CN116702093A

CN116702093A - Marine target positioning method based on big data fusion

Info

Publication number: CN116702093A
Application number: CN202310988115.5A
Authority: CN
Inventors: 张晓杰; 王冰; 韩晓强; 王建平; 刘刚
Original assignee: Hainan Smart Maritime Technology Co ltd
Current assignee: Hainan Smart Maritime Technology Co ltd
Priority date: 2023-08-08
Filing date: 2023-08-08
Publication date: 2023-09-05
Anticipated expiration: 2043-08-08
Also published as: CN116702093B

Abstract

The invention relates to the technical field of marine target positioning methods, in particular to a marine target positioning method based on big data fusion, which comprises the following steps of collecting marine data and GNSS data from sensors and equipment as acquisition data; and cleaning and processing the acquired data, eliminating noise and filling in missing values. According to the invention, the integrity and coverage area of the data are improved by collecting multi-source marine data and GNSS data, the collected data are cleaned and processed, the data quality is improved, multi-scale characteristic information is extracted from the collected data based on a space time-frequency analysis method, a dynamic environment model is established, the influence of marine environment factors on a target is comprehensively considered, the target is positioned under the guidance of the dynamic environment model through a self-adaptive positioning method, so that the accuracy and stability of positioning are improved, and in addition, the selection, weight distribution and data association of an algorithm are improved by utilizing a data fusion and processing method, so that the accuracy and stability of a target positioning result are improved.

Description

Marine target positioning method based on big data fusion

Technical Field

The invention relates to the technical field of offshore target positioning methods, in particular to an offshore target positioning method based on big data fusion.

Background

The marine target positioning method is a technical method for positioning and tracking targets by utilizing related data in the marine field, and the marine targets are accurately positioned by combining multi-source marine data, such as satellite remote sensing data, marine meteorological data, marine observation data, radar data, AIS data and the like and using a data fusion and processing algorithm. The method is realized through four steps of data acquisition, data preprocessing, data fusion and a target positioning algorithm, so that the accuracy and reliability of target positioning can be improved, and important support is provided for offshore safety, resource development, environmental protection and other works.

In the existing offshore target positioning method, due to the complexity of the marine environment and the limitation of marine data acquisition, the data may be incomplete, resulting in a decrease in positioning accuracy. Secondly, the establishment of the target model is influenced by the uncertainty of the marine environment and the variability of the target behavior, the model is difficult to accurately establish, and the positioning accuracy is influenced. In addition, the data fusion and processing method faces technical challenges such as algorithm selection, weight distribution and data association, and needs further research and improvement. Positioning accuracy and instantaneity are also a balance problem, and complex sea conditions and low signal-to-noise environments can affect positioning accuracy.

Disclosure of Invention

The invention aims to solve the defects in the prior art, and provides an offshore target positioning method based on big data fusion.

In order to achieve the above purpose, the present invention adopts the following technical scheme: an offshore target positioning method based on big data fusion comprises the following steps:

collecting ocean data and GNSS data from sensors and devices as acquisition data;

cleaning and processing the acquired data, eliminating noise and filling missing values;

extracting multi-scale characteristic information from the acquired data by adopting space time-frequency analysis;

based on the characteristic information and the historical data, integrating and acquiring a total data set, dividing a training set, a verification set and a test set, and establishing a flexible target model;

integrating the acquired data, the historical data and the characteristic information, and associating by using a data fusion algorithm to obtain fusion data;

training a target model based on the fusion data, predicting new data by using the target model, and providing a preliminary judgment result of a target position;

based on the acquired data acquired in real time, a dynamic environment model is established, and the preliminary judgment result is finely adjusted by adopting a self-adaptive positioning method to acquire a final result;

And feeding the final result back to a dynamic target model, and retraining and optimizing the target model.

As a further scheme of the invention, the sensor comprises a sonar sensor, a radar sensor and a hydroacoustic sensor, and the device is particularly a GNSS receiver;

the step of collecting ocean data and GNSS data from the sensors and the devices as collected data specifically comprises the following steps:

selecting an installation position, integrating the target of data acquisition and the environmental requirement, and establishing installation points of the sensor and the equipment;

the sensor and the equipment are assembled on the mounting point, and a waterproof shell, an anti-corrosion coating and an anti-biological adhesion material are adopted as protective measures;

adjusting the sensitivity of the sensor, correcting errors or correcting instrument deviations, and calibrating the sensor and the equipment;

starting the sensor and the equipment to acquire data, setting parameters and sampling rate, and acquiring the acquired data;

based on the acquired data, MATLAB is used to convert it into processable digital class data.

As a further aspect of the present invention, the cleaning and processing specifically refers to data cleaning and missing value processing;

the data cleaning comprises outlier detection, noise elimination and data normalization;

The abnormal value detection is specifically that a standard deviation statistical method is used for detecting and eliminating abnormal values in the acquired data, and data points based on a threshold value or a rule are regarded as abnormal values and are subjected to replacement processing;

the noise elimination is specifically to apply signal processing technologies including moving average filtering, median filtering and Kalman filtering to eliminate noise in the acquired data;

the data normalization is specifically based on minimum-maximum normalization processing, and the collected data and the measurement unit are ensured to be in the same scale range;

the missing value processing comprises missing value detection, missing value filling and multiple interpolation;

the missing value detection is specifically to detect the existence of the missing value in the acquired data by using a statistical analysis and data visualization method;

the missing value filling is specifically to select one of a mean value, a median value, a mode filling, a regression filling, an interpolation method and a filling method based on machine learning as a filling method based on the characteristics of the acquired data and the type of the missing value, and fill the missing value;

the multiple interpolation is specifically to fill up the missing value by using a multiple interpolation method specifically a matrix decomposition method for the acquired data with the missing value proportion exceeding 10%.

As a further scheme of the invention, the steps of extracting the multi-scale characteristic information from the acquired data by adopting space time-frequency analysis are specifically as follows:

adopting a short-time Fourier transform method as an analysis method of the space time-frequency analysis;

applying the analysis method, performing short-time Fourier transform by using sliding windows with different sizes, and performing multi-scale analysis on the acquired data to explore the time-frequency characteristics of the signals;

based on the time-frequency characteristics as a data source, adopting an extraction method comprising energy calculation, spectrum peak value extraction and wavelet coefficient statistics to extract spectrum energy, frequency profile and waveform shape characteristics of signals and generate a characteristic set;

based on the feature set, a principal component analysis and feature selection algorithm is adopted, the dimension of the feature is reduced, and important information is reserved and used as the feature information.

As a further scheme of the invention, the steps of integrating and acquiring the total data set based on the characteristic information and the historical data and dividing the training set, the verification set and the test set are specifically as follows:

taking the characteristic information of the past period as the historical data, and integrating the historical data and the characteristic information as a total data set;

Performing dimension reduction processing on the features in the total data set based on linear discriminant analysis again to reduce the dimension of the features and eliminate redundant information, and acquiring the processed total data set, wherein the total data set comprises classified variables or non-numerical type features;

performing tag coding based on the classification variable or the non-numerical feature;

the total dataset was divided into a training set of 70%, a validation set of 15% and a test set of 15%.

As a further aspect of the present invention, the step of building a flexible target model specifically includes:

selecting a logistic regression model as the target model;

training a target model based on the training set, fitting the training set by applying a training algorithm and an optimization method, and learning the mode and rule of data;

using the verification set to verify and tune the trained target model, evaluating the performance of the target model on the verification set and generating an evaluation result, and performing model parameter adjustment and feature selection based on the evaluation result to optimize the target model;

using the test set, evaluating the optimized target model by calculating evaluation indexes including accuracy, precision, recall and F1 score;

And deploying the target model into practical application to perform target prediction tasks.

As a further scheme of the invention, the step of acquiring the fusion data by associating with the data fusion algorithm comprises the following steps:

integrating the acquired data, the historical data and the characteristic information to generate a unified data set;

selecting data with important significance to the association result from the unified data set through feature importance evaluation, and extracting key features;

performing association, matching, aligning and merging the key features by using a data fusion algorithm to obtain associated data, wherein the data fusion algorithm comprises association based on key value matching, association based on similarity matching and association based on a time window;

and merging the associated data through weighted average, decision rules and aggregation functions to generate fusion data.

As a further aspect of the present invention, the step of training a target model based on the fused data and predicting new data using the target model to provide a preliminary determination result of a target position specifically includes:

dividing the fusion data into a 70% post training set and a 30% post testing set, and calling the target model;

Training the target model based on the post training set, and adjusting the super parameters and model fitting degree of the target model based on an optimization method comprising cross verification and grid search;

using the post-test set, evaluating the target model again according to evaluation indexes including mean square error, accuracy and recall rate, and selecting an optimal target model based on an evaluation result;

and applying the target model to the latest acquired data, inputting feature data to be predicted, and generating a corresponding target position prediction result by using the trained target model as the preliminary judgment result.

As a further scheme of the present invention, the steps of establishing a dynamic environment model based on the collected data collected in real time, and performing fine adjustment on the preliminary judgment result by adopting an adaptive positioning method to obtain a final result are specifically as follows:

based on the acquired data, a Bayesian filter is used for establishing a dynamic environment model;

adopting a self-adaptive positioning method comprising weighted least square positioning, particle filtering positioning and conditional random field positioning to more accurately estimate the preliminary judgment result and generate a fine adjustment result;

Based on the fine adjustment result, the position information of the target position is reflected more accurately as a final result.

As a further scheme of the present invention, the step of feeding back the final result to a dynamic target model and retraining and optimizing the target model specifically includes:

combining the final result with the fusion data to construct a new training set, wherein the new training set comprises characteristic data and corresponding target position labels;

adopting the original target model as an initial model;

retraining the target model by using a new training set, inputting characteristic data and corresponding target position labels, and iteratively adjusting parameters of the target model to enable the target model to accurately predict the target position;

and (3) retraining and optimizing iteration is carried out based on the steps, and the performance and accuracy of the target model are gradually improved through continuous feedback of a final result, retraining and optimizing.

Compared with the prior art, the invention has the advantages and positive effects that:

according to the invention, the integrity and coverage of the data are improved by collecting the multi-source ocean data and GNSS data, the collected data are cleaned and processed, and the data quality is improved. Based on a space time-frequency analysis method, multi-scale characteristic information is extracted from acquired data, a dynamic environment model is established, and the influence of marine environment factors on a target is comprehensively considered. And (3) performing target positioning under the guidance of a dynamic environment model by a self-adaptive positioning method so as to improve the accuracy and stability of positioning. In addition, the data fusion and processing method is utilized to improve the selection, weight distribution and data association of the algorithm so as to improve the accuracy and stability of the target positioning result. Machine learning and data driving method research are carried out by utilizing large-scale historical marine data, and an accurate prediction model and a positioning model are constructed so as to realize the prediction and positioning of marine target behaviors.

Drawings

FIG. 1 is a schematic diagram showing the main steps of a marine target positioning method based on big data fusion;

FIG. 2 is a detailed schematic diagram of step 1 of the marine target positioning method based on big data fusion;

FIG. 3 is a detailed schematic diagram of step 2 of the marine target positioning method based on big data fusion;

FIG. 4 is a detailed schematic diagram of step 3 of the marine target positioning method based on big data fusion;

FIG. 5 is a detailed schematic diagram of a part of the process of step 4 of the marine target positioning method based on big data fusion;

FIG. 6 is a detailed schematic diagram of another part of the process of step 4 of the marine target positioning method based on big data fusion;

FIG. 7 is a detailed schematic diagram of step 5 of the marine target positioning method based on big data fusion;

FIG. 8 is a detailed schematic diagram of step 6 of the marine target positioning method based on big data fusion;

FIG. 9 is a detailed schematic diagram of step 7 of the marine target positioning method based on big data fusion;

Fig. 10 is a detailed schematic diagram of step 8 of the marine target positioning method based on big data fusion.

Detailed Description

The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.

In the description of the present invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like indicate orientations or positional relationships based on the orientation or positional relationships shown in the drawings, merely to facilitate describing the present invention and simplify the description, and do not indicate or imply that the devices or elements referred to must have a specific orientation, be configured and operated in a specific orientation, and therefore should not be construed as limiting the present invention. Furthermore, in the description of the present invention, the meaning of "a plurality" is two or more, unless explicitly defined otherwise.

Embodiment one: referring to fig. 1, the present invention provides a technical solution: an offshore target positioning method based on big data fusion comprises the following steps:

cleaning and processing the acquired data, eliminating noise and filling up missing values;

integrating the acquired data, the historical data and the characteristic information, and associating by using a data fusion algorithm to acquire fusion data;

training a target model based on the fusion data, predicting new data by using the target model, and providing a preliminary judgment result of the target position;

based on the acquired data acquired in real time, a dynamic environment model is established, and a self-adaptive positioning method is adopted to finely adjust the preliminary judgment result, so that a final result is obtained;

and feeding back the final result to the dynamic target model, and retraining and optimizing the target model.

First, ocean data and GNSS data are collected as acquisition data from different sensors and devices. The acquired data is then cleaned and processed to eliminate noise and fill in missing values. Next, a spatial time-frequency analysis technique is employed to extract multi-scale feature information from the acquired data. On the basis, feature information and historical data are integrated, a total data set is constructed and divided into a training set, a verification set and a test set, and a flexible target model is built. And then, integrating the acquired data, the historical data and the characteristic information, and correlating by using a data fusion algorithm to acquire fusion data. Based on the fusion data, training a target model, and predicting new data by using the model, thereby obtaining a preliminary judgment result of the target position. And then, according to the acquired data acquired in real time, establishing a dynamic environment model, and adopting a self-adaptive positioning method to finely adjust the preliminary judgment result so as to obtain a final result. And finally, feeding back the final result to the dynamic target model, and retraining and optimizing the target model to continuously improve the performance and accuracy of the model.

By collecting data of various sensors and devices and comprehensively analyzing the data by combining historical data, the accuracy of offshore target positioning can be improved. The data fusion algorithm can correlate and integrate information of various data sources, so that more accurate target position information is obtained. By adopting the big data fusion method, the target model can be trained by utilizing the multi-scale characteristic information. By integrating the collected data, the historical data and the characteristic information and continuously feeding back the final result to the target model for retraining and optimization, the performance and accuracy of the model can be gradually improved. The use of a dynamic environment model and an adaptive positioning method enables the method to adapt to changes and complexities of the offshore environment. By means of the data acquired in real time and fine adjustment according to the dynamic environment model, a more accurate final result can be obtained.

Referring to fig. 2, the sensors include a sonar sensor, a radar sensor, a hydroacoustic sensor, and a GNSS receiver;

the steps of collecting ocean data and GNSS data from sensors and devices as collected data are specifically as follows:

selecting an installation position, integrating the target of data acquisition and the environmental requirement, and establishing installation points of a sensor and equipment;

adjusting the sensitivity of the sensor, correcting errors or correcting instrument deviation, and calibrating the sensor and the equipment;

starting a sensor and equipment to acquire data, setting parameters and sampling rate, and acquiring acquired data;

First, a proper installation position is selected, and the installation points of the sensor and the equipment are determined by comprehensively considering the target of data acquisition and the environmental requirement. This helps ensure that the acquired data accurately reflects the position and state of the target, providing an effective input source. Secondly, the sensor and the equipment are assembled on the mounting point, and protective measures such as a waterproof shell, an anti-corrosion coating and an anti-biological adhesion material are adopted, so that the sensor and the equipment can be protected from being damaged by the marine environment. Doing so may increase the durability and reliability of the sensor and device, ensuring continuous stable data acquisition. Third, the sensor and device are calibrated by adjusting the sensitivity of the sensor, correcting errors or correcting instrument deviations. The calibration process can reduce measurement errors of the sensor and the equipment and improve the accuracy and reliability of data. Accurate data is the basis for subsequent data processing and fusion, and is beneficial to improving the accuracy of the positioning method. Furthermore, based on the acquired data, it is converted into processable digital class data using MATLAB or the like tools. Such digital data processing can help to quickly analyze and extract characteristic information in the data, providing powerful support for subsequent data fusion and model building. Proper processing of the data is critical to achieving efficient localization and prediction. In summary, through the fine steps of acquiring the ocean data and the GNSS data, an accurate and complete data source can be obtained, and multiple benefits are brought to the offshore target positioning method based on big data fusion. These benefits include improved positioning accuracy and model performance, improved method flexibility and stability, improved offshore safety and application performance. Effective data acquisition and processing provide powerful technical support for offshore target positioning, and promote development and application of the ocean field.

Referring to fig. 3, cleaning and processing data specifically refers to cleaning and missing value processing;

the abnormal value detection is specifically to detect and exclude abnormal values in the acquired data by using a standard deviation statistical method, and the data points based on a threshold value or a rule are regarded as the abnormal values and are subjected to replacement processing;

the missing value detection is specifically to detect the existence of missing values in the acquired data by using a statistical analysis and data visualization method;

the multiple interpolation is specifically to fill up the missing value by using a multiple interpolation method, specifically a matrix decomposition method, for the acquired data with the missing value proportion exceeding 10%.

First, data cleaning includes outlier detection, noise cancellation, and data normalization. By detecting and replacing outliers using standard deviation statistics, anomalies in the acquired data can be eliminated. Noise in the data can be removed using signal processing techniques such as moving average filtering and median filtering. Furthermore, by a data normalization method such as min-max normalization, data can be mapped to the same scale range, eliminating the influence due to unit difference.

Second, missing value processing is an important step in data cleansing. The method comprises missing value detection, missing value filling and multiple interpolation. Missing values in the acquired data may be detected using statistical analysis and data visualization methods. For different types of missing values, suitable padding methods may be selected, such as mean, median, mode padding, regression padding, interpolation, and machine learning based padding methods. For the case of higher proportion of missing values, multiple interpolation methods, such as matrix decomposition, can be used to fill in missing values.

By performing data cleaning and processing, the reliability, consistency and accuracy of the data can be improved, and a high-quality data base is provided for subsequent data analysis and model training. Cleaning outliers, eliminating noise, normalizing data, and processing missing values helps to ensure data quality and provides more accurate data input for offshore object localization methods. Effective implementation of these steps will enhance the performance and accuracy of the positioning method, providing beneficial support for offshore safety and application efficiency.

Referring to fig. 4, the steps of extracting multi-scale feature information from the collected data by using space time-frequency analysis are specifically as follows:

a short-time Fourier transform method is adopted as an analysis method of space time-frequency analysis;

applying an analysis method, performing short-time Fourier transform by using sliding windows with different sizes, and performing multi-scale analysis on the acquired data to explore the time-frequency characteristics of the signals;

based on time-frequency characteristics as a data source, extracting spectral energy, frequency contour and waveform shape characteristics of signals by adopting an extraction method comprising energy calculation, spectral peak value extraction and wavelet coefficient statistics to generate a characteristic set;

based on the feature set, a principal component analysis and feature selection algorithm is adopted, the dimension of the features is reduced, and important information is reserved and used as feature information.

Firstly, the short-time Fourier transform method is adopted as an analysis method of space time-frequency analysis, so that the time-frequency characteristic of the signal can be effectively analyzed. By applying short-time fourier transforms over sliding windows of different sizes, the signal variations at different times and frequencies can be explored. This helps to capture transient spectral features of the signal, allowing for multi-scale analysis of the signal.

And secondly, the characteristics of spectrum energy, frequency contour, waveform shape and the like of the signals can be extracted by applying a multi-scale analysis method. By calculating the energy of the signal, extracting the spectrum peak value and counting the wavelet coefficient, the key time-frequency characteristic information can be obtained. This helps describe the energy distribution, frequency characteristics and waveform shape of the signal over different time periods and frequency bands, providing an important feature set for subsequent data analysis and modeling.

Based on the extracted feature set, the feature dimension can be reduced and important information can be retained using principal component analysis and feature selection algorithms. Principal component analysis can convert high-dimensional features into low-dimensional feature space through linear transformation, reduce redundancy of data, and preserve main variances in the data. The feature selection algorithm may help to screen out features that are most useful for the target task, further reducing the feature set. This helps to improve the efficiency and accuracy of data analysis.

From the implementation point of view, the adoption of a space time-frequency analysis method and a multi-scale feature extraction step can extract abundant feature information from the acquired data. These characteristic information fully reveals the time-frequency characteristics of the signal and can provide valuable inputs for subsequent data analysis, model training and target positioning. By reducing the dimensionality of the features and selecting important features, the efficiency of data processing and the interpretability of the model can also be improved. Therefore, the implementation of the steps has beneficial effects on the marine target positioning method based on big data fusion, and the accuracy and the reliability of positioning are improved.

Referring to fig. 5, based on the feature information and the history data, the steps of integrating and obtaining the total data set and dividing the training set, the verification set and the test set are specifically as follows:

taking the characteristic information of the past period as historical data, and integrating the historical data and the characteristic information as a total data set;

performing tag coding based on the classified variable or the non-numerical type feature;

Firstly, characteristic information of past period is taken as historical data, and the historical data and the current characteristic information are integrated into a total data set. This total dataset will contain all the data used to train and evaluate the model.

Next, for the features in the total data set, the dimension reduction process may be performed again using a method based on linear discriminant analysis or the like. Dimension reduction can reduce the dimension of the features and eliminate redundant information, and the features with important influence on classification tasks are extracted. This allows a processed total data set to be obtained, including classification variables or non-numeric features.

Next, if there are classification variables or non-numeric features in the total dataset, tag encoding is required. Tag coding is the process of converting a classification variable or non-numeric feature into a numeric representation. For example, the classification variables may be converted to numerical values using a one-hot encoding or tag encoding or the like, so that the machine learning model can handle these features.

And finally, dividing the processed total data set into a training set, a verification set and a test set. A common division ratio is 70% data used as training set and 15% data used as validation set and test set. The training set is used for learning and training model parameters, the verification set is used for adjusting the super parameters of the model and evaluating the performance of the model, and the test set is used for finally evaluating the performance of the model on unseen data.

From an implementation perspective, the feature information and the historical data are integrated into a total data set, and the processing of the dimension reduction and the label coding is carried out, so that key features can be extracted, the data types can be converted, and preparation is provided for modeling and evaluation. The total data set is divided into a training set, a verification set and a test set, which are beneficial to training, tuning and evaluation of the model, and meanwhile, the generalization capability of the model on unknown data can be evaluated. Such a step implementation can effectively utilize the data, improve the performance of the model, and reduce the risk of overfitting, thereby providing benefits to the application and performance of the offshore target positioning method.

Referring to fig. 6, the steps for building a flexible target model are specifically:

selecting a logistic regression model as a target model;

training the target model based on the training set, fitting the training set by applying a training algorithm and an optimization method, and learning the mode and rule of the data;

using the verification set to verify and tune the trained target model, evaluating the performance of the target model on the verification set, generating an evaluation result, adjusting model parameters and selecting characteristics based on the evaluation result, and optimizing the target model;

using a test set, and evaluating the optimized target model by calculating evaluation indexes including accuracy, precision, recall and F1 score;

and deploying the target model into the actual application to perform a target prediction task.

First, a model, such as a logistic regression model, is selected that is appropriate for the target problem. Logistic regression is a commonly used classification algorithm that works well when dealing with two classification problems. The selection of the appropriate model may be based on the requirements of the problem, the characteristics of the data, and prior experience knowledge.

Second, the training set is used to train the target model. Fitting the model to the training set by using a training algorithm and an optimization method, and learning the pattern and rule of the data. In the training process, optimization algorithms such as gradient descent and the like can be adopted, and model parameters are updated through iteration to minimize a loss function, so that the accuracy of the model is improved.

Next, the trained object model is validated and optimized using the validation set. By evaluating the performance of the target model on the validation set, the generalization ability and effect of the model can be known. According to the evaluation result, model parameter adjustment, feature selection and other optimization strategies can be performed to improve the performance and adaptability of the model.

And finally evaluating the optimized target model by using the test set. By calculating evaluation metrics such as accuracy, precision, recall, and F1 score, the predictive power and performance of the model can be objectively evaluated. This assessment will provide a more comprehensive understanding of the behavior of the model on unseen data.

And finally, deploying the optimized target model into the actual application, and carrying out a target prediction task. This may be the integration of the model into a real-time or offline system, with the model being used to make classification predictions or decisions on the unknown data. After the model is deployed, the monitoring and iterative improvement of the model can be further performed so as to ensure the continuous effectiveness and performance stability of the model.

Referring to fig. 7, the step of obtaining the fusion data by associating with the data fusion algorithm specifically includes:

Through feature importance assessment, selecting data with important significance to the association result from the unified data set, and extracting key features;

carrying out association by utilizing a data fusion algorithm, and carrying out matching, alignment and combination on key features to obtain associated data, wherein the data fusion algorithm comprises association based on key value matching, association based on similarity matching and association based on a time window;

and combining the associated data through weighted average, decision rule and aggregation function to generate fusion data.

Firstly, integrating the collected data, the historical data and the characteristic information to generate a unified data set. This may be accomplished by integrating, cleansing and converting the data of the different data sources. And ensuring the uniformity and consistency of the data format so as to facilitate subsequent association and fusion operations.

And secondly, selecting data with important significance for the association result from the unified data set through feature importance evaluation, and extracting key features. Feature importance assessment may be based on statistical analysis, machine learning models, and the like. By selecting features with higher information content and important impact on the target task, the redundancy of the data can be reduced and the accuracy of the correlation result can be improved.

Next, a correlation operation is performed using a data fusion algorithm. The data fusion algorithm may include a key-value-matching-based association, a similarity-matching-based association, a time window-based association, and the like. The algorithms match, align and merge key features according to the association rules and the matching conditions of the data, so that association among different data sources is realized.

And finally, combining the associated data by a weighted average method, a decision rule method or an aggregation function method and the like to generate fusion data. This may be a weighted average of the associated data to obtain a composite result; or making a data decision through a decision rule, and making a decision according to weights of different data sources; aggregation functions, such as summing, averaging, etc., may also be applied to aggregate the associated data into an overall result.

From an implementation point of view, the use of data fusion algorithms to correlate and obtain fusion data can provide more comprehensive, accurate and decision-making information. The process can solve the problems of data isomerism, missing property, inconsistency and the like by integrating a plurality of data sources, extracting key features and applying a fusion algorithm, thereby providing a more reliable and beneficial data basis for offshore target positioning and other applications.

Referring to fig. 8, training a target model based on the fused data, and predicting new data with the target model, the steps of providing a preliminary determination result of the target position are specifically:

dividing the fusion data into a 70% post training set and a 30% post testing set, and calling a target model;

training the target model based on the post training set, and adjusting the super parameters and the model fitting degree of the target model based on an optimization method comprising cross verification and grid search;

using the post-test set, evaluating the target model again according to evaluation indexes including mean square error, accuracy and recall rate, and selecting the optimal target model based on the evaluation result;

and applying the target model to the latest acquired data, inputting the characteristic data to be predicted, and generating a corresponding target position prediction result by using the trained target model as a preliminary judgment result.

First, the fusion data is divided into a training set and a test set according to a certain proportion. 70% of the data was used as training set and 30% of the data was used as test set. This ensures that the training set has a sufficient amount of data to train the target model, and that the test set can be used to evaluate the predictive capabilities of the model.

Next, the target model is trained using the training set. And fitting the target model to the training set by using the characteristics of the fusion data as input, and learning the mode and rule of the data. The performance of the model on the training set can be evaluated by applying the technologies of cross verification and the like, and the super parameters are optimized based on the methods of grid search and the like so as to improve the fitting degree and generalization capability of the model.

The trained target model is then evaluated using the test set. And inputting the characteristics of the test set into the target model for prediction, and comparing with the real target position of the test set. The prediction accuracy and performance of the model can be objectively evaluated by calculating evaluation indexes such as mean square error, accuracy, recall rate and the like. And selecting a target model with the best performance according to the evaluation result.

Finally, the trained object model is applied to the latest acquired data. And inputting the characteristic data to be predicted into a target model, predicting the target position by using the model, and generating a preliminary judgment result. This result can be used to initially locate the target location and provide a reference for subsequent analysis, decision making and further processing.

From an implementation point of view, by training the target model and predicting using the fusion data, a more accurate and reliable preliminary determination result of the target position can be provided by means of various data information. The step can fully utilize the fusion data and the trained model to provide preliminary prediction and decision support for offshore target positioning.

Referring to fig. 9, based on the collected data collected in real time, a dynamic environment model is established, and the preliminary judgment result is fine-tuned by adopting an adaptive positioning method, and the step of obtaining the final result is specifically as follows:

based on the trimming result, the position information of the target position is reflected more accurately as a final result.

First, a dynamic environment model is built using the acquired data. This can be achieved by means of a bayesian filter or the like. The Bayesian filter is a recursive probability filtering method, and can perform state estimation and posterior update according to prior information and data acquired in real time, so as to establish a dynamic environment model. The dynamic environment model may help understand the position and motion state of the target at different points in time.

And then, fine tuning the preliminary judgment result by adopting an adaptive positioning method. The self-adaptive positioning method comprises weighted least square positioning, particle filter positioning, conditional random field positioning and the like. The method can update and correct the initial position estimation according to the data acquired in real time and the dynamic environment model so as to obtain a more accurate target position estimation result. The self-adaptive positioning method considers different environmental factors and data characteristics, and can be better adapted to different scenes and conditions.

Finally, based on the fine adjustment result, the information of the target position is reflected as a final position estimation result. The fine tuning result considers the influence of the dynamic environment model and the real-time data, and can reflect the state and the change of the target position more accurately. This step will provide a more accurate and reliable estimate of the target location, providing a more accurate basis for subsequent analysis, decision-making and manipulation.

From the implementation point of view, a dynamic environment model is established according to the data acquired in real time, and the preliminary judgment result is finely adjusted by adopting a self-adaptive positioning method, so that the accuracy and the precision of target position estimation can be improved. The method combines the establishment of an environment model and the application of the self-adaptive positioning method, fully utilizes real-time data and previous information, and provides a more reliable and accurate final result for the offshore target positioning.

Referring to fig. 10, the final result is fed back to the dynamic target model, and the steps of retraining and optimizing the target model specifically include:

an original target model is adopted as an initial model;

Retraining the target model by utilizing the new training set, inputting characteristic data and corresponding target position labels, and iteratively adjusting parameters of the target model to enable the target model to accurately predict the target position;

and (3) retraining and optimizing iteration are carried out based on the steps, and the performance and accuracy of the target model are gradually improved through continuous feedback of the final result, retraining and optimizing.

First, the final position estimation result is combined with the fusion data to construct a new training set. The new training set includes the feature data and the corresponding target location tags. The characteristic data may include collected data, historical data and other relevant characteristics, and the target location tag is the final location estimation result.

Next, the original target model is used as an initial model. The initial model may be a model that has been previously trained, or a model that has been subjected to a series of training and optimization.

The target model is then retrained with the new training set. The characteristic data of the new training set and the corresponding target position label are input into the target model, and the parameters of the model are iteratively adjusted to enable the model to accurately predict the target position. This may employ common machine learning algorithms such as neural networks, support vector machines, etc., and use optimization methods to update the weights and biases of the model.

Retraining and optimization iteration is performed based on the steps. The performance and accuracy of the target model are gradually improved by continuously feeding back the final result, retraining and optimizing. This step may be repeated a number of times until a satisfactory performance index is reached.

From the implementation point of view, the final result is fed back to the dynamic target model for retraining and optimization, so that the prediction capability and accuracy of the model can be continuously improved. By using the final results as training data, the model can learn more accurate target locations and dynamic environmental patterns, thereby improving reliability of predictions. The feedback information is effectively utilized in the step, and is applied to updating and optimizing the model, so that the overall performance of the target positioning system is improved.

Working principle: first, the position, motion trajectory and other relevant data of the offshore objects are collected from the sea and the air by means of sonar sensors, radar sensors, hydroacoustic sensors, GNSS receivers, etc. The data is subjected to cleaning and processing including outlier detection, noise cancellation, data normalization and missing value processing. Next, multi-scale feature information is extracted from the cleaned data using techniques such as spatial time-frequency analysis. The feature information is integrated with the historical data to construct a total data set, and the total data set is divided into a training set, a verification set and a test set. And selecting a proper target model, such as a logistic regression model, on the training set, and training and optimizing the target model to improve the performance and the prediction capability of the model. And the data fusion algorithm correlates and merges the acquired data, the historical data and the characteristic information to generate fusion data. And predicting new data by using the trained target model to obtain a preliminary judgment result of the target position. Then, a dynamic environment model is established, and the preliminary judgment result is finely adjusted by adopting a self-adaptive positioning method in combination with data acquired in real time, so that a more accurate final result is obtained. And the final result is fed back to the dynamic target model, retraining and optimizing are carried out on the dynamic target model, and the performance and accuracy of the model are gradually improved. The cyclic process is iterated continuously, so that the model can be better adapted to the change of the offshore target and the complex environment. By integrating a plurality of key steps of big data fusion, feature extraction, model training and dynamic environment modeling, the method can realize accurate offshore target positioning and prediction, and is beneficial to improving offshore safety and application efficiency.

The present invention is not limited to the above embodiments, and any equivalent embodiments which can be changed or modified by the technical disclosure described above can be applied to other fields, but any simple modification, equivalent changes and modification made to the above embodiments according to the technical matter of the present invention will still fall within the scope of the technical disclosure.

Claims

1. The marine target positioning method based on big data fusion is characterized by comprising the following steps of:

2. The marine target positioning method based on big data fusion according to claim 1, wherein the sensor comprises a sonar sensor, a radar sensor, a hydroacoustic sensor, and the device is in particular a GNSS receiver;

3. The marine target positioning method based on big data fusion according to claim 1, wherein the cleaning and processing is specific to data cleaning and missing value processing;

4. The marine target positioning method based on big data fusion according to claim 1, wherein the step of extracting multi-scale feature information from the acquired data by spatial time-frequency analysis specifically comprises:

5. The marine target positioning method based on big data fusion according to claim 1, wherein the step of integrating and obtaining a total data set based on the characteristic information and the history data, and dividing a training set, a verification set and a test set specifically comprises the following steps:

6. The marine target positioning method based on big data fusion according to claim 1, wherein the step of establishing a flexible target model specifically comprises:

selecting a logistic regression model as the target model;

7. The marine target positioning method based on big data fusion according to claim 1, wherein the step of obtaining the fusion data by associating with the data fusion algorithm specifically comprises:

8. The marine target positioning method based on big data fusion according to claim 1, wherein the step of training a target model based on the fusion data and predicting new data with the target model to provide a preliminary determination result of a target position specifically comprises:

9. The marine target positioning method based on big data fusion according to claim 1, wherein the steps of establishing a dynamic environment model based on the collected data collected in real time, and performing fine adjustment on the preliminary judgment result by adopting an adaptive positioning method to obtain a final result are specifically as follows:

10. The marine target positioning method based on big data fusion according to claim 1, wherein the step of feeding the final result back to a dynamic target model and retraining and optimizing the target model specifically comprises:

adopting the original target model as an initial model;