Sea surface temperature prediction method and system

Abstract

The invention discloses a sea surface temperature prediction method and system. The method comprises the steps of obtaining first historical data, and performing first preprocessing on the first historical data; performing first decomposition on the first historical data after the first preprocessing to obtain first decomposed data; and taking the first decomposition data as the input of a pre-trained first prediction model, and performing sea surface temperature prediction according to the output of the first prediction model. The accuracy and efficiency of sea surface temperature prediction can be improved through advanced data processing and analysis technologies. By using the deep learning model, a complex mode in historical data can be learned, and the future sea surface temperature change can be accurately predicted. In addition, the method can adapt to specific conditions of different sea areas, provides customized prediction services, and meets the requirements of different users.

Description

Technical Field

[0001] The present invention relates to the technical field of sea surface temperature prediction, and particularly to a sea surface temperature prediction method and system. Background Art

[0002] Sea surface temperature (SST) is a crucial indicator in oceanography and climate science. It not only directly reflects the ability of the ocean to absorb and release heat, but also plays a core role in the global climate system. Changes in SST can indicate the dynamics of ocean circulation, affect marine biodiversity and the balance of ecosystems, and thus have a profound impact on fishery resources, coral reef health, and the marine food chain.

[0003] In climate change research, SST data is crucial for understanding the trends and impacts of global warming. It helps scientists evaluate the potential impacts of climate change on marine ecosystems and how these changes feedback into the atmosphere to affect global climate patterns.

[0004] Therefore, continuous monitoring and accurate prediction of SST are of inestimable value for promoting environmental science, protecting marine resources, addressing climate change, and promoting sustainable development. With the development of remote sensing technology and numerical models, scientists can measure and predict changes in SST more precisely, thus providing strong support for research and decision-making in related fields. Summary of the Invention

[0005] The purpose of this part is to outline some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this part, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this part, the abstract, and the title, but such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the above existing problems, the present invention is proposed.

[0007] Therefore, the present invention provides a sea surface temperature prediction method and system, which can solve the problems mentioned in the background art.

[0008] To solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides a sea surface temperature prediction method, comprising:

[0010] Obtaining first historical data and performing first preprocessing on the first historical data;

[0011] Performing first decomposition on the first historical data after the first preprocessing to obtain first decomposition data;

[0012] Use the first decomposed data as the input for pre-training the first prediction model, and perform sea surface temperature prediction based on the output of the first prediction model.

[0013] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the first pre-processed first historical data is subjected to a first decomposition to obtain the first decomposed data, including:

[0014] Perform a first decomposition on the first pre-processed first historical data;

[0015] The first decomposition includes decomposing the first pre-processed first historical data into a plurality of data components;

[0016] Form the first decomposed data from the plurality of data components.

[0017] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the first prediction model includes: the first pre-training model is any model with the first decomposed data as the input and the sea surface temperature or relevant parameters that can directly or indirectly obtain the sea surface temperature as the output.

[0018] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the step of using the first decomposed data as the input for pre-training the first prediction model includes:

[0019] Perform a second pre-processing on the first decomposed data;

[0020] Use the second pre-processed first decomposed data as the input for pre-training the first prediction model.

[0021] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the first prediction model further includes at least three network structures, and the three network structures include a convolutional network layer, a multi-scale transformation layer, and a connection layer.

[0022] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the step of performing a first decomposition on the first pre-processed first historical data includes:

[0023] Decompose the first pre-processed first historical data into several components through a first decomposition operation;

[0024] The several components at least include a seasonal term component, a trend term component, and a residual term component;

[0025] Use all of the several components as the input for pre-training the first prediction model.

[0026] As a preferred embodiment of the sea surface temperature prediction method of the present invention, wherein: the convolutional network layer includes at least one two-dimensional convolutional layer, one normalization layer, and one ReLU activation function.

[0027] In a second aspect, the present invention provides a sea surface temperature prediction system, including:

[0028] A preprocessing module, configured to obtain first historical data and perform first preprocessing on the first historical data;

[0029] A decomposition module, configured to perform first decomposition on the first historical data after the first preprocessing to obtain first decomposition data;

[0030] A prediction module, configured to use the first decomposition data as the input of a pre-trained first prediction model and perform sea surface temperature prediction according to the output of the first prediction model.

[0031] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the method described above are implemented.

[0032] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described above are implemented.

[0033] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention provides a sea surface temperature prediction method and system, which obtains first historical data and performs first preprocessing on the first historical data; performs first decomposition on the first historical data after the first preprocessing to obtain first decomposition data; uses the first decomposition data as the input of a pre-trained first prediction model and performs sea surface temperature prediction according to the output of the first prediction model. It can improve the accuracy and efficiency of sea surface temperature prediction through advanced data processing and analysis techniques. By using a deep learning model, it can learn complex patterns in historical data and accurately predict future sea surface temperature changes. In addition, it can also adapt to the specific conditions of different sea areas, provide customized prediction services, and meet the needs of different users. The multi-scale transformation layer of the prediction module can process data changes on different time scales, while the connection layer ensures smooth information transmission between different network layers, enhancing the generalization ability of the model. Through these technical means, the sea surface temperature prediction method and system of the present application provide strong technical support for marine environmental monitoring, climate change research, and related industries. Description of the Drawings

[0034] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings. Among them:

[0035] Figure 1 It is a flowchart of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0036] Figure 2 It is a schematic diagram of the specific algorithm framework of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0037] Figure 3 It is a schematic diagram of the STL inner loop process of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0038] Figure 4 It is a schematic diagram of a hybrid quantum-classical neural network prediction model of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0039] Figure 5 It is a schematic diagram of a Multi-scale Qtransformer of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0040] Figure 6 It is a schematic diagram of the VQC architecture of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0041] Figure 7 It is a schematic diagram of the quantum circuit of the quantum network layer of a Multi-scale Qtransformer of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0042] Figure 8 It is a schematic diagram of the quantum circuit design of the quantum fully connected layer of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0043] Figure 9 It is a schematic diagram of the quantum attention mechanism of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0044] Figure 10 It is a schematic diagram of the variational quantum circuit design of the quantum attention mechanism layer of a sea surface temperature prediction method and system provided by an embodiment of the present invention;

[0045] Figure 11Internal structure diagram of a computer device for a sea surface temperature prediction method and system provided by an embodiment of the present invention. Detailed implementation manners

[0046] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following will describe the detailed implementation manners of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0047] Embodiment 1

[0048] Referring to Figures 1 - 11 , which is the first embodiment of the present invention. This embodiment provides a sea surface temperature prediction method and system, including:

[0049] In the existing related technologies, there are some problems. For example, traditional prediction models often rely on a large amount of historical data, which not only increases the complexity of data processing, but also seriously affects the prediction accuracy in the case of missing or incomplete data. In addition, traditional models usually lack the ability to deeply explore potential patterns in the data, which limits the further improvement of prediction accuracy.

[0050] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail how to implement the sea surface temperature prediction method;

[0051] Figure 1 shows a method flowchart of a sea surface temperature prediction method and system, including:

[0052] S101, obtain first historical data and perform first preprocessing on the first historical data;

[0053] In an optional embodiment, the prediction method of this application can be used in different scenarios. For example, it can be integrated into the existing systems of institutions such as ocean monitoring stations, weather forecasting centers, and ocean resource development companies. In this way, users can monitor the changes in sea surface temperature in real time and make corresponding decisions based on the prediction results. The prediction system can be designed modularly to be compatible with different types of hardware and software, thereby improving the flexibility and applicability of the system.

[0054] In an alternative embodiment, the prediction method of the present application can also be used for temperature prediction in other scenarios, including but not limited to fields such as agriculture, fishery, tourism, and environmental protection. For example, in the agricultural field, the prediction of sea surface temperature can help farmers understand the impact of ocean climate on crop growth, thereby optimizing planting plans and irrigation strategies. In the fishery field, accurate sea surface temperature prediction is crucial for determining the migration routes and fishing seasons of fish. The tourism industry can utilize sea surface temperature data to evaluate the safety of beaches and water activities, while the environmental protection department can use this data to monitor and protect the marine ecosystem.

[0055] In an alternative embodiment, in addition to sea surface temperature prediction, other scenario temperatures can also be predicted. For example, the prediction system can be extended to monitor and predict atmospheric temperature, soil temperature, etc. By integrating advanced sensors and data processing technologies, the system can collect and analyze temperature data from different environments, providing users with a comprehensive temperature change trend. In addition, the system can also customize specific temperature prediction models according to user needs to adapt to the climate characteristics and seasonal changes in different regions. In this way, users can obtain more accurate and personalized temperature prediction services, thereby making more scientific decisions in multiple aspects such as agricultural planting, urban planning, and disaster warning.

[0056] In the embodiment of the present application, for the high-precision prediction of sea surface temperature changes, the first historical data regarding the ocean in the target area is obtained.

[0057] In an alternative embodiment, the first historical data may include historical measurement values of sea surface temperature, historical satellite remote sensing data of sea surface temperature, and historical model simulation data of sea surface temperature.

[0058] In an alternative embodiment, these data can cover a time range of several decades, providing rich historical background information for the system, thereby improving the accuracy of prediction. By analyzing these historical data, the system can identify the patterns and trends of temperature changes, providing a basis for predicting future temperature changes.

[0059] In an alternative embodiment, real-time data such as satellite remote sensing data and ocean buoy data can also be combined to further improve the timeliness and accuracy of prediction. Through this method of integrating historical and real-time data, the sea surface temperature prediction system can provide users with more reliable and timely temperature change information.

[0060] In the embodiments of the present application, there is no limitation on the first historical data, and relevant technical personnel can select it according to actual needs. For example, a data set matching the ocean environmental characteristics of the target area can be selected to ensure the applicability and accuracy of the prediction model. In addition, the system can also integrate other types of data as needed, such as ocean salinity, sea current speed and direction, etc. These factors may all affect the change of ocean surface temperature. By comprehensively considering various environmental factors, the prediction system can more comprehensively reflect the dynamic changes of ocean surface temperature and provide more accurate prediction results for users.

[0061] In an optional embodiment, the first preprocessing is to process the first historical data into an original sequence convenient for decomposition. The methods adopted in the first preprocessing may include smoothing processing, denoising processing, normalization processing, etc. These preprocessing steps help to reduce noise and outliers in the data, making subsequent decomposition and analysis more accurate. For example, smoothing processing can be achieved by the moving average method or the exponential smoothing method to reduce the influence of short-term fluctuations; denoising processing may use filters or wavelet transforms to remove high-frequency noise in the data; normalization processing ensures that the data is compared and analyzed on the same scale, avoiding analysis biases caused by differences in dimensions or numerical ranges. Through these preprocessing steps, the first historical data is converted into a format more suitable for time series analysis, providing a solid data foundation for subsequent identification and prediction of temperature change patterns.

[0062] In the embodiments of the present application, the first preprocessing requires at least obtaining the original sea surface temperature (SST) sequence data of the historical time period according to the first historical data. For example, satellite remote sensing data, ocean buoy records or ship observation records and other sources can be used. These data sources can provide continuous temperature information covering a wide sea area, helping to capture the spatio-temporal change characteristics of sea surface temperature.

[0063] In the embodiments of the present application, after obtaining the original SST sequence data, the first preprocessing step will preliminarily organize and format these data to ensure the integrity and consistency of the data, laying a foundation for subsequent analysis and prediction. For example, it may be necessary to synchronize the time of the data to ensure that data from different sources is compared and analyzed on the same time scale. In addition, the processing of missing values is also an indispensable part of the preprocessing. Missing data is filled by interpolation or estimation methods to ensure the continuity of the data sequence. Through these meticulous preprocessing efforts, the accuracy of subsequent analysis and the reliability of prediction results can be ensured.

[0064] It should be noted that obtaining the first historical data and performing the first preprocessing on the first historical data can provide an accurate baseline for sea surface temperature changes. This step is crucial for establishing an effective prediction model because it ensures the quality and relevance of the input data. In this way, the prediction system can more accurately capture the patterns and trends of temperature changes, thereby improving the accuracy of the prediction. In addition, good preprocessing can also reduce the complexity of subsequent calculations and improve the efficiency of the entire prediction process.

[0065] S102, perform the first decomposition on the first historical data after the first preprocessing to obtain the first decomposition data;

[0066] In the embodiment of the present application, performing the first decomposition on the first historical data after the first preprocessing to obtain the first decomposition data includes:

[0067] Perform the first decomposition on the first historical data after the first preprocessing;

[0068] The first decomposition includes decomposing the first historical data after the first preprocessing into several data components;

[0069] Form the first decomposition data from the several data components.

[0070] In an optional embodiment, the first decomposition can use the empirical mode decomposition (EMD) method, which can decompose a complex signal into a series of intrinsic mode functions (IMFs). In this way, the local characteristics of the signal can be analyzed more carefully, thereby providing richer information for the prediction of sea surface temperature changes. In addition, the EMD method can adapt to the nonlinear and non-stationary characteristics of the signal, enabling the prediction model to better handle the complexity of actual sea surface temperature data. After obtaining the first decomposition data, these data components can be further analyzed and processed to extract the most valuable information for predicting sea surface temperature changes.

[0071] In an optional embodiment, the first decomposition can also use the wavelet transform (WT) method. The wavelet transform is an effective time-frequency analysis tool. By selecting an appropriate wavelet basis function, it can decompose a signal into wavelet components at different scales and positions. This method is particularly suitable for processing signals with local characteristics, such as mutations and periodic changes in sea surface temperature data. Through the wavelet transform, the signal can be expanded at different scales, thereby revealing the detailed characteristics of the signal, which is of great significance for improving the accuracy of sea surface temperature change prediction. In addition, the wavelet transform also has good denoising ability and can effectively remove the noise in the data, further improving the reliability of the prediction results.

[0072] In an alternative embodiment, the first decomposition may also use the Adaptive Noise Decomposition (AD) method. The AD method is a data-driven decomposition technique that can automatically adjust the decomposition process according to the characteristics of the data itself to adapt to different signal features. This method is particularly suitable for processing sea surface temperature data with complex noise characteristics because it can effectively separate the useful components and noise components in the signal. By adaptively adjusting the decomposition parameters, the AD method can ensure the accuracy and reliability of the decomposition results, providing high-quality input data for subsequent prediction models.

[0073] In the embodiment of the present application, performing the first decomposition on the first preprocessed first historical data includes:

[0074] Decomposing the first preprocessed first historical data into several components through a first decomposition operation;

[0075] The several components at least include a seasonal term component, a trend term component, and a residual term component;

[0076] Using all the several components as the input for pre-training the first prediction model.

[0077] In the embodiment of the present application, specifically, the STL decomposition is used for a decomposition operation. SST is single-element time series data, so SST prediction can be regarded as a single-element time series regression problem. Therefore, many researchers have tried to use time series analysis methods to predict SST, hoping to obtain higher SST prediction accuracy.

[0078] It should be noted that most current prediction methods achieve prediction by fitting the variation law of SST historical data, only considering the non-linearity of SST, and not fully utilizing the periodicity, persistence, and non-stationarity of SST, which limits the SST prediction accuracy to a certain extent. The Seasonal-Trend Decomposition Procedure based on Loess (STL) is a common time series decomposition algorithm that can not only explore the law of historical data but also be used for prediction. It is applicable to any periodic data and has good robustness. However, it can only decompose the time series into additive change components and is currently widely used in ocean research and the climate field.

[0079] In an alternative embodiment, in order to fully utilize the periodicity, persistence, and non-stationarity of SST and achieve the purpose of improving SST prediction accuracy.

[0080] In the embodiments of the present application, this technology uses STL to decompose the SST original sequence T into a seasonal component S, a trend component C, and a remainder component R, and its decomposition expression is as follows:

[0081] T t = S t + C t + R t (t|0 ≤ t ≤ |T|, t ∈ Z)

[0082] STL was originally proposed by Cleveland et al. It is a filtering process that decomposes a time series into additive change components based on Loess. Loess extracts partial local data and uses the local data to fit a polynomial regression curve, making the curve smoother and making the trends and patterns in the data within the local range easier to observe.

[0083] The calculation process of its decomposition consists of two parts: an inner loop and an outer loop. Each iteration of the inner loop includes seasonal smoothing for updating the seasonal component. After the inner loop is completed, robust weights are calculated in the outer loop and are used in the next inner loop to reduce the influence of outliers on updating the trend component and the seasonal component in the subsequent inner loop.

[0084] In the embodiments of the present application, Figure 3 is a schematic diagram of the STL inner loop process. In the outer loop, the remainder component is calculated using the seasonal component and the trend component obtained in the inner loop. Larger values in the remainder component are regarded as outliers in the data. By introducing robust weights, in the next iteration of the inner loop, the weights are used to reduce the influence of the outliers identified in the previous iteration of the outer loop.

[0085] Step (1): Detrending. In the (i + 1)-th iteration of the inner loop, subtract the estimated trend component obtained in the i-th iteration from the original sequence T Detrending,

[0086] Step (2): Smoothing of the periodic subsequence. Use Loess to smooth the periodic subsequence to obtain a preliminary seasonal component

[0087] Step (3): Low-pass filtering of the smoothed periodic subsequence. Process the preliminary seasonal component obtained in step (2) using a low-pass filter Then use Loess to obtain

[0088] Step (4): Detrending of the smoothed periodic subsequence. The seasonal component is the difference between the low-pass value and the preliminary seasonal component,

[0089] Step (5): Seasonal adjustment, subtracting the seasonal component from the original sequence T Obtain

[0090] Step (6): Trend smoothing, using Loess for After smoothing, the trend component is obtained

[0091] It should be noted that decomposing the first preprocessed first historical data to obtain the first decomposition data can decompose the complex sea surface temperature changes into more easily analyzable and predictable components. In this way, the characteristics at different time scales such as seasonal changes, long-term trends, and short-term fluctuations can be more clearly identified. For example, the seasonal component can reveal the regular fluctuations of the sea surface temperature with seasons, the trend component can reflect the long-term climate change trend, and the remaining component may contain the temperature fluctuations caused by weather events or other short-term factors. This decomposition method not only helps to understand the internal mechanism of the sea surface temperature changes, but also provides a solid foundation for establishing a more accurate prediction model. In practical applications, these decomposed data components can be used as input variables to train and optimize the prediction model, thereby improving the prediction ability of the model for future changes in the sea surface temperature. In addition, by analyzing these components, researchers and decision-makers can better evaluate and respond to the impacts of climate change on the marine environment and related industries.

[0092] S103. Use the first decomposition data as the input for pre-training the first prediction model, and perform sea surface temperature prediction according to the output of the first prediction model.

[0093] In the embodiment of the present application, the first prediction model includes: The first pre-training model is any model with the first decomposition data as the input and the sea surface temperature or relevant parameters that can directly or indirectly obtain the sea surface temperature as the output.

[0094] In an optional embodiment, the first prediction model can be constructed using a deep learning network, which can automatically learn and extract the complex features in the sea surface temperature data. Through training, the network can identify the seasonal patterns, trend changes, and abnormal fluctuations in the data, so as to take these factors into account during prediction. In addition, the prediction system can also integrate multiple data sources, such as satellite remote sensing data, buoy observation data, and historical climate data, to enhance the accuracy and reliability of the prediction. In this way, the prediction system can not only provide short-term sea surface temperature predictions, but also support long-term climate change research.

[0095] In an alternative embodiment, the first prediction model can also be constructed using a Support Vector Machine (SVM). SVM is a powerful machine learning algorithm, especially suitable for dealing with high-dimensional data and non-linear problems. In sea surface temperature prediction, SVM can map data into a high-dimensional space through a kernel function, thereby finding the optimal decision boundary in this space. This model is particularly suitable for dealing with the non-linear characteristics of sea surface temperature data and can effectively identify and predict complex patterns of temperature changes. In addition, the SVM model also has good generalization ability and can provide relatively accurate prediction results even when the amount of data is limited. By adjusting the parameters of the SVM model, such as the penalty factor and the type of kernel function, the prediction performance can be further optimized to adapt to different prediction requirements and data characteristics.

[0096] In an alternative embodiment, the first prediction model can also be constructed using the Random Forest algorithm. Random Forest is an ensemble learning method that improves the accuracy and stability of prediction by constructing multiple decision trees and aggregating their prediction results. In sea surface temperature prediction, Random Forest can handle a large number of features and samples while reducing the risk of overfitting. By randomly selecting a subset of features at each node split, this algorithm can capture complex relationships and non-linear patterns in the data. In addition, the Random Forest model also has good interpretability and can provide an assessment of feature importance to help researchers understand which factors have a greater impact on sea surface temperature changes. By adjusting parameters such as the number of trees and the depth of the trees, the performance of the Random Forest model can be further optimized to meet the specific requirements of sea surface temperature prediction.

[0097] In the embodiment of the present application, a hybrid quantum-classical neural network prediction model is designed as the first prediction model.

[0098] In the embodiment of the present application, including the first decomposed data as the input for pre-training the first prediction model:

[0099] Perform a second preprocessing on the first decomposed data;

[0100] Use the first decomposed data after the second preprocessing as the input for pre-training the first prediction model.

[0101] In an alternative embodiment, before the second preprocessing is used to input the first decomposed data into the prediction model, it is necessary to perform min-max normalization on the three components to eliminate the dimensional influence between indicators, reduce the influence of extreme values, and improve the robustness of the model.

[0102] In an optional embodiment, the second preprocessing can be performed by normalizing the first decomposed data to ensure that the data is on a unified scale, facilitating model processing and learning. Normalization scales the data proportionally so that it falls within a small specific interval. Usually, it performs a linear transformation on the original data to make it fall within the [0,1] interval. This processing method helps to accelerate the convergence speed of the model and can prevent the problems of vanishing gradients or exploding gradients during training. In addition, the normalized data can make the differences between different features more obvious, thereby improving the model's ability to identify the importance of features. In sea surface temperature prediction, the normalized data can help the model more accurately capture the subtle differences in temperature changes, thus improving the prediction accuracy.

[0103] In the embodiment of the present application, the first prediction model further includes at least three layers of network structures, and the three-layer network structure includes a convolutional network layer, a multi-scale transformation layer, and a connection layer.

[0104] In the embodiment of the present application, the convolutional network layer includes at least one two-dimensional convolutional layer, a normalization layer, and a ReLU activation function.

[0105] In the embodiment of the present application, a hybrid quantum-classical neural network prediction model is designed as the first prediction model. Specifically:

[0106] Since the Transformer model was proposed, it has achieved remarkable achievements in the fields of natural language processing (NLP) and computer vision (CV). As time goes by, researchers have begun to explore its application in time series prediction. Time series prediction is a task of using historical data to predict future values, which is common in fields such as finance, meteorology, and energy.

[0107] The application of the Transformer model in time series prediction mainly benefits from its self-attention mechanism, which allows the model to capture long-range dependencies when processing sequence data. In response to the challenges of time series prediction, such as seasonality, trends, and periodicity, researchers have made various improvements to the standard Transformer model. For example, some models introduce positional encoding to consider the temporal order information in the time series; some models modify the attention module to more effectively process the specific characteristics of time series data. In addition, some models attempt to combine the Transformer with other models (such as state space models or graph neural networks) to improve the prediction accuracy and robustness.

[0108] In the embodiments of the present application, the present technology also makes improvements to the Transformer model, introduces a quantum neural network layer, and constructs an Adaptive Multi-Scale Quantum Block (AMS QBlock). This design realizes adaptive multi-scale modeling through multi-scale QTransformer blocks and adaptive paths, can capture changes in features at different scales, and improve prediction accuracy. The structure of the hybrid quantum-classical neural network prediction model is as Figure 4 shown, mainly composed of four groups of AMS modules. Each AMS module is mainly composed of multiple stacked multi-scale quantum Transformers (Multi-scale Qtransformer) and a quantum fully connected layer and a quantum attention mechanism module. The structure of the Multi-scale Qtransformer is as Figure 5 shown.

[0109] In the embodiments of the present application, all the quantum network layer parts in the model are implemented based on the Variational Quantum Circuit (VQC). VQC is a quantum circuit containing parameterized quantum logic gates, whose parameters are adjustable and can be iteratively optimized. The general VQC architecture is as Figure 6 shown. Among them, the U(x) block is used for state preparation to encode classical data x into the quantum state of the circuit and is not affected by optimization. The V(θ) block represents the variational part with learnable parameters θ, which will be optimized by gradient methods. Some existing research results show that this kind of circuit has strong robustness to quantum noise, so it is suitable for NISQ devices. VQC has been successfully applied to tasks such as function approximation, classification, generative modeling, deep reinforcement learning, and transfer learning.

[0110] In an optional embodiment, VQC is more expressive than classical neural networks and may therefore be better than the latter. The expression ability here refers to the ability to represent certain functions or distributions with a finite number of parameters. The three quantum network layers in the Multi-scale Qtransformer module are all implemented using the same variational quantum circuit. The input data is angle-encoded using RY and RX logic gates, and then quantum entanglement is performed using RY parameterized logic gates and CNOT logic gates. Four output vectors are obtained after quantum measurement. The specific circuit design is as Figure 7 shown.

[0111] In an alternative embodiment, angle encoding is a technique used for data representation in quantum machine learning. It utilizes the rotations of quantum gates (RX, RY, and RZ) to encode classical information. This method encodes N features of classical data as the angles of n input qubits between quantum states. In this approach, N is kept equal to n to enable the quantum network layer to use the maximum size of classical features as much as possible. The quantum state generated by performing angle encoding on the input qubits can be represented by the following formula:

[0112]

[0113] where R(.) can be any one of the RX, RY, and RZ logic gates. In angle encoding, the angles between quantum states can vary continuously to capture complex data. This leads to a more precise and detailed data representation and can improve the performance of certain types of quantum machine learning models. Although it can only encode one eigenvalue into one qubit, it reduces noise, which makes it particularly advantageous in NISQ computing.

[0114] In an alternative embodiment, the implementation of the quantum fully connected layer is to input the N features of the input feature vector into N corresponding qubits respectively, perform angle encoding with the RY logic gate, then perform transformation through the CNOT gate and the RY gate with optimizable parameters, and finally obtain the output feature vector of the required M features through quantum measurement. The specific circuit structure is as Figure 8 shown.

[0115] In an alternative embodiment, the quantum attention mechanism mainly extracts features from the Patch components after Patch segmentation through three variational quantum layers, fuses the feature information of different components, improves the generalization ability of the model, and reduces the computational complexity of the model. The specific structure is as Figure 9 shown. The three variational quantum layers are implemented using the same variational quantum circuit. Different from the Figure 7 and Figure 8 two quantum circuit designs, this quantum circuit introduces amplitude encoding and uses n qubits to encode 2 n feature vectors. The specific circuit design is as Figure 10 shown. The quantum logic gates used include the RX, RYY, and RZ gates, all of which are parameterized quantum logic gates. Amplitude encoding is another technique for data representation. It represents a normalized classical data point as the amplitude of a quantum state. After amplitude encoding the 2 n feature vectors, the quantum state of the qubit becomes:

[0116]

[0117] where |ψ x> is the quantum state corresponding to the N-dimensional classical data X, where N = 2 n , x i is the i-th element of X, and |i> is the i-th computational basis state. In a classical neural network, each binary value requires an explicit trainable weight or bias, resulting in a significant number of parameters. In contrast, amplitude encoding allows data to be represented by the amplitudes of a finite number of quantum states, enabling a more compact representation. This has been shown to lead to a significant reduction in the number of trainable parameters, helping to simplify the model and improve its performance. While this approach offers this benefit, if the number of qubits in the layer increases, it also increases the depth of the quantum circuit by O(poly(n)) or O(n).

[0118] In summary, the present invention proposes a sea surface temperature prediction method, which obtains first historical data and performs first preprocessing on the first historical data; decomposes the first preprocessed first historical data to obtain first decomposition data; uses the first decomposition data as the input for pre-training a first prediction model, and predicts the sea surface temperature based on the output of the first prediction model. It can improve the accuracy and efficiency of sea surface temperature prediction through advanced data processing and analysis techniques. By using a deep learning model, it can learn complex patterns in historical data and accurately predict future sea surface temperature changes. In addition, it can also adapt to the specific conditions of different sea areas, provide customized prediction services, and meet the needs of different users. The multi-scale transformation layer of the prediction module can handle data changes at different time scales, while the connection layer ensures smooth information transfer between different network layers, enhancing the generalization ability of the model. Through these technical means, the sea surface temperature prediction method and system of the present application provide strong technical support for marine environmental monitoring, climate change research, and related industries.

[0119] Embodiment 2

[0120] In a preferred embodiment, the present technical solution realizes the prediction of the sea surface temperature at a future moment through an algorithm that combines seasonal time series decomposition and a hybrid quantum-classical neural network prediction model. The specific framework of this algorithm is as Figure 2 shown. The input of the algorithm is the original sea surface temperature (SST) sequence data for a historical time period. First, it performs STL decomposition on it to obtain the seasonal component, trend component, and residual component, and then uses these three components as the input for the hybrid quantum-classical neural network prediction model. Before inputting into the prediction model, it is necessary to perform min-max normalization on the three components to eliminate the dimensional influence between indicators, reduce the influence of extreme values, and improve the robustness of the model.

[0121] After that, the prediction result is obtained through a hybrid quantum-classical neural network time series prediction model. The first layer of the model is a set of standard convolutional network layers, including a two-dimensional convolutional layer (Conv2D), a normalization layer (BatchNormlization2D), and a ReLU activation function; the second layer is an adaptive multi-scale Qtransformer layer. Qtransformer is a hybrid neural network architecture composed of multiple adaptive multi-scale quantum modules, which uses the long-distance dependence capture advantage of Qtransformer to extract complex time series features; the third layer is a fully connected layer, which is the input layer to obtain the SST prediction result within the next few days (the prediction time step is variable for the model). Next, the STL decomposition and the hybrid quantum-classical neural network prediction model will be introduced separately.

[0122] Embodiment 3

[0123] This embodiment also provides a sea surface temperature prediction system, which is characterized by including:

[0124] A preprocessing module for obtaining first historical data and performing first preprocessing on the first historical data;

[0125] A decomposition module for performing first decomposition on the first historical data after the first preprocessing to obtain first decomposition data;

[0126] A prediction module for using the first decomposition data as the input of a pre-trained first prediction model and predicting the sea surface temperature according to the output of the first prediction model.

[0127] The above unit modules can be embedded in or independent of the processor in the computer device in the form of hardware, or stored in the memory of the computer device in the form of software, so as to facilitate the processor to call and execute the operations corresponding to the above modules.

[0128] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as Figure 11As shown in the figure. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a sea surface temperature prediction method. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the computer device housing, or an external keyboard, touchpad, or mouse, etc.

[0129] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0130] Obtain first historical data and perform first preprocessing on the first historical data;

[0131] Perform first decomposition on the first preprocessed first historical data to obtain first decomposition data;

[0132] Use the first decomposition data as the input for pre-training the first prediction model, and perform sea surface temperature prediction according to the output of the first prediction model.

[0133] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0134] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages. For example, object-oriented programming languages such as Java and interpreted scripting languages such as JavaScript.

[0135] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices produce a means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0136] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction means that implements the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0137] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, so that the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0138] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they know the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0139] Obviously, those skilled in the art can make various changes and variations to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these changes and variations.

Claims

1. A method for predicting sea surface temperature, characterized in that: include: Acquire first historical data, and perform first preprocessing on the first historical data; Performing a first decomposition on the first historical data after the first preprocessing to obtain first decomposed data; The first decomposed data is used as an input of a pre-trained first prediction model, and sea surface temperature prediction is performed according to an output of the first prediction model.

2. The sea surface temperature prediction method according to claim 1, characterized in that: The first decomposing the first historical data after the first preprocessing to obtain the first decomposed data comprises: Performing a first decomposition on the first historical data after the first preprocessing; The first decomposition includes decomposing the first preprocessed first historical data into a plurality of data components; The plurality of data components are combined into first decomposed data.

3. The sea surface temperature prediction method according to claim 2, characterized in that: The first prediction model includes: the first pre-trained model is any model whose input is the first decomposed data and whose output is the sea surface temperature or related parameters of the sea surface temperature that can be directly or indirectly obtained.

4. The sea surface temperature prediction method according to claim 3, characterized in that: The using the first decomposed data as the input of the pre-trained first prediction model comprises: performing a second preprocessing on the first decomposed data; The first decomposed data after the second preprocessing is used as input of the pre-trained first prediction model.

5. The sea surface temperature prediction method according to claim 4, characterized in that: The first prediction model also includes at least three layers of network structure, and the three layers of network structure include a convolutional network layer, a multi-scale transformation layer and a connection layer.

6. The sea surface temperature prediction method according to claim 5, characterized in that: The first decomposition of the first historical data after the first preprocessing comprises: Decomposing the first historical data after the first preprocessing into a plurality of components through a first decomposition operation; The several components include at least a seasonal component, a trend component and a residual component; The several components are all used as inputs of the pre-trained first prediction model.

7. The sea surface temperature prediction method according to claim 6, characterized in that: The convolutional network layer includes at least a two-dimensional convolutional layer, a normalization layer and a ReLU activation function.

8. A sea surface temperature prediction system, characterized in that: include: A preprocessing module, used for acquiring first historical data and performing first preprocessing on the first historical data; A decomposition module, used for performing a first decomposition on the first historical data after the first preprocessing to obtain first decomposed data; The prediction module is used to use the first decomposed data as an input of a pre-trained first prediction model and predict the sea surface temperature according to the output of the first prediction model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images