Bayesian-based multi-AUV (Autonomous Underwater Vehicle) collaborative dynamic target searching method
Based on Bayesian -based multi -AUV collaborative dynamic target search methods, establish target movement and detection models, use the maximum discount time income algorithm and inspiration factor, to achieve the coordinated search and target tracking of the AUV cluster, solving the dynamic target search in the marine environment The problem of low efficiency and complexity has improved search efficiency and robustness.
Patent Information
- Application Number
- CN202411544506.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-01-28
AI Technical Summary
In the marine environment, dynamic target search faces problems such as target motion uncertainty, underwater environment complexity and sensor measurement errors. The search efficiency of single AUVs is low, and the searcher cluster collaborative operation is complex.
The multi -AUV collaborative dynamic target search method based on Bayesian is used to establish a target motion model and detection model. Through the maximum discount time income algorithm and inspiration factor, build a cluster decision model to achieve the AUV cluster collaborative search and target tracking.
The efficiency and robustness of dynamic target search in water can quickly discover dynamic targets and reduce the blindness in search in the case of limited thermal initial location information information.
Smart Images
Figure CN119376421A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of target search, and in particular to a Bayesian-based multi-AUV collaborative dynamic target search method and a system thereof. Background Art
[0002] Autonomous Underwater Vehicles (AUVs) and other carrier tools play an important role in tasks such as marine environmental surveys and underwater target detection, and are widely used in marine scientific research and military application research. Affected by factors such as the complexity of the underwater environment, the uncertainty of target motion, and the measurement errors of the sensors themselves, dynamic target search faces many difficulties. Compared with the search of a single AUV, the collaborative operation of a searcher cluster can greatly improve the efficiency and robustness of underwater target search and tracking tasks, effectively expand the task area and coverage, and significantly overcome the various difficulties faced by a single searcher in the ocean. However, the searcher cluster needs to solve problems such as information fusion and task allocation between AUVs, and the collaborative execution of search tasks is more difficult and complex.
[0003] At present, when dealing with search problems with multiple uncertainties and constraints, traditional search methods have the advantage of simple calculations, but they do not give full play to the autonomy of searcher clusters and have the defects of low efficiency, such as lawnmower search, snake search, spiral search, etc. With the improvement of computing power and information processing capabilities, it is very necessary to add a search decision layer, improve the collaborative search performance of searcher clusters, and improve the autonomy of collaborative search. Summary of the invention
[0004] In order to solve the above technical problems, the present invention proposes a Bayesian-based multi-AUV collaborative dynamic target search method to solve the problems of target uncertainty, search time constraints and detection probability constraints. On the basis of only the prior information of the target's initial position, the Bayesian method is used to study the search problem of a searcher cluster for a dynamic target in a certain sea area. The search behavior is decided based on the cluster's perception of the environment. In the process of searching for the target, the tracking and estimation of the target's motion state are synchronously realized.
[0005] The first object of the present invention is to provide a Bayesian-based multi-AUV collaborative dynamic target search method for underwater dynamic target search, comprising: Establish target motion model; A detection model of a single searcher is established based on the target motion model and the state of the single searcher; According to the target motion model, the detection model of a single searcher and the target confidence, a cluster decision model is constructed to expand the search of a single searcher to multiple searchers, i.e., searcher clustering. The optimal action sequence in the process of searcher cluster searching for dynamic targets is solved by the maximum discounted time-benefit algorithm; When the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
[0006] Preferably, the target motion model is established based on the Singer motion model, and the target motion equation is:
[0007] In the formula, Indicates the target's motion state; represents Gaussian white noise; represents the state transfer matrix; is the noise driving matrix.
[0008] Preferably, the detection model is the likelihood probability that the AUV can detect the target in the current state.
[0009] Preferably, the detection model is as follows:
[0010] In the formula, represents the maximum detection probability of the sensor; Indicates the maximum detection range of the sensor; Represents parameters; Indicates the target's motion state; Indicates the searcher status;
[0011] in, Indicates the event of detecting a target. Indicates an event in which a target was not detected; Represents the observation results.
[0012] Preferably, the target confidence is as follows:
[0013] In the formula, Indicated in Confidence in goals at all times; From the initial time to All observations at time instant; From the initial time to The status of all searchers at the moment; Indicates the motion state of the target.
[0014] Preferably, the decision model of the cluster includes a logical model including 7 elements, wherein the 7 elements correspond to the environment, the searcher state, the action space, the detection model, the target state, the target confidence, and the decision strategy, respectively.
[0015] Preferably, when solving by the maximum discounted time benefit algorithm, a heuristic factor is introduced to obtain the utility function of the heuristic maximum discounted time benefit algorithm; The optimal action sequence of the searcher cluster in the process of searching for dynamic targets is solved according to the utility function of the heuristic maximum discounted time-benefit algorithm.
[0016] The second object of the present invention is to provide a Bayesian-based multi-AUV collaborative dynamic target search system, comprising: A model building module is used to build a target motion model; build a detection model of a single searcher according to the target motion model and the state of the single searcher; obtain target confidence according to the target motion model, the state of the single searcher and the detection model; A clustering module is used to build a clustering decision model based on the target motion model, the detection model of a single searcher, and the target confidence, so as to expand the search of a single searcher to multiple searchers, i.e., searcher clustering; The target detection module is used to solve the optimal action sequence of the searcher cluster in the process of searching for dynamic targets through the maximum discounted time-benefit algorithm; when the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
[0017] The third object of the present invention is to provide an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the above method when executed by the processor.
[0018] A fourth object of the present invention is to provide a storage medium having a computer program stored thereon, which is used to execute the steps of the above method when the computer program is executed.
[0019] The present invention has at least the following beneficial effects: The present invention provides a Bayesian-based multi-AUV collaborative dynamic target search method. Based on only the prior information of the target's initial position, the present invention adopts a Bayesian method to study the search problem of an AUV cluster for a dynamic target in a certain sea area, relies on the cluster's perception of the environment to decide the search behavior, and synchronously realizes the tracking and estimation of the target's motion state during the target search process.
[0020] The present invention provides a time-decision processable model for solving the multi-target search problem of an AUV cluster under target state uncertainty under spatial constraints.
[0021] The present invention introduces a discount factor, weighs the benefits of different time periods, and can detect areas with higher target confidence probability faster and earlier. The present invention designs a heuristic factor to improve the short-sightedness problem of dynamic target search and reduce the blindness in the search. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] Figure 1 Search the 10th second process graph for the cluster; Figure 2 Search the 20th second process graph for the cluster; Figure 3 30-second process graph for cluster search; Figure 4 Search for the 44th second process graph for the cluster; Figure 5 is the overall search trajectory; Figure 6 Estimated map for target position; Figure 7 This is a schematic diagram of the scope of the profit function; Figure 8 A diagram showing the search effect of the algorithm with heuristic factors. DETAILED DESCRIPTION
[0023] In order to illustrate the technical means and effects adopted by the present invention to achieve the predetermined purpose of the invention, the following is a detailed description in conjunction with embodiments.
[0024] Based on only the prior information of the target's initial position, the present invention adopts the Bayesian method to study the search problem of a searcher cluster for a dynamic target in a certain sea area, relies on the cluster's perception of the environment to decide the search behavior, and synchronously realizes the tracking and estimation of the target's motion state during the target search process.
[0025] The dynamic target targeted by the present invention exists in a closed and limited environment, and the searcher cluster only has a probabilistic prior information about its initial position. In other words, the position of the target is uncertain and has dynamic characteristics that are independent of the searcher. A Bayesian-based multi-AUV collaborative dynamic target search method is proposed for the unilateral cluster search problem of such unrelated search objects. The searcher involved in the present invention is an autonomous underwater vehicle (AUV).
[0026] In order to achieve the above object, the present invention provides a Bayesian-based multi-AUV collaborative dynamic target search method for underwater dynamic target search, comprising: S1, establish target motion model; The target motion model is established based on the Singer motion model, and the target motion equation is:
[0027] In the formula, Indicates the target's motion state; represents Gaussian white noise; represents the state transfer matrix; is the noise driving matrix.
[0028] S2, establishing a detection model of a single searcher according to the target motion model and the state of the single searcher; The detection model is the likelihood probability that the AUV can detect the target in the current state.
[0029] The detection model is as follows:
[0030] In the formula, represents the maximum detection probability of the sensor; Indicates the maximum detection range of the sensor; Represents parameters; Indicates the target's motion state; Indicates the searcher status; Indicates that the target motion state at time k is 、AUV status is The likelihood probability that the AUV detects the target when .
[0031]
[0032] in, Indicates the event of detecting a target. Indicates an event in which a target was not detected; Represents the observation results.
[0033] S3, obtaining target confidence according to the target motion model and the state and detection model of a single searcher; The target confidence is as follows:
[0034] In the formula, Indicated in Confidence in goals at all times; From the initial time to All observations at time instant; From the initial time to The status of all searchers at the moment; Indicates the target's motion state; Indicates that given the time from the initial moment to The subjective probability distribution of the AUV to the target state under the observation information and state information of the AUV at the moment.
[0035] S4, building a cluster decision model based on the target motion model, the detection model of a single searcher, and the target confidence, and extending the search of a single searcher to multiple searchers, i.e., searcher clustering; The decision model of the cluster includes a logical model including 7 elements, wherein the 7 elements correspond to the environment, the searcher state, the action space, the detection model, the target state, the target confidence, and the decision strategy, respectively.
[0036] The optimal action sequence in the process of searcher cluster searching for dynamic targets is solved by the maximum discounted time-benefit algorithm; When the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
[0037] Among them, when solving by the maximum discounted time benefit algorithm, the heuristic factor is introduced to obtain the utility function of the heuristic maximum discounted time benefit algorithm; The optimal action sequence of the searcher cluster in the process of searching for dynamic targets is solved according to the utility function of the heuristic maximum discounted time-benefit algorithm.
[0038] In order to illustrate a Bayesian-based multi-AUV collaborative dynamic target search method provided by the present invention, the following specific steps are described in conjunction with the accompanying drawings: Step 1: Establish a motion model of a dynamic target: Based on the Singer motion model, Taking the motion on the coordinate axis as an example, the target motion equation is:
[0039] In the formula, Indicates the target's motion state. represents Gaussian white noise, represents the state transfer matrix, is the noise driving matrix.
[0040] Step 2: Establish the confidence of dynamic targets: Target confidence is the subjective perception of the target information by the searcher. It represents the task environment Within , the AUV believes that the target exists in the specific probability distribution of the location. At this moment, the target's real motion state Yes, but the AUV cannot be sure during the search process when the target has not been found. The specific value of , so the probability form is used to fit it, that is:
[0041] In the formula, From the initial time to All observations at time From the initial time to All searcher states at the moment, It means in Here, the observations are obtained based on the detection model in step 3.
[0042] Given a target Probability distribution of state at a given moment The prediction step predicts the target at the first State confidence at the moment The specific principle is:
[0043] After getting the latest observation, update the posterior target confidence:
[0044] Summarizing the above prediction update process, we obtain the target confidence update algorithm, as shown in Table 1.
[0045] Table 1 Target confidence update algorithm
[0046] Step 3: Establish the detection model of AUV. The detection model represents the likelihood probability that AUV can detect the target in the current state. Indicates that Indicates the event of detecting a target. represents the event of not detecting the target, and the two constitute the observation space, namely:
[0047] in, express The state of the target at the moment, due to The sensor states and agents are unified by default, so they are all used express.
[0048] The detection model of AUV in the present invention adopts an exponential detection function, and its mathematical expression is:
[0049] in represents the maximum detection probability of the sensor, It indicates the maximum detection range of the sensor. It is a parameter given by humans.
[0050] Step 4: Establish the cluster decision model as a logical model including 7 elements: , which correspond to the seven parts of environment, searcher state, action space, detection model, target state, target confidence, and decision strategy. When there are multiple searchers forming a searcher cluster, the cluster search and tracking problem model (Search and Tracking POMDP, SAT-POMDP) is obtained by expanding the dimensions of the agent state, action, and observation results. In the cluster SAT-POMDP model, the control decision layer makes the search decision of the cluster as a whole based on the dynamic target confidence and controls the search motion of the AUV cluster. This step solves the target search problem of the searcher cluster under the target state uncertainty under spatial constraints by designing a time-decision processable model.
[0051] It should be noted that the logical model is aimed at the physical scenario of underwater dynamic target search and tracking. The search and tracking problem is theoretically modeled in mathematical language to obtain the search and tracking problem model in the form of POMDP under probability theory (Search and Tracking POMDP, SAT-POMDP).
[0052] use represents the search cluster composed of all agents. Therefore, in the search tracking problem model, the AUV state , Observation And actions All are rewritten to the status of the entire cluster , Observation and actions , the information related to the target remains unchanged, and the following assumptions are made about the cluster as a whole: (1) Observation independence assumption. The detection process of each searcher in the cluster is independent of each other, and the state and observation of each searcher are also independent of each other, that is:
[0053] (2) Markov assumption. The movement and detection of each searcher in the cluster are Markov-like, that is, the state at the next moment is determined only by the current moment and has nothing to do with past information.
[0054] Step 5: Design the Maximum Discounted Time Reward (MDTR) algorithm. Assume that the initial state of the cluster is , given Step-length action sequence , and the target confidence is ,use Represents the discount factor and establishes the discounted time benefit function of the cluster for:
[0055] Since the discounted time benefit strategy comes from the AUV cluster The joint probability of detection events within steps can be maximized To solve the action sequence. Maximize It may not provide a globally optimal strategy, but it can ensure that Optimality within.
[0056] In order to further illustrate the technical means adopted by the present invention to achieve the predetermined purpose, a simulation of the AUV cluster collaborative search for dynamic targets is carried out, which is described in detail as follows in conjunction with the accompanying drawings.
[0057] Assuming that the target motion state is still as shown in Table 2, the prediction step of data fusion uses the Singer model to provide predictions, and the parameters are selected as , , ; When searching without observation information, the update steps in Table 1 are used to provide the latest target confidence. The detection model parameters are selected as , , Take 0.7, that is, the effective detection range of the sensor is 50 meters. The AUV cluster consists of two AUVs, and their partial information is shown in Table 3. .
[0058] Table 2 Target movement process
[0059] Table 3 Searcher AUV cluster information
[0060] The simulation duration is limited to 200 seconds. When the shortest distance between the target and the cluster is within 50 meters, the target is considered to be found and the search process ends immediately.
[0061] like Figures 1 to 4 The figure shows the search process of the searcher cluster from the initial moment until the target is found at intervals of 10 seconds. It can be seen that the AUV cluster found the target at the 44th second.
[0062] The overall search process trajectory and target position estimation are as follows: Figures 5 and 6 As shown, combined Figure 1It can be seen that due to the lack of observed target information, the target position estimation obtained based on the detection model still retains a large error. However, in this case, the search algorithm still plays a good role, enabling the AUV cluster to search around the high-probability area and successfully find the target in the 44th second.
[0063] Step 6: Introduce a heuristic factor to fully consider the impact of long-term benefits and build a non-short-sighted search strategy. Definition is the expectation of future returns, and the expression of long-term returns is:
[0064] In the formula, is Always have expectations for future earnings; is a given time step; is the number of AUVs; represents the discount factor; represents the likelihood probability when the AUV does not detect the target; Indicates the status of the AUV; represents the AUV observation results; Indicates the state of the target; is an estimate of the target confidence.
[0065] In this formula, infinite series calculations are required, which is difficult to handle in most cases. Figure 7 It can be seen that the long-term benefit is the searcher cluster in state After that, we assume that the cluster is in state When there is a detection sensor with infinite distance, the detection model of this sensor is used to construct the expected return in the future, and it is combined in the framework of the utility function to form a heuristic search strategy.
[0066] Define the infinite distance detection model as:
[0067] in represents the Euclidean distance between the target and the searcher, represents the searcher's speed, represents the possibility of detecting the target near the current state of the searcher, It is still a normalization factor, but in order to increase the importance of long-term returns, we can also let , no normalization is performed.
[0068] According to the infinite distance detection model, the expression of undetected events at infinite distance can also be obtained:
[0069] By replacing the detection probability and non-detection probability in the utility function (the benefit function in step 5) with the above two functions, we can get the utility function of the heuristic maximum discounted time benefit (Heuristic MDTR, H-MDTR) algorithm. for:
[0070] In the formula, represents the discount factor; is a given time step; is the number of AUVs; represents the likelihood probability when the AUV does not detect the target; Indicates the status of the AUV; represents the AUV observation results; is the estimate of the target confidence; Indicates the state of the target; Represents the probability function of an undetected event at infinite distance.
[0071] It can be seen that by introducing the infinite distance detection model as a hypothetical detection device, the infinity in time is transformed into the finiteness in space, so as to construct a heuristic search strategy without calculating infinite time benefits, in order to reduce the short-sighted characteristics of the original strategy.
[0072] Parameters , , To verify the influence of the heuristic factor, a simulation experiment was conducted on the search algorithm containing the heuristic factor under the same conditions, and the final search trajectory was obtained as follows: Figure 8 As shown in the figure, the target is found in the 27th second. It can be seen that the long-term benefit is added to the maximum discounted time benefit algorithm, which only has immediate benefit and recent benefit. The heuristic factor can effectively improve the short-sightedness problem of the maximum discounted time benefit algorithm, further shorten the search time, and achieve better search results.
[0073] In summary, the present invention first establishes a motion model of a dynamic target, uses the target motion model as a prediction model in the target confidence, calculates the target confidence, and then combines the detection model to obtain the likelihood probability that the AUV can detect the target in the current state, thereby obtaining the AUV's estimate of the target information. By constructing a cluster decision model, the search of a single AUV is extended to multiple AUVs, namely an AUV cluster. And the optimal action sequence in the process of AUV cluster searching for dynamic targets is solved by the maximum discounted time benefit algorithm. When the distance between the AUV cluster and the dynamic target is less than the maximum detection range of the sensor, it is considered that the target has been detected. Finally, by introducing a heuristic factor, the above search algorithm is further optimized to shorten the search time.
[0074] The present invention provides a Bayesian-based multi-AUV collaborative dynamic target search system, comprising: A model building module is used to build a target motion model; build a detection model of a single searcher according to the target motion model and the state of the single searcher; obtain target confidence according to the target motion model, the state of the single searcher and the detection model; A clustering module is used to build a clustering decision model based on the target motion model, the detection model of a single searcher, and the target confidence, so as to expand the search of a single searcher to multiple searchers, i.e., searcher clustering; The target detection module is used to solve the optimal action sequence of the searcher cluster in the process of searching for dynamic targets through the maximum discounted time-benefit algorithm; when the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
[0075] The present invention provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the steps of the above method are implemented when the computer program is executed by the processor.
[0076] The present invention provides a storage medium on which a computer program is stored. When the computer program is run, it is used to execute the steps of the above method.
[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A Bayesian-based multi-AUV collaborative dynamic target search method, characterized in that: Used for dynamic target search in water, including: Establish target motion model; A detection model of a single searcher is established based on the target motion model and the state of the single searcher; According to the target motion model, the detection model of a single searcher and the target confidence, a cluster decision model is constructed to expand the search of a single searcher to multiple searchers, i.e., searcher clustering. The optimal action sequence in the process of searcher cluster searching for dynamic targets is solved through the maximum discounted time-benefit algorithm; When the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
2. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 1 is characterized in that: The target motion model is established based on the Singer motion model, and the target motion equation is: In the formula, Indicates the target's motion state; represents Gaussian white noise; represents the state transfer matrix; is the noise driving matrix.
3. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 1 is characterized in that: The detection model is the likelihood probability that the AUV can detect the target in the current state.
4. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 3 is characterized in that: The detection model is as follows: In the formula, represents the maximum detection probability of the sensor; Indicates the maximum detection range of the sensor; Represents parameters; Indicates the target's motion state; Indicates the searcher status; in, Indicates the event of detecting a target. Indicates an event in which a target was not detected; Represents the observation results.
5. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 1 is characterized in that: The target confidence is as follows: In the formula, Indicated in Confidence in goals at all times; From the initial time to All observations at time instant; From the initial time to The state of all searchers at the moment; Indicates the motion state of the target.
6. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 1 is characterized in that: The decision model of the cluster includes a logical model including 7 elements, wherein the 7 elements correspond to the environment, the searcher state, the action space, the detection model, the target state, the target confidence, and the decision strategy, respectively.
7. The Bayesian-based multi-AUV collaborative dynamic target search method according to claim 1 is characterized in that: When solving the problem through the maximum discounted time benefit algorithm, a heuristic factor is introduced to obtain the utility function of the heuristic maximum discounted time benefit algorithm; The optimal action sequence of the searcher cluster in the process of searching for dynamic targets is solved according to the utility function of the heuristic maximum discounted time-benefit algorithm.
8. A Bayesian-based multi-AUV collaborative dynamic target search system, characterized in that: include: A model building module is used to build a target motion model; a detection model of a single searcher is built according to the target motion model and the state of the single searcher; Obtain target confidence based on the target motion model and the state and detection model of a single searcher; A clustering module is used to build a clustering decision model based on the target motion model, the detection model of a single searcher, and the target confidence, so as to expand the search of a single searcher to multiple searchers, i.e., searcher clustering; The target detection module is used to solve the optimal action sequence of the searcher cluster in the process of searching for dynamic targets through the maximum discounted time benefit algorithm; When the distance between the searcher cluster and the dynamic target is less than the maximum detection range of the sensor, the target is detected.
9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the steps of the method according to any one of claims 1 to 6 when executed by the processor.
10. A storage medium, characterized in that: A computer program is stored thereon, and when the computer program is run, it is used to execute the steps of the method according to any one of claims 1 to 6.