Purse net breeding water temperature prediction method and electronic device

Abstract

This application discloses a method and electronic device for predicting water temperature in enclosure aquaculture. The method includes: obtaining first data; the first data includes multiple input parameters corresponding to the enclosure aquaculture area at multiple historical moments, the input parameters including at least water quality parameters and climate parameters, the water quality parameters including at least water temperature parameters; using a multi-step prediction model to extract global feature information and local feature information corresponding to the first data, the global feature information representing the relationship between the input parameters; the local feature information representing the relationship between the input parameters and the water temperature parameters; using the multi-step prediction model, at least based on the global feature information and the local feature information, to obtain second data, the second data including: predicted water temperature values ​​corresponding to the enclosure aquaculture area at multiple future moments.

Description

Technical Field

[0001] This application relates to the field of net enclosure aquaculture technology, and in particular to a method and electronic device for predicting water temperature in net enclosure aquaculture. Background Technology

[0002] When aquaculture is carried out in nearshore net enclosures, the openness of the aquaculture environment makes the water temperature susceptible to the temporal and spatial influence of external factors, and the correlation between each external factor and the water temperature is also different.

[0003] However, existing deep neural network models cannot fully analyze the correlation between external factors and water temperature in water temperature prediction schemes, which can lead to inaccurate water temperature predictions. Summary of the Invention

[0004] In view of the above problems, this application provides a method and electronic device for predicting water temperature in net enclosure aquaculture, aiming to improve the accuracy of the predicted water temperature. The specific solution is as follows:

[0005] The first aspect of this application provides a method for predicting water temperature in net enclosure aquaculture, the method comprising:

[0006] Obtain first data; the first data includes: multiple input parameters corresponding to the fenced aquaculture area at multiple historical moments, the input parameters including at least: water quality parameters and climate parameters, the water quality parameters including at least water temperature parameters;

[0007] Using a multi-step prediction model, global and local feature information corresponding to the first data are extracted. The global feature information represents the relationship between the input parameters; the local feature information represents the relationship between the input parameters and the water temperature parameter.

[0008] Using the multi-step prediction model, second data is obtained based at least on the global feature information and the local feature information. The second data includes: predicted water temperature values ​​corresponding to the fenced aquaculture area at multiple future time points.

[0009] In one possible implementation, a multi-step prediction model is used to extract global feature information corresponding to the first data, including:

[0010] The first data is subjected to convolution processing to obtain a data convolution result, which represents the local correlation changes of the first data in the feature dimension;

[0011] Temporal position information is added to the data convolution result to obtain the position encoding result; the temporal position information represents the historical moment.

[0012] The data convolution results and the position encoding results are processed to obtain the data embedding results;

[0013] The data embedding results are processed at least to obtain the global feature information.

[0014] In one possible implementation, a multi-step prediction model is used to extract local feature information corresponding to the first data, including at least:

[0015] The first data is processed to obtain strongly correlated variables; the strongly correlated variables represent the relationship between the input parameters and the water temperature parameters that satisfy the strongly correlated condition.

[0016] The strongly correlated variables are processed to obtain the local feature information.

[0017] In one possible implementation, a multi-step prediction model is used to extract local feature information corresponding to the first data, including at least:

[0018] The first data is processed to obtain weakly correlated variables; the weakly correlated variables represent the relationship between the input parameters and the water temperature parameters that satisfy the weak correlation condition with the water temperature parameters.

[0019] The weakly correlated variables are processed to obtain the local feature information.

[0020] In one possible implementation, the strongly correlated variables are processed to obtain the local feature information, including:

[0021] The strongly correlated variables are convolved to obtain the variable convolution result;

[0022] The results of the variable convolution are processed to obtain the local feature information.

[0023] In one possible implementation, the second data is obtained using the multi-step prediction model, based at least on the global feature information and the local feature information, including:

[0024] The global feature information and the local feature information are fused to obtain fused feature information;

[0025] The predicted water temperature value is obtained based on the fused feature information.

[0026] In one possible implementation, the method further includes:

[0027] The first data is processed to obtain autoregressive results, which at least characterize the fluctuation of the water temperature parameter over time.

[0028] Specifically, the second data is obtained using the multi-step prediction model, based at least on the global feature information and the local feature information, including:

[0029] The predicted water temperature value is obtained based on the global feature information, the local feature information, and the autoregression result.

[0030] In one possible implementation, obtaining the predicted water temperature value based on the global feature information, the local feature information, and the autoregression result includes:

[0031] The global feature information and the local feature information are fused to obtain fused feature information;

[0032] The fused feature information is processed to obtain a first predicted value;

[0033] The autoregression results are processed to obtain a second predicted value;

[0034] The predicted water temperature value is obtained based on the first predicted value and the second predicted value.

[0035] In one possible implementation, the water quality parameters include at least: water temperature values ​​at multiple water depths in the enclosed aquaculture area, dissolved oxygen values ​​at each water depth, and pH values ​​at each water depth;

[0036] The climate parameters include at least: air temperature, relative humidity, surface wind speed, air pressure, precipitation, and radiation.

[0037] A second aspect of this application provides a device for predicting water temperature in enclosed aquaculture, comprising:

[0038] A data acquisition unit is used to acquire first data; the first data includes: multiple input parameters corresponding to the fenced aquaculture area at multiple historical moments, the input parameters including at least: water quality parameters and climate parameters, the water quality parameters including at least water temperature parameters;

[0039] The data control unit is used to extract global feature information and local feature information corresponding to the first data using a multi-step prediction model. The global feature information represents the relationship between the input parameters; the local feature information represents the relationship between the input parameters and the water temperature parameter; and using the multi-step prediction model, at least based on the global feature information and the local feature information, obtains second data, which includes: predicted water temperature values ​​corresponding to the fenced aquaculture area at multiple future times.

[0040] A third aspect of this application provides a computer program product, including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the method for predicting water temperature in enclosed aquaculture as described in the first aspect or any implementation thereof.

[0041] A fourth aspect of this application provides an electronic device, including at least one processor and a memory connected to the processor, wherein:

[0042] The memory is used to store computer programs and data generated by the execution of the computer programs;

[0043] The processor is configured to execute the computer program to enable the electronic device to implement the method for predicting water temperature in net enclosure aquaculture as described in the first aspect or any implementation thereof.

[0044] The fifth aspect of this application provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to perform a method for predicting the water temperature in enclosed aquaculture as described in the first aspect or any implementation thereof.

[0045] By employing the above technical solution, the method and electronic device for predicting water temperature in enclosure aquaculture provided in this application extracts global and local feature information corresponding to multiple input parameters through a multi-step prediction model. This comprehensively considers at least the spatiotemporal correlation between climate parameters and water quality parameters in the enclosure aquaculture area, the spatial distribution pattern of water bodies, and the time dependence of historical time series data. The global feature information focuses on the complex relationships between various input parameters, while the local feature information overcomes the defect that the global feature information may ignore the influence of some input parameters on water temperature. Therefore, water temperature prediction can be performed by combining global and local feature information, which can improve the accuracy of water temperature prediction. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart illustrating a method for predicting water temperature in enclosed aquaculture, as provided in an embodiment of this application;

[0048] Figure 2 Example diagram of an enclosed aquaculture area;

[0049] Figure 3This is a flowchart illustrating the process of obtaining global feature information in this application embodiment;

[0050] Figure 4 This is a schematic diagram of the global feature extraction module in an embodiment of this application;

[0051] Figure 5 This is a flowchart illustrating the process of obtaining local feature information in an embodiment of this application;

[0052] Figure 6 This is a schematic diagram of the local feature extraction module in an embodiment of this application;

[0053] Figure 7 This is a flowchart illustrating the process of obtaining local feature information corresponding to strongly correlated variables in this application embodiment;

[0054] Figure 8 This is a schematic diagram of the structure of the first coordinated attention module in the embodiments of this application;

[0055] Figure 9 This is another flowchart illustrating the process of obtaining local feature information in this application embodiment;

[0056] Figure 10 This is another structural schematic diagram of the local feature extraction module in this application embodiment;

[0057] Figure 11 This is a flowchart illustrating the process of obtaining local feature information corresponding to weakly correlated variables in this application embodiment;

[0058] Figure 12 This is a schematic diagram of the structure of the second coordinated attention module in the embodiments of this application;

[0059] Figure 13 This is a flowchart illustrating the process of obtaining the predicted water temperature in an embodiment of this application;

[0060] Figure 14 This is a schematic diagram of the structure of the multi-step prediction model in the embodiments of this application;

[0061] Figure 15 A flowchart illustrating a method for predicting water temperature in enclosed aquaculture, as provided in an embodiment of this application;

[0062] Figure 16 This is another structural diagram of the multi-step prediction model in the embodiments of this application;

[0063] Figure 17 This is another flowchart illustrating the process of obtaining predicted water temperature values ​​in this application embodiment;

[0064] Figure 18 This is another structural diagram of the multi-step prediction model in the embodiments of this application;

[0065] Figure 19A schematic diagram of the structure of a device for predicting water temperature in enclosed aquaculture, provided in an embodiment of this application;

[0066] Figure 20 To realize the structural schematic diagram of the electronic device in the embodiments of this application;

[0067] Figure 21 An example diagram illustrating water temperature prediction using the multi-step prediction model in this application;

[0068] Figure 22 This is a schematic diagram of the CA module in the local feature extraction module of this application. Detailed Implementation

[0069] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0070] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0071] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0072] The feed intake, growth rate, quality, and safety of farmed aquatic products are all closely related to water temperature. Meanwhile, marine aquaculture is a vulnerable industry, characterized by a high frequency of disasters and significant losses. In particular, prolonged periods of high temperatures in summer can lead to a decrease in seawater salinity and dissolved oxygen, thereby inhibiting the growth of farmed products or causing widespread disease and increased mortality. The open nature of net-enclosed aquaculture environments makes water temperature susceptible to the spatiotemporal influence of external factors, including the spatiotemporal correlation between climate conditions and water quality parameters in the enclosure area, the spatial distribution patterns of water bodies, and the time dependence of historical data. These factors exhibit nonlinearity, dynamism, and complexity, and each parameter has a different correlation with water temperature. Therefore, comprehensively analyzing historical data from multiple factors and the weighting relationships between different parameters to accurately and timely grasp future water temperature trends is of paramount importance.

[0073] Deep Neural Networks (DNNs) can approximate solutions to complex multidimensional functions, uncover hidden information in large datasets, and possess strong feature extraction capabilities. Compared to shallow machine learning methods, they exhibit stronger data learning and generalization abilities, leading many researchers both domestically and internationally to apply DNN models to time series prediction of water quality factors. Long Short-Term Memory (LSTM) networks can effectively analyze historical data in time series, but their long-term memory capacity is limited. Temporal Convolutional Networks (TCNs) can achieve a larger receptive field and perceive longer historical data through stacking depth. However, neither of these network structures can analyze the correlation weights between variables. Self-attention mechanisms have the ability to capture key information and are an effective tool for extracting data features. Researchers optimized LSTM models with two attention mechanisms for sea surface temperature prediction, showing that the model with the added attention mechanism outperformed the simple LSTM model. However, self-attention mechanisms may cause the model to overemphasize certain parts of the input sequence while ignoring others, resulting in lower prediction accuracy. Therefore, this issue needs to be addressed for predicting highly complex variables.

[0074] This application primarily addresses three key issues in water temperature prediction during net enclosure aquaculture:

[0075] First, a multi-step prediction scheme for water temperature series under spatiotemporal fusion: Based on the data characteristics of seawater temperature time series, the multi-step prediction model fully considers the spatiotemporal correlation between climate conditions and water quality parameters in the enclosure aquaculture area, the spatial distribution pattern of water bodies, and the time dependence of the series data in historical periods.

[0076] Secondly, it overcomes the inherent problems of the self-attention mechanism itself, namely that factors that have little impact on water temperature prediction in most cases can have a significant impact at some special times, thereby enhancing the robustness of the multi-step prediction model.

[0077] Third, an effective multi-step prediction model for water temperature in net enclosure aquaculture: A multi-step prediction model is designed based on the environmental characteristics of net enclosure aquaculture areas to make it more consistent with the actual situation and to effectively apply the multi-step prediction model to the prediction of water temperature in net enclosure aquaculture environments.

[0078] In practical applications, accurate and meaningful prediction results can be obtained for different temperature conditions (high and low temperatures) in the enclosure aquaculture area.

[0079] refer to Figure 1This document presents a flowchart illustrating a method for predicting water temperature in enclosed aquaculture, as provided in this embodiment. This method is applicable to electronic devices capable of data processing, such as computers or servers. The electronic devices can be locally deployed or deployed in the cloud. The technical solution in this embodiment primarily aims to improve the accuracy of water temperature prediction in enclosed aquaculture areas.

[0080] Specifically, the method in this embodiment may include the following steps:

[0081] Step 101: Obtain the first data.

[0082] The first data may include multiple input parameters corresponding to the enclosed aquaculture area at multiple historical moments. The input parameters include at least water quality parameters and climate parameters, with water quality parameters including at least water temperature. The historical moments can be any time prior to the current moment, and the historical moments corresponding to the first data are consecutive. For example, the first data includes multiple input parameters corresponding to the enclosed aquaculture area for each hour of the previous week's 84 consecutive hours. The first data can be represented as X = [X1, X2, ..., X...]. T [] indicates that T is the number of historical moments, and X is... T This represents the input parameters at time T.

[0083] In specific implementation, the water temperature parameter can at least include: water temperature detection values ​​corresponding to multiple water depths in the enclosed aquaculture area. These water depths can be multiple, such as 1 meter, 5 meters, and 10 meters in the enclosed aquaculture area. In this embodiment, temperature sensors can be used to detect the water temperature values ​​corresponding to each water depth in the enclosed aquaculture area. For example... Figure 2 As shown in the figure, according to the scheme in this embodiment, in the fenced aquaculture platform (i.e., the fenced aquaculture area), with sea level as the reference, water temperature collection points A, B, and C are set at depths of 1 meter, 5 meters, and 10 meters respectively. Temperature sensors are deployed at the water temperature collection points, and the temperature sensors collect the water temperature detection values ​​at the corresponding locations.

[0084] In addition to the water temperature parameters mentioned above, water quality parameters may also include: dissolved oxygen levels and pH levels at each water depth. In this embodiment, water quality sensors can be used to detect the dissolved oxygen and pH levels at each water depth within the enclosed aquaculture area. For example, ... Figure 2 As shown in the figure, according to the scheme in this embodiment, in the enclosed aquaculture platform, water quality collection points are set at depths of 1 meter, 5 meters and 10 meters respectively, with sea level as the reference. Water quality sensors are deployed at the water quality collection points, and the water quality sensors collect the dissolved oxygen value and pH value at the corresponding locations.

[0085] Climate parameters may include at least: temperature, relative humidity, surface wind speed, air pressure, precipitation, and radiation.

[0086] Step 102: Using a multi-step prediction model, extract the global and local feature information corresponding to the first data.

[0087] Step 103: Using a multi-step prediction model, obtain the second data based at least on global and local feature information.

[0088] The second data includes predicted water temperatures for the fenced aquaculture area at multiple future times. Specifically, the second data may include predicted water temperatures for one or more water depths within the fenced aquaculture area at multiple future times. For example, in this embodiment, water temperature, dissolved oxygen, pH, water quality parameters, and climate parameters for the fenced aquaculture area at multiple historical times at multiple water depths are input into a multi-step prediction model, causing the model to output predicted water temperatures for one or more water depths within the fenced aquaculture area at multiple future times.

[0089] It should be noted that "future time" refers to a time after the current time. For example, the second set of data includes: the predicted water temperature of the enclosed aquaculture area at a depth of 1 meter for each hour over the next 12 consecutive hours.

[0090] The multi-step prediction model includes at least a global feature extraction module and a local feature extraction module. The global feature extraction module is used to extract global feature information corresponding to the first data; the global feature information represents the relationship between the input parameters.

[0091] The local feature extraction module is used to extract local feature information corresponding to the first data; the local feature information represents the relationship between the input parameters and the water temperature parameters. Among them, global feature information and local feature information are used to obtain the predicted water temperature value.

[0092] In its implementation, the multi-step prediction model includes an input module, also known as an input layer. The first data is input into this layer, which preprocesses it. Then, a global feature extraction module and a local feature extraction module extract features from the first data to obtain global and local feature information. Based on this, the multi-step prediction model predicts the water temperature at one or more depths within the enclosed aquaculture area, using at least the global and local feature information, to obtain the second data.

[0093] As can be seen from the above technical solution, in the water temperature prediction method for enclosure aquaculture provided in Embodiment 1 of this application, a multi-step prediction model is used to extract global and local feature information corresponding to multiple input parameters. This comprehensively considers at least the spatiotemporal correlation between climate parameters and water quality parameters in the enclosure aquaculture area, the spatial distribution pattern of water bodies, and the time dependence of historical time series data. Among them, global feature information focuses on the complex relationship between each input parameter, while local feature information can overcome the defect that global feature information ignores the influence of some input parameters on water temperature. Therefore, water temperature prediction can be performed by combining global and local feature information, which can improve the accuracy of water temperature prediction.

[0094] In one possible implementation, extracting global feature information in step 102 can be achieved in the following way: Figure 3 As shown:

[0095] Step 301: Perform convolution processing on the first data to obtain the data convolution result. The data convolution result represents the local correlation changes of the first data in the feature dimension.

[0096] Step 302: Add temporal position information to the data convolution results to obtain the position encoding results; the temporal position information represents historical moments;

[0097] Step 303: Process the data convolution results and positional encoding results to obtain the data embedding results;

[0098] Step 304: At least process the data embedding results to obtain global feature information.

[0099] In a specific implementation, the global feature extraction module may include at least: a data embedding module and a self-attention module; wherein, the data embedding module includes at least a convolutional submodule, a positional encoding submodule, and a random dropout submodule; such as Figure 4 As shown:

[0100] The convolution submodule is used to perform convolution processing on the first data, obtaining the data convolution result. The data convolution result represents the local correlation changes of the first data in the feature dimension. The convolution submodule can use a 1*3 convolution kernel f to perform convolution processing on the first data X. T Local correlation changes in the feature dimension can be represented by f. 1×3 (X) represents.

[0101] The positional encoding submodule is used to add temporal positional information to the data convolution results to obtain positional encoding results. The temporal positional information represents historical moments. That is, the positional encoding submodule adds temporal positional information to the data convolution results according to the corresponding historical moments, so that the data convolution results correspond to historical moments. The positional encoding submodule uses positional embedding to make the data convolution results carry temporal positional information, thereby performing secondary representation on each input parameter in the data convolution results, so that the data convolution results can better understand the relative positional information PE(pos,d) between temporal data, where pos is the temporal positional information and d is the input parameter. As shown in formulas (1) and (2), the odd-numbered positions 2d+1 and even-numbered positions 2d of the temporal positional information pos are encoded by the cosine function and the sine function, respectively, to obtain the corresponding position vector values ​​in the embedding dimension dmodel, i.e., the positional encoding results are PE(pos,2d+1) and PE(pos,2d):

[0102]

[0103] The random dropout submodule, also known as the dropout layer, processes the data convolution and positional encoding results to obtain the data embedding result. Specifically, the random dropout submodule processes the data convolution result f... 1×3 (X) and the position encoding result PE(pos,d) are used as inputs, and the data embedding result Embedding is output as shown in formula (3):

[0104] Embedding=Dropout[PE(pos,d)+f 1×3 (X)] (3)

[0105] The self-attention module is built on the self-attention mechanism. The self-attention module is used to process at least the data embedding results to obtain global feature information.

[0106] As can be seen, the global feature extraction module in the multi-step prediction model can effectively obtain the correlation weights between the input parameters in the first data by adding temporal position information to the data convolution results, so as to characterize the relationship between the input parameters, i.e. global feature information. Based on this, the accuracy of water temperature prediction can be improved.

[0107] In one possible implementation, extracting local feature information in step 102 can be achieved in the following way: Figure 5 As shown:

[0108] Step 501: Process the first data to obtain strongly correlated variables.

[0109] Among them, strongly correlated variables represent the relationship between input parameters and water temperature parameters that satisfy the strong correlation condition.

[0110] Step 502: Process the strongly correlated variables to obtain local feature information.

[0111] In a specific implementation, the local feature extraction module may include at least: a strong correlation processing module and a first coordinated attention module, such as... Figure 6 As shown:

[0112] The strong correlation processing module processes the first data to obtain strongly correlated variables. These variables characterize the correlation between the input parameters and the water temperature parameter that satisfy the strong correlation condition. The strong correlation condition can be that the absolute value of the correlation coefficient between the input parameter and the water temperature parameter is greater than or equal to a correlation threshold. In other words, in this embodiment, input parameters that satisfy the strong correlation condition are first selected according to the strong correlation condition, and then the correlation between these input parameters and the water temperature parameter is analyzed to obtain the strongly correlated variables.

[0113] In practice, the correlation coefficient in the strong correlation condition can be obtained through corresponding algorithms, such as the Pearson correlation coefficient.

[0114] The first coordination attention module is used to process strongly correlated variables to obtain local feature information. This local feature information represents the relationship between the input parameters and the water temperature parameters that satisfy the strong correlation condition.

[0115] In step 502, when processing strongly correlated variables to obtain the local feature information, this can be achieved in the following way: Figure 7 As shown:

[0116] Step 701: Perform convolution on strongly correlated variables to obtain the variable convolution result.

[0117] Step 702: Process the result of variable convolution to obtain local feature information.

[0118] In specific implementation, the first coordinating attention module can also be called the coordinating attention (CA) module. The first coordinating attention module can include a convolutional submodule and an attention submodule, such as... Figure 8As shown in the diagram. The convolutional submodule in the first coordinated attention module can be constructed based on a convolutional neural network (CNN) to perform convolution processing on strongly correlated variables to obtain the variable convolution result. The attention submodule in the first coordinated attention module can be constructed based on an attention mechanism to process the variable convolution result to obtain local feature information.

[0119] In one possible implementation, step 102, extracting local feature information, can also be achieved in the following ways: Figure 9 As shown:

[0120] In step 901: The first data is processed to obtain weakly correlated variables.

[0121] Among them, weakly correlated variables represent the relationship between input parameters and water temperature parameters that satisfy the weak correlation condition.

[0122] Step 902: Process the weakly correlated variables to obtain local feature information.

[0123] In a specific implementation, the local feature extraction module may also include: a weak correlation processing module and a second coordinated attention module, such as... Figure 10 As shown:

[0124] The weak correlation processing module processes the first data to obtain weakly correlated variables. These variables characterize the correlation between the input parameters and the water temperature parameter that satisfy the weak correlation condition. The weak correlation condition can be that the absolute value of the correlation coefficient between the input parameter and the water temperature parameter is less than a correlation threshold. In other words, in this embodiment, input parameters that satisfy the weak correlation condition are first selected according to the weak correlation condition, and then the correlation between these input parameters and the water temperature parameter is analyzed to obtain the weakly correlated variables.

[0125] In practice, the correlation coefficient in the weak correlation condition can be obtained through corresponding algorithms, such as the Pearson correlation coefficient.

[0126] The second coordination attention module is used to process weakly correlated variables to obtain local feature information. This local feature information represents the relationship between the input parameters and the water temperature parameters that satisfy the weak correlation condition.

[0127] In step 902, when processing weakly correlated variables to obtain local feature information, this can be achieved in the following way: Figure 11 As shown:

[0128] Step 1101: Perform convolution on weakly correlated variables to obtain the variable convolution result.

[0129] Step 1102: Construct an attention mechanism to process the results of variable convolution to obtain local feature information.

[0130] In the specific implementation, the second coordination attention module is used to process weakly correlated variables to obtain local feature information. Here, the local feature information represents the relationship between the input parameters and the water temperature parameters that satisfy the weak correlation condition.

[0131] The second coordinating attention module, also known as the coordinating attention (CA) module, may include convolutional submodules and attention submodules, such as... Figure 12 As shown in the diagram. The convolutional submodule in the second coordinated attention module can be built based on a convolutional neural network (CNN) to perform convolution processing on weakly correlated variables to obtain the variable convolution result. The attention submodule in the second coordinated attention module can be built based on an attention mechanism to process the variable convolution result to obtain local feature information.

[0132] Based on the above scheme, this embodiment also takes into account the relationship between the input parameters that are weakly correlated with the water temperature parameters and the water temperature parameters in the water temperature prediction, so as to avoid the situation where the water temperature prediction is inaccurate due to only considering the relationship between the input parameters that are strongly correlated with the water temperature parameters, thereby improving the accuracy of water temperature prediction.

[0133] In one possible implementation, obtaining the second data in step 103 can be achieved in the following way: Figure 13 As shown:

[0134] Step 1301: Fuse global feature information and local feature information to obtain fused feature information.

[0135] Step 1302: Obtain the predicted water temperature value based on the fusion feature information.

[0136] In a specific implementation, the multi-step prediction model may also include: a data fusion module, which connects the global feature extraction module and the local feature extraction module, such as... Figure 14 As shown:

[0137] The data fusion module is used to fuse global and local feature information to obtain fused feature information, which is then used to obtain water temperature prediction values.

[0138] Based on this, the fused feature information is processed through a fully connected module in the multi-step prediction model to obtain the predicted water temperature value.

[0139] In one possible implementation, this embodiment may further include the following processes before step 103, such as... Figure 15 As shown:

[0140] Step 104: Process the first data to obtain the autoregression results.

[0141] Among them, the autoregression results at least characterize the fluctuation of water temperature parameters over time.

[0142] Based on this, obtaining the second data in step 103 can be achieved in the following way:

[0143] Based on global feature information, local feature information, and autoregression results, the predicted water temperature value is obtained.

[0144] In practical implementation, the multi-step prediction model may also include: an autoregressive module, such as... Figure 16 As shown:

[0145] The autoregressive module, also known as the autoregressive AR module, is used to process the first data to obtain autoregressive results. The autoregressive results characterize the volatility of water temperature parameters over time, such as the volatility of water temperature, water quality, and climate parameters in the first data over time. Correspondingly, this characterizes the volatility of the predicted water temperature value over time.

[0146] The autoregression results are used to combine global and local feature information to obtain the predicted water temperature value. For example, in this embodiment, the predicted water temperature value is obtained based on global and local feature information and the autoregression results.

[0147] Based on this, this embodiment performs autoregression on the input parameters, thereby incorporating the temporal volatility of the input parameters into the water temperature prediction, in order to improve the accuracy of the water temperature prediction.

[0148] Specifically, in step 103, when obtaining the predicted water temperature value based on global feature information, local feature information, and autoregression results, this can be achieved in the following ways: Figure 17 As shown:

[0149] Step 1701: Fuse global feature information and local feature information to obtain fused feature information.

[0150] Step 1702: Process the fused feature information to obtain the first predicted value;

[0151] Step 1703: Process the autoregression results to obtain the second predicted value;

[0152] Step 1704: Obtain the water temperature prediction value based on the first and second prediction values.

[0153] In a specific implementation, the multi-step prediction model in this embodiment may further include a data fusion module, a first fully connected module, a second fully connected module, and a prediction fusion module, such as... Figure 18 As shown:

[0154] The data fusion module is used to fuse global and local feature information to obtain fused feature information.

[0155] The first fully connected module processes the fused feature information to obtain the first predicted value; the fused feature information is the feature information obtained by the data fusion module based on global and local feature information. The first fully connected module can also be called the fully connected layer FC1 (Fully Connected Layer).

[0156] The second fully connected module is used to process the autoregression results to obtain a second predicted value. The second fully connected module can also be called the fully connected layer FC2.

[0157] The prediction fusion module is used to obtain the water temperature prediction value based on the first prediction value and the second prediction value.

[0158] The prediction fusion module processes the corresponding first and second predicted values ​​for each future time moment to obtain the water temperature prediction value for that time moment. Specifically, for each future time moment, the prediction fusion module either directly adds the corresponding first and second predicted values ​​or performs a weighted summation according to appropriate weights to obtain the water temperature prediction value for that time moment.

[0159] The above describes a method for predicting water temperature in enclosed aquaculture provided by the embodiments of this application. The following will describe the apparatus for performing the above-described method for predicting water temperature in enclosed aquaculture.

[0160] Please see Figure 19 , Figure 19 This is a schematic diagram of a device for predicting water temperature in enclosed aquaculture, provided in an embodiment of this application. This device can be configured on an electronic device capable of data processing, such as a computer or server. The electronic device can be a locally deployed device or a cloud-deployed device. The technical solution in this embodiment is mainly used to improve the accuracy of water temperature prediction in enclosed aquaculture areas.

[0161] Specifically, the apparatus in this embodiment may include the following units:

[0162] The data acquisition unit 1901 is used to acquire first data; the first data includes: multiple input parameters corresponding to the fenced aquaculture area at multiple historical moments, the input parameters include at least: water quality parameters and climate parameters, the water quality parameters include at least water temperature parameters;

[0163] The data control unit 1902 is used to extract global feature information and local feature information corresponding to the first data using a multi-step prediction model. The global feature information represents the relationship between the input parameters; the local feature information represents the relationship between the input parameters and the water temperature parameter. Using the multi-step prediction model, at least based on the global feature information and the local feature information, the second data includes: predicted water temperature values ​​corresponding to the fenced aquaculture area at multiple future times.

[0164] As can be seen from the above technical solution, in the water temperature prediction device for enclosure aquaculture provided in this application embodiment, global and local feature information corresponding to multiple input parameters are extracted through a multi-step prediction model. Thus, at least the spatiotemporal correlation between climate parameters and water quality parameters in the enclosure aquaculture area, the spatial distribution pattern of water bodies, and the time dependence of historical time series data are comprehensively considered. Two key modules are established in the multi-step prediction model. The global feature information focuses on the complex relationship between each input parameter, while the local feature information can overcome the defect that the global feature information ignores the influence of some input parameters on water temperature. Therefore, water temperature prediction can be performed by combining global and local feature information, which can improve the accuracy of water temperature prediction.

[0165] This application embodiment also provides an electronic device, which may include a memory and a processor. The memory is used to store computer programs and data generated by the execution of the computer programs, and the processor is used to execute the computer programs to implement the above-described method for predicting water temperature in enclosed aquaculture. (Reference) Figure 20 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this application. The electronic device in the embodiments of this application may include, but is not limited to, fixed terminals such as mobile phones, laptops, PDAs (personal digital assistants), PADs (tablet computers), desktop computers, etc. Figure 20 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0166] like Figure 20As shown, the electronic device may include a processing unit (e.g., a central processing unit, a graphics processor, etc.) 2001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 2002 or a program loaded from a storage device 2008 into a random access memory (RAM) 2003. When the electronic device is powered on, the RAM 2003 also stores various programs and data required for the operation of the electronic device. The processing unit 2001, ROM 2002, and RAM 2003 are interconnected via a bus 2004. An input / output (I / O) interface 2005 is also connected to the bus 2004.

[0167] Typically, the following devices can be connected to the I / O interface 2005: input devices 2006 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 2007 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 2008 including, for example, memory cards, hard drives, etc.; and communication devices 2009. The communication device 2009 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although... Figure 20 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have instead.

[0168] This application also provides a computer program product including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the methods for predicting water temperature in enclosed aquaculture provided in this application.

[0169] This application also provides a computer-readable storage medium carrying one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the methods for predicting water temperature in enclosed aquaculture provided in this application.

[0170] Taking electronic devices as cloud servers as an example, the following is an illustration of a water temperature prediction scenario in nearshore aquaculture enclosures:

[0171] This application proposes a multi-step prediction scheme for water temperature in net-enclosed aquaculture by fusing global and local features over time. The main steps are as follows:

[0172] (1) Deploy sensors at different depths in nearshore net cage aquaculture areas to collect data, such as water quality parameters and climate parameters;

[0173] (2) Collect meteorological data from meteorological stations in the enclosure aquaculture area and perform data cleaning: including data transformation, data standardization, handling outliers, handling missing values ​​and data integration;

[0174] (3) Set up a global feature extraction module and a local feature extraction module to analyze the feature information of sequence data (i.e., the input parameters corresponding to multiple historical moments mentioned above), such as global feature information and local feature information;

[0175] (4) Introduce an autoregressive module to obtain the volatility of the input sequence data and the changes in the output predicted data;

[0176] (5) Perform fitting verification and performance evaluation on the test set.

[0177] The following combination Figure 21 The example diagram shown illustrates the multi-step prediction model for water temperature prediction, and the steps described above are explained in detail:

[0178] Step (1): The data collection site is located 10 kilometers from the coast. The experimental aquaculture system is a circular structure with a circumference of 160 meters and a water volume of 200,000 cubic meters. 3 To fully explore the temperature variations of seawater at different depths, three parameter collection points were set up at depths of 1m, 5m, and 10m underwater, corresponding to... Figure 2 Points A, B, and C are located within the enclosure. To capture water temperature differences in different areas of the aquaculture zone and reduce sensor errors, two sets of sensor equipment are deployed, situated symmetrically within the enclosure. All collected sensor data is transmitted via a wireless sensor network to the application layer for data acquisition, intelligent information processing, and other operations.

[0179] The multi-step prediction model for the enclosure aquaculture area requires 15 input parameters: water temperature, dissolved oxygen, and pH values ​​(i.e., acidity / alkalinity) at depths of 1m, 5m, and 10m, as well as air temperature, relative humidity, surface wind speed, air pressure, precipitation, and direct radiation. The water temperature, dissolved oxygen, and pH values ​​(i.e., acidity / alkalinity) can be obtained through sensors at the parameter acquisition points.

[0180] Step (2): Utilize a meteorological monitoring system to perceive local fishery climate changes, including meteorological factors such as air temperature, relative humidity, surface wind speed, air pressure, precipitation, and direct radiation. Since seawater has a high specific heat capacity and its temperature changes slowly, a time interval of 30 minutes was set. Data was collected for two time periods, for example, 15 input parameters at each collection time point from July 22, 2023 to August 31, 2023, and from September 15, 2023 to November 1, 2023. The input sequence data for the multi-step prediction model consists of multiple sets of exogenous variables corresponding to time window T (each set of exogenous variables is a set of input parameters), including the target variable (i.e., the detected water temperature value). The output sequence data consists of predicted water temperature values ​​corresponding to multiple future times.

[0181] Step (3): In the multi-step prediction model, a global feature extraction module and a local feature extraction module are set up to analyze the feature information of the input sequence data, such as global feature information and local feature information. The following is an explanation of the global feature extraction module and the local extraction module:

[0182] 1) The global feature extraction module is... Figure 21 The data embedding module, the cat (CrossAttention in Vision Transformer) module, and the self-attention module are all included in the model.

[0183] Due to the complex water flow, climate variations, and physical environmental differences in netted aquaculture areas, multi-step prediction models need to consider historical data of global variables simultaneously. The data embedding module, also known as the embedding layer, associates the input parameters at each time step (i.e., each historical moment) with other relevant features, visually demonstrating the relationships between data from different time steps. The attention module in the global feature extraction module effectively learns the historical data information corresponding to all input parameters through a self-attention network, capturing the correlation weights between input parameters.

[0184] like Figure 21 The global feature extraction module and the data embedding module include the Convolution submodule, the Positional Encoding submodule, and the Dropout submodule.

[0185] The convolution submodule uses a 1×3 convolution kernel f to process the input data X=[X1,X2,...,X... T Perform convolution, where X T ∈R H Data X represents the measured data at time T. T The local correlation change in the feature dimension is f 1×3 (X). Convolutional submodules enhance the model's generalization ability and help the embedding layer better capture local features of the input data.

[0186] The Positional Encoding submodule uses positional embedding to carry positional information in the data, performing a secondary representation on each word in the input sequence, so that the embedding layer can better understand the relative positional information between temporal data. As shown in formulas (1) and (2), the odd-numbered position pos (2d+1) and even-numbered position pos (2d) are encoded by the cosine and sine functions respectively, and the corresponding position vectors in the embedding dimension dmodel are PE(pos,2d+1) and PE(pos,2d).

[0187] Then, the weights of the convolutional submodules are output as f. 1×3 (X) is added to the output PE(pos,d) of the position encoding submodule and passed through the random dropout submodule, i.e. the Dropout layer, which improves the generalization ability of the embedding layer, as shown in formula (3).

[0188] 2) The local feature extraction module includes a strong correlation processing module, a first CA module, a weak correlation processing module, and a second CA module:

[0189] The CA module includes a convolutional submodule and an attention submodule. The convolutional neural network in the convolutional submodule has many advantages. In time series prediction tasks, it can automatically learn and extract the spatial relationships between different feature values ​​of data, capture and describe the details of local time series features, and perceive multivariate correlations, making it suitable for extracting local features from time series.

[0190] Subsequently, the data fusion module uses a suitable fusion algorithm to fuse global and local feature information to obtain fused feature information. This enables better analysis of the multi-level and multi-scale feature information of the water temperature sequence of the enclosure system, while taking into account both the global structure and local details, and better adapting to different data distributions and changes in the sequence.

[0191] like Figure 22 This diagram illustrates the CA module within the local feature extraction module, where K is the kernel size, and 1D K-size conv is a one-dimensional kernel-size convolution. The CA module includes a convolutional submodule (CNN) and an attention submodule (SelfAttn). The SelfAttn submodule comprises the following layers: convolutional layers (conv), pooling layers, linear layers, and four encoder layers. The attention mechanism within the attention submodule complements the CNN's difficulty in capturing long-term temporal dependencies by weighting the entire sequence at different time steps to capture these dependencies.

[0192] The data output from the CNN layer is X' = [X1', X2', ..., X...]. T '](X T '∈R dm As shown in formula (4), after passing through the linear layer [W] q W k W v ](W q W k W v ∈R dm×dk), calculate the queries, keys, and values ​​of the attention module at each time step, and form a matrix Q for each set of queries, keys, and values. G ,K G V G (Q G ,K G V G ∈R dm×dk ), as in formula (5). Q G With (K) G ) T Multiply by V to obtain the correlation of time series data at different time intervals, then multiply by V. G Multiplying yields the attention weight Wa G As shown in formula (6), where Wa G ∈R dm×dk It is used to represent the correlation characteristics between time series.

[0193] X'=W(ReLU(Conv(X))) (4)

[0194] [Q G ,K G V G ] = X'[W q W k W v (5)

[0195]

[0196] The softmax function is used to measure Q. G ,K G V G Regularization yields the final attention weight Wa. G When the vector dimension used to calculate similarity is long, the softmax value will approach 1 or 0 extremely closely. In this case, This can effectively avoid gradient vanishing or gradient exploding problems, which is beneficial for training network models.

[0197] Step (4): Introduce an autoregressive module to obtain the volatility of the sequence input (i.e., the input parameters) and the pattern changes of the output.

[0198] Among them, the global feature extraction module and the local feature extraction module enable the multi-step prediction model to have good nonlinear representation capabilities and can parse more comprehensive information from the training data. However, for actual observed time series, the data has the characteristics of noise and irregular changes. It is difficult to obtain the fluctuation of the input and the pattern change of the output by relying solely on the neural network, which makes the model insensitive to the fluctuation of the dataset. Therefore, the AR module is introduced. The AR module introduces information from historical moments, adds a scale modeling method, improves the ability of the multi-step prediction model to express noisy and complex fence time series data, enables the model to better integrate historical information, obtain more comprehensive background information, adapt to changes in different time steps, and improve the generalization ability of the model. The AR model is expressed as shown in formula (7):

[0199]

[0200] in, and b ar ∈R represents the parameters to be learned.

[0201] Based on this, FC1 is set for global and local feature information, and FC2 is set for the autoregressive module. FC1 obtains the first predicted value, and FC2 obtains the second predicted value. Finally, prediction fusion is performed to obtain multi-step prediction results, namely, water temperature prediction values ​​at multiple future time points:

[0202] Taking FC1 as an example, the batch size of the input data for FC1 is M1*N1, such as 84*15, which represents the N1 input parameters corresponding to M1 historical time points. The batch size of the output data for FC1 and FC2 is M2*N2, such as 12*1, which represents the N2 predicted water temperature values ​​corresponding to M2 future time points. The processing stages of FC1 include FC1-1 and FC1-2. In the FC1-1 stage, M1 is reduced to M2, and in the FC1-2 stage, N1 is reduced to N2. The implementation principle of FC2 is the same as that of FC1.

[0203] Step (5): Perform test set fitting verification and performance evaluation.

[0204] To evaluate the effectiveness of the multivariate time series multi-step prediction model, both the overall prediction accuracy and local performance of the multi-step prediction model are assessed. Based on the hyperparameters of the multi-step prediction model shown in Table 1, the root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) can be used as evaluation indicators to assess the overall performance, while the empirical correlation coefficient (CORR) and root relative squared error (RRSE) can be used as evaluation indicators to measure the local goodness of fit. Specifically, as shown in formulas (8)-(12):

[0205]

[0206]

[0207] in, For the true value, Let W be the predicted value, and W be the length of the multi-step prediction window, i.e., the number of future time steps. Based on this, the smaller the values ​​of RMSE, MAE, and RRSE, the higher the prediction accuracy; the larger the CORR value, the better the local fitting performance of the multi-step prediction model.

[0208] Table 1 Hyperparameter Settings for Multistep Prediction Model

[0209] Hyperparameters set up Predicting time steps 12 Global module window size 84 Local module window size 84 Data batch size 128 Optimizer Adam Learning rate 0.001 Gradient clipping size 5 Dropout ratio 0.3 Global CNN convolutional kernel count 32 Number of local CNN convolution kernels 64 Encoder Layer 6 Global module CNN convolution kernel size 1×3 Local module CNN convolution kernel size 1×1 Weight matrix dimension kernel 64 Scaling factor Inner 2048

[0210] In summary, this application analyzes the characteristics of seawater temperature in net-enclosed aquaculture, studies the information extraction capabilities of models with different structures for historical water temperature data, and proposes a multivariate time series prediction framework based on grouped dual-attention convolution, such as... Figure 21 As shown in the diagram, the embedding layer is used for global feature extraction of multivariate time series data; the convolutional neural network combined with a self-attention mechanism serves as a local feature extraction module; and the autoregressive module is used to model the trend and autocorrelation of time series data, addressing the problem of prediction lag. By comparing with other benchmark models, the following conclusions are drawn:

[0211] The multi-step prediction model achieved good results in multi-step prediction of seawater temperature in net cage aquaculture. Compared with the best results of other benchmark models, the RMSE was reduced by 10.43%, 29.3%, and 31.55% in the predictions at 1h, 3h, and 6h, respectively. In particular, the accuracy of the model was significantly improved in multi-step prediction.

[0212] The global and local information extraction mode of seawater temperature time series data in net enclosure aquaculture enables the model to have the ability to analyze from coarse-grained to fine-grained, and can fully explore the trend and fluctuation information of the data at the same time.

[0213] Complex spatiotemporal patterns exist in seawater temperature data from net cage aquaculture. By utilizing the characteristics of different neural networks, targeted processing and effective integration of multivariate sequences can successfully compensate for the shortcomings of individual modules, which is of great significance.

[0214] Furthermore, it should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Additionally, in the accompanying drawings of the device embodiments provided in this application, the connection relationships between modules indicate that they have communication connections, which can specifically be implemented as one or more communication buses or signal lines.

[0215] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0216] In the above embodiments, the implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, in the form of a computer program product.

[0217] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for predicting water temperature in net enclosure aquaculture, characterized in that, The method includes: Obtain first data; the first data includes: multiple input parameters corresponding to the fenced aquaculture area at multiple historical moments, the input parameters including at least: water quality parameters and climate parameters, the water quality parameters including at least water temperature parameters; Using a multi-step prediction model, global and local feature information corresponding to the first data are extracted. The global feature information represents the relationship between the input parameters; the local feature information represents the relationship between the input parameters and the water temperature parameter. The first data is processed to obtain autoregressive results, which at least characterize the fluctuation of the water temperature parameter over time. Using the multi-step prediction model, second data is obtained based at least on the global feature information and the local feature information. The second data includes: predicted water temperature values ​​corresponding to the fenced aquaculture area at multiple future time points. Specifically, a multi-step prediction model is used to extract global and local feature information corresponding to the first data, including: The first data is subjected to convolution processing to obtain a data convolution result, which represents the local correlation changes of the first data in the feature dimension; Temporal position information is added to the data convolution result to obtain the position encoding result; the temporal position information represents the historical moment. The data convolution results and the position encoding results are processed to obtain the data embedding results; The data embedding results are processed at least to obtain the global feature information; The first data is processed to obtain strongly correlated variables; the strongly correlated variables represent the relationship between the input parameters and the water temperature parameters that satisfy the strongly correlated condition. The strongly correlated variables are processed to obtain the local feature information; The first data is processed to obtain weakly correlated variables; the weakly correlated variables represent the relationship between the input parameters and the water temperature parameters that satisfy the weak correlation condition with the water temperature parameters. The weakly correlated variables are processed to obtain the local feature information.

2. The method according to claim 1, characterized in that, The strongly correlated variables are processed to obtain the local feature information, including: The strongly correlated variables are convolved to obtain the variable convolution result; The results of the variable convolution are processed to obtain the local feature information.

3. The method according to claim 1, characterized in that, Using the multi-step prediction model, second data is obtained based at least on the global feature information and the local feature information, including: The global feature information and the local feature information are fused to obtain fused feature information; The predicted water temperature value is obtained based on the fused feature information.

4. The method according to claim 1, characterized in that, Using the multi-step prediction model, second data is obtained based at least on the global feature information and the local feature information, including: The predicted water temperature value is obtained based on the global feature information, the local feature information, and the autoregression result.

5. The method according to claim 4, characterized in that, Based on the global feature information, the local feature information, and the autoregression result, the predicted water temperature value is obtained, including: The global feature information and the local feature information are fused to obtain fused feature information; The fused feature information is processed to obtain a first predicted value; The autoregression results are processed to obtain a second predicted value; The predicted water temperature value is obtained based on the first predicted value and the second predicted value.

6. The method according to claim 1, characterized in that, The water quality parameters include at least: water temperature values ​​at multiple water depths in the enclosed aquaculture area, dissolved oxygen values ​​at each water depth, and pH values ​​at each water depth; The climate parameters include at least: air temperature, relative humidity, surface wind speed, air pressure, precipitation, and radiation.

7. An electronic device, characterized in that, include: A memory used to store computer programs and the data generated by the execution of said computer programs; A processor for executing the computer program to achieve: obtaining first data; The first data includes: multiple input parameters corresponding to the fenced aquaculture area at multiple historical moments, the input parameters including at least: water quality parameters and climate parameters, the water quality parameters including at least water temperature parameters; using a multi-step prediction model, extracting global feature information and local feature information corresponding to the first data, the global feature information representing the relationship between the input parameters; processing the first data to obtain an autoregressive result, the autoregressive result representing at least the temporal volatility of the water temperature parameter; the local feature information representing the relationship between the input parameters and the water temperature parameter; using the multi-step prediction model, at least based on the global feature information and the local feature information, obtaining second data, the second data including: predicted water temperature values ​​corresponding to the fenced aquaculture area at multiple future moments; Specifically, a multi-step prediction model is used to extract global and local feature information corresponding to the first data, including: The first data is convolved to obtain a data convolution result, which represents the local correlation changes of the first data in the feature dimension. Temporal position information is added to the data convolution result to obtain a position encoding result, which represents the historical time. The data convolution result and the position encoding result are processed to obtain a data embedding result. At least the data embedding result is processed to obtain the global feature information. The first data is processed to obtain strongly correlated variables, which represent the relationship between the input parameter and the water temperature parameter that satisfy a strong correlation condition. The strongly correlated variables are processed to obtain the local feature information. The first data is processed to obtain weakly correlated variables, which represent the relationship between the input parameter and the water temperature parameter that satisfy a weak correlation condition. The weakly correlated variables are processed to obtain the local feature information.

Images