A multi-source heterogeneous-based fishing boat trajectory data interpolation method and device
By combining improved generative adversarial networks (GANs) with multi-source heterogeneous fishing vessel trajectory data, the problem of missing and anomaly fishing vessel trajectory data in smart ocean scenarios was solved, achieving high-precision trajectory interpolation and improving the accuracy of fishing vessel trajectory analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HANGZHOU DIANZI UNIV
- Filing Date
- 2023-01-10
- Publication Date
- 2026-05-15
AI Technical Summary
In smart ocean scenarios, existing technologies are susceptible to the impact of severe weather and equipment errors on fishing vessel trajectory data, resulting in missing and abnormal data. Traditional fishing vessel trajectory interpolation methods fail to effectively utilize multi-source heterogeneous data, making it difficult to improve interpolation accuracy.
An improved generative adversarial networks (GANs) are used in conjunction with multi-source heterogeneous fishing vessel trajectory data. Through data preprocessing and iterative training of the generator and discriminator, fishing vessel trajectory data that conforms to the distribution of the original data is generated, and missing values are imputed.
It improves the accuracy of fishing vessel trajectory data interpolation, optimizes trajectory quality, avoids the impact of irregular data distribution on interpolation, and enhances the accuracy of fishing vessel trajectory analysis.
Smart Images
Figure CN116841993B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of fishing vessel trajectory interpolation in smart ocean scenarios, and particularly relates to a multi-source heterogeneous fishing vessel trajectory data interpolation technology. Background Technology
[0002] Fishing vessel trajectory data is an important analytical object in maritime intelligent transportation and maritime geographic information. In the smart ocean scenario, the collection of fishing vessel trajectory data mainly relies on AIS (Automatic Identification System) and Beidou fishing vessel control system (VMS) to obtain data such as fishing vessel position and speed. Fishing vessel trajectory data can be obtained by receiving data at different frequencies. Trajectory data has a wide range of applications in fishing vessel traffic engineering and has important application value.
[0003] In this era of rapid information technology development, historical trajectory analysis of fishing vessels has become a crucial component of smart ocean scenarios. Since fishing vessel dynamic information is measured by different devices or sensors at varying frequencies, one of the most popular areas of research in fishing vessel trajectory data is its analysis, as it provides a wealth of accurate navigation data. However, AIS and BeiDou data are easily affected by adverse weather, equipment malfunctions, and communication channel congestion. Analysis using raw data often results in missing and abnormal data, leading to errors in fishing vessel trajectory data analysis. Reconstructing fishing vessel trajectories to improve data quality is of great significance for further analysis and application. Missing values are mainly divided into three groups: Completely Random Missing (MACA), Random Missing (MAR), and Non-Random Missing (MNAR). In smart ocean scenarios, fishing vessel behavior data typically exhibits completely random missing values.
[0004] The purpose of fishing vessel trajectory interpolation technology is to improve data quality using techniques such as interpolation. However, traditional fishing vessel trajectory interpolation only targets data measured by a single device or sensor, neglecting the crucial role of combining and analyzing data from multiple devices in improving the accuracy of fishing vessel trajectory interpolation. Effectively interpolating fishing vessel trajectory data in smart ocean scenarios is a significant challenge. On the one hand, unlike vehicle or pedestrian trajectories, moving targets at sea are not constrained by geometric structures, and their motion behavior and patterns are more complex than those of targets on land. On the other hand, the trajectory distribution of fishing vessels at sea is irregular, making it crucial to understand their movement. Deep learning models, such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory Networks (LSTMs), can learn the irregular distribution of fishing vessel trajectories.
[0005] In recent years, Generative Adversarial Networks (GANs) have learned the irregular distribution of data through their generators and generated data that conforms to the original data distribution, effectively improving the accuracy of fishing boat trajectory interpolation. As a generative model, GANs are trained using an unsupervised learning approach, which can produce more realistic samples; at the same time, by combining multi-source heterogeneous fishing boat trajectory data, they can improve the accuracy of interpolating missing fishing boat trajectory data. Summary of the Invention
[0006] The purpose of this invention is to provide a multi-source heterogeneous fishing vessel trajectory data interpolation scheme, which can improve the accuracy of trajectory interpolation by combining multi-source heterogeneous fishing vessel trajectory data with large amounts of fishing vessel data. The technical solution adopted by this invention is as follows:
[0007] In a first aspect, the present invention provides a method for interpolating fishing vessel trajectory data based on multi-source heterogeneity, comprising the following steps:
[0008] S1: Data Acquisition
[0009] The fishing vessel sensors transmit basic information about the fishing vessel to the AIS and Beidou databases; a spatial index is established in the AIS and Beidou databases, and the spatial inclusion search method is used to obtain fishing vessel trajectory data in the near-shore area, resulting in multi-source heterogeneous fishing vessel trajectory data; the fishing vessel trajectory data includes Maritime Mobility Service Representation (MMSI), timestamp (t), longitude (lon), latitude (lat), ground speed (Sog), and ground heading (Cog).
[0010] S2: Data Preprocessing
[0011] 2-1 Multi-source heterogeneous fishing vessel trajectory data with different return frequencies Perform time interval standardization analysis, specifically:
[0012] Each fishing boat has multiple different fishing boat sensors. For each different fishing boat sensor, there is a set of data collection time intervals. ; This represents the time interval for data collection by the fishing vessel's sensor at position k on the hull.
[0013] Traversing all multi-source heterogeneous fishing vessel trajectory data For the data measured by sensor k on the fishing boat j N represents the number of data points measured by sensor k on fishing boat j, and D represents the data dimension. The data are judged sequentially. If the time t of each data point satisfies formula (1), retain it; otherwise, discard it to obtain the data. ;
[0014]
[0015] in This represents the threshold value, which can be 0.1.
[0016] 2-2 Number of multi-source heterogeneous fishing vessel trajectories after processing in step 2-1 Missing data is marked, and after the following processing, the result is obtained. ;
[0017]
[0018] in Representation matrix Elements in; express Elements in the matrix; n=1,2,…,N; d=1,2,…,D;
[0019] 2-3 Normalize the multi-source heterogeneous fishing vessel trajectory data processed in step 2-1;
[0020]
[0021] in This represents the data after normalization;
[0022] S3: Imput missing values in the multi-source heterogeneous fishing vessel trajectory data after preprocessing in step S2 using improved GANs; the improved GANs include a discriminator and a generator; the generator includes two fully connected layers, a ReLU activation function, and a sigmoid activation function, and the discriminator includes two fully connected layers, a ReLU activation function, and a sigmoid activation function; specifically:
[0023] 3-1 Data Generation:
[0024] First, random noise is randomly sampled from a uniform distribution [0, 0.01). , recorded as ,Will Input into the generator to get data :
[0025]
[0026]
[0027]
[0028]
[0029] in For the corresponding hyperparameters, and Let be the activation function, where, This represents the output of the first fully connected layer in the generator. This represents the output of the second fully connected layer. express The result after the activation function;
[0030] 3-2 Regarding the data from step 3-1 and step S2 data After processing with equation (7), we obtain :
[0031]
[0032] in This represents a combination of the missing data generated by the generator and the non-missing data.
[0033] Then After processing by the discriminator, the following is obtained: :
[0034]
[0035]
[0036]
[0037]
[0038] in For the corresponding hyperparameters, and For activation function, This represents the output of the first fully connected layer in the discriminator. This represents the output of the second fully connected layer. express The result after the activation function.
[0039] 3-3 Iteratively train steps 3-1 and 3-2 to finally obtain the output of GANs. ;
[0040] S4: Improve the accuracy of interpolated fishing vessel trajectory data by combining multi-source heterogeneous fishing vessel trajectory data.
[0041] 4-1 Record the actual data obtained by the vessel j from the sensor o on the fishing boat as follows: , where o k; The data from the fishing boat sensor k after interpolation of boat j. Update and get :
[0042]
[0043] in express Time, t represents The time corresponding to the interpolated position data; express Elements in a matrix;
[0044] 4-2 will be updated Combined with the original data, complete fishing boat trajectory data is obtained. The details are as follows:
[0045]
[0046] in Represents the original data. This indicates dot product.
[0047] Preferably, the generator loss function is:
[0048]
[0049] in This represents the output of the generator. express Output, Represents random noise The expected value of the distribution.
[0050] Preferably, the loss function of the discriminator is:
[0051]
[0052] in This indicates the output that the discriminator determines to be true. This indicates the output that the discriminator determines to be false. This represents the expected value that follows the true data distribution. This represents the expectation that the data follows a spurious distribution.
[0053] Secondly, the present invention provides a fishing vessel trajectory data interpolation device based on multi-source heterogeneity, comprising:
[0054] The data acquisition module is used to collect multi-source heterogeneous fishing vessel trajectory data;
[0055] The data preprocessing module is used to perform time interval standardization analysis, missing data marking, and normalization on multi-source heterogeneous fishing vessel trajectory data with different return frequencies.
[0056] The missing value imputation module is used to impute missing values in preprocessed multi-source heterogeneous fishing vessel trajectory data using improved GANs.
[0057] The data integration module is used to integrate the data after interpolation processing of missing values from all sensors on the same fishing vessel.
[0058] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method described thereon.
[0059] Fourthly, the present invention provides a computing device, including a memory and a processor, characterized in that the memory stores executable code, and when the processor executes the executable code, it implements the method described above.
[0060] The beneficial effects of this invention are as follows:
[0061] This invention provides a method for interpolating fishing vessel trajectory data based on multi-source heterogeneity. It can improve the interpolation accuracy of missing trajectory data of fishing vessels in nearshore areas under smart ocean scenarios by combining multi-source heterogeneous fishing vessel trajectory data, and optimize the quality of fishing vessel trajectories. The trajectory interpolation method adopts improved GANs, which effectively avoids the difficulty in capturing irregular distribution of original data, thereby improving the interpolation accuracy of missing trajectories and avoiding errors caused by fishing vessel trajectory data analysis. Attached Figure Description
[0062] Figure 1 This is a schematic diagram of the structure of the fishing vessel trajectory data interpolation method of the present invention;
[0063] Figure 2 A schematic diagram of the improved GANs architecture provided by this invention;
[0064] Figure 3 This is a flowchart of the fishing vessel trajectory data interpolation method of the present invention. Detailed Implementation
[0065] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.
[0066] A multi-source heterogeneous fishing vessel trajectory data interpolation device, the specific steps of which are as follows: Figure 1 As shown, where:
[0067] The data acquisition module includes establishing a spatial index for the MPP database storing AIS data and BeiDou data, using spatial inclusion search to obtain fishing vessel trajectory data in the nearshore area, and obtaining multi-source heterogeneous fishing vessel trajectory data.
[0068] The MPP database is used to store fishing vessel movement information, and can quickly search and extract fishing vessel trajectory data for a specific area and time.
[0069] The fishing vessel trajectory data includes Maritime Mobility Service (MMSI), timestamp t, longitude lon, latitude lat, ground speed Sog, and ground heading Cog.
[0070] The multi-source heterogeneous fishing vessel trajectory data includes fishing vessel trajectory data with different data structures measured on different sensors;
[0071] The data preprocessing module processes the multi-source heterogeneous fishing vessel trajectory data to obtain a multi-source heterogeneous fishing vessel trajectory dataset.
[0072] The processing of multi-source heterogeneous fishing vessel trajectory data includes:
[0073] S21 uses multi-source heterogeneous fishing vessel trajectory data with different return frequencies. Perform time interval standardization analysis, specifically:
[0074] Each fishing boat has multiple different fishing boat sensors. For each different fishing boat sensor, there is a set of data collection time intervals. ; This represents the time interval for data collection by the fishing vessel's sensor at position k on the hull.
[0075] Traversing all multi-source heterogeneous fishing vessel trajectory data For the data measured by sensor k on the fishing boat j N represents the number of data points measured by sensor k on fishing boat j, and D represents the data dimension. The data are judged sequentially. If the time t of each data point satisfies formula (1), retain it; otherwise, discard it to obtain the data. ;
[0076]
[0077] in This represents the threshold value, which can be 0.1.
[0078] S22 is the number of multi-source heterogeneous fishing vessel trajectories processed in step S21. Missing data is marked, and after the following processing, the result is obtained. ;
[0079]
[0080] in Representation matrix Elements in; express Elements in the matrix; n=1,2,…,N; d=1,2,…,D;
[0081] S23 normalizes the multi-source heterogeneous fishing vessel trajectory data processed in step S21;
[0082]
[0083] in This represents the data after normalization;
[0084] The missing value imputation module is used to impute missing values in preprocessed multi-source heterogeneous fishing vessel trajectory data using improved GANs to obtain complete fishing vessel trajectory data.
[0085] The improved GANs structure is as follows: Figure 2 As shown, it contains two networks: a generator and a discriminator.
[0086] Both the generator and the discriminator consist of two fully connected layers, a ReLU activation function, and a sigmoid activation function.
[0087] The processing procedure in the generator is as follows:
[0088] First, random noise is randomly sampled from a uniform distribution [0, 0.01). , recorded as Input the data into the generator to get the data. :
[0089]
[0090]
[0091]
[0092]
[0093] in For the corresponding hyperparameters. and Let be the activation function; where This represents the output of the first fully connected layer in the generator. This represents the output of the second fully connected layer. express The result after the activation function;
[0094] 3-2 Regarding the data obtained in step 3-1 and the data obtained from S2 processing After processing with Equation 7, we obtain :
[0095]
[0096] in This represents a combination of the missing data generated by the generator and the data that is not missing.
[0097] Then After processing by the discriminator, the following is obtained: ;
[0098]
[0099]
[0100]
[0101]
[0102] in For the corresponding hyperparameters. and Activation function This represents the output of the first fully connected layer in the discriminator. This represents the output of the second fully connected layer. express The result after the activation function.
[0103] During GAN training, the generator loss function is:
[0104]
[0105] in This represents the output of the generator. express Output, Represents random noise The expected value of the distribution.
[0106] The loss function of the discriminator is:
[0107]
[0108] in This indicates the output that the discriminator determines to be true. The output indicates that the discriminator determines the value to be false. This represents the expected value that follows the true data distribution. This represents the expectation that the data follows a spurious distribution.
[0109] The generator and discriminator are trained iteratively to obtain the output of GANs. .
[0110] The data integration module is used to integrate the imputed data from all fishing boat sensors on the same fishing vessel; and to integrate the complete fishing boat trajectory data output by the improved GANs model. To improve the accuracy of fishing vessel trajectory interpolation;
[0111] 4-1 Record the actual data obtained by the vessel j from the sensor o on the fishing boat as follows: , where o k; The data from the fishing boat sensor k after interpolation of boat j. Update and get :
[0112]
[0113] in express Time, t represents The time corresponding to the interpolated position data; express Elements in a matrix;
[0114] 4-2 will be updated Combined with the original data, complete fishing boat trajectory data is obtained. The details are as follows:
[0115]
[0116] in Represents the original data. This indicates dot product.
[0117] like Figure 3 As shown, the workflow of a multi-source heterogeneous fishing vessel trajectory data interpolation method is as follows:
[0118] Step S1: The data acquisition module collects fishing vessel trajectory information from the MPP database. The fishing vessel sensors transmit the basic information of the fishing vessel to the AIS and Beidou databases. A spatial index is established in the AIS and Beidou databases, and the spatial inclusion search method is used to obtain fishing vessel trajectory data in the nearshore area, resulting in multi-source heterogeneous fishing vessel trajectory data. The fishing vessel trajectory information includes Maritime Mobility Service Representation (MMSI), timestamp t, longitude lon, latitude lat, ground speed Sog, and ground heading Cog.
[0119] Step S2: The data preprocessing module processes the multi-source heterogeneous fishing vessel trajectory data with different return frequencies. Perform time interval standardization analysis, missing data marking, and data normalization.
[0120] Step S3: The missing value imputation module uses improved GANs to imput missing values in the preprocessed multi-source heterogeneous fishing vessel trajectory data into complete fishing vessel trajectory data.
[0121] Step S4: The data integration module integrates the missing value interpolated data from all fishing boat sensors on the same fishing boat, improving the accuracy of the complete fishing boat trajectory data generated in step S3.
[0122] Functions and effects of the embodiments
[0123] Based on the multi-source heterogeneous fishing vessel trajectory data interpolation method provided in this example, a spatial index is first established in the MPP database to search for fishing vessel trajectory data in nearshore areas, obtaining multi-source heterogeneous fishing vessel trajectory data. Secondly, the fishing vessel trajectory data is normalized by time interval standardization and data normalization to obtain incomplete data. Then, improved GANs are used to interpolate the incomplete fishing vessel trajectory data, and the accuracy of the interpolation is improved by referring to the distribution of trajectory data generated by different fishing vessel sensors. This method is useful for historical trajectory playback, prediction, and analysis of fishing vessels in marine scenarios.
[0124] The above examples are only used to illustrate specific embodiments of the present invention, and the present invention is not limited to the scope of the above examples.
Claims
1. A method for interpolating fishing vessel trajectory data based on multi-source heterogeneity, characterized in that... The method includes the following steps: S1: Data Acquisition The fishing vessel sensors transmit basic information about the fishing vessel to the AIS and Beidou databases; a spatial index is established in the AIS and Beidou databases, and the spatial inclusion search method is used to obtain fishing vessel trajectory data in the near-shore area, resulting in multi-source heterogeneous fishing vessel trajectory data; the fishing vessel trajectory data includes Maritime Mobility Service Representation (MMSI), timestamp (t), longitude (lon), latitude (lat), ground speed (Sog), and ground heading (Cog). S2: Data Preprocessing 2-1 Multi-source heterogeneous fishing vessel trajectory data with different return frequencies Perform time interval standardization analysis, specifically: Each fishing boat has multiple different fishing boat sensors. For each different fishing boat sensor, there is a set of data collection time intervals. ; This represents the time interval for data collection by the fishing vessel's sensor at position k on the hull. Traverse all multi-source heterogeneous fishing vessel trajectory data For the data measured by sensor k on the fishing boat j N represents the number of data points measured by sensor k on fishing boat j, and D represents the data dimension. The data are judged sequentially. If the time t of each data point satisfies formula (1), retain it; otherwise, discard it to obtain the data. ; ； in This represents the threshold value, which is 0.
1. 2-2 Number of multi-source heterogeneous fishing vessel trajectories after processing in step 2-1 Missing data is marked, and after the following processing, the result is obtained. ; ； in Representation matrix Elements in; express Elements in the matrix; n=1,2,…,N; d=1,2,…,D; 2-3 Normalize the multi-source heterogeneous fishing vessel trajectory data processed in step 2-1; ； in This represents the data after normalization; S3: Imput missing values in the multi-source heterogeneous fishing vessel trajectory data after preprocessing in step S2 using improved GANs; the improved GANs include a discriminator and a generator; the generator includes two fully connected layers, a ReLU activation function, and a sigmoid activation function, and the discriminator includes two fully connected layers, a ReLU activation function, and a sigmoid activation function; specifically: 3-1 Data Generation: First, random noise is randomly sampled from a uniform distribution [0, 0.01). , recorded as ,Will Input into the generator to get data : ； ； ； ； in For the corresponding hyperparameters, and Let be the activation function, where, This represents the output of the first fully connected layer in the generator. This represents the output of the second fully connected layer. express The result after the activation function; 3-2 Regarding the data from step 3-1 and step S2 data After processing with equation (7), we obtain : ； in This represents a combination of the missing data generated by the generator and the non-missing data. Then After processing by the discriminator, the following is obtained: : ； ； ； ； in For the corresponding hyperparameters, and For activation function, This represents the output of the first fully connected layer in the discriminator. This represents the output of the second fully connected layer. express The result after the activation function; 3-3 Iteratively train steps 3-1 and 3-2 to finally obtain the output of GANs. ; S4: Improve the accuracy of interpolated fishing vessel trajectory data by combining multi-source heterogeneous fishing vessel trajectory data. 4-1 Record the actual data obtained by the vessel j from the sensor o on the fishing boat as follows: , where o k; The data from the fishing boat sensor k after interpolation of boat j. Update and get : ； in express Time, t represents The time corresponding to the interpolated position data; express Elements in a matrix; 4-2 will be updated Combined with the original data, complete fishing boat trajectory data is obtained. The details are as follows: ； in Represents the original data. This indicates dot product.
2. The method according to claim 1, characterized in that... The generator loss function is: ； in This represents the output of the generator. express Output, Represents random noise The expected value of the distribution.
3. The method according to claim 1 or 2, characterized in that... The loss function of the discriminator is: ； in This indicates the output that the discriminator determines to be true. This indicates the output that the discriminator determines to be false. This represents the expected value that follows the true data distribution. This represents the expectation that the data follows a spurious distribution.
4. A fishing vessel trajectory data interpolation device based on multi-source heterogeneity that implements the method of any one of claims 1-3, characterized in that... include: The data acquisition module is used to collect multi-source heterogeneous fishing vessel trajectory data; The data preprocessing module is used to perform time interval standardization analysis, missing data marking, and normalization on multi-source heterogeneous fishing vessel trajectory data with different return frequencies. The missing value imputation module is used to impute missing values in preprocessed multi-source heterogeneous fishing vessel trajectory data using improved GANs. The data integration module is used to integrate the data after interpolation processing of missing values from all sensors on the same fishing vessel.
5. A computer-readable storage medium having a computer program stored thereon, characterized in that... When the computer program is executed in the computer, it causes the computer to perform the method according to any one of claims 1-3.
6. A computing device, comprising a memory and a processor, characterized in that... The memory stores executable code, and when the processor executes the executable code, it implements the method of any one of claims 1-3.