A method for underwater acoustic environment decision support using artificial intelligence

Abstract

The application belongs to the technical field of marine sound field application, and discloses a kind of underwater acoustic environment decision support method using artificial intelligence.The method comprises obtaining current underwater acoustic environment data, sonar current working state data and historical detection feedback;The obtained underwater acoustic environment data and sonar current working state data are integrated with historical detection feedback as state space vector;State space vector is input into trained AI sonar parameter recommendation model for inference calculation to obtain optimal sonar working parameter combination under current environment;According to the optimal sonar working parameter combination under current environment and the physical environment model, auxiliary decision information is formulated.The application realizes the collaborative optimization of sonar working depth, working frequency and beam elevation angle, breaks through the limitation of single parameter optimization in the prior art, forms the optimal parameter combination adapted to complex environment;The global optimality and convergence speed of parameter optimization are improved, and optimal decision can be made in high dynamic and high complexity underwater acoustic environment.

Description

Technical Field

[0001] This invention relates to the field of marine acoustic field application technology, and more specifically to a decision support method for underwater acoustic environment using artificial intelligence. Background Technology

[0002] As a core technology in fields such as underwater detection, active sonar's operating parameters directly determine key performance indicators such as detection range and detection probability. Currently, there are three main types of conventional methods for optimizing sonar operating parameters: first, manual setting based on engineering experience, relying on empirical formulas; second, traditional algorithms used for preliminary optimization of multiple parameters, such as genetic algorithms and particle swarm optimization algorithms; and third, traditional adaptive signal processing techniques, which can dynamically adjust single or partial parameters by combining real-time environmental data.

[0003] Existing technologies mainly employ multi-parameter adaptive adjustment techniques, but they utilize traditional adaptive algorithms without incorporating artificial intelligence mechanisms, and their real-time performance and global optimality in parameter adjustment are insufficient. The following are the core problems: 1) Insufficient multi-parameter collaborative optimization capability. Existing technologies mostly focus on optimizing single or partial parameters, which is difficult to adapt to the detection needs in complex marine environments; 2) The optimization accuracy and efficiency are out of balance. Traditional optimization algorithms are prone to getting stuck in local optima, have slow convergence speed, and cannot quickly respond to dynamic changes in the environment. 3) Weak decision support capabilities: Existing technologies primarily output isolated sonar physical parameters, lacking integration with the actual situation. It is difficult to intuitively assess the benefits of parameter adjustments, and a systematic set of auxiliary visual outputs has not been developed, resulting in a lack of a "human-in-the-loop" decision support closed loop. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by providing a decision support method for underwater acoustic environments that utilizes artificial intelligence.

[0005] To achieve the above objectives, the present invention provides a decision support method for underwater acoustic environments utilizing artificial intelligence, comprising: Step 1: Acquire current underwater acoustic environment data, current sonar operating status data, and historical detection feedback; Step 2: Integrate the acquired underwater acoustic environment data and the current working status data of the sonar with historical detection feedback into a state space vector; Step 3: Input the state space vector into the trained AI sonar parameter recommendation model for inference calculation to obtain the optimal combination of sonar operating parameters in the current environment; Step 4: Develop auxiliary decision-making information based on the optimal combination of sonar operating parameters and the physical environment model under the current environment.

[0006] Furthermore, the underwater acoustic environment data includes hydrological parameters, ocean noise, and target information.

[0007] Furthermore, the hydrological parameters include temperature. ,salinity and depth The marine noise includes a noise intensity of 5 kHz. 10kHz noise intensity Noise intensity at 15kHz The target information includes the target depth. and target reflectivity .

[0008] Furthermore, the sonar's current operating status data includes the current operating depth. Current operating frequency and current beam pitch angle .

[0009] Furthermore, the historical detection feedback includes the detection probability of the most recent M detections, the sonar range, and the parameter adjustment range of the previous round.

[0010] Furthermore, the value of M is 3, and the parameter adjustment range includes the working depth adjustment range. Operating frequency adjustment range Beam pitch adjustment range .

[0011] Furthermore, the training process of the AI ​​sonar parameter recommendation model is as follows: Constructing state space vectors for: ; in, These represent the detection probabilities for the first, second, and third detections, respectively. These are the sonar ranges for the first, second, and third detections, respectively. A physical environment model based on sonar equations is constructed as follows: ; in, This refers to the signal margin for an active sonar system with both transceiver and receiver integrated. At the source level, To spread the loss, For target strength, Noise level, As a directional index, The detection threshold; The reward function is designed with the effective range R and the detection probability P as follows: ; in, The instant reward obtained at time t. To detect probability, To detect probability weights, This refers to the effective range of sonar. This represents the maximum effective range of the sonar. For the effective distance weight, For constraint terms, As a penalty for failure, For indicator functions; Constructing Action Space for: ; Using the state space vector Physical environment model, motion space The model is trained on an AI sonar parameter recommendation model using an Actor-Critic network framework and a reward function. During training, the action results of the AI ​​sonar parameter recommendation model are evaluated using either a proximal policy optimization algorithm or a deep deterministic policy gradient algorithm. The Actor and Critic networks are continuously iterated and updated until the task level meets the preset convergence conditions.

[0012] Furthermore, the method for evaluating the action results of the AI ​​sonar parameter recommendation model using the proximal policy optimization algorithm is as follows: The generalized advantage estimation is used to calculate the action advantage function and the generalized advantage estimate of the action to be taken at time t. for: ; ; in, As a discount factor, To estimate hyperparameters for generalized dominance, For TD error, For action index variables, For the objective value function, Let t+1 be the state space vector. If the generalized dominance estimate is If the generalized dominance estimate is greater than the average at time t, it means that taking the action at time t is better than the average, increasing the probability of choosing that action; If the generalized dominance estimate is below average at time t, it indicates that the action taken at time t is worse than average, thus reducing the probability of its selection; If , it means that the action taken at time t is at the average level and no adjustment is needed.

[0013] Furthermore, the preset convergence condition is the detection probability. And detection range .

[0014] Beneficial effects: 1. This invention can directly output the optimal combination of parameters that can be implemented, and present indicators such as the detection probability coverage map and the maximum effective distance driven by the optimal parameters, providing relevant personnel with an intuitive situational reference of "human in the loop" and improving the scientific nature of decision-making; 2. This invention breaks through the limitations of existing technologies that optimize only one or some parameters, and achieves joint optimization of the three core parameters of sonar: working depth, working frequency, and beam pitch angle. It can output more adaptable parameter combinations according to the complex and ever-changing marine acoustic field environment, and meet the usage requirements of high dynamic underwater detection scenarios. 3. This invention adopts an AI-assisted decision-making mode, which greatly improves the computing speed and global optimization capability compared with traditional algorithms, and achieves rapid decision response. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating the underwater acoustic environment decision support method utilizing artificial intelligence according to an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the training of the AI ​​sonar parameter recommendation model according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the training process of the AI ​​sonar parameter recommendation model according to an embodiment of the present invention. Detailed Implementation

[0016] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. These embodiments are implemented based on the technical solutions of the present invention, and it should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention.

[0017] like Figure 1 As shown, this embodiment of the invention provides a decision support method for underwater acoustic environments utilizing artificial intelligence, including: Step 1: Acquire current underwater acoustic environment data, sonar current operating status data, and historical detection feedback. The underwater acoustic environment data is 8-dimensional, including hydrological parameters, ocean noise, and target information. Hydrological parameters include temperature. ,salinity and depth Ocean noise includes noise levels up to 5 kHz. 10kHz noise intensity Noise intensity at 15kHz Target information includes target depth. and target reflectivity The current sonar operating status data is 3D data, including the current operating depth. Current operating frequency and current beam pitch angle Historical detection feedback is presented as 7-dimensional data, including the detection probability of the most recent M detections, the sonar range, and the parameter adjustment range of the previous round. The preferred value for M is 3, corresponding to the detection probabilities of the three detections as follows: the detection probability of the first detection... Detection probability of the second detection And the detection probability of the third detection The sonar ranges for the three detections were as follows: [Sonar range for the first detection] The effective range of the sonar during the second detection The effective range of the sonar during the third detection The adjustment range of the above parameters includes the adjustment range of the working depth. Operating frequency adjustment range Beam pitch adjustment range .

[0018] Step 2: Integrate the acquired underwater acoustic environment data, current sonar operating status data, and historical detection feedback into a state space vector. Specifically, the integrated state space vector can be represented as: ; in, The state space vector at time t (current) has a total of 18 dimensions, but it is not limited to this. Parameters such as ocean current velocity and suspected target location can be added on this basis to further enrich the state input dimensions, so that the AI ​​sonar parameter recommendation model can combine more environmental and target information to optimize parameters and improve the parameter adaptability in complex adversarial scenarios.

[0019] Step 3: Input the state space vector into the trained AI sonar parameter recommendation model for inference calculation to obtain the optimal combination of sonar operating parameters in the current environment. Specifically, the optimal combination of sonar operating parameters includes operating depth, operating frequency, and beam elevation angle.

[0020] See Figure 2 and Figure 3 The training method for the above AI sonar parameter recommendation model is as follows: Constructing state space vectors .

[0021] A physical environment model based on sonar equations is constructed as follows: ; in, This refers to the signal margin for an active sonar system with both transceiver and receiver integrated. At the source level, To spread the loss, For target strength, Noise level, As a directional index, This is the detection threshold.

[0022] The reward function is designed with the effective range R and the detection probability P as follows: ; in, The instant reward obtained at time t. To detect probability, To detect probability weights, This refers to the effective range of sonar. This represents the maximum effective range of the sonar. For the effective distance weight, For constraint terms, As a penalty for failure, This is an indicator function.

[0023] Constructing Action Space for: ; Using the above state space vector Physical environment model, motion space A reward function is defined, and an AI sonar parameter recommendation model is trained based on the Actor network-Critic network framework. The AI ​​sonar parameter recommendation model is preferably a Markov reinforcement learning model. During training, the PPO algorithm (Proximal Policy Optimization) or DDPG algorithm (Deep Deterministic Policy Gradient Algorithm) can be used to evaluate the action results of the AI ​​sonar parameter recommendation model. Taking the PPO algorithm as an example, the evaluation method for the action results of the AI ​​sonar parameter recommendation model is as follows: The generalized advantage estimation is used to calculate the action advantage function and the generalized advantage estimate of the action to be taken at time t. for: ; ; in, As a discount factor, To estimate hyperparameters for generalized dominance, For TD error, For action index variables, For the objective value function, Let be the state-space vector at time t+1. If the generalized dominance estimate is... If the generalized advantage estimate is 0, it means that taking the action at time t is better than the average level, increasing the probability of choosing that action; This indicates that taking an action at time t is worse than average, thus reducing its selection probability; if the generalized dominance estimate is... If , it means that the action taken at time t is at the average level and no adjustment is needed.

[0024] The Actor and Critic networks are continuously iterated and updated until the task level meets a preset convergence condition. This preset convergence condition is the probe probability. And detection range .

[0025] Step 4: Develop auxiliary decision-making information based on the optimal combination of sonar operating parameters and the physical environment model under the current conditions. This auxiliary decision-making information can be manually developed by the user and includes the maximum sonar range and detection probability map. The optimal combination of sonar operating parameters and the auxiliary decision-making information can then be graphically displayed on the control panel for relevant operators to use as a reference for decision-making.

[0026] The above description is merely a preferred embodiment of the present invention. It should be noted that for those skilled in the art, other parts not specifically described are existing technology or common knowledge. Several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A decision support method for underwater acoustic environments utilizing artificial intelligence, characterized in that, include: Step 1: Acquire current underwater acoustic environment data, current sonar operating status data, and historical detection feedback; Step 2: Integrate the acquired underwater acoustic environment data and the current working status data of the sonar with historical detection feedback into a state space vector; Step 3: Input the state space vector into the trained AI sonar parameter recommendation model for inference calculation to obtain the optimal combination of sonar operating parameters in the current environment; Step 4: Develop auxiliary decision-making information based on the optimal combination of sonar operating parameters and the physical environment model under the current environment; The underwater acoustic environment data includes hydrological parameters, ocean noise, and target information; The hydrological parameters include temperature. T ,salinity S and depth h The marine noise includes a noise intensity of 5 kHz. SN1 10kHz noise intensity SN2 Noise intensity at 15kHz SN3 The target information includes the target depth. l and target reflectivity β ; The sonar's current operating status data includes the current operating depth. d Current operating frequency f and the current beam elevation angle θ; The historical detection feedback includes the detection probability of the most recent M detections, the sonar range, and the parameter adjustment range of the previous round; The value of M is 3, and the parameter adjustment range includes the working depth adjustment range Δ. d , Operating frequency adjustment range Δ f Beam pitch adjustment amplitude Δ θ ; The training process of the AI ​​sonar parameter recommendation model is as follows: Constructing state space vectors st for: ； in, P 1 ,P 2 ,P 3 These represent the detection probabilities for the first, second, and third detections, respectively. R 1 ,R 2 ,R 3 These are the sonar ranges for the first, second, and third detections, respectively. A physical environment model based on sonar equations is constructed as follows: SE = SL - 2TL + TS - (NL - DI) - DT; in, SE This refers to the signal margin for an active sonar system with both transceiver and receiver integrated. SL At the sound source level, TL To spread the loss, TS For target strength, NL Noise level, DI As a directional index, DT The detection threshold; The reward function is designed with the effective range R and the detection probability P as follows: ； in, r t The instant reward obtained at time t. P To detect probability, α To detect probability weights, R This refers to the effective range of the sonar. R max This represents the maximum effective range of the sonar. σ For the effective distance weight, λ Here, γ represents the constraint term, and γ represents the penalty for failure. I(.) For indicator functions; Constructing Action Space at for: at→[d,f,θ] ； Using the state space vector st Physical environment model, motion space at The model is designed with a reward function and is trained on an AI sonar parameter recommendation model based on the Actor network-Critic network framework. During training, the action results of the AI ​​sonar parameter recommendation model are evaluated using a proximal policy optimization algorithm or a deep deterministic policy gradient algorithm. The Actor and Critic networks are continuously iterated and updated until the task level meets the preset convergence conditions.

2. The underwater acoustic environment decision support method using artificial intelligence according to claim 1, characterized in that, The method for evaluating the action results of the AI ​​sonar parameter recommendation model using the proximal policy optimization algorithm is as follows: The generalized advantage estimation is used to calculate the action advantage function and the generalized advantage estimate of the action to be taken at time t. At GAE for: ； ； in, As a discount factor, To estimate hyperparameters for generalized dominance, δ t+l For TD error, l For action index variables, For the objective value function, s t+1 Let t+1 be the state space vector. If the generalized dominance estimate is At GAE If the generalized dominance estimate is greater than 0, it means that taking the action at time t is better than the average level, thus increasing the probability of choosing that action; if the generalized dominance estimate is greater than 0, it means that taking the action at time t is better than the average level, thus increasing the probability of choosing that action; At GAE If the value is less than 0, it indicates that the action taken at time t is worse than the average, thus reducing the probability of its selection; if the generalized dominance estimate is... At GAE =0 indicates that the action taken at time t is at the average level and no adjustment is needed.

3. The underwater acoustic environment decision support method using artificial intelligence according to claim 1, characterized in that, The preset convergence condition is the detection probability. P≥75% And detection range R≥6km .

Images