] The data features include ship position, ship speed and ship attitude;

[0141] The double-layer GRU network layer is used to further extract the data features of the ship time series trajectory sequence from the output of the convolutional neural network layer, and perform feature encoding on it to obtain an intermediate vector;

[0142] The position Dropout network layer is used to ignore some hidden neurons in proportion during the training process of the CNN+GRU network model to obtain an uncertainty estimation network model for the obstacle ship motion information;

[0143] The decoder unit is specifically a cyclic network structure composed of a single-layer GRU and a fully connected layer;

[0144] The single-layer GRU is used to perform a feature decoding operation on the intermediate vector;

[0145] The fully connected layer is used to extract the output of the single-layer GRU to obtain the state prediction information of the obstacle ship at the next moment;

[0146] S52: The mean square error function is used as a loss function to determine whether the output of the trained CNN+GRU network model converges, and the trained CNN+GRU network model with converged output is used as the final CNN+GRU network model.

[0147] In this embodiment, the CNN-GRU prediction model mainly includes two parts: an encoder and a decoder. In the encoder, the feature codes {position, speed, posture} in the input time series data are extracted and converted into intermediate vectors, and used as the input of the encoder. The function of the decoder is to decode the intermediate vectors, combine them with the input data at the current moment, and predict the state sequence at the next moment. In this embodiment, CNN+GRU is used in the encoder structure to extract features of the ship's motion state. CNN has the characteristics of parameter sharing and memory when processing time series. In this application, the kernel size of the convolution kernel layer is 3×3, the output channel is 512, the step size is 1, the expansion rate is 2, and the Relu activation function is used. In the encoder, this application adopts a double-layer GRU network structure such as Figure 6 , the hidden layer neurons in the first layer of GRU are 128, and the second layer of GRU is composed of 64 hidden layer neurons. The decoder part uses a recurrent network structure composed of a single-layer GRU and a fully connected layer, in which the number of GRU network layers is set to 128, and the number of neurons in the fully connected layer is set to 128. The tanh function is used as the activation function, the mean square error function is used as the loss function, and Adam is used as the optimization function. The ship state prediction information predicted by the above method is affected by data noise, the accuracy of the prediction model itself, and the uncertainty of ship motion, and the result is bound to have a certain error. At present, due to the lack of estimation means for the uncertainty of ship motion state prediction, it is difficult to quantify this uncertainty. In order to ensure the navigation safety of intelligent ships at sea, the usual technical means is to leave a large safety threshold to avoid potential risks. However, this approach will increase the area of ​​non-navigable waters in ship collision avoidance and affect collision avoidance decisions. This embodiment intends to set a priori probability distribution on the model parameter weights, and then estimate the uncertainty of model output changes caused by changes in input data. Secondly, the Dropout algorithm (position Dropout network layer) is used to estimate the uncertainty of the prediction results, and then the ship trajectory area in the future time period (such as Figure 8 The uncertainty of the prediction result, the predicted value of heading and speed are shown in Figure 1. The Dropout algorithm is used to discard some neurons in the training network according to a certain probability, so that the network structure in the training process is different, which leads to certain differences in the results of network training. The complete prediction result uncertainty estimation network model is as follows: Figure 7 The Dropout algorithm in the network ignores some hidden neurons in the GRU network in proportion, which causes the ship motion state information extracted by the convolution layer to not enter the same sub-network every time, reducing the dependence on fixed nodes when updating parameters, thereby obtaining the uncertain distribution of prediction results of different sub-networks at future time points, so as to obtain the prediction information of the obstacle ship state at the next moment;

[0148] The specific expression for obtaining the uncertain distribution at a future time point is:

[0149]

[0150] Where: E(x t+1:t+k ) represents the expected value of N predictions; x t+1:t+k represents the predicted position coordinates; σ t+1:k+1 Represents the uncertainty position of the prediction result, which is used to calculate the covariance matrix Q t+i ;

[0151]

[0152] Where: t+i,x , σ t+i,y Respectively represent the position abscissa and position ordinate corresponding to the uncertainty position;

[0153] According to the covariance matrix Q t+i Obtain the ellipse confidence rate of the ship position prediction result;

[0154] That is, assume that the uncertainty follows a multivariate normal distribution, the ellipse confidence rate is s, and the center of the ellipse is obtained by taking the average of N prediction results as Then the ellipse confidence rate s of the ship position prediction result is obtained, and then the prediction result of the ship's motion state in the future time is obtained;

[0155] And the expression of the ellipse confidence rate s of the ship position prediction result is

[0156]

[0157] Where: Indicates the x-axis coordinate at time t+1; Indicates the y-axis coordinate at time t+1;

[0158] In the process of autonomous navigation, the smart ship of this embodiment needs to continuously use a variety of sensors to perceive the motion status of itself and other ships. However, in the process of collision avoidance, the detection signal may be lost, making it difficult for the smart ship to obtain the motion status information of the obstacle ship in real time. At this time, the motion state of the obstacle ship is highly uncertain. At this time, the smart ship collision avoidance algorithm under the condition of known obstacle motion information will no longer be applicable. To this end, this embodiment has predicted the uncertainty of the obstacle ship's trajectory and the ship's speed and track direction that are highly related to the collision avoidance decision. This embodiment will make full use of the prediction results to further optimize the collision avoidance decision algorithm studied in the previous article. Taking into account that the above-mentioned method has the assumption of "maintaining direction and speed" for the moving obstacle ship, but in fact the movement speed v of the obstacle ship TS It is time-varying, and its moving speed can be sensed by the smart ship in real time.

[0159] S6: Based on the real-time ship collision risk / current ship collision risk, and in combination with the ship collision risk stage obtained according to the preset ship collision risk range, it is determined whether the target smart ship and the obstacle ship are in an urgent danger stage;

[0160] If yes, the current achievable speed space of the target smart ship is obtained based on the DWA algorithm; and the optimal speed pair in the current achievable speed space is selected based on the objective function of the DWA algorithm;

[0161] Specifically, the DWA algorithm in this embodiment takes into account the acceleration and speed limit of the ship. Its principle is to search for feasible solutions for the planned path in the speed space composed of the plane speed and the rotational speed. Multiple groups of reachable speed vector pairs (v, w) constitute the reachable speed space of the smart ship. Each group of speed pairs contains the posture information within the motion cycle of the smart ship. The smart ship sets certain constraints to ensure that the only feasible solution is searched in the speed space, and then obtains the corresponding motion instructions, thereby ensuring the safety of the smart ship's navigation. This application mainly makes the following restrictions, so that the sampling space is limited to a specific window, which is used for ship avoidance under urgent and dangerous conditions, that is, when CRI ≥ 0.7, such as Figure 3 As shown, specifically including

[0162] (1) Intelligent ship speed sampling space

[0163] The achievable speed vector pairs in the speed space of the smart ship are sampled multiple times, and the maximum achievable speed constraint can be expressed as V s =(v,ω)∈{(v,ω)|v∈[v min ,v max ],ω∈[ψ min ,ω max ]}, where v min With v max are the minimum and maximum achievable linear speeds of the smart ship, respectively. Similarly, ω min With ω max is the minimum and maximum feasible angular velocity of the smart ship. The above indicators are directly determined by the maneuverability and decision-making requirements of the smart ship when the smart ship is modeled.

[0164] (2) Smart ship velocity vector variable constraint

[0165] Affected by the environment and its own maneuverability, the speed change of the smart ship should also be constrained within a certain space V d The mathematical description is V d ∈{(v,w)|v∈[v o -Δv b t,v o +Δv ht],ω∈[ω o -Δω b t,v o +Δω h t]}, in the above expression, v o With ω o is the current linear velocity and angular velocity of the smart ship, Δv b With Δv h are the maximum linear accelerations of the smart ship when moving backward and forward, respectively, and Δω b With Δω h The above constraints are determined by the hull performance when the smart ship is modeled, and are also restricted by the smart ship's navigation environment.

[0166] (3) Speed ​​limit allowed by safety

[0167] In the collision avoidance decision of smart ships, safety is always the most important factor to be considered. When a ship encounters an obstacle, it should take timely measures to avoid danger. For this reason, the speed of the smart ship is constrained to ensure that the smart ship can always pass by the obstacle at a safe distance. For this reason, the allowed speed space V is designed. t , that is, the speed before collision with the static obstacle is Where dist(v,ω) refers to the shortest distance between the end of the simulated trajectory and the obstacle, v smax and ω smax They refer to the maximum linear acceleration and maximum bow angular velocity of the smart ship respectively. The speed space is determined by multiple factors such as the hydrodynamic characteristics of the smart ship, the position and shape of obstacles, and the motion state of the smart ship. The allowable speed space needs to be used as a constraint in path planning.

[0168] (4) Evaluation function

[0169] After setting the constraints, the DWA algorithm can calculate the actual trajectory of the ship according to the current achievable speed space (v, w). The speed search space V of the smart ship is m It can be expressed as the formula V m =V s ∩V d ∩V t .like Figure 3 shown.

[0170] After the search space is determined, the optimal velocity vector pair needs to be searched in the velocity space. In the DWA algorithm, the planned linear velocity and angular velocity remain unchanged during the planning cycle. Therefore, during the planning cycle, the local trajectory of the smart ship consists of straight lines and circular arcs. m In space, the value of the optimal speed pair (v, w) of the classic DWA algorithm can be determined by the objective function of the corresponding local trajectory, and its expression is:

[0171]

[0172] Among them, dist(v,w) is the evaluation factor of the distance between the local section and the obstacle, α, β and γ are weight coefficients, vel(v,w) is the evaluation factor of (v,w) to the water speed, head(v,w) is the deviation value of the heading angle of the smart ship to the critical target point. θ is the angle between the smart ship and the target point, R is the turning radius of the smart ship, and dist is the distance between the current position of the car and the target point. When there is no obstacle between the current position of the smart ship and the local target point, n = 0, otherwise n = 1. This function will be used for ship decision-making under the condition of failure of collision avoidance rules.

[0173] If not, the DWA algorithm is improved based on the VO algorithm to obtain the current achievable speed space of the target smart ship, which includes the following steps:

[0174] S61: Define the position of the obstacle ship at the current time t as P t , whose expression is

[0175]

[0176] Where: N represents the acquisition period and N = (t-t0) / Δt; Δt represents the duration of the intelligent ship control period; t0 represents the start time of obstacle ship prediction; P TS Indicates the predicted starting position of the obstacle ship;

[0177] The speed interval v where the target smart ship OS collides with the obstacle ship TS is OS , expressed as

[0178]

[0179] Where: D(P, r s um / (t-t0)) represents the overlapping area of ​​the target smart ship and the obstacle ship, and the two ships will collide in the time interval [t0,+∞); Represents the Minkowski vector operator symbol; r sum represents the sum of the ship areas of the target smart ship and the obstacle ship; P OS (t0) represents the position of the target smart ship at time t0; P TS (t0) represents the position of the obstacle ship at time t0;

[0180] S62: The next moment’s ship speed v of the obstacle ship predicted by the CNN+GRU network model TS It obeys Gaussian distribution, that is

[0181] Then the position distribution of the obstacle ship is [P TS (t)-(v e -1.96σ x )Δt,P TS (t)+(v e +1.96σ x )Δt]

[0182] Among them, v e represents the mean of the navigation speed obtained from multiple prediction results; σ represents the standard deviation of the navigation speed; the confidence level of the obstacle ship navigation speed prediction result is set to 95%, and the distribution range of the navigation speed at time t+1 can be obtained, such as Fig.10 As shown, the red dashed line is the predicted uncertainty range of the obstacle ship position distribution;

[0183] According to the position distribution of the obstacle ship, determine t n The constraints satisfied by the target smart ship position and the obstacle ship position at the moment are:

[0184] S63: Based on the constraint conditions, the target smart ship OS and the obstacle ship TS collide according to the speed interval v OS , obtain the speed optional interval VNLHO considering the uncertainty of the dynamic motion information of the obstacle ship, and use it as the current achievable speed space of the target smart ship, and its expression is:

[0185]

[0186] The objective function of the local trajectory planning of the intelligent ship is constructed to select the optimal speed pair in the current achievable speed space. The objective function J of the local trajectory planning of the intelligent ship is constructed, and its expression is:

[0187] min J=ω v J v +ω ψ J ψ

[0188] J v =|Δv OS |=|v OS (t+Δt)-v OS (t)|

[0189] J ψ =|Δψ OS |=|ψ OS (t+Δt)-ψ OS (t)|

[0190] Where: J v ,J ψThey represent the defined speed change and heading change respectively; v represents the achievable linear velocity; ψ represents the feasible angular velocity; ω v ,ω ψ represents the weight parameter; Δv OS Indicates the change in achievable linear velocity; Δψ OS Represents the possible angular velocity change.

[0191] S7: Based on the optimal speed pair selected in the current achievable speed space, ship collision avoidance under the condition of uncertain motion information of the obstacle ship is realized. That is, based on the position uncertainty prediction result, the position uncertainty ellipse and the ship field are numerically superimposed. Fig. 9 As shown, the updated ship domain is input into the improved DWA and VO hybrid collision avoidance decision method, and the intelligent ship collision avoidance algorithm under the condition of uncertain obstacle ship motion information can be obtained. Based on the intelligent ship collision avoidance strategy, the ship collision avoidance process under the condition of uncertain obstacle ship motion information is realized, and the basis for determining safe passage is passing outside the superimposed domain.

[0192] In this embodiment, the DWA collision avoidance algorithm and the VO collision avoidance algorithm are combined to provide support for the formulation of autonomous collision avoidance strategies for smart ships in different collision avoidance situations. In view of the discontinuous output of the ship's navigation status, a CNN-GRU ship navigation status prediction network is proposed to predict the ship's navigation status and the uncertainty of the status based on the AIS historical data, and obtain the uncertainty distribution of its position, so that when the moving obstacles cannot be continuously perceived, the intelligent decision-making system can still provide effective collision avoidance decision support for the smart ship, fully consider the uncertainty of the ship's position, and optimize the ship field.

[0193] The experimental verification result of this embodiment is that after the target smart ship passes the yellow and green area ships and passes the red turning point, it will cross with the blue and red area ships. At this time, assuming that the ship loses the blue ship signal, the ship begins to predict the blue ship's position uncertainty. The safe passing area after the position uncertainty ellipse and its own ship area are superimposed is as follows: Fig.11 As shown in the figure, the distance between the ship and the predicted blue area ship gradually approaches, the position uncertainty ellipse area of ​​the blue ship gradually increases, and the red ship signal is also lost at this time. The ship begins to predict the position uncertainty of the red ship as shown in the figure. Fig.12 As the own ship and the predicted blue ship area approach, until the own ship takes a sharp right turn to avoid the situation, as shown in Fig.13 When the ship passes by the blue area, the ship Fig.14As shown in the figure, after returning to the original route, the ship encounters the ship in the red area. At this time, the blue ellipse area used to predict the position uncertainty is much larger than the ship's own area, and its ship area is closer to the ellipse. It is worth noting that in this process, the actual position of the ship with the blue historical track is always within the predicted blue ellipse dashed line, which proves the effectiveness of the algorithm. Fig.15 As shown in the figure, after the ship has completed the avoidance of the blue ship, the position uncertainty ellipse area of ​​the red ship has exceeded its inherent ship area, and the avoidance safety area formed is already large. Despite this, the ship still took a sharp right turn to complete the avoidance, and finally returned to the original route after passing the blue ship. Fig.16 shown.

[0194] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information, characterized in that: The specific steps include: S1: Detect and obtain the status information of the obstacle ship and the target smart ship in the sensing area of ​​the navigable waters of the target smart ship; The state information includes at least the ship position, ship speed and ship attitude; S2: According to the status information of the obstacle ship and the target intelligent ship, a membership function for measuring the ship collision risk is constructed, and a ship collision risk mathematical model is constructed based on the membership function to obtain the current ship collision risk; S3: Based on the International Maritime Collision Avoidance Regulations, the preset ship collision risk range is divided and the ship collision avoidance risk stage and the ship encounter situation and give way relationship corresponding to the ship collision avoidance risk stage are obtained; The ship collision avoidance danger stage includes the safety stage, the collision danger stage, the emergency situation stage and the imminent danger stage; S4: Based on the ship collision avoidance danger stage, confirm whether the target smart ship and the obstacle ship are in a safe stage according to the current ship collision risk; If yes, the target smart ship is controlled to sail according to the originally set route; Otherwise, execute step S5; S5: confirm whether the obstacle ship cannot be stably tracked; If so, the state information of the obstacle ship that is difficult for the smart ship to obtain in real time due to the loss of detection signals is obtained through the CNN+GRU network model constructed by combining the trained convolutional neural network CNN and the gated recurrent neural network GRU; And according to the state information of the obstacle ship obtained in real time, the real-time ship collision risk with the target smart ship is obtained, and step S6 is continued; If not, directly execute step S6; S6: Based on the real-time ship collision risk level / current ship collision risk level, determine whether the target smart ship and the obstacle ship are in an urgent danger stage; If so, the current achievable speed space of the target smart ship is obtained based on the DWA algorithm; And based on the objective function of the DWA algorithm, the optimal speed pair in the current achievable speed space is selected; If not, the DWA algorithm is improved based on the VO algorithm to obtain the current achievable speed space of the target smart ship, and the objective function of the local trajectory planning of the smart ship is constructed to select the optimal speed pair in the current achievable speed space; S7: Based on the selected optimal speed pair in the current achievable speed space, ship collision avoidance is achieved under the condition of uncertain motion information of the obstacle ship.

2. The ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information according to claim 1 is characterized in that: The membership functions constructed in S2 include a membership function of the closest encounter distance of a ship, a membership function of the closest encounter time, a membership function of the distance to a dangerous encounter area, and a membership function of a relative true azimuth angle and a membership function of a relative ship speed ratio that affect the navigation safety of the smart ship; The membership function D(DCPA) of the closest encounter distance of ships is: Where: d1 represents the distance of the ship area that can perform collision avoidance maneuvers, d2 represents the distance of the area where the collision avoidance maneuvers are performed at the latest, and d2 = nd1 (n = 1, 2, ..., k); The closest encounter time membership function D(TCPA) is: Where: t1 and t2 represent the lower and upper bounds of the closest encounter time TCPA respectively; represents the relative velocity modulus of the obstacle ship relative to the own ship, and The dangerous encounter area distance membership function D(d) is: The relative true azimuth membership function D(R c ) and the relative ship speed ratio membership function D(K) is: Where: R c represents the relative true position of the obstacle ship and the target smart ship; K represents the relative speed ratio of the obstacle ship and the target smart ship; Q c Indicates that the obstacle ship is located at the true position of the target smart ship; The mathematical model of ship collision risk constructed based on the membership function is: CRI=ω1D(DCPA)+ω2D(TCPA)+ω3D(d)+ω4D(R c )+ω5D(K) Where: CRI represents the risk of ship collision; ω1~ω5 represent the preset weight values.

3. The ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information according to claim 2 is characterized in that: As described in S3, based on the international maritime collision avoidance regulations, the ship collision avoidance risk stage Θ is divided and obtained according to the preset ship collision risk range, and its expression is: The ship encounter situation and the relationship between giving way corresponding to the dangerous stage of ship collision avoidance is: Others = {Q c , R c Not included in the above cases} OS: Go straight.

4. The ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information according to claim 3 is characterized in that: The training process of the CNN+GRU network model constructed based on the combination of the convolutional neural network CNN and the gated recurrent neural network GRU in S5 is as follows: S51: obtaining historical ship time series trajectory sequence data, normalizing it to obtain a standard sample set, and inputting the standard sample set into the CNN+GRU network model for training; And the CNN+GRU network model includes an encoder unit and a decoder unit; The encoder unit includes a convolutional neural network layer, a double-layer GRU network layer, and a positional Dropout network layer connected in sequence; The convolutional neural network layer is used to extract data features from the ship time-series trajectory sequence data in the standard sample set; The data features include ship position, ship speed and ship attitude; The double-layer GRU network layer is used to further extract the data features of the ship time series trajectory sequence from the output of the convolutional neural network layer, and perform feature encoding on it to obtain an intermediate vector; The position Dropout network layer is used to ignore some hidden neurons in proportion during the training process of the CNN+GRU network model to obtain an uncertainty estimation network model for the obstacle ship motion information; The decoder unit is specifically a cyclic network structure composed of a single-layer GRU and a fully connected layer; The single-layer GRU is used to perform a feature decoding operation on the intermediate vector; The fully connected layer is used to extract the output of the single-layer GRU to obtain the state prediction information of the obstacle ship at the next moment; S52: The mean square error function is used as a loss function to determine whether the output of the trained CNN+GRU network model converges, and the trained CNN+GRU network model with converged output is used as the final CNN+GRU network model.

5. The ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information according to claim 4 is characterized in that: The DWA algorithm is improved based on the VO algorithm as described in S6 to obtain the current achievable speed space of the target smart ship, specifically: S61: Define the position of the obstacle ship at the current time t as P t , whose expression is Where: N represents the acquisition period and N = (t-t0) / Δt; Δt represents the duration of the intelligent ship control period; t0 represents the starting time of obstacle ship prediction; P TS Indicates the predicted starting position of the obstacle ship; The speed interval v where the target smart ship OS collides with the obstacle ship TS is OS , expressed as Where: D(P,r sum / (t-t0)) represents the overlapping area between the target smart ship and the obstacle ship. Represents the Minkowski vector operator symbol; r sum represents the sum of the ship areas of the target smart ship and the obstacle ship; P OS (t0) represents the position of the target smart ship at time t0; P TS (t0) represents the position of the obstacle ship at time t0; S62: The next moment’s ship speed v of the obstacle ship predicted by the CNN+GRU network model TS It obeys Gaussian distribution, that is Then the position distribution of the obstacle ship is [P TS (t)-(v e -1.96σ x )Δt,P TS (t)+(v e +1.96σ x )Δt] Among them, v e represents the mean of the navigation speed obtained from multiple prediction results; σ represents the standard deviation of the navigation speed; According to the position distribution of the obstacle ship, determine t n The constraints satisfied by the target smart ship position and the obstacle ship position at the moment are: S63: Based on the constraint conditions, according to the speed interval v where the target smart ship OS collides with the obstacle ship TS OS , obtain the speed optional interval VNLHO considering the uncertainty of the dynamic motion information of the obstacle ship, and use it as the current achievable speed space of the target smart ship, and its expression is:

6. The ship collision avoidance optimization method under the condition of uncertain obstacle ship motion information according to claim 5, characterized in that: The objective function J of the local trajectory planning of the intelligent ship is constructed in S6, and its expression is: min J=ω v J v +oh ψ J ψ J v =|Δv OS |=|v OS (t+Δt)-v OS (t)| J ψ =|Dψ OS |=|ψ OS (t+Δt)-ψ OS (t)| Where: J v ,J ψ They represent the defined speed change and heading change respectively; v represents the achievable linear velocity; ψ represents the feasible angular velocity; ω v ,ω ψ represents the weight parameter; Δv OS Indicates the change in achievable linear velocity; Δψ OS Represents the possible angular velocity change.

Images