Maritime information broadcasting method based on multi-source data fusion

Abstract

A maritime information broadcasting method based on multi-source data fusion comprises the following steps: S1, setting a data interface, and receiving maritime safety information data sent by different mechanisms in time; s2, analyzing the acquired maritime affair safety information data, and converting the data into a standardized data format for subsequent processing; s3, fusing the data from different sources, and removing repeated information; s4, merging the information with the same information content and different expression modes, and saving broadcast resources; s5, establishing a danger early warning model, and dividing danger levels according to the acquired maritime information; and S6, performing priority ranking according to the time and the danger level of the information, and broadcasting the information according to the priority sequence. According to the maritime information broadcasting method based on multi-source data fusion, information data of different data sources can be fused and optimized, the manual inspection time is shortened, and broadcasted information can be more reliable and accurate.

Description

Technical Field

[0001] The present invention belongs to the field of maritime safety information broadcasting, and in particular relates to a maritime information broadcasting method based on multi-source data fusion. Background Art

[0002] Maritime safety information broadcasting is the transmission and broadcasting of information to ensure the safety of ships' navigation at sea. It is used to ensure that maritime safety information can be delivered to the target audience quickly and accurately to improve the safety of maritime navigation. An effective broadcasting mechanism can enhance the safety of maritime navigation and provide accurate and timely key information support for ships and maritime personnel.

[0003] There is a lot of manual intervention in the current maritime safety information broadcasting system. For example, after receiving maritime safety information, the coastal radio station staff on duty must check the information in a timely manner, such as checking whether the text, number, time, etc. in the information are correct, which increases labor costs and consumes a lot of time. Moreover, with the continuous upgrading of the scale of maritime business, especially for future maritime application scenarios, the data sources of maritime safety information are likely to increase. Different data sources may generate duplicate and overlapping information, and different data sources may have the same content but different expressions, which will increase the workload of staff and consume more time. Therefore, it is necessary to merge and optimize the information data of different data sources to provide a more concise and accurate information presentation, so as to provide more accurate and reliable maritime safety information to the broadcast objects. The present invention solves this technical problem. Summary of the Invention

[0004] The present invention provides a maritime information broadcasting method based on multi-source data fusion, which can fuse and optimize information data from different data sources, reduce the time of manual inspection, and make the broadcasted information more reliable and accurate.

[0005] A method for broadcasting maritime information based on multi-source data fusion, comprising the following steps:

[0006] S1. Set up data interfaces to promptly receive maritime safety information data sent by different agencies;

[0007] S2. Analyze the acquired maritime safety information data and convert it into a standardized data format for subsequent processing;

[0008] S3. Fusion of data from different sources, removal of duplicate information, and calibration in time and space;

[0009] S4. Merge information with the same content but different expressions to save broadcasting resources;

[0010] S5. Establish a hazard warning model. Based on the acquired maritime information, extract meteorological information and input it into the model to achieve hazard level classification.

[0011] S6. Prioritize the information based on its time and danger level, and broadcast the information in order of priority.

[0012] Furthermore, step S2 includes the following steps:

[0013] S21. Convert the text into a vector through the word embedding layer, and obtain the Q / K / V matrix through different linear transformation layers. Then, process the text information through the multi-head attention mechanism:

[0014]

[0015] Among them, Q is the query vector, K is the key vector, V is the numerical vector, d k Represents the dimension of the key vector, and extracts key data from the text information in the above way;

[0016] S22. Compare the extracted key data with maritime regulations, and record the data that does not comply with maritime regulations for subsequent hazard level classification;

[0017] S23. For uncertain data, calculate the credibility of the data through a probabilistic logic reasoning model:

[0018]

[0019] Among them, λ r represents the confidence weight of rule r, Represents an exponential function, used to reflect the entity e i Whether rule r is satisfied;

[0020] S24. When the acquired data are contradictory, the data are compromised by the following methods:

[0021]

[0022] Among them, a speed and b speed They represent data on the same aspect from different sources, and α represents the reliability coefficient;

[0023] S25. Select a corresponding template and output the obtained data as standardized JSON.

[0024] Furthermore, step S3 includes the following steps:

[0025] S31. Map data from different sources to a unified spatiotemporal coordinate system:

[0026] S32. Determine whether different data describe the same entity by calculating similarity;

[0027] S33. De-duplication and fusion of data from multiple different sources describing the same entity.

[0028] Furthermore, the step S31 includes the following steps:

[0029] S311. Establish a unified space-time coordinate system:

[0030]

[0031] Where, Δt clock represents the clock deviation, R(θ) represents the rotation matrix, and (x0, y0) represents the coordinate translation;

[0032] S312: Add the clock calibration value to the timestamp t of each data to obtain the calibrated time t ‘ ;

[0033] S313. Compensate the angle between the sensor installation direction and the ship's forward direction through a rotation matrix, offset the origin of the sensor in the global coordinate system, and convert the relative coordinates into global coordinates.

[0034] Furthermore, the step S32 includes the following steps:

[0035] S321. Calculate the time similarity of different data:

[0036]

[0037] Among them, w t represents the time weight, σ t represents the time tolerance threshold. The absolute value of the time difference is calculated by the above method and then substituted into the Gaussian decay function to obtain the time similarity. The higher the time similarity, the closer the two entities corresponding to the data are in time.

[0038] S322. Calculate the spatial similarity of different data:

[0039]

[0040] Among them, σ p represents the spatial tolerance threshold, w p Represents the spatial weight. By calculating the Euclidean distance in the above way and introducing Gaussian decay, we can get the spatial similarity. The higher the spatial similarity, the closer the two entities corresponding to the data are in terms of geographical location.

[0041] S323. Calculate the attribute similarity of different data:

[0042]

[0043] Among them, w a represents the attribute weight, Represents attribute vectors. Through the above method, discrete attributes are encoded into numbers, continuous attributes are normalized, and the cosine similarity of different attribute vectors is calculated. The higher the similarity, the closer the entity identity corresponding to the data is.

[0044] Furthermore, step S33 includes the following steps:

[0045] S331. Establish a weighted fusion formula:

[0046]

[0047] in, Represents the output value after fusion, v k represents data from different sources, median(v) represents the median of all data, ∈ represents the smoothing factor, c k Represents dynamic weight;

[0048] S332. Calculate the median of different data and calculate the weight c of the corresponding data based on the obtained median k ;

[0049] S333. Substitute the data from different sources and the corresponding weight values obtained into the above formula to perform weighted fusion calculation so that the obtained data is close to the true value.

[0050] Furthermore, step S4 includes the following steps:

[0051] S41. Use the pre-trained language model to convert the text information in the data into semantic vectors:

[0052]

[0053] And use named entity recognition to extract fixed fields and generate standardized templates;

[0054] S42. Quantify the semantic similarity between different texts:

[0055]

[0056] Among them, α represents the semantic vector weight, β represents the degree of repetition of key fields, and γ represents the time decay factor;

[0057] S43. When there are multiple expressions for the same field, the more general term shall be retained;

[0058] S44. For similar events in consecutive time periods, merge them into one dynamic information.

[0059] Furthermore, step S5 includes the following steps:

[0060] S51. Extracting data of various indicators from structured data;

[0061] S52. Establish a weighted calculation formula:

[0062]

[0063] Among them, w i represents the weight, f i (x i ) represents the scoring function;

[0064] S53. Substitute the data of different indicators into the above formula to calculate the corresponding output value. When the output value exceeds the threshold, it is determined to be a dangerous event.

[0065] The technical effects of the present invention are as follows:

[0066] (1) This solution can analyze data from different sources and convert them into a standardized format. It can also integrate data from different sources, remove duplicate and redundant data, make the information data more concise, and merge information with the same content but different expressions, thereby further saving information broadcasting resources and effectively reducing the inspection work of staff. It can also judge the important and dangerous information according to the hazard warning model, and then broadcast the information according to the degree of danger, thereby improving practicality and effectively reducing the harm to ships or equipment at sea;

[0067] (2) This solution calculates the similarity of time, space, and identity attributes of entities corresponding to data from different sources to determine whether the entities corresponding to different data are the same entity and whether the corresponding events are the same event. For the same entity or the same event, only one piece of relevant information can be retained, thereby saving broadcast resources;

[0068] (3) This solution can convert text data into vectors, extract key information from the text, and then calculate the similarity between vectors, so as to determine information with the same content but different expressions. In this way, the information can be merged or one of the descriptive information can be retained, thereby avoiding the broadcast of multiple identical information and improving the simplicity and accuracy of information broadcasting. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0070] The technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments and drawings.

[0071] See also Figure 1 , a maritime information broadcasting method based on multi-source data fusion, comprising the following steps:

[0072] S1. Set up data interfaces to promptly receive maritime safety information data sent by different agencies;

[0073] S2. Analyze the acquired maritime safety information data and convert it into a standardized data format for subsequent processing;

[0074] S3. Fusion of data from different sources, removal of duplicate information, and calibration in time and space;

[0075] S4. Merge information with the same content but different expressions to save broadcasting resources;

[0076] S5. Establish a hazard warning model. Based on the acquired maritime information, extract meteorological information and input it into the model to achieve hazard level classification.

[0077] S6. Prioritize the information based on its time and danger level, and broadcast the information in order of priority.

[0078] Furthermore, step S2 includes the following steps:

[0079] S21. Convert the text into a vector through the word embedding layer, and obtain the Q / K / V matrix through different linear transformation layers. Then, process the text information through the multi-head attention mechanism:

[0080]

[0081] Among them, Q is the query vector, K is the key vector, V is the numerical vector, d k Represents the dimension of the key vector, which is used to extract key data from the text information, such as the ship name, coordinates, speed, etc.

[0082] S22. Compare the extracted key data with maritime regulations and record the data that does not comply with maritime regulations for subsequent classification of danger levels; for example, if the cargo ship is speeding or the actual wind speed is greater than the normal wind speed, record it.

[0083] S23. For uncertain data, calculate the credibility of the data through a probabilistic logic reasoning model:

[0084]

[0085] Among them, λr represents the confidence weight of rule r, Represents an exponential function, used to reflect the entity e i (i.e. actual situation) whether rule r is satisfied;

[0086] S24. When there is a contradiction in the acquired data, for example, when the data acquired by the radar differs from the data reported by the AIS (Automatic Identification System) in the same respect, the data shall be compromised in the following manner:

[0087]

[0088] Among them, a speed and b speed They represent the same aspect of data from different sources, such as the ship speed obtained by radar and the ship speed reported by AIS, respectively. α represents the reliability coefficient;

[0089] S25. Select a corresponding template and output the obtained data as standardized JSON.

[0090] Furthermore, step S3 includes the following steps:

[0091] S31. Map data from different sources to a unified spatiotemporal coordinate system:

[0092] S32. Determine whether different data describe the same entity by calculating similarity;

[0093] S33. De-duplication and fusion of data from multiple different sources describing the same entity.

[0094] Furthermore, the step S31 includes the following steps:

[0095] S311. Establish a unified space-time coordinate system:

[0096]

[0097] Where, Δt clock represents the clock deviation, R(θ) represents the rotation matrix, and (x0, y0) represents the coordinate translation;

[0098] The clock deviation can be measured by the NTP protocol. Taking radar as an example, when the radar is 2 seconds ahead of the standard time, the clock deviation is -2. The θ in the rotation matrix represents the yaw angle of the sensor installation. For example, the radar antenna orientation deviates by 30° from the ship's forward direction. The translation represents the offset of the sensor's origin in the global coordinate system. For example, if the radar is installed at the stern, 50 meters from the center of the ship, the position detected by the radar will be based on the radar itself as the origin. Therefore, the ship coordinates collected by different devices are different, but the ship's position is the same. Therefore, the coordinates provided by different devices need to be converted into a global coordinate system through the above method. A global coordinate system can be established with the ship as the origin, which makes it easier to understand what happened in what direction or at what position.

[0099] S312: Add the clock calibration value to the timestamp t of each data to obtain the calibrated time t ‘ ;

[0100] S313. Compensate the angle between the sensor installation direction and the ship's forward direction through a rotation matrix, offset the origin of the sensor in the global coordinate system, and convert the relative coordinates into global coordinates.

[0101] Furthermore, the step S32 includes the following steps:

[0102] S321. Calculate the time similarity of different data:

[0103]

[0104] Among them, w t represents the time weight (empirical value, i.e., the update interval of data acquisition by different devices), σ t represents the time tolerance threshold (assuming that the ship's position does not change significantly within 10 seconds, and the position changes significantly after 10 seconds, 10 seconds can be used as this threshold). The absolute value of the time difference is calculated using the above method, and then substituted into the Gaussian decay function to obtain the temporal similarity. The higher the temporal similarity, the closer the two entities corresponding to the data are in time, avoiding misidentification of historical data as current data.

[0105] S322. Calculate the spatial similarity of different data:

[0106]

[0107] Among them, σ p represents the space tolerance threshold (which can be the maximum safe distance between ships), w p Represents the spatial weight. By calculating the Euclidean distance in the above way and introducing Gaussian decay, we can get the spatial similarity. The higher the spatial similarity, the closer the two entities corresponding to the data are in terms of geographical location.

[0108] S323. Calculate the attribute similarity of different data:

[0109]

[0110] Among them, w a represents the attribute weight, Representing attribute vectors, such as ship model, ship length, etc., through the above method, discrete attributes (such as ship model) are encoded into numbers, continuous attributes (such as ship length) are normalized, and the cosine similarity of different attribute vectors is calculated. The higher the similarity, the closer the entity identity (such as ship) corresponding to the data is.

[0111] Furthermore, step S33 includes the following steps:

[0112] S331. Establish a weighted fusion formula:

[0113]

[0114] in, Indicates the output value after fusion, such as speed, longitude and latitude, etc.; v k represents data from different sources, such as AIS, radar, VHF, etc.; median(v) represents the median of all data; ∈ represents the smoothing factor; c k Represents dynamic weight;

[0115] S332. Calculate the median of different data and calculate the weight c of the corresponding data based on the obtained median k ;

[0116] S333. Substitute the data from different sources and the corresponding weight values obtained into the above formula to perform weighted fusion calculation so that the obtained data is close to the true value.

[0117] In this way, multiple observations (such as speed and position) of the same entity (such as the ship itself) can be robustly fused to suppress the influence of outliers. Since the median is much more robust to outliers than the mean, it can maintain a stable output when the sensor fails or is disturbed.

[0118] Furthermore, step S4 includes the following steps:

[0119] S41. Use the pre-trained language model to convert the text information in the data into semantic vectors:

[0120]

[0121] And use named entity recognition to extract fixed fields and generate standardized templates;

[0122] S42. Quantify the semantic similarity between different texts:

[0123]

[0124] Among them, α represents the semantic vector weight, β represents the degree of repetition of key fields, and γ represents the time decay factor;

[0125] S43. When multiple expressions exist for the same field, retain the more general term. For example, if "engine failure" or "main engine failure" refer to the same event, retain only the more general term or combine them into "engine / main engine failure."

[0126] S44. For similar events in consecutive time periods, combine them into one dynamic message. For example, if a ship updates its position multiple times within a few minutes, the multiple updated information can be combined into one message and then broadcast.

[0127] Furthermore, step S5 includes the following steps:

[0128] S51. Extracting data of various indicators from structured data;

[0129] S52. Establish a weighted calculation formula:

[0130]

[0131] Among them, w i represents the weight, f i (x i ) represents the scoring function;

[0132] S53. Substitute the data of different indicators into the above formula to calculate the corresponding output value. When the output value exceeds the threshold, it is determined to be a dangerous event.

[0133] The indicators mentioned above are specifically wind speed, wave height, visibility, ship density, current speed and other indicators in the information. For different indicators, the scoring function is different:

[0134] The scoring function for wind speed is: min(x / 20, 1.0), where 20 represents a wind speed of 20 m / s;

[0135] The scoring function of wave height is: 1-e -0.5x , the unit of wave height is meters, indicating an exponentially increasing risk;

[0136] The scoring function for visibility is: max(0, 1-x / 5), which decreases linearly within 5 nautical miles;

[0137] The scoring function for ship density is: The number of x is the number of ships per square nautical mile;

[0138] The scoring function for ocean current velocity is: A current speed of 3 knots is the danger threshold;

[0139] And give each indicator a different weight. When the information reflected by the above indicators is within the same sea area, the output values of the scoring functions of each indicator can be added together. When it exceeds the set value, it is judged as a dangerous event. Preferably, multiple thresholds of different sizes can be set to divide the danger levels into different degrees.

[0140] When broadcasting information, the danger level of the information and the time it takes to obtain the information can be considered at the same time. When the time it takes to obtain the information is long and it is a dangerous event, this information can be broadcast first.

[0141] The above description is merely an exemplary embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformation made by utilizing the contents of the present invention's description and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields, is included in the patent protection scope of the present invention.

Claims

1. A method for disseminating maritime information based on multi-source data fusion, characterized in that: The following steps are involved: S1. Set up data interfaces to promptly receive maritime safety information data sent by different agencies; S2. Analyze the acquired maritime safety information data and convert it into a standardized data format for subsequent processing; S3. Fusion of data from different sources, removal of duplicate information, and calibration in time and space; S4. Merge information with the same content but different expressions to save broadcasting resources; S5. Establish a hazard warning model. Based on the acquired maritime information, extract the information and input it into the model to achieve the classification of hazard levels. S6. Prioritize the information based on its time and danger level, and broadcast the information in order of priority.

2. The method for disseminating maritime information based on multi-source data fusion according to claim 1, characterized in that: The step S2 comprises the following steps: S21. Convert the text into a vector through the word embedding layer, and obtain the Q / K / V matrix through different linear transformation layers. Then, process the text information through the multi-head attention mechanism: Among them, Q is the query vector, K is the key vector, V is the numerical vector, d k Represents the dimension of the key vector, and extracts key data from the text information in the above way; S22. Compare the extracted key data with maritime regulations, and record the data that does not comply with maritime regulations for subsequent hazard level classification; S23. For uncertain data, calculate the credibility of the data through a probabilistic logic reasoning model: Among them, λ r represents the confidence weight of rule r, Represents an exponential function, used to reflect the entity e i Whether rule r is satisfied; S24. When the acquired data are contradictory, the data are compromised by the following methods: Among them, a speed and b speed They represent data on the same aspect from different sources, and α represents the reliability coefficient; S25. Select a corresponding template and output the obtained data as standardized JSON.

3. The method for disseminating maritime information based on multi-source data fusion according to claim 1, characterized in that: The step S3 comprises the following steps: S31. Map data from different sources to a unified spatiotemporal coordinate system: S32. Determine whether different data describe the same entity by calculating similarity; S33. De-duplication and fusion of data from multiple different sources describing the same entity.

4. The method for disseminating maritime information based on multi-source data fusion according to claim 3, characterized in that: The step S31 includes the following steps: S311. Establish a unified space-time coordinate system: Where, Δt clock represents the clock deviation, R(θ) represents the rotation matrix, and (x0, y0) represents the coordinate translation; S312: Add the clock calibration value to the timestamp t of each data to obtain the calibrated time t ‘ ; S313. Compensate the angle between the sensor installation direction and the ship's forward direction through a rotation matrix, offset the origin of the sensor in the global coordinate system, and convert the relative coordinates into global coordinates.

5. The method for disseminating maritime information based on multi-source data fusion according to claim 3, characterized in that: The step S32 includes the following steps: S321. Calculate the time similarity of different data: Among them, w t represents the time weight, σ t represents the time tolerance threshold. The absolute value of the time difference is calculated by the above method and then substituted into the Gaussian decay function to obtain the time similarity. The higher the time similarity, the closer the two entities corresponding to the data are in time. S322. Calculate the spatial similarity of different data: Among them, σ p represents the spatial tolerance threshold, w p Represents the spatial weight. By calculating the Euclidean distance in the above way and introducing Gaussian decay, we can get the spatial similarity. The higher the spatial similarity, the closer the two entities corresponding to the data are in terms of geographical location. S323. Calculate the attribute similarity of different data: Among them, w a represents the attribute weight, Represents attribute vectors. Through the above method, discrete attributes are encoded into numbers, continuous attributes are normalized, and the cosine similarity of different attribute vectors is calculated. The higher the similarity, the closer the entity identity corresponding to the data is.

6. The method for disseminating maritime information based on multi-source data fusion according to claim 3, characterized in that: The step S33 includes the following steps: S331. Establish a weighted fusion formula: in, Represents the output value after fusion, v k represents data from different sources, median(v) represents the median of all data, ∈ represents the smoothing factor, c k Represents dynamic weight; S332. Calculate the median of different data and calculate the weight c of the corresponding data based on the obtained median k ; S333. Substitute the data from different sources and the corresponding weight values obtained into the above formula to perform weighted fusion calculation so that the obtained data is close to the true value.

7. The method for disseminating maritime information based on multi-source data fusion according to claim 1, characterized in that: The step S4 comprises the following steps: S41. Use the pre-trained language model to convert the text information in the data into semantic vectors: And use named entity recognition to extract fixed fields and generate standardized templates; S42. Quantify the semantic similarity between different texts: Among them, α represents the semantic vector weight, β represents the degree of repetition of key fields, and γ represents the time decay factor; S43. When there are multiple expressions for the same field, the more general term shall be retained; S44. For similar events in consecutive time periods, merge them into one dynamic information.

8. The method for disseminating maritime information based on multi-source data fusion according to claim 1, characterized in that: The step S5 comprises the following steps: S51. Extracting data of various indicators from structured data; S52. Establish a weighted calculation formula: Among them, w i represents the weight, f i (x i ) represents the scoring function; S53. Substitute the data of different indicators into the above formula to calculate the corresponding output value. When the output value exceeds the threshold, it is determined to be a dangerous event.

Images