A cross-modal feature fusion method for predicting tidal bore height in Qiantang River

Abstract

The present invention discloses a Qiantang River tidal height prediction method using cross-modal feature fusion. The Qiantang River tidal height prediction method of the present invention uses a cross-modal fusion method to perform feature fusion between different data sources and a generative decoder for prediction. This method effectively combines tidal height data with meteorological data, utilizes the unique properties and complementary information of each data, and improves the model's understanding and prediction capabilities of tidal dynamics. The application of a generative decoder allows the model to predict future tidal height trends at one time, which not only optimizes the efficiency of the prediction process, but also significantly reduces the error accumulation that may occur in step-by-step predictions.

Description

Technical Field

[0001] The present invention relates to the field of Qiantang River tidal bore height prediction, and in particular to a Qiantang River tidal bore height prediction method based on cross-modal feature fusion. Background Art

[0002] With the widespread application and maturity of deep learning technology in the field of time series prediction, its significant advantages in processing complex, nonlinear and high-dimensional data have been widely recognized. This technology has been successfully applied in many important fields such as finance, energy, transportation, climate and environment, providing new solutions and perspectives for these fields. As the largest river in Zhejiang Province, Qiantang River accounts for about half of the province's basin area, basin population and GDP. The tidal bore phenomenon of Qiantang River has a very important impact on environmental facilities and the daily life of local residents. The accurate prediction of tidal bore is of great significance in terms of tide prevention safety, river-related engineering construction, tourism services, shipping and water resource utilization.

[0003] In the prediction of complex phenomena such as tidal bores, the interaction of multiple influencing factors constitutes a complex nonlinear relationship. It is difficult to fully capture the dynamic interaction between these factors using only a single data source or modality, which limits the effectiveness of traditional prediction methods. In order to overcome this challenge, multimodal fusion strategies are particularly important. This method enriches the expression of input data by integrating data from multiple different modalities, thereby providing more comprehensive information for the prediction model. This enables the model to have a deeper understanding and accurately capture the various factors affecting tidal bores and their interactions. In addition, traditional autoregressive prediction methods often face serious cumulative error problems when dealing with multi-step prediction tasks. Since the prediction of each step depends on the output of the previous step, any small initial error may be amplified in the subsequent prediction of the sequence, thereby affecting the accuracy and reliability of the overall prediction. Therefore, developing new prediction methods to reduce cumulative errors and improve the accuracy of multi-step predictions is an important direction of current research. Summary of the invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention aims to propose a Qiantang River tidal height prediction method based on cross-modal feature fusion, which integrates the tidal sequence with other key meteorological data by adopting the cross-modal feature fusion method, and avoids the problem of cumulative error in multi-step prediction by using the design of a generative decoder.

[0005] In order to achieve the above object, the technical solution adopted by the present invention includes the following steps:

[0006] In a first aspect, the present invention provides a Qiantang River tidal height prediction method based on cross-modal feature fusion, which comprises:

[0007] S1. Obtain the historical sequence of Qiantang River tidal height and the historical sequence of Qiantang River regional meteorological data before the target period, and preprocess the data to meet the model input requirements. The historical meteorological data types include four modes: wind speed data, temperature data, precipitation data, and air pressure data;

[0008] S2. Obtain a pre-trained Qiantang River tidal height prediction model, which is composed of an encoder based on cross-modal feature fusion and a generative decoder; in the encoder, each type of historical sequence of Qiantang River regional meteorological data in the model input is first processed by a convolution layer to extract key spatial features of meteorological data, and then a self-attention mechanism is applied to each key spatial feature of meteorological data processed by the convolution layer and the Qiantang River tidal height historical sequence in the model input through a self-attention layer to obtain multimodal features after adjusting the importance of features, and finally the multimodal features processed by the self-attention are spliced ​​and fused along the feature dimension to obtain fused features; in the decoder, the fused features obtained by the encoder are first spliced ​​with a placeholder embedding having a length equal to the sequence to be predicted, and the spliced ​​input features are used to obtain an output sequence through a fully connected network;

[0009] S3, input the Qiantang River tidal height historical sequence after data preprocessing in S1 and the Qiantang River regional meteorological data historical sequence into the pre-trained Qiantang River tidal height prediction model to obtain the Qiantang River tidal height prediction sequence corresponding to the target period.

[0010] As a preferred embodiment of the first aspect, the wind force and speed data adopts the wind speed vector u10 indicator in the east-west direction.

[0011] As a preferred embodiment of the first aspect, the data preprocessing includes sequentially performing missing value filling, resampling and normalization on the historical sequences of each modality data.

[0012] As a preferred embodiment of the above-mentioned first aspect, when the Qiantang River area meteorological data historical series is resampled, the spatial distribution map of the Qiantang River area meteorological data at each historical moment needs to be resampled to the same spatial resolution, and at the same time, the temporal resolution of the Qiantang River area meteorological data needs to be resampled to the same as the temporal resolution of the Qiantang River tidal height historical series.

[0013] As a preferred embodiment of the first aspect, the normalization adopts maximum and minimum value normalization.

[0014] As a preferred embodiment of the first aspect above, the Qiantang River tidal height prediction model needs to be supervised and trained with a labeled data set before being used for actual reasoning. During the training process, the goal is to minimize the loss function value, Adam is used as the optimizer of the model to adjust the network parameters, and an early stopping mechanism is introduced to control the iterative cycle of training.

[0015] As a preferred embodiment of the above-mentioned first aspect, the loss function adopted for training the Qiantang River tidal height prediction model is the mean square error (MSE) between the Qiantang River tidal height prediction sequence and the true value label.

[0016] In a second aspect, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the Qiantang River tidal height prediction method using cross-modal feature fusion as described in any one of the first aspects above.

[0017] In a third aspect, the present invention provides a computer-readable storage medium, characterized in that a computer program is stored on the storage medium, and when the computer program is executed by a processor, it can implement the Qiantang River tidal height prediction method based on cross-modal feature fusion as described in any one of the first aspects above.

[0018] In a fourth aspect, the present invention provides a computer electronic device, characterized in that it includes a memory and a processor;

[0019] The memory is used to store computer programs;

[0020] The processor is used to implement the Qiantang River tidal height prediction method based on cross-modal feature fusion as described in any one of the first aspects above when executing the computer program.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] The present invention uses a prediction method of cross-modal fusion and generative decoder to implement the prediction of Qiantang River tidal height. By effectively fusing different modes of tidal height data and meteorological data, this method provides a more comprehensive data perspective and enhances the model's understanding and processing capabilities of tidal influencing factors. In addition, the use of a generative decoder allows the present invention to predict tidal heights at multiple future time points at one time, significantly improving prediction efficiency and reducing error accumulation caused by step-by-step prediction. While ensuring prediction accuracy, this method also optimizes the use of computing resources, making tidal prediction faster and more accurate, and providing important decision support for related work. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1A flowchart of the steps of the Qiantang River tidal height prediction method based on cross-modal feature fusion;

[0024] Figure 2 It is a structural schematic diagram of the Qiantang River tidal bore height prediction model of the present invention.

[0025] Figure 3 This is a structural diagram of the generative decoder of the present invention.

[0026] Figure 4 Graph showing the error distribution of the Qiantang River tidal bore height prediction model in an embodiment of the present invention.

[0027] Figure 5 This is an error distribution diagram of the Transformer model used as a comparison in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the above-mentioned purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation mode of the present invention is described in detail below in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below. The technical features in each embodiment of the present invention can be combined accordingly without conflicting with each other.

[0029] In the description of the present invention, it is to be understood that when an element is considered to be "connected" to another element, it may be directly connected to the other element or indirectly connected, that is, there are intermediate elements. On the contrary, when an element is said to be "directly" connected to another element, there are no intermediate elements.

[0030] In the description of the present invention, it should be understood that the terms "first" and "second" are only used for the purpose of distinguishing descriptions, and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features.

[0031] In a preferred embodiment of the present invention, a method for predicting the tidal height of the Qiantang River using cross-modal feature fusion is provided. The method uses a prediction method of cross-modal fusion and a generative decoder to achieve accurate prediction of the tidal height of the Qiantang River by effectively fusing different modalities of tidal height data and meteorological data. Figure 1 As shown, the Qiantang River tidal height prediction method based on cross-modal feature fusion includes the following steps:

[0032] S1. Obtain the historical sequence of Qiantang River tidal height and the historical sequence of Qiantang River regional meteorological data before the target period, and preprocess the data to meet the model input requirements. The historical meteorological data types include four modes: wind speed data, temperature data, precipitation data, and air pressure data.

[0033] It should be noted that the wind force and speed data can generally be characterized by the two parameters u10 and v10. These two parameters represent the components of the eastward and southward wind speeds at a height of 10 meters above the earth's surface, respectively, measured in meters per second (m / s). The u10 parameter indicates the wind speed component in the east-west direction, a positive value means the wind is in the east, and a negative value means the wind is in the west; while the v10 parameter reflects the wind speed component in the north-south direction, wherein a positive value indicates the wind is in the north, and a negative value indicates the wind is in the south. Due to the Qiantang River tidal bore phenomenon, the positive eastward wind (i.e., the positive value of the u10 parameter) has a significant boosting effect, while the positive westward wind (the negative value of the u10 parameter) has a certain inhibitory effect, the wind force and speed data in the present invention preferably adopts the wind speed vector u10 parameter in the east-west direction, as the main basis for evaluating the influence of wind factors on the Qiantang River tide.

[0034] It should be noted that the specific data preprocessing needs to be selected according to the actual data obtained, so as to meet the input requirements of the model. In an embodiment of the present invention, the data preprocessing includes filling missing values, resampling and normalizing the historical sequences of each modal data in sequence. Among them, when the Qiantang River regional meteorological data historical sequence is resampled, the spatial distribution map of the Qiantang River regional meteorological data at each historical moment can be resampled to the same spatial resolution. At the same time, the time resolution of the Qiantang River regional meteorological data needs to be resampled to the same time resolution as the Qiantang River tidal height historical sequence, and the sequence lengths of each model input, if different, also need to be resampled to the same. In addition, the above normalization can adopt maximum and minimum value normalization.

[0035] S2. Obtain a pre-trained Qiantang River tidal bore height prediction model.

[0036] like Figure 2 As shown in the figure, the structure of the Qiantang River tidal height prediction model is presented, which consists of an encoder based on cross-modal feature fusion and a generative decoder.

[0037] In the encoder of the model, each type of historical sequence of meteorological data in the Qiantang River area in the model input is first processed by a convolution layer to extract the key spatial features of the meteorological data. Then, the self-attention mechanism is applied to each key spatial feature of the meteorological data processed by the convolution layer and the historical sequence of the Qiantang River tidal height in the model input. The data of each modality will obtain the corresponding single-modal features after adjusting the feature importance through the self-attention layer. All modalities are finally aggregated to form multi-modal features after adjusting the feature importance. Finally, the multi-modal features after self-attention processing are spliced ​​and fused along the feature dimension to obtain the fused features.

[0038] In the decoder of the model, the fused features obtained by the encoder are first concatenated with a placeholder embedding whose length is equal to the sequence to be predicted, and the concatenated input features are passed through a fully connected network to obtain an output sequence.

[0039] S3, input the Qiantang River tidal height historical sequence after data preprocessing in S1 and the Qiantang River regional meteorological data historical sequence into the pre-trained Qiantang River tidal height prediction model to obtain the Qiantang River tidal height prediction sequence corresponding to the target period.

[0040] It should be noted that the above steps S1 to S3 describe the process of predicting the Qiantang River tidal height in the actual reasoning link. The above Qiantang River tidal height prediction model is pre-trained, that is, the Qiantang River tidal height prediction model needs to be pre-trained before being used for reasoning. The Qiantang River tidal height prediction model can be supervised by using a labeled data set. The goal during the training process is to minimize the loss function value. Adam is used as the optimizer of the model to adjust the network parameters, and an early stopping mechanism is introduced to control the iterative cycle of the training. The loss function used for training the Qiantang River tidal height prediction model is the mean square error (MSE) between the Qiantang River tidal height prediction sequence and the true value label. The specific model training method belongs to the prior art and will not be described in detail.

[0041] Similarly, based on the same inventive concept, the present invention also provides a computer electronic device corresponding to the Qiantang River tidal height prediction method of cross-modal feature fusion provided in the above embodiment, which includes a memory and a processor;

[0042] The memory is used to store computer programs;

[0043] The processor is used to implement the Qiantang River tidal height prediction method based on cross-modal feature fusion as described in the above-mentioned embodiment when executing the computer program.

[0044] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0045] Therefore, based on the same inventive concept, the present invention provides a computer-readable storage medium corresponding to a Qiantang River tidal height prediction method with cross-modal feature fusion, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it can implement the Qiantang River tidal height prediction method with cross-modal feature fusion as described in the aforementioned embodiment.

[0046] Therefore, based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, can implement the Qiantang River tidal height prediction method using cross-modal feature fusion as described in the aforementioned embodiments.

[0047] It is understandable that the above storage medium may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage. The storage medium may also be a U disk, a mobile hard disk, a magnetic disk or an optical disk, etc., which can store program codes.

[0048] It can be understood that the above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0049] It should also be noted that those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. In the various embodiments provided in this application, the division of steps or modules in the system and method is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules or steps can be combined or integrated together, and a module or step can also be split.

[0050] The Qiantang River tidal height prediction method based on cross-modal feature fusion mentioned above is applied to a specific example to demonstrate the construction, training process and test results of the specific Qiantang River tidal height prediction model.

[0051] Example

[0052] In this embodiment, the Qiantang River tidal height prediction method based on cross-modal feature fusion adopts a technical solution comprising the following steps:

[0053] Step 1: Obtain the tidal height and meteorological datasets of Qiantang River and perform preprocessing, including filling in the null and missing values ​​in the tidal height data and meteorological datasets to construct training sample data.

[0054] Among them, meteorological data mainly include multimodal meteorological data covering the area around Qiantang River, including wind force and speed data, temperature data, precipitation data and air pressure data. The meteorological data at each historical moment are recorded in the form of raster spatial distribution.

[0055] The mean interpolation method is used to fill in the null values ​​and missing values. The specific formula is as follows:

[0056]

[0057] Among them, y i is a missing value, y i-1 and i+1 Corresponding to the values ​​of the day before and the day after the missing value respectively.

[0058] As mentioned above, in this embodiment, the wind speed data selects the u10 parameter because this parameter has an important impact on the Qiantang River tide. In the preprocessing of precipitation data, since the directly acquired data is the accumulated precipitation data every 12 hours, the cumulative step is marked by the step parameter that cycles from 0 to 12, so the data is processed into hourly precipitation data according to the step parameter. Meteorological data of all modal types are processed into data with hourly scale, and the spatial resolution is also ensured to be the same through resampling.

[0059] Finally, the historical series of meteorological data of the Qiantang River region for each mode of input is in three-dimensional form Where L represents the time length of the historical sequence, λ and φ represent the longitude and latitude ranges of the spatial area considered in the meteorological data. The historical sequence of the Qiantang River tidal height is presented in two-dimensional form. Input, L also represents the time length of the historical sequence.

[0060] Since different modal element data have different size ranges, the normalization method is used to preprocess the input factors. The processing formula is as follows:

[0061]

[0062] Among them, x′ is the normalized data, x is the original data of different input factors, and x max , x min are the maximum and minimum values ​​of the corresponding input factors, respectively.

[0063] Finally, the processed training sample data set is divided into training set, validation set and test set in the ratio of 6:2:2.

[0064] Step 2: Construct an encoder based on cross-modal feature fusion, encode data of different modalities separately and then perform data cascading to achieve effective feature fusion.

[0065] The encoder consists of a convolutional layer, a self-attention layer, and an output layer. The historical sequence of meteorological data in the Qiantang River region of each modality is processed by a convolutional layer C to reduce its dimension and extract key spatial features:

[0066] X′=ReLU(W*X+b)

[0067] Where * represents the convolution operation, W and b are the weight matrix and bias term respectively, and ReLU represents the activation function.

[0068] Then, the self-attention mechanism is applied to each modality of meteorological data feature X′ and Qiantang River tidal height history sequence T after convolutional layer processing to adjust the importance of features, thereby optimizing feature representation. Let the self-attention function be Attention(·), then:

[0069] X″=Attention(X′)

[0070] T′=Attention(T)

[0071] Finally, the processed tidal height data T′ and the meteorological data features X″ of each mode are merged along the feature dimension to obtain the final fusion feature F:

[0072] F = concat(T′,X 1 ″,X 2″,X 3 ″,X 4 ″)

[0073] Where concat represents a cascade operation, which combines the features of the two modalities into a single high-dimensional feature vector. Note that in this formula, X 1 ″,X 2 ″,X 3 ″,X 4 ″ represents the meteorological data features X output by the attention function corresponding to the four modes of wind speed data, temperature data, precipitation data and air pressure data.

[0074] Step 3: Construct a generative decoder and use direct mapping to predict tidal height.

[0075] The decoder mainly consists of an input layer and an output layer. Figure 3 As shown in the figure, the input layer consists of two key parts: the target sequence to be predicted and the known sequence before the target sequence, as shown in the following formula:

[0076]

[0077] in, is the fusion feature of the known sequence, i.e. the encoder output, L is the time length, F t +F s is the feature dimension of the fusion feature F, X placeholder is a placeholder embedding of the target sequence, which serves to provide a structural framework for the decoder, indicating the length L of the future tidal height sequence to be predicted y , so that the decoder can generate predictions based on this. These placeholders do not carry actual future information, but provide the necessary spatial dimensions for model prediction. y is the sum of the length of the known sequence and the length of the future sequence.

[0078] The output layer includes several fully connected layers, which can transform the input feature X decoder Convert to predicted output Y pred , the process can be expressed as:

[0079] Y pred =f(W·X decoder +b)

[0080] Among them, W and b represent the weight and bias parameters of the fully connected layer respectively.

[0081] Step 4: Use the preprocessed data set for training to obtain a prediction model.

[0082] During the training phase, the mean square error loss function (MSE) is used to calculate the model loss and evaluate the difference between the model prediction effect and the actual data. The calculation formula is as follows:

[0083]

[0084] Among them, y i Represents real data, Represents the predicted data, and n represents the number of samples. At the same time, Adam is used as the optimizer of the model to adjust the network parameters to minimize the loss function value, thereby optimizing the prediction performance. In addition, in order to prevent the model from overfitting and ensure the effectiveness of the training, the EarlyStopping technology is used. This technology controls the training process by monitoring the loss value on the validation set. If the loss on the validation set does not drop significantly in a certain number of consecutive training cycles, it can be considered that the model has begun to overfit or no longer learns new patterns, and the training will be stopped at this time. This method can not only save computing resources and avoid meaningless training time, but also improve the generalization ability of the model to a certain extent, and control the training process more accurately to ensure that the model achieves the best prediction effect without failure due to overfitting.

[0085] After training, the Qiantang River tidal bore height prediction model can be used for inference applications on the test set or the time series that actually needs to be predicted. In the same way as the training samples, the Qiantang River tidal bore height history sequence and the Qiantang River regional meteorological data history sequence before the target period can be constructed, and then input into the pre-trained Qiantang River tidal bore height prediction model to obtain the Qiantang River tidal bore height prediction sequence corresponding to the target period.

[0086] This example compares the performance of the Qiantang River tidal height prediction model (referred to as the present invention model) with the traditional Transformer model on the same test set. Table 1 shows the distribution of the difference between the model prediction error and the actual value:

[0087] Table 1 Comparison of indicators between the proposed model and the Transformer model

[0088]

[0089] and Figure 4 and Figure 5 The error distribution of the model of the present invention and the Transformer model is shown by histograms. By comparison, it can be seen that the error distribution of the model of the present invention is concentrated near 0, indicating that most of the predicted values ​​are very close to the true values, and the shape of the error distribution is close to the normal distribution, indicating that the prediction error has good symmetry and no obvious skewness.

[0090] In summary, the Qiantang River tidal height prediction method of the present invention uses a cross-modal fusion method to perform feature fusion between different data sources and a generative decoder for prediction. This method effectively combines tidal height data with meteorological data, utilizes the unique properties and complementary information of each data, and improves the model's understanding and prediction capabilities of tidal dynamics. The application of a generative decoder allows the model to predict future tidal height trends at one time, which not only optimizes the efficiency of the prediction process, but also significantly reduces the error accumulation that may occur in step-by-step prediction.

[0091] The above-described embodiment is only a preferred solution of the present invention, but it is not intended to limit the present invention. A person skilled in the relevant technical field may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, any technical solution obtained by equivalent replacement or equivalent transformation falls within the protection scope of the present invention.

Claims

1. A Qiantang River tidal height prediction method based on cross-modal feature fusion, characterized in that: include: S1. Obtain the historical sequence of Qiantang River tidal height and the historical sequence of Qiantang River regional meteorological data before the target period, and preprocess the data to meet the model input requirements. The historical meteorological data types include four modes: wind speed data, temperature data, precipitation data, and air pressure data; S2. Obtain a pre-trained Qiantang River tidal height prediction model, which is composed of an encoder based on cross-modal feature fusion and a generative decoder; in the encoder, each type of historical sequence of Qiantang River regional meteorological data in the model input is first processed by a convolution layer to extract key spatial features of meteorological data, and then a self-attention mechanism is applied to each key spatial feature of meteorological data processed by the convolution layer and the Qiantang River tidal height historical sequence in the model input through a self-attention layer to obtain multimodal features after adjusting the importance of features, and finally the multimodal features processed by the self-attention are spliced ​​and fused along the feature dimension to obtain fused features; in the decoder, the fused features obtained by the encoder are first spliced ​​with a placeholder embedding having a length equal to the sequence to be predicted, and the spliced ​​input features are used to obtain an output sequence through a fully connected network; S3, input the Qiantang River tidal height historical sequence after data preprocessing in S1 and the Qiantang River regional meteorological data historical sequence into the pre-trained Qiantang River tidal height prediction model to obtain the Qiantang River tidal height prediction sequence corresponding to the target period.

2. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 1 is characterized in that: The wind force and speed data adopts the wind speed vector u10 index in the east-west direction.

3. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 1 is characterized in that: The data preprocessing includes sequentially performing missing value filling, resampling and normalization on the historical sequences of each modality data.

4. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 3 is characterized in that: When resampling the historical series of meteorological data in the Qiantang River region, the spatial distribution map of the meteorological data in the Qiantang River region at each historical moment needs to be resampled to the same spatial resolution, and at the same time, the temporal resolution of the meteorological data in the Qiantang River region needs to be resampled to the same temporal resolution as the historical series of the Qiantang River tidal height.

5. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 3 is characterized in that: The normalization adopts maximum and minimum value normalization.

6. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 1, characterized in that: The Qiantang River tidal height prediction model needs to be supervised and trained with a labeled data set before being used for actual reasoning. During the training process, the goal is to minimize the loss function value. Adam is used as the optimizer of the model to adjust the network parameters, and an early stopping mechanism is introduced to control the iterative cycle of training.

7. The Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in claim 6, characterized in that: The loss function used in training the Qiantang River tidal height prediction model is the mean square error (MSE) between the Qiantang River tidal height prediction sequence and the true value label.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by a processor, the Qiantang River tidal height prediction method based on cross-modal feature fusion as claimed in any one of claims 1 to 7 can be implemented.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the Qiantang River tidal height prediction method based on cross-modal feature fusion according to any one of claims 1 to 7 is implemented.

10. A computer electronic device, characterized in that: including memory and processor; The memory is used to store computer programs; The processor is used to implement the Qiantang River tidal height prediction method based on cross-modal feature fusion according to any one of claims 1 to 7 when executing the computer program.

Images