Arctic sea ice prediction algorithm based on deep learning model

By constructing a deep learning model and feature extraction network based on convolutional neural networks and combining them with seasonal variation patterns, the problems of high computational resources and inaccurate prediction in Arctic sea ice forecasting were solved, and efficient and accurate real-time sea ice forecasting was achieved.

CN119004982APending Publication Date: 2024-11-22TIANJIN RES INST FOR WATER TRANSPORT ENG M O T
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411093716.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing methods for predicting Arctic sea ice suffer from high computational resource requirements, insufficient consideration of meteorological factors in the models, unstable prediction results, and long computation time, making them difficult to apply effectively in lightweight equipment or resource-constrained environments.

Method used

We employ a deep learning model based on convolutional neural networks. By collecting and preprocessing historical Arctic sea ice data, we construct a feature extraction network model, analyze seasonal variation patterns, optimize the loss using the Stochastic Gradient Descent method, introduce dynamic learning rate and momentum matrix to adjust the direction of the loss gradient, and perform group processing of multi-source fused data for prediction.

Benefits of technology

It achieves more efficient and accurate Arctic sea ice prediction, reduces computing resource requirements, improves prediction accuracy and speed, and is suitable for lightweight equipment and resource-constrained environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119004982A_ABST
    Figure CN119004982A_ABST
Patent Text Reader

Abstract

The invention relates to a north pole sea ice prediction algorithm, in particular to a north pole sea ice prediction algorithm based on a deep learning model. The invention discloses a deep learning model-based north pole sea ice prediction algorithm. The algorithm comprises the following steps of 1, collecting related historical data of a to-be-predicted north pole sea ice region to form a historical data set; the objective of the invention is to provide a more comprehensive, more efficient and more accurate north pole sea ice prediction algorithm based on a deep learning model.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to an Arctic sea ice prediction algorithm, and in particular to an Arctic sea ice prediction algorithm based on a deep learning model. Background Art

[0002] In recent decades, global warming has become a common topic of concern for many countries around the world and the entire scientific community. The Arctic climate system has also undergone a series of changes. According to the future predictions of the CMIP6 model, due to continued anthropogenic warming, the Arctic Ocean may be ice-free for the first time in September 2050 (sea ice area is less than 1 million square kilometers). The melting of Arctic sea ice directly affects human production and life, and also brings unprecedented opportunities for humans to develop Arctic resources. In areas with low sea ice density, the navigable time of waterways can be effectively extended, which is conducive to scientific exploration and commercial development. Therefore, the accuracy of the prediction of sea ice density in the Arctic region is crucial.

[0003] Traditional sea ice density prediction methods are mainly numerical models and statistical models. Numerical models can simulate the formation and evolution of sea ice according to the physical processes of the ocean and atmosphere. These models are based on physical equations and parameterization methods, which require high computing resources and cannot achieve lightweight predictions. Statistical methods need to process each grid one by one. In order to reduce computing costs, researchers reduced the grid resolution of sea ice density by nearly 10 times (from 25km to 225km) to make coarse-grained predictions. Therefore, in practical applications, statistical methods are difficult to support predictions for all grids in the Arctic region. Deep learning models use the powerful learning ability of neural networks to extract features from a large amount of sea ice observation data and meteorological factors, automatically capture the nonlinear relationship between them, and thus make accurate predictions. There are currently many deep learning-based methods for predicting sea ice density, but these methods still face some challenges. Current sea ice density prediction models have some limitations in capturing the spatiotemporal changes of sea ice, resulting in large errors and low accuracy. One of the main problems is that the model does not take into account meteorological factors, which play an important role in the melting and freezing of sea ice. The lack of comprehensive consideration of meteorological factors can lead to instability in forecast results. In addition, deep learning-based models usually require a lot of computing resources and a long training time, which makes them demanding on equipment. This is a challenge for lightweight forecasting or for deploying models in resource-constrained environments. Summary of the invention

[0004] In view of this, the present invention provides an Arctic sea ice prediction algorithm based on a deep learning model. The purpose is to provide a more comprehensive, more efficient and more accurate Arctic sea ice prediction algorithm based on a deep learning model.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] An Arctic sea ice prediction algorithm based on a deep learning model includes the following steps:

[0007] Step 1: Collect relevant historical data of the Arctic sea ice area to be predicted to form a historical data set;

[0008] Step 2: Preprocess the historical data set in a targeted manner to extract the time data set and the space data set from the relevant historical data of the Arctic sea ice area to be predicted;

[0009] Step 3: Build a deep learning model based on convolutional neural network;

[0010] Step 4: importing the time dataset and the space dataset in the relevant historical dataset of the Arctic sea ice area to be predicted into the deep learning model based on the convolutional neural network for training, training the time dataset and the space dataset separately during the training process, and finally fusing the two to obtain multi-source fusion data;

[0011] Step 5: Divide the multi-source fusion data in step 4 into n groups according to time, and establish a feature extraction network model;

[0012] Step 6: Use the feature extraction network model to make predictions and output the final prediction results.

[0013] Preferably, in step 1, a data set is downloaded from the official website of OS I-SAF, and the naming format of the downloaded data is modified in batches using Python coding.

[0014] Preferably, in step 2, the preprocessing method for the relevant historical data set of the Arctic sea ice area to be predicted is the GMT polar projection method.

[0015] Preferably, the feature extraction network model comprises n groups of sequentially connected input modules, model calculation modules and output modules, a group of sequentially connected input modules, model calculation modules and output modules corresponds to a group of multi-source fusion data, and the multi-source fusion data are respectively input into the corresponding sequentially connected input modules, model calculation modules and output modules in chronological order.

[0016] Preferably, in step 5, the method of dividing the multi-source fusion data in step 4 into n groups according to time is:

[0017] Pst=Ps+Pt

[0018] Among them, Pst is the historical Arctic sea ice area series, Ps is the seasonal cycle series, and Pt is the trend of Arctic sea ice area changes.

[0019] Preferably, the feature extraction network model uses a Stochastic Gradient Descent method to perform loss optimization, and during the loss optimization process, dynamically adjusts the direction of the loss gradient by introducing a first-order momentum matrix and a second-order momentum matrix.

[0020] Preferably, the expression of the first-order momentum matrix is:

[0021] m t =α1m t-1 +(1-α1)g t ,

[0022] In the formula, m t is the first-order momentum estimate at time t, α1 is the first-order momentum parameter, g t is the loss gradient at time t;

[0023] The expression of the second-order momentum matrix is:

[0024] q t =α2p t-1 +(1-α2)g t 2 ,

[0025] In the formula, q t is the second-order momentum estimate at time t, α2 is the second-order momentum parameter, g t 2 is the square of the loss gradient at time t.

[0026] The beneficial effects of the present invention are:

[0027] 1. The present invention comprises the following steps: Step 1: Collect relevant historical data of the Arctic sea ice area to be predicted to form a historical data set; Step 2: Perform targeted preprocessing on the historical data set, and extract the time data set and the space data set from the relevant historical data of the Arctic sea ice area to be predicted; Step 3: Construct a deep learning model based on a convolutional neural network; Step 4: Import the time data set and the space data set in the relevant historical data set of the Arctic sea ice area to be predicted into the deep learning model based on the convolutional neural network for training, and train the time data set and the space data set respectively during the training process, and finally fuse the two to obtain multi-source fusion data; Step 5: Divide the multi-source fusion data in the step 4 into n groups according to time, and establish a feature extraction network model; Step 6: Use the feature extraction network model to perform prediction and output the final prediction result. Compared with the traditional Arctic sea ice prediction algorithm, this Arctic sea ice prediction algorithm groups the relevant historical data of the Arctic sea ice area to be predicted in chronological order, and uses the established feature extraction network model combined with the seasonal variation law for analysis and prediction, so that the output result is more accurate, the entire analysis process is faster, and it is conducive to the display of real-time prediction results of Arctic sea ice.

[0028] 2. In step 1 of the present invention, a data set is downloaded from the OSI-SAF official website. The data set provides information such as global sea ice density from 1978 to the present. The data set is processed according to the official document introduction, and the naming format of the downloaded data is modified in batches using Python coding to facilitate its reading and processing in the next stage.

[0029] 3. In step 2 of the present invention, the preprocessing method for the relevant historical data set to be predicted in the Arctic sea ice area is the GMT polar projection method. The GMT polar projection method is divided into conical projection, azimuthal projection, cylindrical projection and mixed projection, with a total of more than 30 projection methods. The polar regions of the earth mostly use polar azimuthal projection. This method can quickly and intuitively analyze and process the historical data related to the Arctic sea ice.

[0030] 4. The feature extraction network model of the present invention includes n groups of sequentially connected input modules, model calculation modules and output modules, one group of sequentially connected input modules, model calculation modules and output modules corresponds to a group of multi-source fusion data, and the multi-source fusion data are respectively input into the corresponding sequentially connected input modules, model calculation modules and output modules in chronological order. By grouping the relevant historical data of the Arctic sea ice area to be predicted in chronological order and using the established feature extraction network model in combination with the law of seasonal changes for analysis and prediction, the output result is more accurate and the entire analysis process is faster, which is conducive to the presentation of real-time prediction results of Arctic sea ice.

[0031] 5. In step 5 of the present invention, the method for dividing the multi-source fusion data in step 4 into n groups according to time is: Pst=Ps+Pt, wherein Pst is the historical Arctic sea ice area sequence, Ps is the seasonal cycle sequence, and Pt is the trend of Arctic sea ice area changes. These two parameters more intuitively reflect the basic seasonal change characteristics of Arctic sea ice caused by altitude, latitude, earth rotation, climate differences, etc.

[0032] 6. The feature extraction network model of the present invention adopts the Stochastic Gradient Descent method for loss optimization, and dynamically adjusts the direction of the loss gradient by introducing the first-order momentum matrix and the second-order momentum matrix during the loss optimization process. Compared with other schemes, it is not easy to fall into the local minimum. At the same time, a dynamically decaying learning rate is introduced in the loss optimization. Compared with the previous fixed learning rate, it can not only quickly realize the gradient descent, save computing resources, but also better achieve convergence.

[0033] 7. The expression of the first-order momentum matrix of the present invention is: t =α1m t-1 +(1-α1)g t ,

[0034] In the formula, m t is the first-order momentum estimate at time t, α1 is the first-order momentum parameter, g t is the loss gradient at time t;

[0035] The expression of the 2nd-order momentum matrix is: t =α2p t-1 +(1-α2)g t 2 ,

[0036] In the formula, q t is the second-order momentum estimate at time t, α2 is the second-order momentum parameter, g t 2 is the square of the loss gradient at time t. This embodiment uses different loss functions for optimization according to the change of Arctic sea ice area in different seasons, which can more accurately predict the Arctic sea ice. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of the Arctic sea ice prediction algorithm based on the deep learning model of the present invention;

[0038] Figure 2 A histogram of the prediction of Arctic sea ice area changes based on the feature extraction network model of the present invention;

[0039] Figure 3 It is a structural schematic diagram of the feature extraction network model of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0042] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "install", "connect", "connect", and "set" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0043] like Figure 1 , Figure 2 and Figure 3As shown, the following steps are included: Step 1: Collect relevant historical data of the Arctic sea ice area to be predicted to form a historical data set; Step 2: Perform targeted preprocessing on the historical data set, and extract the time data set and space data set from the relevant historical data of the Arctic sea ice area to be predicted; Step 3: Construct a deep learning model based on a convolutional neural network; Step 4: Import the time data set and the space data set in the relevant historical data set of the Arctic sea ice area to be predicted into the deep learning model based on the convolutional neural network for training, and train the time data set and the space data set respectively during the training process, and finally fuse the two to obtain multi-source fusion data; Step 5: Divide the multi-source fusion data in the step 4 into n groups according to time, and establish a feature extraction network model; Step 6: Use the feature extraction network model to predict and output the final prediction result. Compared with the traditional Arctic sea ice prediction algorithm, this Arctic sea ice prediction algorithm groups the relevant historical data of the Arctic sea ice area to be predicted in chronological order, and uses the established feature extraction network model combined with the seasonal change law for analysis and prediction, so that the output result is more accurate, the entire analysis process is faster, and it is conducive to the display of real-time prediction results of Arctic sea ice.

[0044] like Figure 1 As shown, in step 1, download the data set from the official website of OS I-SAF. The data set provides information such as global sea ice density from 1978 to the present. Process the data set according to the official document, and use Python coding to modify the naming format of the downloaded data in batches to facilitate its reading and processing in the next stage.

[0045] like Figure 1 As shown in FIG. 2 , the preprocessing method for the relevant historical data set to be predicted in the Arctic sea ice area in step 2 is the GMT polar projection method. The GMT polar projection method is divided into conical projection, azimuthal projection, cylindrical projection and mixed projection, with a total of more than 30 projection methods. The polar regions of the earth mostly use polar azimuthal projection. This method can quickly and intuitively analyze and process the historical data related to the Arctic sea ice.

[0046] like Figure 1 and Figure 3As shown, the feature extraction network model includes n groups of sequentially connected input modules, model calculation modules and output modules. A group of sequentially connected input modules, model calculation modules and output modules corresponds to a group of multi-source fusion data. The multi-source fusion data are input into the corresponding sequentially connected input modules, model calculation modules and output modules in chronological order. The relevant historical data of the Arctic sea ice area to be predicted are grouped in chronological order, and the established feature extraction network model is used to analyze and predict in combination with the law of seasonal changes, so that the output results are more accurate and the whole analysis process is faster, which is conducive to the display of real-time prediction results of Arctic sea ice.

[0047] like Figure 1 As shown, in step 5, the method of dividing the multi-source fusion data in step 4 into n groups according to time is: Pst=Ps+Pt, wherein Pst is the historical Arctic sea ice area sequence, Ps is the seasonal cycle sequence, and Pt is the trend of Arctic sea ice area changes. These two parameters more intuitively reflect the basic seasonal change characteristics of Arctic sea ice caused by altitude, latitude, earth rotation, climate differences, etc.

[0048] like Figure 1 and Figure 3 As shown in the figure, the feature extraction network model uses the Stochastic Gradient Descent method for loss optimization, and dynamically adjusts the direction of the loss gradient by introducing the first-order momentum matrix and the second-order momentum matrix during the loss optimization process. Compared with other schemes, it is not easy to fall into the local minimum. At the same time, the dynamically decaying learning rate is introduced in the loss optimization. Compared with the previous fixed learning rate, it can not only quickly realize the gradient descent, save computing resources, but also better achieve convergence.

[0049] The expression of the first-order momentum matrix in this embodiment is: t =α1m t-1 +(1-α1)g t ,

[0050] In the formula, m t is the first-order momentum estimate at time t, α1 is the first-order momentum parameter, g t is the loss gradient at time t;

[0051] The expression of the 2nd-order momentum matrix is: t =α2p t-1 +(1-α2)g t 2 ,

[0052] In the formula, q t is the second-order momentum estimate at time t, α2 is the second-order momentum parameter, g t 2is the square of the loss gradient at time t. This embodiment uses different loss functions for optimization according to the change of Arctic sea ice area in different seasons, which can more accurately predict the Arctic sea ice.

[0053] The specific operating principle is:

[0054] Step 1: Collect relevant historical data of the Arctic sea ice area to be predicted to form a historical data set;

[0055] Step 2: Preprocess the historical data set in a targeted manner to extract the time data set and the space data set from the relevant historical data of the Arctic sea ice area to be predicted;

[0056] Step 3: Build a deep learning model based on convolutional neural network;

[0057] Step 4: importing the time dataset and the space dataset in the relevant historical dataset of the Arctic sea ice area to be predicted into the deep learning model based on the convolutional neural network for training, training the time dataset and the space dataset separately during the training process, and finally fusing the two to obtain multi-source fusion data;

[0058] Step 5: Divide the multi-source fusion data in step 4 into n groups according to time, and establish a feature extraction network model;

[0059] Step 6: Use the feature extraction network model to make predictions and output the final prediction results.

[0060] The electrical components appearing in this article can all be connected to the external main controller and 220V AC power through a transformer, and the main controller can be a conventional known device such as a computer that plays a control role. The conventional known device such as a computer that plays a control role can receive various data signals monitored by this device in real time. The electrical components provided by the present invention are only used for the technical solution according to the structural characteristics of the product. The product will be adjusted and modified after purchase to make it more matching and consistent with the technical solution of the present invention. It is an optimal application technical solution of the present technical solution. The model of the product can be replaced and modified according to the required technical parameters. It is well known to technical personnel in this field. Therefore, technical personnel in this field can clearly obtain the corresponding use effect through the technical solution provided by the present invention.

[0061] The control method of the present invention is automatic control through a controller. The control circuit of the controller can be realized by simple programming by technicians in this field. The provision of power is also common knowledge in this field. The present invention is mainly used to protect mechanical devices, so the present invention will not explain the control method and circuit connection in detail.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An Arctic sea ice prediction algorithm based on a deep learning model, characterized in that: The following steps are involved: Step 1: Collect relevant historical data of the Arctic sea ice area to be predicted to form a historical data set; Step 2: Preprocess the historical data set in a targeted manner to extract the time data set and the space data set from the relevant historical data of the Arctic sea ice area to be predicted; Step 3: Build a deep learning model based on convolutional neural network; Step 4: importing the time dataset and the space dataset in the relevant historical dataset of the Arctic sea ice area to be predicted into the deep learning model based on the convolutional neural network for training, training the time dataset and the space dataset separately during the training process, and finally fusing the two to obtain multi-source fusion data; Step 5: Divide the multi-source fusion data in step 4 into n groups according to time, and establish a feature extraction network model; Step 6: Use the feature extraction network model to make predictions and output the final prediction results.

2. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 1, characterized in that: In step 1, the data set is downloaded from the OSI-SAF official website, and the naming format of the downloaded data is modified in batches using Python coding.

3. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 1, characterized in that: In step 2, the preprocessing method for the relevant historical data set of the Arctic sea ice area to be predicted is the GMT polar projection method.

4. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 1, characterized in that: The feature extraction network model includes n groups of sequentially connected input modules, model calculation modules and output modules, one group of sequentially connected input modules, model calculation modules and output modules corresponds to one group of multi-source fusion data, and the multi-source fusion data are respectively input into the corresponding sequentially connected input modules, model calculation modules and output modules in chronological order.

5. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 1, characterized in that: In step 5, the method for dividing the multi-source fusion data in step 4 into n groups according to time is: Pst=Ps+Pt Among them, Pst is the historical Arctic sea ice area series, Ps is the seasonal cycle series, and Pt is the trend of Arctic sea ice area changes.

6. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 1, characterized in that: The feature extraction network model adopts the Stochastic Gradient Descent method to perform loss optimization, and dynamically adjusts the direction of the loss gradient by introducing the first-order momentum matrix and the second-order momentum matrix during the loss optimization process.

7. The Arctic sea ice prediction algorithm based on a deep learning model according to claim 6, characterized in that: The expression of the first-order momentum matrix is: m t =α1m t-1 +(1-α1)g t , In the formula, m t is the first-order momentum estimate at time t, α1 is the first-order momentum parameter, g t is the loss gradient at time t; The expression of the second-order momentum matrix is: q t =α2p t-1 +(1-α2)g t 2 , In the formula, q t is the second-order momentum estimate at time t, α2 is the second-order momentum parameter, g t2 is the square of the loss gradient at time t.