Fishery resource dynamic trend prediction system and method based on time sequence analysis

Abstract

The invention discloses a fishery resource dynamic trend prediction system and method based on time sequence analysis. The method comprises the following steps: S1, constructing a fishery resource time sequence data set; s2, forming a standardized fishery resource time sequence data set; s3, constructing a self-supervised learning model comprising an automatic encoder module and a contrast learning module by using the fishery resource time sequence data set; s4, screening high-correlation features to construct a feature space; s5, constructing a time sequence dynamic prediction model taking a self-attention mechanism as a core based on the feature space; and S6, generating prediction results of the spatial position of resource distribution, the resource abundance change trend and the time range of the peak period and the valley period. According to the method, accurate prediction of the dynamic trend of the fishery resources is realized, and powerful technical support is provided for sustainable management of the fishery resources.

Description

Technical Field

[0001] The present invention relates to the technical field of fishery resources, and in particular to a system and method for predicting the dynamic trend of fishery resources based on time series analysis. Background Art

[0002] With the rapid development of big data technology and artificial intelligence technology, data-driven methods have been gradually introduced into the field of fishery resource management for resource monitoring and prediction. Accurate prediction of the dynamic trend of fishery resources is an important basis for achieving sustainable fishery management and optimal resource allocation. By predicting the distribution of fishery resources, abundance trends, and peak and trough time ranges, it can provide a scientific basis for policy makers, fishery practitioners, and environmental protection organizations. However, the current fishery resource prediction methods still have many technical bottlenecks in practical applications.

[0003] At present, fishery resource prediction mainly relies on traditional statistical models and empirical analysis methods. Common techniques include prediction methods based on regression analysis, time series modeling methods based on ARIMA models, and experts' empirical inference based on historical data. Traditional methods can reflect the overall change trend of resources to a certain extent, but they have obvious limitations. First, traditional methods are highly sensitive to data quality and have difficulty dealing with the noise, incompleteness and nonlinear characteristics that are prevalent in fishery data. Second, traditional statistical models find it difficult to capture the comprehensive impact of complex environmental variables on resource changes, and the prediction effect is significantly reduced in scenarios where seasonal fluctuations and emergencies occur frequently.

[0004] In recent years, with the rise of artificial intelligence technology, some machine learning and deep learning methods have gradually been applied to the field of fishery resource prediction, such as the use of neural networks for nonlinear modeling and pattern recognition. However, machine learning and deep learning methods also have significant shortcomings. On the one hand, most of the existing deep learning models rely on a large amount of labeled data for training, while in the fishery field, labeled data is difficult and costly to obtain, resulting in insufficient model training and unstable prediction results. On the other hand, existing models often lack effective modeling capabilities for time dependence and spatial distribution characteristics, and cannot achieve a comprehensive analysis of the complex dynamics of fishery resources. In addition, although some methods integrate multi-source data, they cannot fully utilize the potential correlation of multi-source data due to the lack of a unified feature extraction mechanism, which makes it difficult to improve the model prediction accuracy.

[0005] In summary, the existing technologies have significant defects in data quality adaptability, multidimensional data association modeling, prediction accuracy and robustness, and are difficult to meet the actual needs of accurate prediction of dynamic trends of fishery resources. These technical deficiencies have seriously restricted the efficiency and scientificity of modern fishery management and optimal resource allocation. There is an urgent need for a new method that can integrate multi-source data, adapt to dynamic changes and have high prediction accuracy to solve the above problems. Summary of the invention

[0006] One purpose of the present invention is to propose a system and method for predicting the dynamic trend of fishery resources based on time series analysis. The present invention realizes accurate prediction of the dynamic trend of fishery resources and provides strong technical support for the sustainable management of fishery resources.

[0007] A method for predicting the dynamic trend of fishery resources based on time series analysis according to an embodiment of the present invention comprises the following steps: S1. Collect multi-source fishery resources and construct a fishery resource time series dataset; S2, standardizing the fishery resources time series data set to form a standardized fishery resources time series data set; S3. Using the fishery resource time series data set, a self-supervised learning model including an autoencoder module and a contrastive learning module is constructed. The autoencoder is used to learn the characteristic distribution of the time series data, and the contrastive learning module improves the adaptability of the self-supervised learning model to different fishery resource time series data sets. S4. Extract multi-layer features including short-term trend features, long-term trend features, periodic features and environmental response features from standardized fishery resource time series data based on the pre-trained self-supervised learning model, perform correlation analysis on the multi-layer features, and select highly correlated features to construct a feature space; S5. Based on the feature space, a time series dynamic prediction model with self-attention mechanism as the core is constructed, and the feature representation extracted by the self-supervised learning model is used to fine-tune the time series dynamic prediction model to optimize the prediction ability; S6. Use the time series dynamic prediction model to predict the future trends of the current standardized fishery resource time series data set, and generate prediction results for the spatial location of resource distribution, resource abundance change trend, and peak and trough time ranges.

[0008] Optionally, the S1 specifically includes the following contents: S11. Collect time series data of fishery resources from multiple sources, including: Historical fishing data ,Historical fishing data include fishing time, catch volume and fishing location in fishing records; Environmental monitoring data ,Environmental monitoring data include seawater temperature, salinity and dissolved oxygen content at the monitoring time point; Data on reproductive cycles of target species , the target species reproduction cycle data include the reproduction time cycle points and the reproduction size of the target species; Ship track data: ; in, represents the geographical coordinates of the ship’s trajectory, t is the collection time point, and p represents the number of time points in the ship’s trajectory data; S12. Time alignment of multi-source data: using a unified time reference frame T to perform time alignment on the collected multi-source fishery resource time series data to generate a fishery resource time series data set : ; in, represents the time point of the time series, and q represents the total number of time points of the fishery resources time series data.

[0009] Optionally, S2 specifically includes the following contents: S21. Using wavelet transform denoising algorithm to analyze the fishery resources time series data set De-noising and decomposition of fishery resource time series dataset For different frequency components, high-frequency noise is suppressed by threshold method, and the denoised fishery resource time series dataset is reconstructed; S22, using the double mean method to detect outliers on the denoised fishery resource time series data set, setting the time window W to calculate each time point The mean value in the window and standard deviation , marking data points that meet the following conditions as outliers: ; in, represents a single feature value of time series data, is the threshold coefficient, and the abnormal points are marked as NULL, generating a fishery resource time series dataset containing abnormal marks; S23, performing linear interpolation completion on the fishery resource time series data set with the outlier marked as NULL, estimating the data value at the outlier position by using the weighted average of the values ​​before and after the time series data point, and generating a completed fishery resource time series data set; S24. Normalize the completed fishery resource time series dataset, map the values ​​of different features to a unified interval [0,1], and finally form a standardized fishery resource time series dataset. .

[0010] Optionally, S3 specifically includes the following contents: S31. Using the standardized fishery resource time series dataset Constructing a self-supervised learning model including an autoencoder module and a contrastive learning module, wherein the autoencoder module includes an encoder network and a decoder network; S32, encoder network transforms the standardized fishery resources time series dataset through nonlinear mapping function The time point feature vector in Mapping to latent feature space representation ; S33, the decoder network represents the potential features through a nonlinear mapping function Reconstructed as an estimate of the original eigenvector ; S34, by minimizing the reconstruction error loss function Train the autoencoder module: ; Where n is the number of time points, express norm; Reconstruction error loss function Reflects the original feature vector and reconstructed eigenvector The difference between them is minimized by minimizing the reconstruction error loss function Optimize the parameters of the encoder and decoder networks so that the self-supervised learning model can capture the characteristic distribution of fishery resource time series data; S35. Standardized fishery resource time series dataset Perform data enhancement, build a comparative learning module, and optimize the adaptability of the self-supervised learning model to different fishery resource time series data sets; S36. For each time point, the original feature vector Apply random data augmentation operations to generate a set of positive samples And negative sample set ; S37, encoder network Applied to the time point feature vector, positive sample set and negative sample set to obtain the corresponding potential feature representation and : ; ; ; S38. Define contrast loss function Optimize the encoder network so that the potential feature representation of positive samples is closer to the original samples, and the potential feature representation of negative samples is farther away from the original samples: ; in, is the cosine similarity between latent feature representations, is the temperature parameter, K is the number of negative samples for each sample; S39. By jointly optimizing the loss function Train the self-supervised learning model and improve its adaptability to different fishery resource time series data sets: ; in, and is the weight coefficient used to balance the reconstruction loss and contrast loss , by minimizing the joint optimization loss function Optimize encoder network parameters at the same time and decoder network parameters , enabling the self-supervised learning model to learn the characteristic distribution of fishery resource time series data.

[0011] Optionally, the S4 specifically includes the following contents: S41. Using the pre-trained self-supervised learning model to analyze the standardized fishery resources time series dataset Extract multi-layer features and use the encoder network to generate feature vectors at each time point Perform feature extraction and generate potential feature vectors , and decompose it into short-term trend characteristics, long-term trend characteristics, cycle characteristics and environmental response characteristics: ; in, Indicates short-term trend characteristics, reflecting the short-term fluctuation trend of catch and environmental indicators. It indicates the long-term trend characteristics and reflects the changes of fishery resources on a long-term scale. Indicates periodic characteristics, reveals the reproductive cycle and seasonal fluctuations of target species, It represents the environmental response characteristics and is associated with changes in the external environment; S42. Use the correlation analysis method to calculate the correlation coefficient R between different features and establish a feature correlation matrix : }; in, is the feature vector and the eigenvector The covariance of and is the standard deviation of the corresponding feature, is the correlation coefficient between two features; S43. According to the correlation matrix The analysis results screen for feature sets that are highly correlated with future catch or fishery resource abundance , and finally construct the feature space.

[0012] Optionally, the S5 specifically includes the following contents: S51. Features extracted using self-supervised learning models As input features, a time series dynamic prediction model with self-attention mechanism as the core is constructed The time series dynamic prediction model models the temporal dependency of input features through the self-attention mechanism. For the time series feature X, the attention weight of each time point is calculated. ; The features at time point i are calculated by weighted summation to obtain the self-attention representation : ; in, is a value vector, is the learnable parameter matrix; S52. Feature representation after modeling using self-attention mechanism Enter the fully connected network , generating forecast values ​​for future time points : ; in, are the parameters of the fully connected network; S53. Features extracted using self-supervised learning models and the known target variable Fine-tune the time series dynamic prediction model by minimizing the prediction error loss function Optimize model parameters: ; Where N is the number of samples, Dynamic forecasting model for time series The predicted value of is the actual value; S54. In the fine-tuning process, the prediction loss of the time series dynamic prediction model is combined with the feature extraction loss of the self-supervised learning model to define the joint optimization objective function: ; in, is the weight coefficient, which is used to balance the optimization objectives of the prediction model and the self-supervised learning model.

[0013] Optionally, the S6 specifically includes the following contents: S61. Dynamic prediction model of time series Current time point feature representation applied to standardized fishery resource time series datasets , predict the resource status of the next h time steps, and get the predicted value of the next h time steps ; S62. Using predicted values and geographical distribution characteristics of historical catch data Calculate the resource distribution position at time point t+h ; S63. Based on the predicted value Calculate the overall trend of future resource abundance changes: ; in, represents the future rate of change of resource abundance, is the actual value of historical resource abundance; S64, based on the predicted value Identify the extreme points of resource changes and determine the time ranges of peak and trough periods: ; ; in, Indicates the predicted peak time point of resource abundance, Indicates the predicted time point of low resource abundance; S65, the predicted value , the spatial location of resource distribution , resource abundance change rate and peak time points and low point Output as a forecast report through visualization tools: Time series graph, intuitively showing the changing trend of resource abundance; Geographic information heat map, marking the location of future resource distribution; Annotations of peak and trough time points provide suggestions for optimizing fishing plans.

[0014] A fishery resource dynamic trend prediction system based on time series analysis, including the following modules: The data collection module is used to collect multi-source fishery resource time series data, including historical fishing data, environmental monitoring data, target species reproduction cycle data and ship trajectory data, to form a fishery resource time series data set and provide real-time update capabilities to support dynamic prediction needs; The data preprocessing module is used to perform denoising, outlier detection, data completion and normalization on the collected fishery resource time series data to form a standardized fishery resource time series data set; The self-supervised learning model construction module learns the data feature distribution based on the standardized fishery resource time series data by constructing an autoencoder and a contrastive learning module. The autoencoder is used to extract the potential feature space representation. The contrastive learning module improves the model's adaptability to different data sets and realizes the pre-training of the self-supervised learning model. The multi-layer feature extraction and analysis module extracts short-term trend features, long-term trend features, periodic features, and environmental response features based on a self-supervised learning model, establishes a feature association matrix through correlation analysis, selects feature sets with high correlation with the target variable, and constructs a feature space for dynamic prediction; The dynamic prediction model construction and optimization module uses the constructed feature space to input the time series dynamic prediction model. The time series dynamic prediction model is based on the self-attention mechanism and predicts the future resource status by modeling time dependencies. At the same time, the time series dynamic prediction model is fine-tuned and trained by combining the feature representation extracted by the self-supervised learning model. The future trend prediction module predicts the future trend of the standardized fishery resource time series data set based on the time series dynamic prediction model, including the spatial location of resource distribution, resource abundance change trend, and peak and trough time range, and generates prediction results; The visualization and result output module outputs the forecast results in an intuitive form, including time series curve graphs, geographic information heat maps, and peak and trough time point annotations, forming a comprehensive forecast report to provide decision support for fishery resource management and fishing plan optimization.

[0015] The beneficial effects of the present invention are: (1) The self-supervised learning model proposed in the present invention can realize feature learning of complex fishery resource time series data without relying on a large amount of labeled data by combining an autoencoder and a contrastive learning module. The autoencoder module captures the characteristic distribution of time series data and optimizes the model structure by minimizing the reconstruction error. The contrastive learning module generates positive and negative sample pairs to improve the adaptability and generalization ability of the model to different data sets. It can efficiently process nonlinear and non-stationary fishery time series data and significantly reduce the data annotation cost.

[0016] (2) The present invention extracts short-term trend features, long-term trend features, periodic features and environmental response features from fishery resource data through a multi-layer feature extraction module, and uses correlation analysis to construct a feature association matrix to screen high-correlation features and construct a feature space. Different from the existing methods that rely solely on historical catch or environmental data, the present invention innovatively integrates multi-source data to fully capture the coupling relationship between short-term fluctuations, long-term trends and external environment in resource changes, providing multi-dimensional and dynamic input support for the prediction model, solving the problem of insufficient utilization of multi-source data in traditional methods.

[0017] (3) The time series dynamic prediction model constructed by the present invention is based on the self-attention mechanism, which can model the time dependency of input features and dynamically capture the key features at different time points, and optimize the model performance by combining fine-tuning training. Compared with traditional prediction models based on simple regression or fixed weights, the present invention can more effectively identify important time points in resource changes and adjust the weight distribution, and exhibits excellent prediction capabilities when processing data with complex seasonal fluctuations and the influence of multi-source environmental variables. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 A flow chart of a system and method for predicting the dynamic trend of fishery resources based on time series analysis proposed by the present invention; Figure 2 It is a structural schematic diagram of a time series dynamic prediction model constructed based on feature space in a fishery resources dynamic trend prediction system and method based on time series analysis proposed by the present invention. DETAILED DESCRIPTION

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0020] refer to Figure 1-Figure 2 , a method for predicting the dynamic trend of fishery resources based on time series analysis, comprising the following steps: S1. Collect multi-source fishery resources and construct a fishery resource time series dataset; S2, standardizing the fishery resources time series data set to form a standardized fishery resources time series data set; S3. Using the fishery resource time series data set, a self-supervised learning model including an autoencoder module and a contrastive learning module is constructed. The autoencoder is used to learn the characteristic distribution of the time series data, and the contrastive learning module improves the adaptability of the self-supervised learning model to different fishery resource time series data sets. S4. Extract multi-layer features including short-term trend features, long-term trend features, periodic features and environmental response features from standardized fishery resource time series data based on the pre-trained self-supervised learning model, perform correlation analysis on the multi-layer features, and select highly correlated features to construct a feature space; S5. Based on the feature space, a time series dynamic prediction model with self-attention mechanism as the core is constructed, and the feature representation extracted by the self-supervised learning model is used to fine-tune the time series dynamic prediction model to optimize the prediction ability; S6. Use the time series dynamic prediction model to predict the future trends of the current standardized fishery resource time series data set, and generate prediction results for the spatial location of resource distribution, resource abundance change trend, and peak and trough time ranges.

[0021] In this implementation, S1 specifically includes the following contents: S11. Collect time series data of fishery resources from multiple sources, including: Historical fishing data ,Historical fishing data include fishing time, catch volume and fishing location in fishing records; Environmental monitoring data ,Environmental monitoring data include seawater temperature, salinity and dissolved oxygen content at the monitoring time point; Data on reproductive cycles of target species , the target species reproduction cycle data include the reproduction time cycle points and the reproduction size of the target species; Ship track data: ; in, represents the geographical coordinates of the ship’s trajectory, t is the collection time point, and p represents the number of time points in the ship’s trajectory data; S12. Time alignment of multi-source data: using a unified time reference frame T to perform time alignment on the collected multi-source fishery resource time series data to generate a fishery resource time series data set : ; in, represents the time point of the time series, and q represents the total number of time points of the fishery resources time series data.

[0022] In this implementation, S2 specifically includes the following contents: S21. Using wavelet transform denoising algorithm to analyze the fishery resources time series data set De-noising and decomposition of fishery resource time series dataset For different frequency components, high-frequency noise is suppressed by threshold method, and the denoised fishery resource time series dataset is reconstructed; S22, using the double mean method to detect outliers on the denoised fishery resource time series data set, setting the time window W to calculate each time point The mean value in the window and standard deviation , marking data points that meet the following conditions as outliers: ; in, represents a single feature value of time series data, is the threshold coefficient, and the abnormal points are marked as NULL, generating a fishery resource time series dataset containing abnormal marks; S23, performing linear interpolation completion on the fishery resource time series data set with the outlier marked as NULL, estimating the data value at the outlier position by using the weighted average of the values ​​before and after the time series data point, and generating a completed fishery resource time series data set; S24. Normalize the completed fishery resource time series dataset, map the values ​​of different features to a unified interval [0,1], and finally form a standardized fishery resource time series dataset. .

[0023] In this implementation, S3 specifically includes the following contents: S31. Using the standardized fishery resource time series dataset Constructing a self-supervised learning model including an autoencoder module and a contrastive learning module, wherein the autoencoder module includes an encoder network and a decoder network; S32, encoder network transforms the standardized fishery resources time series dataset through nonlinear mapping function The time point feature vector in Mapping to latent feature space representation ; S33, the decoder network represents the potential features through a nonlinear mapping function Reconstructed as an estimate of the original eigenvector ; S34, by minimizing the reconstruction error loss function Train the autoencoder module: ; Where n is the number of time points, express norm; Reconstruction error loss function Reflects the original feature vector and reconstructed eigenvector The difference between them is minimized by minimizing the reconstruction error loss function Optimize the parameters of the encoder and decoder networks so that the self-supervised learning model can capture the characteristic distribution of fishery resource time series data; S35. Standardized fishery resource time series dataset Perform data enhancement, build a comparative learning module, and optimize the adaptability of the self-supervised learning model to different fishery resource time series data sets; S36. For each time point, the original feature vector Apply random data augmentation operations to generate a set of positive samples And negative sample set ; S37, encoder network Applied to the time point feature vector, positive sample set and negative sample set to obtain the corresponding potential feature representation and : ; ; ; S38. Define contrast loss function Optimize the encoder network so that the potential feature representation of positive samples is closer to the original samples, and the potential feature representation of negative samples is farther away from the original samples: ; in, is the cosine similarity between latent feature representations, is the temperature parameter, K is the number of negative samples for each sample; S39. By jointly optimizing the loss function Train the self-supervised learning model and improve its adaptability to different fishery resource time series data sets: ; in, and is the weight coefficient used to balance the reconstruction loss and contrast loss , by minimizing the joint optimization loss function Optimize encoder network parameters at the same time and decoder network parameters , enabling the self-supervised learning model to learn the characteristic distribution of fishery resource time series data.

[0024] In this implementation, S4 specifically includes the following contents: S41. Using the pre-trained self-supervised learning model to analyze the standardized fishery resources time series dataset Extract multi-layer features and use the encoder network to generate feature vectors at each time point Perform feature extraction and generate potential feature vectors , and decompose it into short-term trend characteristics, long-term trend characteristics, cycle characteristics and environmental response characteristics: ; in, Indicates short-term trend characteristics, reflecting the short-term fluctuation trend of catch and environmental indicators. It indicates the long-term trend characteristics and reflects the changes of fishery resources on a long-term scale. Indicates periodic characteristics, reveals the reproductive cycle and seasonal fluctuations of target species, It represents the environmental response characteristics and is associated with changes in the external environment; S42. Use the correlation analysis method to calculate the correlation coefficient R between different features and establish a feature correlation matrix : }; in, is the feature vector and the eigenvector The covariance of and is the standard deviation of the corresponding feature, is the correlation coefficient between two features; S43. According to the correlation matrix The analysis results screen for feature sets that are highly correlated with future catch or fishery resource abundance , and finally construct the feature space.

[0025] In this implementation, S5 specifically includes the following contents: S51. Features extracted using self-supervised learning models As input features, a time series dynamic prediction model with self-attention mechanism as the core is constructed The time series dynamic prediction model models the temporal dependency of input features through the self-attention mechanism. For the time series feature X, the attention weight of each time point is calculated. ; The features at time point i are calculated by weighted summation to obtain the self-attention representation : ; in, is a value vector, is the learnable parameter matrix; S52. Feature representation after modeling using self-attention mechanism Enter the fully connected network , generating forecast values ​​for future time points : ; in, are the parameters of the fully connected network; S53. Features extracted using self-supervised learning models and the known target variable Fine-tune the time series dynamic prediction model by minimizing the prediction error loss function Optimize model parameters: ; Where N is the number of samples, Dynamic forecasting model for time series The predicted value of is the actual value; S54. In the fine-tuning process, the prediction loss of the time series dynamic prediction model is combined with the feature extraction loss of the self-supervised learning model to define the joint optimization objective function: ; in, is the weight coefficient, which is used to balance the optimization objectives of the prediction model and the self-supervised learning model.

[0026] In this implementation, S6 specifically includes the following contents: S61. Dynamic prediction model of time series Current time point feature representation applied to standardized fishery resource time series datasets , predict the resource status of the next h time steps, and get the predicted value of the next h time steps ; S62. Using predicted values and geographical distribution characteristics of historical catch data Calculate the resource distribution position at time point t+h ; S63. Based on the predicted value Calculate the overall trend of future resource abundance changes: ; in, represents the future rate of change of resource abundance, is the actual value of historical resource abundance; S64, based on the predicted value Identify the extreme points of resource changes and determine the time ranges of peak and trough periods: ; ; in, Indicates the predicted peak time point of resource abundance, Indicates the predicted time point of low resource abundance; S65, the predicted value , the spatial location of resource distribution , resource abundance change rate and peak time points and low point Output as a forecast report through visualization tools: Time series graph, intuitively showing the changing trend of resource abundance; Geographic information heat map, marking the location of future resource distribution; Annotations of peak and trough time points provide suggestions for optimizing fishing plans.

[0027] A fishery resource dynamic trend prediction system based on time series analysis, including the following modules: The data collection module is used to collect multi-source fishery resource time series data, including historical fishing data, environmental monitoring data, target species reproduction cycle data and ship trajectory data, to form a fishery resource time series data set and provide real-time update capabilities to support dynamic prediction needs; The data preprocessing module is used to perform denoising, outlier detection, data completion and normalization on the collected fishery resource time series data to form a standardized fishery resource time series data set; The self-supervised learning model construction module learns the data feature distribution based on the standardized fishery resource time series data by constructing an autoencoder and a contrastive learning module. The autoencoder is used to extract the potential feature space representation. The contrastive learning module improves the model's adaptability to different data sets and realizes the pre-training of the self-supervised learning model. The multi-layer feature extraction and analysis module extracts short-term trend features, long-term trend features, periodic features, and environmental response features based on a self-supervised learning model, establishes a feature association matrix through correlation analysis, selects feature sets with high correlation with the target variable, and constructs a feature space for dynamic prediction; The dynamic prediction model construction and optimization module uses the constructed feature space to input the time series dynamic prediction model. The time series dynamic prediction model is based on the self-attention mechanism and predicts the future resource status by modeling time dependencies. At the same time, the time series dynamic prediction model is fine-tuned and trained by combining the feature representation extracted by the self-supervised learning model. The future trend prediction module predicts the future trend of the standardized fishery resource time series data set based on the time series dynamic prediction model, including the spatial location of resource distribution, resource abundance change trend, and peak and trough time range, and generates prediction results; The visualization and result output module outputs the forecast results in an intuitive form, including time series curve graphs, geographic information heat maps, and peak and trough time point annotations, forming a comprehensive forecast report to provide decision support for fishery resource management and fishing plan optimization. Example

[0028] From March to May 2023, the coastal waters management department of Province A conducted a comprehensive dynamic trend forecast and tracking management of mantis shrimp resources. This forecast and research took the southern waters of Shandong Peninsula as the core area, and used the method of the present invention to integrate multi-source data, apply dynamic prediction models, scientifically plan fishing strategies and optimize management decisions.

[0029] Mantis shrimp is an important economic species in the coastal waters of Province A. It is widely distributed in muddy or silty seabed areas with a depth of 5-60 meters. Due to the continuous increase in fishing pressure and the impact of climate change in recent years, its natural resource quantity and population structure have fluctuated greatly. In the spring of 2023, the management department received feedback from fishermen that the catch of mantis shrimp in the coastal area of ​​Haiyang has decreased significantly, and fishermen have found that the migration path may have shifted. This phenomenon has triggered the need for research on the dynamic trend of mantis shrimp resources.

[0030] In March 2023, the project team conducted comprehensive data collection at 90 stations in the coastal area of ​​Haiyang. The collected data include: Historical catch data: Mantis shrimp catches from 2013 to 2022, about 50,000 records.

[0031] Environmental monitoring data: daily water temperature (15°C-25°C), salinity (30‰-34‰), dissolved oxygen (4 mg / L-8mg / L), etc.

[0032] Fishing vessel trajectory data: Fishing paths and location data of over 1,000 fishing vessels.

[0033] Reproductive cycle data: time and area of ​​peak spring spawning of mantis shrimp.

[0034] Through the data preprocessing module of the present invention, abnormal data are eliminated, data normalization and linear interpolation are performed, and a standardized fishery resource time series data set is formed.

[0035] Through the self-supervised learning model, the following multi-layer features are extracted: Short-term trends: Between the third week of March and the first week of April, the abundance of mantis shrimp decreased by 15%.

[0036] Long-term trend: Data from the past 10 years show that mantis shrimp catches peak in the third week of May each year.

[0037] Periodic characteristics: Strongly correlated with periodic changes in water temperature and salinity.

[0038] Environmental response characteristics: Salinity and dissolved oxygen significantly affect resource distribution.

[0039] Based on multi-layer features, a time series dynamic prediction model with self-attention mechanism as the core was constructed. The model identified the key time points of resource changes by learning the time dependencies in historical data, and fine-tuned the predictions based on real-time data.

[0040] In April 2023, the model outputted a resource dynamic trend forecast for the next 8 weeks, mainly including: 1. Resource distribution prediction: It is expected that the migration path of mantis shrimp will shift from the coast of Haiyang to the area east of Weihai.

[0041] 2. Abundance change trend: Abundance will begin to rise in the second week of April and reach a peak in the third week of May, with an average daily catch of 18 tons.

[0042] 3. Peak and trough time points: The peak time is May 20 to May 25, and the trough time is June 10 to June 15.

[0043] To verify the accuracy of the model, the project team monitored the actual fishing data and compared the results with the traditional statistical method (ARIMA model). In peak prediction: the actual monitoring data showed that the catch from May 20 to May 25, 2023 was 17.8 tons, close to the model prediction of 18 tons; the traditional method predicted 15 tons, with a large error. In distribution prediction: the actual overlap rate of the model-predicted migration path was 91%, significantly higher than the 67% of the traditional method. In abundance changes: the model captured short-term fluctuations in abundance changes, and the degree of agreement with the actual data was 93.5%.

[0044] Through this example, the efficiency and accuracy of the present invention in predicting the dynamic trend of mantis shrimp have been verified. The present invention not only solves the problem of insufficient integration of complex environmental changes and multi-source data in traditional methods, but also significantly improves the prediction accuracy, providing strong technical support for fishery management.

[0045] The self-supervised learning model proposed in the present invention combines an autoencoder and a contrastive learning module to achieve feature learning of complex fishery resource time series data without relying on a large amount of labeled data. The autoencoder module captures the characteristic distribution of time series data and optimizes the model structure by minimizing the reconstruction error. The contrastive learning module generates positive and negative sample pairs to improve the adaptability and generalization ability of the model to different data sets. It can efficiently process nonlinear and non-stationary fishery time series data and significantly reduce data annotation costs.

[0046] The present invention extracts short-term trend features, long-term trend features, periodic features and environmental response features from fishery resource data through a multi-layer feature extraction module, and uses correlation analysis to construct a feature association matrix to screen high-correlation features and construct a feature space. Unlike the existing methods that solely rely on historical catch or environmental data, the present invention innovatively integrates multi-source data to fully capture the coupling relationship between short-term fluctuations, long-term trends and external environment in resource changes, provides multi-dimensional and dynamic input support for the prediction model, and solves the problem of insufficient utilization of multi-source data in traditional methods.

[0047] The time series dynamic prediction model constructed by the present invention takes the self-attention mechanism as its core, can model the time dependency of input features, dynamically capture the key features at different time points, and optimize the model performance in combination with fine-tuning training. Compared with traditional prediction models based on simple regression or fixed weights, the present invention can more effectively identify important time points in resource changes and adjust the weight distribution, and exhibits excellent prediction capabilities when processing data with complex seasonal fluctuations and the influence of multi-source environmental variables.

[0048] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for predicting the dynamic trend of fishery resources based on time series analysis, characterized in that: The steps include: S1. Collect multi-source fishery resources and construct a fishery resource time series dataset; S2, standardizing the fishery resources time series data set to form a standardized fishery resources time series data set; S3. Using the fishery resource time series data set, a self-supervised learning model including an autoencoder module and a contrastive learning module is constructed. The autoencoder is used to learn the characteristic distribution of the time series data, and the contrastive learning module improves the adaptability of the self-supervised learning model to different fishery resource time series data sets. S4. Extract multi-layer features including short-term trend features, long-term trend features, periodic features and environmental response features from standardized fishery resource time series data based on the pre-trained self-supervised learning model, perform correlation analysis on the multi-layer features, and select highly correlated features to construct feature space; S5. Based on the feature space, a time series dynamic prediction model with self-attention mechanism as the core is constructed, and the feature representation extracted by the self-supervised learning model is used to fine-tune the time series dynamic prediction model to optimize the prediction ability; S6. Use the time series dynamic prediction model to predict the future trends of the current standardized fishery resource time series data set, and generate prediction results for the spatial location of resource distribution, resource abundance change trend, and peak and trough time ranges.

2. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 1, characterized in that: The S1 specifically includes the following contents: S11. Collect time series data of fishery resources from multiple sources, including: Historical fishing data ,Historical fishing data include fishing time, catch volume and fishing location in fishing records; Environmental monitoring data ,Environmental monitoring data include seawater temperature, salinity and dissolved oxygen content at the monitoring time point; Data on reproductive cycles of target species , the target species reproduction cycle data include the reproduction time cycle points and the reproduction size of the target species; Ship track data: ； in, represents the geographical coordinates of the ship’s trajectory, t is the collection time point, and p represents the number of time points in the ship’s trajectory data; S12. Time alignment of multi-source data: using a unified time reference frame T to perform time alignment on the collected multi-source fishery resource time series data to generate a fishery resource time series data set : ； in, represents the time point of the time series, and q represents the total number of time points of the fishery resources time series data.

3. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 2, characterized in that: The S2 specifically includes the following contents: S21. Using wavelet transform denoising algorithm to analyze the fishery resources time series data set De-noising and decomposition of fishery resource time series dataset For different frequency components, high-frequency noise is suppressed by threshold method, and the denoised fishery resource time series dataset is reconstructed; S22, using the double mean method to detect outliers on the denoised fishery resource time series data set, setting the time window W to calculate each time point The mean value in the window and standard deviation , marking data points that meet the following conditions as outliers: ； in, represents a single feature value of time series data, is the threshold coefficient, and the abnormal points are marked as NULL, generating a fishery resource time series dataset containing abnormal marks; S23, performing linear interpolation completion on the fishery resource time series data set with the outlier marked as NULL, estimating the data value at the outlier position by using the weighted average of the values ​​before and after the time series data point, and generating a completed fishery resource time series data set; S24. Normalize the completed fishery resource time series dataset, map the values ​​of different features to a unified interval [0,1], and finally form a standardized fishery resource time series dataset. .

4. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 3, characterized in that: The S3 specifically Includes the following: S31. Using the standardized fishery resource time series dataset Constructing a self-supervised learning model including an autoencoder module and a contrastive learning module, wherein the autoencoder module includes an encoder network and a decoder network; S32, encoder network transforms the standardized fishery resources time series dataset through nonlinear mapping function The time point feature vector in Mapping to latent feature space representation ; S33, the decoder network represents the potential features through a nonlinear mapping function Reconstructed as an estimate of the original eigenvector ; S34, by minimizing the reconstruction error loss function Train the autoencoder module: ； Where n is the number of time points, express norm; Reconstruction error loss function Reflects the original feature vector and reconstructed eigenvector The difference between them is minimized by minimizing the reconstruction error loss function Optimize the parameters of the encoder and decoder networks so that the self-supervised learning model can capture the characteristic distribution of fishery resource time series data; S35. Standardized fishery resource time series dataset Perform data enhancement, build a comparative learning module, and optimize the adaptability of the self-supervised learning model to different fishery resource time series data sets; S36. For each time point feature vector Apply random data augmentation operations to generate a set of positive samples And negative sample set ; S37, encoder network Applied to the time point feature vector, positive sample set and negative sample set to obtain the corresponding potential feature representation and : ； ； ； S38. Define contrast loss function Optimize the encoder network so that the potential feature representation of positive samples is closer to the original samples, and the potential feature representation of negative samples is farther away from the original samples: ； in, is the cosine similarity between latent feature representations, is the temperature parameter, K is the number of negative samples for each sample; S39. By jointly optimizing the loss function Train the self-supervised learning model and improve its adaptability to different fishery resource time series data sets: ； in, and is the weight coefficient used to balance the reconstruction loss and contrast loss , by minimizing the joint optimization loss function Optimize encoder network parameters at the same time and decoder network parameters , enabling the self-supervised learning model to learn the characteristic distribution of fishery resource time series data.

5. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 4, characterized in that: The S4 specifically includes the following contents: S41. Using the pre-trained self-supervised learning model to analyze the standardized fishery resources time series dataset Extract multi-layer features and use the encoder network to generate feature vectors at each time point Perform feature extraction and generate potential feature vectors , and decompose it into short-term trend characteristics, long-term trend characteristics, cycle characteristics and environmental response characteristics: ； in, Indicates short-term trend characteristics, reflecting the short-term fluctuation trend of catch and environmental indicators. It indicates the long-term trend characteristics and reflects the changes of fishery resources on a long-term scale. Indicates periodic characteristics, reveals the reproductive cycle and seasonal fluctuations of target species, It represents the environmental response characteristics and is associated with changes in the external environment; S42. Use the correlation analysis method to calculate the correlation coefficient R between different features and establish a feature correlation matrix : }； in, is the feature vector and the eigenvector The covariance of and is the standard deviation of the corresponding feature, is the correlation coefficient between two features; S43. According to the correlation matrix The analysis results screen for feature sets that are highly correlated with future catch or fishery resource abundance , and finally construct the feature space.

6. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 5, characterized in that: The S5 specifically includes the following contents: S51. Features extracted using self-supervised learning models As input features, a time series dynamic prediction model with self-attention mechanism as the core is constructed The time series dynamic prediction model models the temporal dependency of input features through the self-attention mechanism. For the time series feature X, the attention weight of each time point is calculated. ; The features at time point i are calculated by weighted summation to obtain the self-attention representation : ； in, is a value vector, is the learnable parameter matrix; S52. Feature representation after modeling using self-attention mechanism Enter the fully connected network , generating forecast values ​​for future time points : ； in, are the parameters of the fully connected network; S53. Features extracted using self-supervised learning models and the known target variable Fine-tune the time series dynamic prediction model by minimizing the prediction error loss function Optimize model parameters: ； Where N is the number of samples, Dynamic forecasting model for time series The predicted value of is the actual value; S54. In the fine-tuning process, the prediction loss of the time series dynamic prediction model is combined with the feature extraction loss of the self-supervised learning model to define the joint optimization objective function: ； in, is the weight coefficient, which is used to balance the optimization objectives of the prediction model and the self-supervised learning model.

7. A method for predicting the dynamic trend of fishery resources based on time series analysis according to claim 6, characterized in that: The S6 specifically includes the following contents: S61. Dynamic prediction model of time series Current time point feature representation applied to standardized fishery resource time series datasets , predict the resource status of the next h time steps, and get the predicted value of the next h time steps ; S62. Using predicted values and geographical distribution characteristics of historical catch data Calculate the resource distribution position at time point t+h ; S63. Based on the predicted value Calculate the overall trend of future resource abundance changes: ； in, represents the future rate of change of resource abundance, is the actual value of historical resource abundance; S64, based on the predicted value Identify the extreme points of resource changes and determine the time ranges of peak and trough periods: ； ； in, Indicates the predicted peak time point of resource abundance, Indicates the predicted time point of low resource abundance; S65, the predicted value , the spatial location of resource distribution , resource abundance change rate and peak time points and low point Output as a forecast report through visualization tools: Time series graph, intuitively showing the changing trend of resource abundance; Geographic information heat map, marking the location of future resource distribution; Annotations of peak and trough time points provide suggestions for optimizing fishing plans.

8. A fishery resource dynamic trend prediction system based on time series analysis, comprising the method according to any one of claims 1 to 7, characterized in that: Includes the following modules: The data collection module is used to collect multi-source fishery resource time series data, including historical fishing data, environmental monitoring data, target species reproduction cycle data and ship trajectory data, to form a fishery resource time series data set and provide real-time update capabilities to support dynamic prediction needs; The data preprocessing module is used to perform denoising, outlier detection, data completion and normalization on the collected fishery resource time series data to form a standardized fishery resource time series data set; The self-supervised learning model construction module learns the data feature distribution based on the standardized fishery resource time series data by constructing an autoencoder and a contrastive learning module. The autoencoder is used to extract the potential feature space representation. The contrastive learning module improves the model's adaptability to different data sets and realizes the pre-training of the self-supervised learning model. The multi-layer feature extraction and analysis module extracts short-term trend features, long-term trend features, periodic features, and environmental response features based on a self-supervised learning model, establishes a feature association matrix through correlation analysis, selects feature sets with high correlation with the target variable, and constructs a feature space for dynamic prediction; The dynamic prediction model construction and optimization module uses the constructed feature space to input the time series dynamic prediction model. The time series dynamic prediction model is based on the self-attention mechanism and predicts the future resource status by modeling time dependencies. At the same time, the time series dynamic prediction model is fine-tuned and trained by combining the feature representation extracted by the self-supervised learning model. The future trend prediction module predicts the future trend of the standardized fishery resource time series data set based on the time series dynamic prediction model, including the spatial location of resource distribution, resource abundance change trend, and peak and trough time range, and generates prediction results; The visualization and result output module outputs the forecast results in an intuitive form, including time series curve graphs, geographic information heat maps, and peak and trough time point annotations, forming a comprehensive forecast report to provide decision support for fishery resource management and fishing plan optimization.

Images