Ice layer thickness prediction method and system based on multi-modal sensor data fusion

Abstract

The invention discloses an ice layer thickness prediction method and system based on multi-modal sensor data fusion. According to the invention, by integrating various types of sensors, such as a microwave sensor, an ultrasonic sensor and an infrared thermal imaging sensor, multi-dimensional and omnibearing monitoring of the thickness of the ice layer is realized. The multi-modal data fusion strategy effectively overcomes the limitation of a single sensor in a complex environment, and improves the comprehensiveness and accuracy of data acquisition. In particular, due to the application of a self-adaptive sensor calibration module, sensor data are calibrated in real time by using advanced technologies such as a Kalman filtering algorithm and the like, so that measurement errors and noise interference are remarkably reduced. And the fused data is intelligently analyzed and processed through a deep learning algorithm and a time sequence prediction model, so that the prediction model is further optimized, the accuracy and reliability of ice layer thickness prediction are improved, reliable data support is provided for applications such as ice region navigation and ocean engineering, and the safety risk and operation cost can be reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of ice thickness prediction, and specifically relates to an ice thickness prediction method and system based on multi-modal sensor data fusion. Background Art

[0002] With the intensification of global climate change and the development of polar resources, accurate prediction of ice thickness is particularly important in the fields of navigation safety, marine engineering, climate research, etc. Ice thickness prediction is a method of predicting future changes in ice thickness by monitoring and analyzing the ice layer of frozen water bodies (such as lakes, rivers, oceans, etc.). It is of great significance to climate research, shipping safety, water resources management and other fields. The prediction process involves a variety of data collection methods, including satellite remote sensing, ground observation, drone monitoring and ice drilling. After processing, these data are used to construct physical or statistical models of ice growth and melting. The key factors in ice thickness prediction include air temperature, wind speed, sunshine time, water heat exchange and subglacial water flow, which together affect the formation and melting process of ice.

[0003] However, existing prediction methods often lack effective sensor calibration methods, resulting in inconsistent data and reduced prediction accuracy. At the same time, these systems do not incorporate advanced artificial intelligence algorithms for intelligent decision-making, which limits the adaptability and intelligence level of the system. Summary of the invention

[0004] The purpose of the present invention is to provide an ice thickness prediction method and system based on multimodal sensor data fusion in order to solve the above-mentioned problems.

[0005] The technical solution adopted by the present invention is as follows: an ice thickness prediction system based on multimodal sensor data fusion, the system comprising: a multimodal sensing unit, an adaptive sensor calibration module, a data fusion and processing module, an ice thickness prediction model, an edge computing unit, a cloud analysis and management platform and an artificial intelligence auxiliary decision module;

[0006] The artificial intelligence-assisted decision-making module is internally provided with a model training and optimization module, a decision-making generation module, and a knowledge base management and update module;

[0007] The multimodal sensing unit, as the front end of data collection, integrates microwave sensors, ultrasonic sensors, and infrared thermal imaging sensors to collect ice layer-related data in the environment in real time. These data are first transmitted to the adaptive sensor calibration module;

[0008] The sensor calibration module performs real-time calibration of sensor data through an automatic calibration algorithm and feedback control system to ensure data accuracy. The calibrated data then enters the data fusion and processing module;

[0009] The data fusion and processing module performs data cleaning, feature extraction and fusion processing to provide a high-quality data basis for subsequent analysis. The processed data is transmitted to the edge computing unit.

[0010] The edge computing unit uses deep learning models and time series prediction models to make preliminary predictions on ice thickness and implement real-time monitoring and early warning. The edge computing unit uploads the processed data and prediction results to the cloud analysis and management platform.

[0011] The cloud-based analysis and management platform is responsible for large-scale data storage, model training and updating, advanced data analysis and visualization, and provides a web-based user interface that allows remote monitoring and management.

[0012] The artificial intelligence-assisted decision-making module receives preprocessed data, generates intelligent decision suggestions through its internal model training and optimization submodule, decision generation submodule, and knowledge base management and update submodule, and feeds these suggestions back to users and edge computing units to optimize the prediction model and decision-making process.

[0013] In a preferred embodiment, the multimodal sensing unit integrates a microwave sensor, an ultrasonic sensor, and an infrared thermal imaging sensor. The microwave sensor uses the changes in the propagation characteristics of microwaves in different media to detect ice formation; the ultrasonic sensor accurately calculates the thickness of the ice layer by measuring the propagation time of the sound waves in the ice layer; the infrared thermal imaging sensor is used to capture the surface temperature distribution and identify temperature anomalies in the early stages of ice formation. These three sensors are cleverly integrated into a compact waterproof housing and flexibly installed in key locations on the ship. The sensing unit is also equipped with temperature and humidity sensors to collect environmental parameters and provide auxiliary data for ice prediction.

[0014] In a preferred embodiment, the adaptive sensor calibration module performs sensor calibration by using a Kalman filter algorithm, and the specific method includes:

[0015] Initialization: Set the initial state estimate x^0 and the initial error covariance matrix P0.

[0016] Prediction stage: Use the system dynamic model to predict the state estimate x^k|k-1 and error covariance matrix Pk|k-1 at the next moment.

[0017] Update phase: receiving sensor observations z k , then calculate the Kalman gain K k .

[0018] Update the state estimate x^k|k and the error covariance matrix}P k∣k .

[0019] Output calibrated sensor data: Output the updated state estimate as the calibrated sensor data.

[0020] In a preferred embodiment, the data fusion and processing module: This module uses a deep learning algorithm, in particular a model based on a convolutional neural network (CNN) and a self-attention mechanism, to process and fuse multimodal sensor data. CNN is used to extract spatial features of microwave and infrared images, while the self-attention mechanism is used to capture the correlation between different sensor data. This module also includes a data preprocessing unit for signal denoising, standardization, and feature extraction. Through this advanced data fusion method, the system can comprehensively utilize the advantages of various sensors and significantly improve the accuracy and robustness of ice detection.

[0021] In a preferred embodiment, the ice thickness prediction model is built based on a long short-term memory network (LSTM), which is specifically used for time series data analysis and prediction. It not only considers the current sensor fusion data, but also combines historical ice data, weather forecast information, and sea state data. The model can predict the trend of ice thickness changes in the next 24 hours by learning long-term and short-term time dependencies. In addition, the model also uses an attention mechanism to identify the most critical features and time points for prediction, thereby improving the accuracy and interpretability of the prediction.

[0022] In a preferred embodiment, the edge computing unit uses a low-power, high-performance embedded processor, an ARM Cortex-A series processor, equipped with a dedicated AI acceleration chip. It is responsible for performing real-time data processing, preliminary data fusion and anomaly detection algorithms. The edge computing unit also contains a local cache and a simplified version of the prediction model, which can maintain basic ice monitoring and early warning functions even when the network connection is interrupted. The design of this unit fully considers the particularity of the marine environment and adopts a shock-proof and corrosion-resistant hardware design.

[0023] In a preferred embodiment, the model training and optimization module uses a random forest machine learning algorithm to perform model training optimization, specifically comprising the following steps:

[0024] Data preparation: Input: Feature dataset D processed by the data fusion and processing module, containing n samples, each with m features. Output: Trained random forest model.

[0025] Model initialization: Set the number of trees in the random forest, T. Set the maximum depth and minimum number of leaf node samples for each tree.

[0026] Bootstrap sampling: Perform T bootstrap sampling on the data set D, and each sampling generates a sub-dataset Di.

[0027] Construct a decision tree: For each sub - dataset Di, construct a decision tree ti. During the construction process, randomly select k features (k < m) for splitting node selection. Use information gain, gain ratio, or Gini impurity metrics to select the best splitting feature and threshold.

[0028] Growth of the tree: Repeat splitting the nodes until the stopping condition is met;

[0029] Model integration: Integrate all decision trees ti into a random forest model.

[0030] Model optimization: Adjust hyperparameters through cross - validation methods to optimize the model performance;

[0031] The formula for information gain is:

[0032] Where: IG(D,A) represents the information gain of feature A on dataset D.

[0033] H(D) represents the entropy of dataset D.

[0034] v represents the number of values of feature A.

[0035] |Dj| represents the number of samples in the sub - dataset Dj after splitting.

[0036] |D| represents the total number of samples in dataset D.

[0037] H(Dj) represents the entropy of sub - dataset Dj.

[0038] The formula for entropy is:

[0039] Where: c represents the number of classes. pi represents the probability of the i - th class.

[0040] In a preferred embodiment, the decision - making generation module uses the support vector machine SVM algorithm for decision - making generation. The specific steps include:

[0041] S1: Data input: Receive the pre - processed feature dataset D from the data fusion and processing module, which contains n samples, and each sample has m features.

[0042] S2: Model selection: Select the support vector machine (SVM) as the decision - making generation model.

[0043] S3: Feature selection: Select the most relevant feature subset according to the feature importance score.

[0044] S4: Model training: Use the training dataset D to train the SVM model. Select an appropriate radial basis kernel function and solve for the optimal hyperplane through an optimization algorithm.

[0045] S5: Model validation: Use the validation set to evaluate model performance and adjust model parameters.

[0046] S6: Decision generation: Use the trained SVM model to perform classification or regression prediction on the newly input data points. Output the decision result, i.e. the predicted value of ice thickness;

[0047] The calculation formula of the SVM optimization problem is:

[0048] subject to:

[0049] Where: w represents the normal vector of the separating hyperplane.

[0050] b represents the bias term of the separating hyperplane.

[0051] ξi represents a slack variable that allows some samples to be misclassified.

[0052] C represents the regularization parameter, which controls the degree of penalty for misclassification.

[0053] yi represents the label of the i-th sample (for regression problems, it is a continuous value).

[0054] xi represents the feature vector of the i-th sample;

[0055] The calculation formula of RBF kernel function is: K(x i , x j ) = exp(-γ||x i -x j || 2 )

[0056] Where: K(xi,xj) kernel function value, represents the similarity between samples xi and xj.

[0057] γ represents the kernel function parameter, which controls the complexity of the feature space after mapping.

[0058] In a preferred embodiment, the knowledge base management and update module is responsible for storing, managing and continuously updating knowledge and information related to ice thickness prediction.

[0059] Knowledge storage: Stores icing patterns and optimal response strategies under different environmental conditions, including historical ice thickness data, sensor calibration parameters, and prediction model parameters. Stores expert experience and domain knowledge, including different types of ice characteristics and environmental factors that affect ice thickness.

[0060] Knowledge organization: Organize knowledge in a structured way, including using database management systems to store and manage data. Use ontology or knowledge graph technology to represent and associate different types of knowledge to improve the queryability and reusability of knowledge.

[0061] Knowledge update: Receive new knowledge and updated model parameters from the model training and optimization module through the continuous learning mechanism. Regularly obtain the latest environmental data and prediction results from the cloud analysis and management platform, and update the historical data and analysis results in the knowledge base.

[0062] Knowledge retrieval: Provides an efficient query interface, allowing users and other modules of the system to quickly retrieve the required knowledge and information. Supports multi-dimensional queries, including filtering relevant data based on time, location, and environmental conditions.

[0063] Knowledge verification: Regularly verify and calibrate the content in the knowledge base to ensure the accuracy and reliability of the knowledge. Evaluate the effectiveness of the prediction models and response strategies stored in the knowledge base by comparing with actual observation data.

[0064] Knowledge feedback: Feedback the information and decision suggestions in the knowledge base to the AI-assisted decision-making module to support the generation of intelligent decisions. Feedback the updated knowledge and model parameters to the edge computing unit to optimize the local prediction model.

[0065] Knowledge sharing: Support remote access and sharing of knowledge bases, allowing different users and systems to access and use information in the knowledge base. Provide API interfaces to facilitate data exchange and knowledge sharing with other systems or platforms;

[0066] The cloud-based analysis and management platform is deployed on a high-performance server cluster and is responsible for large-scale data storage, deep learning model training and updating, advanced data analysis and visualization. It provides a web-based user interface that allows remote monitoring and management of ice conditions for multiple ships. The platform also has a knowledge base system that stores ice patterns and optimal response strategies under different environmental conditions. Through continuous learning and model optimization, the cloud-based platform can continuously improve the prediction accuracy and adaptability of the system.

[0067] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0068] 1. In the present invention, by integrating various types of sensors, such as microwave sensors, ultrasonic sensors and infrared thermal imaging sensors, multi-dimensional and all-round monitoring of ice thickness is achieved. This multimodal data fusion strategy effectively overcomes the limitations of a single sensor in a complex environment and improves the comprehensiveness and accuracy of data acquisition. In particular, the application of the adaptive sensor calibration module uses advanced technologies such as the Kalman filter algorithm to calibrate the sensor data in real time, significantly reducing measurement errors and noise interference. In addition, the artificial intelligence-assisted decision-making module uses deep learning algorithms and time series prediction models to intelligently analyze and process the fused data, further optimizes the prediction model, and improves the accuracy and reliability of ice thickness prediction. This precise prediction capability provides reliable data support for applications such as ice navigation and marine engineering, helping to reduce safety risks and operating costs.

[0069] 2. In the present invention, a high degree of adaptability and intelligence is achieved through the artificial intelligence-assisted decision-making module. The artificial intelligence-assisted decision-making module includes multiple sub-modules such as model training and optimization, decision generation, knowledge base management and updating, and human-computer interaction. These sub-modules work together to form a complete intelligent process from data reception, processing, model training to decision generation. The model training and optimization sub-module enables the system to adapt to changes in icing patterns under different environmental conditions through continuous learning and model updating, and maintain the timeliness and accuracy of the prediction model. The decision generation sub-module uses advanced algorithms such as support vector machines to quickly generate intelligent decision-making suggestions, providing users with effective decision support. The knowledge base management and update module ensures the continuous updating and sharing of system knowledge, enables the system to continuously absorb new information and experience, achieve self-optimization and upgrading, and enables the system to maintain efficient operation in a complex and changeable ice environment, providing strong technical support and guarantee for ice thickness prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is the overall system block diagram of the present invention;

[0071] Figure 2 This is a system block diagram of the artificial intelligence assisted decision-making module in the present invention. DETAILED DESCRIPTION

[0072] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0073] Reference Figure 1-2 ,

[0074] Example:

[0075] An ice thickness prediction system based on multimodal sensor data fusion, the system includes: a multimodal sensing unit, an adaptive sensor calibration module, a data fusion and processing module, an ice thickness prediction model, an edge computing unit, a cloud analysis and management platform, and an artificial intelligence-assisted decision-making module;

[0076] The AI-assisted decision-making module is internally configured with a model training and optimization module, a decision-making generation module, and a knowledge base management and update module;

[0077] As the front end of data collection, the multimodal sensing unit integrates multiple sensors such as microwave sensors, ultrasonic sensors, infrared thermal imaging sensors, etc. to collect ice-related data in the environment in real time. These data are first transmitted to the adaptive sensor calibration module;

[0078] The sensor calibration module calibrates the sensor data in real time through automatic calibration algorithms and feedback control systems to ensure data accuracy. The calibrated data then enters the data fusion and processing module;

[0079] The data fusion and processing module performs data cleaning, feature extraction and fusion processing to provide a high-quality data basis for subsequent analysis. The processed data is transmitted to the edge computing unit.

[0080] The edge computing unit uses deep learning models and time series prediction models to make preliminary predictions on ice thickness and implement real-time monitoring and early warning. The edge computing unit uploads the processed data and prediction results to the cloud analysis and management platform.

[0081] The cloud-based analysis and management platform is responsible for large-scale data storage, model training and updating, advanced data analysis and visualization, and provides a web-based user interface that allows remote monitoring and management.

[0082] The AI-assisted decision-making module receives preprocessed data, generates intelligent decision suggestions through its internal model training and optimization submodule, decision generation submodule, and knowledge base management and update submodule, and feeds these suggestions back to users and edge computing units to optimize the prediction model and decision-making process.

[0083] The multimodal sensing unit integrates microwave sensors, ultrasonic sensors and infrared thermal imaging sensors. Microwave sensors use the changes in the propagation characteristics of microwaves in different media to detect ice formation; ultrasonic sensors accurately calculate the thickness of the ice layer by measuring the propagation time of sound waves in the ice layer; infrared thermal imaging sensors are used to capture surface temperature distribution and identify temperature anomalies in the early stages of ice formation. These three sensors are cleverly integrated into a compact waterproof housing and can be flexibly installed in key locations on the ship. The sensing unit is also equipped with temperature and humidity sensors to collect environmental parameters and provide auxiliary data for ice prediction.

[0084] The adaptive sensor calibration module performs sensor calibration through the Kalman filter algorithm. The specific methods include:

[0085] Initialization: Set the initial state estimate x^0 and the initial error covariance matrix P0.

[0086] Prediction stage: Use the system dynamic model to predict the state estimate x^k|k-1 and error covariance matrix Pk|k-1 at the next moment.

[0087] Update phase: receiving sensor observations z k , then calculate the Kalman gain K k .

[0088] Update the state estimate x^k|k and the error covariance matrix}P k∣k .

[0089] Output calibrated sensor data: Output the updated state estimate as the calibrated sensor data.

[0090] Data fusion and processing module: This module uses deep learning algorithms, especially models based on convolutional neural networks (CNNs) and self-attention mechanisms, to process and fuse multimodal sensor data. CNNs are used to extract spatial features of microwave and infrared images, while self-attention mechanisms are used to capture correlations between different sensor data. This module also contains a data preprocessing unit for signal denoising, standardization, and feature extraction. Through this advanced data fusion method, the system is able to comprehensively utilize the advantages of various sensors and significantly improve the accuracy and robustness of ice detection.

[0091] The ice thickness prediction model is built on the long short-term memory network (LSTM), which is specifically used for time series data analysis and prediction. It not only considers the current sensor fusion data, but also combines multi-source information such as historical ice data, weather forecast information, and sea condition data. By learning long-term and short-term time dependencies, the model can predict the trend of ice thickness changes in the next 24 hours. In addition, the model also uses an attention mechanism to identify the most critical features and time points for prediction, thereby improving the accuracy and interpretability of the prediction.

[0092] The edge computing unit uses a low-power, high-performance embedded processor, such as the ARM Cortex-A series processor, equipped with a dedicated AI acceleration chip. It is responsible for performing real-time data processing, preliminary data fusion, and anomaly detection algorithms. The edge computing unit also contains a local cache and a simplified version of the prediction model, which can maintain basic ice monitoring and early warning functions even when the network connection is interrupted. The design of this unit fully considers the particularity of the marine environment and adopts a shock-proof and corrosion-resistant hardware design.

[0093] The model training and optimization module uses the random forest machine learning algorithm for model training and optimization, which specifically includes the following steps:

[0094] Data preparation: Input: The feature dataset D processed by the data fusion and processing module, which contains n samples, and each sample has m features. Output: The trained random forest model.

[0095] Model initialization: Set the number of trees T in the random forest. Set hyperparameters such as the maximum depth of each tree and the minimum number of samples in a leaf node.

[0096] Bootstrap sampling: Perform T times of bootstrap sampling on the dataset D, and generate a sub-dataset Di each time.

[0097] Construct decision trees: For each sub-dataset Di, construct a decision tree ti. During the construction process, randomly select k features (k < m) for splitting node selection. Use metrics such as information gain, gain ratio, or Gini impurity to select the best splitting feature and threshold.

[0098] Growth of the tree: Repeat splitting the nodes until the stopping conditions (such as the maximum depth, the minimum number of samples in a leaf node, etc.) are met.

[0099] Model integration: Integrate all the decision trees ti into a random forest model.

[0100] Model optimization: Adjust the hyperparameters through the cross-validation method to optimize the model performance;

[0101] The calculation formula for information gain is as follows:

[0102] Where: IG(D,A) represents the information gain of feature A on dataset D.

[0103] H(D) represents the entropy of dataset D.

[0104] v represents the number of values of feature A.

[0105] |D_j| represents the number of samples in the sub-dataset Dj after splitting.

[0106] |D| represents the total number of samples in dataset D.

[0107] H(Dj) represents the entropy of the sub-dataset Dj.

[0108] The calculation formula for entropy is as follows:

[0109] Where: c represents the number of categories. pi represents the probability of the i-th category.

[0110] The decision generation module uses the support vector machine (SVM) algorithm to generate decisions. The specific steps include:

[0111] S1: Data input: Receive the preprocessed feature data set D from the data fusion and processing module, which contains n samples, each with m features.

[0112] S2: Model selection: Support vector machine (SVM) is selected as the decision generation model.

[0113] S3: Feature selection: Select the most relevant feature subset based on feature importance scores.

[0114] S4: Model training: Train the SVM model using the training data set D. Select a suitable radial basis kernel function and solve the optimal hyperplane through an optimization algorithm (such as sequential minimum optimization SMO).

[0115] S5: Model validation: Use the validation set to evaluate model performance and adjust model parameters.

[0116] S6: Decision generation: Use the trained SVM model to perform classification or regression prediction on the newly input data points. Output the decision result, i.e. the predicted value of ice thickness;

[0117] The calculation formula of the SVM optimization problem is:

[0118] subject to:

[0119] Where: w represents the normal vector of the separating hyperplane.

[0120] b represents the bias term of the separating hyperplane.

[0121] ξi represents a slack variable that allows some samples to be misclassified.

[0122] C represents the regularization parameter, which controls the degree of penalty for misclassification.

[0123] yi represents the label of the i-th sample (for regression problems, it is a continuous value).

[0124] xi represents the feature vector of the i-th sample;

[0125] The calculation formula of RBF kernel function is: K(x i , x j ) = exp(-γ||x i -x j || 2 )

[0126] Where: K(xi,xj) kernel function value, represents the similarity between samples xi and xj.

[0127] γ represents the kernel function parameter, which controls the complexity of the feature space after mapping.

[0128] The knowledge base management and update module is responsible for storing, managing and continuously updating the knowledge and information related to ice thickness prediction. It includes:

[0129] Knowledge storage: Stores icing patterns and optimal response strategies under different environmental conditions, including historical ice thickness data, sensor calibration parameters, prediction model parameters, etc. Stores expert experience and domain knowledge, such as the characteristics of different types of ice, environmental factors that affect ice thickness, etc.

[0130] Knowledge organization: Organize knowledge in a structured way, such as using a database management system (DBMS) to store and manage data. Use ontology or knowledge graph technology to represent and associate different types of knowledge to improve the queryability and reusability of knowledge.

[0131] Knowledge update: Receive new knowledge and updated model parameters from the model training and optimization module through the continuous learning mechanism. Regularly obtain the latest environmental data and prediction results from the cloud analysis and management platform, and update the historical data and analysis results in the knowledge base.

[0132] Knowledge retrieval: Provides an efficient query interface, allowing users and other modules of the system to quickly retrieve the required knowledge and information. Supports multi-dimensional queries, such as filtering relevant data based on time, location, environmental conditions, etc.

[0133] Knowledge verification: Regularly verify and calibrate the content in the knowledge base to ensure the accuracy and reliability of the knowledge. Evaluate the effectiveness of the prediction models and response strategies stored in the knowledge base by comparing with actual observation data.

[0134] Knowledge feedback: Feedback the information and decision suggestions in the knowledge base to the AI-assisted decision-making module to support the generation of intelligent decisions. Feedback the updated knowledge and model parameters to the edge computing unit to optimize the local prediction model.

[0135] Knowledge sharing: Support remote access and sharing of knowledge bases, allowing different users and systems to access and use information in the knowledge base. Provide API interfaces to facilitate data exchange and knowledge sharing with other systems or platforms;

[0136] The cloud-based analysis and management platform is deployed on a high-performance server cluster and is responsible for large-scale data storage, deep learning model training and updating, advanced data analysis and visualization. It provides a web-based user interface that allows remote monitoring and management of ice conditions for multiple ships. The platform also has a knowledge base system that stores ice patterns and optimal response strategies under different environmental conditions. Through continuous learning and model optimization, the cloud-based platform can continuously improve the prediction accuracy and adaptability of the system.

[0137] The ice thickness prediction method based on multi-modal sensor data fusion runs the ice thickness prediction system based on multi-modal sensor data fusion of the above embodiment when the method is used.

[0138] In the present invention, by integrating various types of sensors, such as microwave sensors, ultrasonic sensors and infrared thermal imaging sensors, multi-dimensional and all-round monitoring of ice thickness is achieved. This multimodal data fusion strategy effectively overcomes the limitations of a single sensor in a complex environment and improves the comprehensiveness and accuracy of data acquisition. In particular, the application of the adaptive sensor calibration module uses advanced technologies such as the Kalman filter algorithm to calibrate the sensor data in real time, significantly reducing measurement errors and noise interference. In addition, the artificial intelligence-assisted decision-making module uses deep learning algorithms and time series prediction models to intelligently analyze and process the fused data, further optimizing the prediction model and improving the accuracy and reliability of ice thickness prediction. This precise prediction capability provides reliable data support for applications such as ice navigation and marine engineering, helping to reduce safety risks and operating costs.

[0139] In the present invention, a high degree of adaptability and intelligence is achieved through an artificial intelligence-assisted decision-making module. The artificial intelligence-assisted decision-making module includes multiple submodules such as model training and optimization, decision generation, knowledge base management and updating, and human-computer interaction. These submodules work together to form a complete intelligent process from data reception, processing, model training to decision generation. The model training and optimization submodule enables the system to adapt to changes in icing patterns under different environmental conditions through continuous learning and model updating, and maintain the timeliness and accuracy of the prediction model. The decision generation submodule uses advanced algorithms such as support vector machines to quickly generate intelligent decision recommendations, providing users with effective decision support. The knowledge base management and update module ensures the continuous updating and sharing of system knowledge, enables the system to continuously absorb new information and experience, achieve self-optimization and upgrading, and enables the system to maintain efficient operation in a complex and changeable ice environment, providing strong technical support and guarantee for ice thickness prediction.

[0140] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0141] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. Ice thickness prediction system based on multimodal sensor data fusion, characterized by: The system includes: a multimodal sensing unit, an adaptive sensor calibration module, a data fusion and processing module, an ice thickness prediction model, an edge computing unit, a cloud-based analysis and management platform, and an artificial intelligence-assisted decision-making module; The artificial intelligence-assisted decision-making module is internally provided with a model training and optimization module, a decision-making generation module, and a knowledge base management and update module; The multimodal sensing unit, as the front end of data collection, integrates microwave sensors, ultrasonic sensors, and infrared thermal imaging sensors to collect ice layer-related data in the environment in real time; these data are first transmitted to the adaptive sensor calibration module; The sensor calibration module performs real-time calibration of sensor data through an automatic calibration algorithm and a feedback control system to ensure data accuracy; the calibrated data then enters the data fusion and processing module; The data fusion and processing module performs data cleaning, feature extraction and fusion processing to provide a high-quality data basis for subsequent analysis. The processed data is transmitted to the edge computing unit. The edge computing unit uses a deep learning model and a time series prediction model to make a preliminary prediction of ice thickness and implement real-time monitoring and early warning. The edge computing unit uploads the processed data and prediction results to a cloud analysis and management platform. The cloud-based analysis and management platform is responsible for large-scale data storage, model training and updating, advanced data analysis and visualization, and provides a web-based user interface that allows remote monitoring and management; The artificial intelligence-assisted decision-making module receives preprocessed data, generates intelligent decision suggestions through its internal model training and optimization submodule, decision generation submodule, and knowledge base management and update submodule, and feeds these suggestions back to users and edge computing units to optimize the prediction model and decision-making process.

2. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The multimodal sensing unit integrates microwave sensors, ultrasonic sensors and infrared thermal imaging sensors; microwave sensors use changes in the propagation characteristics of microwaves in different media to detect icing conditions; ultrasonic sensors accurately calculate the thickness of the ice layer by measuring the propagation time of sound waves in the ice layer; infrared thermal imaging sensors are used to capture surface temperature distribution and identify temperature anomalies in the early stages of icing; these three sensors are cleverly integrated in a compact waterproof casing and flexibly installed in key locations on the ship; the sensing unit is also equipped with temperature and humidity sensors for collecting environmental parameters and providing auxiliary data for icing prediction.

3. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The adaptive sensor calibration module performs sensor calibration through a Kalman filter algorithm, and the specific method includes: Initialization: Set the initial state estimate x^0 and the initial error covariance matrix P0; Prediction stage: Use the system dynamic model to predict the state estimate x^k|k-1 and error covariance matrix Pk|k-1 at the next moment; Update phase: receiving sensor observations z k , then calculate the Kalman gain K k ; Update the state estimate x^k|k and the error covariance matrix}P k∣k ; Output calibrated sensor data: Output the updated state estimate as the calibrated sensor data.

4. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The data fusion and processing module: This module uses deep learning algorithms, especially models based on convolutional neural networks (CNNs) and self-attention mechanisms, to process and fuse multi-modal sensor data; the CNN is used to extract the spatial features of microwave and infrared images, while the self-attention mechanism is used to capture the correlations between different sensor data; This module also includes a data preprocessing unit for signal denoising, normalization, and feature extraction; through this advanced data fusion method, the system can comprehensively utilize the advantages of various sensors and significantly improve the accuracy and robustness of ice detection.

5. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The ice thickness prediction model is constructed based on long short-term memory networks and is specifically used for time series data analysis and prediction; it not only considers the current sensor fusion data but also combines multi-source information such as historical ice data, weather forecast information, and sea condition data; by learning long-term and short-term time dependencies, the model can predict the trend of ice thickness changes within the next 24 hours; in addition, the model also adopts an attention mechanism to identify the features and time points that are most critical for prediction, thereby improving the accuracy and interpretability of the prediction.

6. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The edge computing unit uses a low-power, high-performance embedded processor, the ARM Cortex-A series processor, equipped with a dedicated AI acceleration chip; it is responsible for performing real-time data processing, preliminary data fusion, and anomaly detection algorithms; the edge computing unit also includes a local cache and a simplified version of the prediction model, which can maintain basic ice monitoring and warning functions even in the case of a network connection interruption; the design of this unit fully considers the particularity of the marine environment and adopts a shock-proof and corrosion-proof hardware design.

7. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The model training and optimization module uses the random forest machine learning algorithm for model training and optimization, which specifically includes the following steps: Data preparation: Input: The feature data set D processed by the data fusion and processing module, which contains n samples, and each sample has m features; Output: A trained random forest model; Model initialization: Set the number of trees T in the random forest; Set hyperparameters such as the maximum depth of each tree and the minimum number of samples in the leaf nodes; Bootstrap sampling: Perform T times of bootstrap sampling on the data set D, and each sampling generates a sub-data set Di; Construct decision trees: For each sub-data set Di, construct a decision tree ti; during the construction process, randomly select k features (k < m) for splitting node selection; Use information gain, gain ratio, or Gini impurity metrics to select the best splitting feature and threshold; Growth of the tree: Repeat splitting nodes until the stopping condition is met; Model integration: Integrate all decision trees ti into a random forest model; Model optimization: Adjust hyperparameters through cross-validation methods to optimize model performance; The calculation formula of information gain is: Where: IG(D,A) represents the information gain of feature A on the data set D; H(D) represents the entropy of the data set D; v represents the number of values of feature A; |D_j| represents the number of samples in the sub-data set Dj after splitting; |D| represents the total number of samples in the data set D; H(Dj) represents the entropy of the sub-data set Dj; The calculation formula for entropy is: Where: c represents the number of categories; pi represents the probability of the i-th category.

8. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The decision generation module uses the support vector machine (SVM) algorithm to generate decisions, and the specific steps include: S1: Data input: receiving the preprocessed feature data set D from the data fusion and processing module, which contains n samples, each with m features; S2: Model selection: Support vector machine (SVM) is selected as the decision generation model; S3: Feature selection: Select the most relevant feature subset based on feature importance scores; S4: Model training: Use the training data set D to train the SVM model; select a suitable radial basis kernel function to solve the optimal hyperplane through the optimization algorithm; S5: Model validation: Use the validation set to evaluate model performance and adjust model parameters; S6: Decision generation: For the newly input data points, use the trained SVM model to perform classification or regression prediction; output the decision result, that is, the predicted value of ice thickness; The calculation formula of the SVM optimization problem is: subject to: Where: w represents the normal vector of the separating hyperplane; b represents the bias term of the separating hyperplane; ξi represents a slack variable, which allows some samples to be misclassified; C represents the regularization parameter, which controls the degree of penalty for misclassification; yi represents the label of the i-th sample (for regression problems, it is a continuous value); xi represents the feature vector of the i-th sample; The calculation formula of RBF kernel function is: K(x i , x j )=exp(-γ||x i -x j || 2 ) Among them: K(xi,xj) kernel function value, represents the similarity between samples xi and xj; γ represents the kernel function parameter, which controls the complexity of the feature space after mapping.

9. The ice thickness prediction system based on multimodal sensor data fusion according to claim 1, characterized in that: The knowledge base management and update module is responsible for storing, managing and continuously updating knowledge and information related to ice thickness prediction; it includes: Knowledge storage: Stores icing patterns and optimal response strategies under different environmental conditions, including historical ice thickness data, sensor calibration parameters, and prediction model parameters; stores expert experience and domain knowledge, including different types of ice characteristics and environmental factors that affect ice thickness; Knowledge organization: Organizing knowledge in a structured way, including using database management systems to store and manage data; Use ontology or knowledge graph technology to represent and associate different types of knowledge to improve the queryability and reusability of knowledge; Knowledge update: Receive new knowledge and updated model parameters from the model training and optimization module through a continuous learning mechanism; regularly obtain the latest environmental data and prediction results from the cloud analysis and management platform, and update the historical data and analysis results in the knowledge base; Knowledge retrieval: Provides an efficient query interface, allowing users and other modules of the system to quickly retrieve the required knowledge and information; supports multi-dimensional queries, including filtering relevant data based on time, location, and environmental conditions; Knowledge verification: Regularly verify and calibrate the content in the knowledge base to ensure the accuracy and reliability of the knowledge; evaluate the effectiveness of the prediction models and response strategies stored in the knowledge base by comparing with actual observation data; Knowledge feedback: Feedback the information and decision suggestions in the knowledge base to the AI-assisted decision-making module to support the generation of intelligent decisions; Feedback the updated knowledge and model parameters to the edge computing unit to optimize the local prediction model; Knowledge sharing: Support remote access and sharing of knowledge bases, allowing different users and systems to access and use information in the knowledge base; Provide API interface to facilitate data exchange and knowledge sharing with other systems or platforms; The cloud-based analysis and management platform is deployed on a high-performance server cluster and is responsible for large-scale data storage, deep learning model training and updating, and advanced data analysis and visualization. It provides a web-based user interface that allows remote monitoring and management of the icing conditions of multiple ships. The platform also has a knowledge base system that stores icing patterns and optimal response strategies under different environmental conditions. Through continuous learning and model optimization, the cloud-based platform can continuously improve the system's prediction accuracy and adaptability.

10. Ice thickness prediction method based on multimodal sensor data fusion, characterized by: When in use, the method runs the ice thickness prediction system based on multimodal sensor data fusion as described in any one of claims 1 to 9.

Images