Unmanned ship safety control method based on bidirectional long-short term memory neural network

Abstract

The invention provides an unmanned ship safety control method based on a bidirectional long-short-term memory neural network, and the method comprises the steps: designing a predictor and constructing a detection and compensation mechanism through employing the bidirectional long-short-term memory neural network when an unmanned ship is subjected to disturbance and mixed attack; based on a hybrid attack compensation mechanism, data driving control and sliding mode control are combined to design a safe and effective elastic control strategy. The effective data-driven sliding mode safety control method is provided for the unmanned ship which has ocean disturbance and suffers from DoS attacks and FDI attacks, the influence of mixed attacks on the unmanned ship is reduced, and the unmanned ship can safely and effectively track the expected trajectory. According to the invention, the data driving technology and the sliding mode technology are combined, compared with the traditional data driving technology, the control algorithm is higher in interpretability, and meanwhile, the controller designed based on the sliding mode technology is better in control effect.

Description

Technical Field

[0001] The present invention relates to the technical field of ship data-driven safety control, and in particular to a data-driven sliding mode safety control method based on a bidirectional long short-term memory neural network. Background Art

[0002] Due to their versatility, flexibility, and cost-effectiveness, unmanned vessels are widely used in marine fisheries, maritime rescue, maritime traffic management, and other fields. At the same time, the safety control of unmanned vessels has attracted more and more attention from researchers.

[0003] The development of wireless communication, with its advantages of being environmentally unrestricted and capable of real-time communication, has made wireless communication a crucial means of transmitting information for unmanned vessels. However, its high degree of openness makes the signal transmission process more vulnerable to malicious attacks, which can cause signal transmission failures, leading to loss of control of the unmanned vessel and inability to complete its mission. Examples include denial of service (DoS) attacks and false data injection (FDI) attacks. DoS attacks involve injecting a large number of signals, causing sensors to lose their information processing capabilities or even become paralyzed. FDI attacks involve injecting false data to deceive the target computer. DoS attacks are straightforward and simple, but they consume a lot of energy and are easily detected. FDI attacks, on the other hand, consume less energy. Combining these two attack methods can achieve a more sustained and effective attack. During unmanned vessel operations, a hybrid attack could cause the vessel's communication, navigation, or control systems to fail, resulting in loss of control or inability to complete its mission. Existing literature has limited research on both DoS and Denial of Service attacks. Dan Zhang et al. studied DoS attacks on unmanned vessels, but did not investigate hybrid attacks.

[0004] Furthermore, many mature control methods have been developed for the safety control of unmanned vessels, such as sliding mode control, adaptive control, and model predictive control. However, these control methods all rely on the system model of the unmanned vessel. However, due to the influence of the marine environment, the model information of the unmanned vessel is difficult to determine. Summary of the Invention

[0005] In order to deal with problems such as ocean disturbances, DoS attacks, and FDI attacks during the trajectory tracking of unmanned ships, this paper proposes a data-driven sliding mode security control strategy based on a bidirectional long short-term memory neural network. Compared with traditional control strategies, this algorithm is completely independent of model information and can complete the safety control of unmanned ships when they are subjected to mixed attacks.

[0006] The technical means adopted in the present invention are as follows:

[0007] A safety control method for an unmanned vessel based on a bidirectional long short-term memory neural network comprises the following steps:

[0008] S1: Design of an equivalent data-driven model of an unmanned ship based on a partial format dynamic linearization method;

[0009] S2: Collect control signals, velocity information, and position information during the UAV's motion for the purpose of trajectory tracking, train a bidirectional long-short-term memory neural network model, and use the bidirectional long-short-term memory neural network model to design an UAV output signal predictor;

[0010] S3: Select a detection threshold based on the UAV output signal predictor, calculate the error between the predicted UAV position and speed signals and the actual signals, and determine whether the UAV is under cyber attack by comparing the error with the detection threshold.

[0011] S4: Design an attack compensation mechanism based on the unmanned ship output signal predictor. When the unmanned ship is not attacked, it receives the signal normally. When the unmanned ship is attacked, it automatically calls the unmanned ship output signal predictor to output the predicted signal.

[0012] S5: Based on the attack compensation mechanism, the signal received by the unmanned ship control end is the compensated position signal, thereby changing the unmanned ship equivalent data-driven model, and designing a decoupled unmanned ship equivalent data-driven model based on the unmanned ship output signal predictor;

[0013] S6: Design a discrete integral sliding surface based on the decoupled unmanned ship equivalent data-driven model to obtain a data-driven sliding mode trajectory tracking safety controller.

[0014] Furthermore, suppose the current state of the unmanned ship is related to the state of the unmanned ship in the previous time domain. When the unmanned ship equivalent data-driven model is expressed as

[0015]

[0016] Among them, Δv(k+1)=v(k+1)-v(k), v(k)=[ν(k), θ(k), r(k)] T is the velocity vector; Δτ(k)=τ(k)-τ(k-1), τ(k) is the control signal, ΔΩ(k)=[Δω(k), Δω(k-1),..., Δω(k-l+1)] T , Δω(k)=ω(k)-ω(k-1), ω(k) represents the ocean disturbance, represents the pseudo partial derivative Jacobian matrix, and satisfies

[0017] Furthermore, the normal operation control signal when the unmanned ship is not attacked is {τ1, τ2, ..., τ k ,…,τ n}, speed information {v1, v2, ..., v k ,…,v n}, location information {ζ1, ζ2, ..., ζ k ,…,ζ n};

[0018] The above information is input into the bidirectional long short-term memory neural network model and trained:

[0019] f k =σ(W f ×[τ k , h k-1 ]+b f ),

[0020] i k =σ(W i ×[τ k , h k-1 ]+b i ),

[0021] o k =σ(W o ×[τ k , h k-1 ]+b o ),

[0022] g k =tanh(W c ×[τ k , h k-1 ]+b),

[0023] c k =c k-1 ×f k +g k ×i k ,

[0024] h k =o k ×tanh(c k );

[0025] Among them, f k Forget gate, which determines how much information at time k-1 is retained and transmitted to time k, i k is the input gate, which determines how much information is obtained from the input and retained at the current moment, o k is the output gate, which determines the output information of the control unit at the current moment. The function tanh and σ are nonlinear activation functions;

[0026] Design an unmanned ship output signal predictor:

[0027]

[0028] Defining prediction error

[0029]

[0030] in,

[0031] Furthermore, the following methods are used to determine whether the unmanned ship is under cyber attack:

[0032]

[0033]

[0034] in, is the unmanned ship position prediction signal, and ε is the detection threshold.

[0035] Furthermore, when setting up the attack compensation mechanism, the compensation signal is as follows:

[0036]

[0037] Among them, c (k) is the signal received by the UAV controller. When , γ(k)=1; when When , γ(k)=0.

[0038] Furthermore, assuming that the unmanned ship is attacked, the attack signal is

[0039] ζ A (k)=α(k)(ζ(k)+β(k)ζ′(ζ(k))),

[0040] Where α(k) = 0, 1; β(k) = 0, 1, ζ′(ζ(k)) is the false data related to the position signal of the unmanned ship; when α(k) = 0, the unmanned ship is attacked by DoS; when α(k) = 1, β(k) = 1, the unmanned ship is attacked by false data injection; when α(k) = 1, β(k) = 1, the unmanned ship is not attacked;

[0041] When the unmanned ship is attacked, the unmanned ship equivalent data-driven model becomes as follows:

[0042] ζ c (k+1)=ζ c (k)+T s J(ψ(k))v(k),

[0043] v(k+1)=v(k)+f(v(k),ζ c (k), τ(k), ω(k));

[0044] Based on the predicted speed information, the decoupling process of the equivalent data-driven model of the unmanned ship is as follows:

[0045]

[0046] Its adaptive update rate is as follows:

[0047]

[0048]

[0049]

[0050] in, Respectively The estimated value of Λ(k),

[0051] Based on the error, the equivalent data-driven model of the decoupled unmanned ship is written as:

[0052]

[0053] in,

[0054] Furthermore, the discrete integral sliding mode surface is as follows:

[0055] s(k)=C1e(k)-C1e(1)+C2θ(k),

[0056] in, C1=diag(c 11 , c 12 , c 13 ), C2=diag(c 21 , c 22 , c 23 ) is the parameter matrix;

[0057] The data-driven sliding mode safety controller of the unmanned ship is obtained based on the discrete integral sliding mode surface as follows:

[0058]

[0059] The design challenges of the present invention come from the fact that hybrid attacks cause the unmanned ship controller to not receive signals or receive erroneous signals, and the disturbance of the ocean environment affects the control effect of the unmanned ship, which will make it difficult for the unmanned ship to accurately track the desired trajectory. In order to solve this problem, the present invention discloses an unmanned ship safety control method based on a bidirectional long short-term memory neural network, designs an unmanned ship information predictor based on a Bi-LSTM neural network, and designs a flexible data-driven sliding mode safety controller considering the prediction error. This controller can enable the unmanned ship to still track the desired trajectory when it is subjected to ocean disturbances and hybrid attacks. Combined with data-driven control, the proposed control method is simulated and verified, and the simulation results verify the effectiveness of the proposed data-driven sliding mode safety control mechanism. Compared with the prior art, the present invention has the following advantages:

[0060] 1. The present invention designs an unmanned ship state predictor based on a bidirectional long short-term memory neural network, takes the prediction error into consideration, and designs an unmanned ship safety controller based on the upper bound of the prediction error, thereby further improving the stability of the unmanned ship during movement.

[0061] 2. Based on the bidirectional long short-term memory neural network compensation mechanism, this invention designs a flexible controller for mixed attacks. When the unmanned ship is attacked by mixed attacks, it can automatically adjust the control mechanism.

[0062] 3. The present invention combines data-driven technology with sliding mode technology. Compared with traditional data-driven technology, its control algorithm is more interpretable. At the same time, the controller designed based on sliding mode technology has better control effect.

[0063] Based on the above reasons, the present invention can be widely promoted in fields such as unmanned ship trajectory tracking control. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] 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 any creative labor.

[0065] Figure 1 Flow chart of the method of the present invention.

[0066] Figure 2 This is an estimated signal diagram provided by an embodiment of the present invention.

[0067] Figure 3 Root mean square error and loss function diagram for training a bidirectional long short-term memory neural network provided by an embodiment of the present invention.

[0068] Figure 4This is a simulation diagram of the attack signal provided by an embodiment of the present invention.

[0069] Figure 5 This is a simulation diagram of the timing of different attack signals provided by an embodiment of the present invention.

[0070] Figure 6 This is a simulation diagram of the unmanned ship trajectory tracking provided by an embodiment of the present invention.

[0071] Figure 7 This is a tracking error simulation diagram provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions 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 embodiments described are only 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 making creative efforts should fall within the scope of protection of the present invention.

[0073] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0074] like Figure 1 As shown, the present invention provides a data-driven iterative learning trajectory tracking control method, which is characterized by comprising the following steps:

[0075] S1: Design of an equivalent data-driven model of an unmanned ship based on a partial format dynamic linearization method;

[0076] S2: Collect control signals, velocity information, and position information during the UAV's motion for the purpose of trajectory tracking, train a bidirectional long-short-term memory neural network model, and use the bidirectional long-short-term memory neural network model to design an UAV output signal predictor;

[0077] S3: Select a detection threshold based on the UAV output signal predictor, calculate the error between the predicted UAV position and speed signals and the actual signals, and determine whether the UAV is under cyber attack by comparing the error with the detection threshold.

[0078] S4: Design an attack compensation mechanism based on the unmanned ship output signal predictor. When the unmanned ship is not attacked, it receives the signal normally. When the unmanned ship is attacked, it automatically calls the unmanned ship output signal predictor to output the predicted signal.

[0079] S5: Based on the attack compensation mechanism, the signal received by the unmanned ship control end is the compensated position signal, thereby changing the unmanned ship equivalent data-driven model, and designing a decoupled unmanned ship equivalent data-driven model based on the unmanned ship output signal predictor;

[0080] S6: Design a discrete integral sliding surface based on the decoupled unmanned ship equivalent data-driven model to obtain a data-driven sliding mode trajectory tracking safety controller.

[0081] The specific process of step S1 is as follows:

[0082] S11. Define the position vector of the unmanned ship as ζ(t) = [x(t), y(t), ψ(t)] T , the velocity vector is v(t) = [v(t), θ(t), r(t)] T , where (x(t), y(t)) represents the position information, ψ(t) represents the yaw angle, v(t), θ(t), r(t) represent the surge, sway and yaw angular velocity respectively. Define the reference state ζ r (k) = [x r (k), y r (k), ψ r (k)] T , tracking error: e(k) = ζ r (k)-ζ(k), when the unmanned ship is attacked, the compensated signal received by the unmanned ship controller is ζ c (k);

[0083] S12. Using the first-order Euler method, the system model of the unmanned ship is discretized as:

[0084] ζ(k+1)=ζ(k)+T s J(ψ(k))v(k),

[0085] v(k+1)=v(k)+f(v(k),ζ(k),τ(k),ω(k)),

[0086] T s represents the sampling time, k represents the sampling point, and the function f(·) represents the model uncertainty of the unmanned ship caused by ocean disturbances and other reasons;

[0087] S13, the signal v received by the controller is related to the current state of the unmanned ship and the state of the unmanned ship in the previous l time domain. When , the equivalent data-driven iterative learning model of the unmanned ship can be expressed as

[0088]

[0089] in,

[0090]

[0091]

[0092]

[0093]

[0094] and satisfy

[0095] The specific process of step S2 is as follows:

[0096] S21. Through experiments, obtain the data of normal operation of the unmanned ship when it is not attacked, and control the signal {τ1, τ2, ..., τ k ,…,τ n}, velocity information {v1, v2, ..., v k ,…,v n}, location information {ζ1,ζ2,…,ζ k ,…,ζ n};

[0097] S22. Import the model into the bidirectional long short-term memory neural network to train the network model:

[0098] f k =σ(W f ×[τ k , h k-1 ]+b f ),

[0099] i k =σ(W i ×[τ k , h k-1 ]+b i ),

[0100] o k =σ(W o ×[τ k , h k-1 ]+b o ),

[0101] g k=tanh(W c ×[τ k , h k-1 ]+b),

[0102] c k =c k-1 ×f k +g k ×i k ,

[0103] h k =o k ×tanh(c k );

[0104] Among them, f k is the forget gate, which determines how much information at time k-1 is retained and transmitted to time k; i k The input gate determines how much information is obtained from the input and retained at the current moment; k is the output gate, which determines the output information of the control unit at the current moment. The function tanh and σ are nonlinear activation functions.

[0105] Design an unmanned ship output signal predictor:

[0106]

[0107] Defining prediction error

[0108]

[0109] in

[0110] The specific process of step S3 is as follows:

[0111] S31. Design a detector for the predictor designed in S2. The detection mechanism is as follows:

[0112]

[0113]

[0114] Where ε is the detection threshold.

[0115] The specific process of step S4 is as follows:

[0116] S41. Based on the predictor designed in S2, an attack compensation mechanism is designed. The compensation signal is as follows:

[0117]

[0118] Among them, c (k) is the signal received by the UAV controller. When , γ(k)=1; when When , γ(k)=0.

[0119] The specific process of step S5 is as follows:

[0120] S51. Assume that when the unmanned ship is attacked, the attack signal is

[0121] ζ A (k)=α(k)(ζ(k)+β(k)ζ′(ζ(k)))

[0122] Where α(k) = 0, 1; β(k) = 0, 1, ζ′(ζ(k)) is the false data related to the position signal of the unmanned ship; when α(k) = 0, the unmanned ship is attacked by DoS; when α(k) = 1, β(k) = 1, the unmanned ship is attacked by false data injection; when α(k) = 1, β(k) = 1, the unmanned ship is not attacked;

[0123] S52. When the unmanned ship is attacked, the discrete model of the unmanned ship in S1 becomes as follows:

[0124] ζ c (k+1)=ζ c (k)+T s J(ψ(k))v c (k),

[0125] v(k+1)=v(k)+f(v(k),ζ c (k), τ(k), ω(k));

[0126] S62. Based on the sliding surface designed in S61, the data-driven sliding mode safety controller of the unmanned ship can be obtained as follows:

[0127]

[0128] S54: Based on the equivalent data model of the unmanned ship in step S53, further decoupling processing can be written as:

[0129]

[0130] Its adaptive update rate is as follows:

[0131]

[0132]

[0133]

[0134] in, Respectively Estimated value of Λ(k).

[0135] S55. Based on the error defined in S23, the equivalent data model in S54 can be further written as:

[0136]

[0137] in,

[0138] S61. Design the discrete integral sliding surface as follows:

[0139] s(k)=C1e(k)-C1e(1)+C2θ(k),

[0140] in, C1=diag(c 11 , c 12 , c 13 ), C2=diag(c 21 , c 22 , c 23 ) is a designable parameter matrix.

[0141] S62. Based on the sliding surface designed in S61, the data-driven sliding mode safety controller of the unmanned ship can be obtained as follows:

[0142]

[0143] The method further includes:

[0144] S7. Simulation experiments are conducted on the equivalent data-driven model of the unmanned ship with environmental disturbances and hybrid attacks, the hybrid attack detection mechanism, the compensation mechanism, and the proposed data-driven sliding mode security control strategy to verify their effectiveness.

[0145] In order to verify the effectiveness of the data-driven iterative learning safety control method provided in this embodiment, a simulation experiment is carried out using MATLAB and a detailed description is given.

[0146] Taking into account factors such as ocean disturbances, DoS attacks, and FDI attacks, data-driven and sliding mode technologies were adopted to design a data-driven sliding mode safety controller based on a bidirectional long short-term memory neural network. The controller is robust to ocean disturbances and can compensate for the impact of mixed attacks, thereby ensuring the safety of the unmanned ship's movement process.

[0147] Specifically, in this embodiment, any unknown unmanned ship can be used. In the simulation, the tracking trajectory is set to

[0148] x r (k)=3sin(k / 50+0.6)+0.2cos(k / 20+0.6),

[0149] y r (k)=0.2sin(k / 20+0.8)+2cos(k / 50+0.8),

[0150] ψ r (k) = atan2(x r (k), y r (k)).

[0151] The initial conditions of the ship's initial state and parameters are set as: ζ(1) = [0; 0; 1], v(1) = [0; 0; 1]. In the simulation, the sampling time is set to 0.1s, L = 3, κ = 0.3, λ = 0.2, C1 = [0.7 0.7 0.7], C2 = [0.001 0.001 0.001]. The ocean disturbance form selected is

[0152] ω(k)=[3sin(2k); 2cos(3k); -1.5cos(1.5k)+2sin(1.5k)],

[0153] The mixed attack signal is as follows:

[0154] ζ A (k) = α(k)(ζ c (k)+β(k)sin(ζ c (k))cos(ζ c (k))).

[0155] Among them, P{α(m)=0}=0.5, P{β(m)=1}=0.6.

[0156] Based on the above parameters, the data-driven iterative learning trajectory tracking control method proposed in this invention is simulated and verified as follows: Figure 2-7 As shown. Among them, Figure 2 is the estimated signal based on the bidirectional long short-term memory neural network estimator; Figure 3 is the root mean square error and loss function value during the training of the bidirectional long short-term memory neural network; Figure 4 are the mixed attack signals received by the unmanned ship, which are correlated with the output signals of the unmanned ship; Figure 5 is the timing of occurrence of different attack signals; Figure 6 This figure shows the trajectory tracking effect of the unmanned ship after being subjected to disturbances and mixed attacks, controlled by a data-driven sliding mode controller designed based on compensation signals. It can be seen that the unmanned ship can accurately track the desired trajectory. Figure 7 The figure shows the change of the maximum error of the system. The tracking errors of the three positions eventually tend to 0, which means that the unmanned ship performs well when subjected to disturbances and mixed attacks.

[0157] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0158] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0159] 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 above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above 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 safety control method for an unmanned vessel based on a bidirectional long short-term memory neural network, characterized in that: The steps include: S1: Design of an equivalent data-driven model of an unmanned ship based on a partial format dynamic linearization method; S2: Collect control signals, velocity information, and position information during the UAV's motion for the purpose of trajectory tracking, train a bidirectional long-short-term memory neural network model, and use the bidirectional long-short-term memory neural network model to design an UAV output signal predictor; S3: Select a detection threshold based on the UAV output signal predictor, calculate the error between the predicted UAV position and speed signals and the actual signals, and determine whether the UAV is under cyber attack by comparing the error with the detection threshold. S4: Design an attack compensation mechanism based on the unmanned ship output signal predictor. When the unmanned ship is not attacked, it receives the signal normally. When the unmanned ship is attacked, it automatically calls the unmanned ship output signal predictor to output the predicted signal. S5: Based on the attack compensation mechanism, the signal received by the unmanned ship control end is the compensated position signal, thereby changing the unmanned ship equivalent data-driven model, and designing a decoupled unmanned ship equivalent data-driven model based on the unmanned ship output signal predictor; S6: Design a discrete integral sliding surface based on the decoupled unmanned ship equivalent data-driven model to obtain a data-driven sliding mode trajectory tracking safety controller.

2. The unmanned vessel safety control method based on a bidirectional long short-term memory neural network according to claim 1 is characterized in that: Assume that the current state of the unmanned ship is related to the state of the unmanned ship l time domain ago. When the unmanned ship equivalent data-driven model is expressed as Among them, Δv(k+1)=v(k+1)-v(k), v(k)=[v(k), θ(k), r(k)] T is the velocity vector; Δτ(k)=τ(k)-τ(k-1), τ(k) is the control signal, ΔΩ(k)=[Δω(k), Δω(k-1),..., Δω(k-l+1)] T , Δω(k)=ω(k)-ω(k-1), ω(k) represents the ocean disturbance, represents the pseudo partial derivative Jacobian matrix, and satisfies 3. The unmanned vessel safety control method based on a bidirectional long short-term memory neural network according to claim 1 is characterized in that: The normal operation control signal when the unmanned ship is not attacked is {τ1, τ2, ..., τ k ,…,τ n }, speed information {v1, v2, ..., v k ,…,v n }, location information {ζ1, ζ2, ..., ζ k ,…,ζ n }; The above information is input into the bidirectional long short-term memory neural network model and trained: f k =σ(W f ×[t k ,h k-1 ]+b f ), I k =σ(W i ×[t k ,h k-1 ]+b i ), the k =σ(W o ×[t k ,h k-1 ]+b o ), g k =tanh(W c ×[τ k ,h k-1 ]+b), c k =c k-1 ×f k +g k ×i k , h k =o k ×tanh(c k ), Among them, f k Forget gate, which determines how much information at time k-1 is retained and transmitted to time k, i k is the input gate, which determines how much information is obtained from the input and retained at the current moment, o k is the output gate, which determines the output information of the control unit at the current moment. The function tanh and σ are nonlinear activation functions; Design an unmanned ship output signal predictor: Defining prediction error in, 4. The unmanned vessel safety control method based on a bidirectional long short-term memory neural network according to claim 3 is characterized by: The following methods are used to determine whether an unmanned ship is under cyber attack: in, is the unmanned ship position prediction signal, and ε is the detection threshold.

5. The unmanned vessel safety control method based on a bidirectional long short-term memory neural network according to claim 4 is characterized in that: When setting up the attack compensation mechanism, the compensation signal is as follows: Among them, c (k) is the signal received by the UAV controller. When , γ(k)=1; when When , γ(k)=0.

6. The unmanned vessel safety control method based on bidirectional long short-term memory neural network according to claim 4 is characterized in that: Assume that when the unmanned ship is attacked, the attack signal is g A (k)=α(k)(ζ(k)+β(k)ζ′(ζ(k))), Where α(k) = 0, 1; β(k) = 0, 1, ζ′(ζ(k)) is the false data related to the position signal of the unmanned ship; when α(k) = 0, the unmanned ship is attacked by DoS; when α(k) = 1, β(k) = 1, the unmanned ship is attacked by false data injection; when α(k) = 1, β(k) = 1, the unmanned ship is not attacked; When the unmanned ship is attacked, the unmanned ship equivalent data-driven model becomes as follows: g c (k+1)=ζ c (k)+T s J(ψ(k))v(k), v(k+1)=v(k)+f(v(k),ζ c (k), τ(k), ω(k)); Based on the predicted speed information, the decoupling process of the equivalent data-driven model of the unmanned ship is as follows: Its adaptive update rate is as follows: in, Respectively The estimated value of Based on the error, the equivalent data-driven model of the decoupled unmanned ship is written as: in, 7. The unmanned vessel safety control method based on a bidirectional long short-term memory neural network according to claim 1 or 6, characterized in that: The discrete integral sliding surface is shown below: s(k)=C1e(k)-C1e(1)+C2θ(k), in, C1=diag(c 11 , c 12 , c 13 ), C2=diag(c 21 , c 22 , c 23 ) is the parameter matrix; The data-driven sliding mode safety controller of the unmanned ship is obtained based on the discrete integral sliding mode surface as follows:

Images