Phytoplankton monitoring scheme design method based on multivariate time series model

Abstract

This invention discloses a method for designing a phytoplankton monitoring scheme based on a multivariate time series model, comprising: dividing all marine monitoring samples into training and testing sets according to chronological order; calculating the correlation between each variable value and the phytoplankton density label using multiple variable filtering methods, filtering out several variable values ​​with lower ranking, and obtaining multiple different variable sets; training a multivariate long short-term memory network model and a multivariate time series model containing an attention prototype network using multiple variable sets and the full variable set, respectively, and selecting the model with optimal prediction performance and training time, along with its corresponding variable set; and generating a monitoring scheme for a future period based on the prediction results of the optimal model and its corresponding variable set, under the constraints of project budget, existing monitoring infrastructure, and management priorities. This invention can support high-frequency rolling prediction based on large-scale, long-term observation data.

Description

Technical Field

[0001] This invention relates to the field of marine environmental monitoring and ecological prediction, specifically to a method for designing phytoplankton monitoring schemes based on multivariate time series models. Background Technology

[0002] Marine phytoplankton, as the foundation of marine food webs, significantly impact not only the stability of regional food webs but also global climate regulation and the sustainable use of fishery resources. In recent years, coastal areas, as transitional zones where land and sea meet, have frequently faced severe natural climate change and anthropogenic eutrophication pressures. Consequently, since the 21st century, these regions have become major areas with a significant increase in the frequency of algal blooms, especially in coastal areas of developing countries. Due to the rapid occurrence and decline of algal blooms, real-time and accurate monitoring is crucial for prediction, early warning, and effective management. However, traditional marine environmental monitoring relies heavily on the deployment of fixed stations and the formulation of empirical rules. The selection of monitoring indicators, sampling frequency, and spatial layout are often based on "empirical judgment," lacking systematic quantitative assessment and data-driven support. On the one hand, with limited project budgets, monitoring resources are easily dispersed in the pursuit of "comprehensive coverage," making it difficult to focus on key influencing factors. On the other hand, existing monitoring data has not been fully utilized to optimize monitoring scheme design, limiting its application in refined and forward-looking ecological management decisions, thus hindering the transformation of environmental monitoring towards "digitalization and intelligence." Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention proposes a phytoplankton monitoring scheme design method based on a multivariate time series model. This method can achieve accurate and efficient prediction and early warning of phytoplankton community dynamics in coastal waters, providing key technical support for marine ecological protection and algal bloom disaster prevention and control.

[0004] The objective of this invention is achieved through the following technical solution:

[0005] A method for designing a phytoplankton monitoring scheme based on a multivariate time series model includes the following steps:

[0006] S1: Divide all marine monitoring samples into training and testing sets according to their chronological order;

[0007] S2: The correlation between each variable value and the phytoplankton density label is calculated using multiple different variable filtering methods. Based on this, all variable values ​​are sorted. Then, based on the sorting results, several variable values ​​that rank low in each variable filtering method are filtered out, thus obtaining multiple different variable sets.

[0008] S3: Using the multiple variable sets and the full variable set obtained in S2, train a multivariate long short-term memory network model and a multivariate time series model containing an attention prototype network, respectively. Compare the prediction performance and training time of the two models on the multiple variable sets and the full variable set, and select the model with the best prediction performance and training time and its corresponding variable set.

[0009] S4: Under the constraints of project budget, existing monitoring foundation and management priorities, the prediction results obtained from the optimal model obtained in S3 and its corresponding variable set are used to generate a phytoplankton monitoring plan for a future period of time.

[0010] Furthermore, in S2, the multivariate time series model containing the attention prototype network includes branch one, which consists of a random dimension permutation module, a local variable extraction module, two high-order local modules with shared parameters and an average pooling layer one, and branch two, which consists of an LSTM model and an average pooling layer two; it also includes a splicing layer connecting branch one and branch two, two fully connected layers, an embedding layer, an attention prototype network module, a gated multiplication module, a similarity transformation module, and a normalized exponential function module.

[0011] Furthermore, the local variable extraction module includes multiple local variable extraction sub-modules consisting of one-dimensional convolutional layers, batch normalization layers, and rectified functions with leakage, which extract local variables from the corresponding multivariate time series.

[0012] Furthermore, in S2, the multivariate long short-term memory network model includes an input layer, a hidden layer, and an output layer.

[0013] Furthermore, the marine monitoring samples in S1 include variables such as silicate, phosphate, nitrate, salinity, sea surface temperature, ocean currents, and dissolved oxygen.

[0014] Furthermore, the variable filtering methods include mutual information method, Pearson correlation coefficient method, and variable importance ranking method based on random forest model.

[0015] Furthermore, in S2, the last two variable values ​​in each variable filtering method are filtered out.

[0016] A device for designing phytoplankton monitoring schemes based on multivariate time series models, comprising one or more processors, for implementing a method for designing phytoplankton monitoring schemes based on multivariate time series models.

[0017] An electronic device, comprising:

[0018] One or more processors;

[0019] A storage device for storing one or more programs, which, when executed by the electronic device, enable the electronic device to implement a phytoplankton monitoring scheme design method based on a multivariate time series model.

[0020] A computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements a method for designing a phytoplankton monitoring scheme based on a multivariate time series model.

[0021] The beneficial effects of this invention are as follows:

[0022] 1. This invention sorts the variables by importance and gradually reduces the dimensionality of the multi-source monitoring indicators, thereby effectively compressing the input dimension and model size while maintaining the prediction performance. This reduces the computational complexity and resource consumption of training and updating, and supports high-frequency rolling prediction based on large-scale, long-term series observation data.

[0023] 2. Under constraints such as monitoring budget, this invention generates future monitoring plans based on the prediction results of the optimal time series model, realizes the optimized configuration of monitoring stations and sampling frequencies in high-risk sea areas and key periods, thereby improving the efficiency of monitoring resource utilization, and constructs a scalable closed-loop marine ecological monitoring decision-making system of "prediction-monitoring-feedback retraining".

[0024] 3. This invention transforms monitoring design from the previous experience-based setting into an iterative, data-driven optimization process, enabling limited monitoring resources to prioritize key driving variables and high-risk areas, enhancing the adaptability and scalability of the monitoring network, providing a replicable and transferable technical path for marine ecological environment governance, and promoting the transformation of environmental monitoring from "experience-driven" to "data-driven and intelligent decision-making". Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for designing a phytoplankton monitoring scheme based on a multivariate time series model, according to an embodiment of the present invention.

[0026] Figure 2 The image shows the correlation ranking results based on three variable filtering methods provided in an embodiment of the present invention, where (a) is the correlation ranking based on Pearson correlation coefficient, (b) is the correlation ranking result based on mutual information, and (c) is the correlation ranking result based on random forest model.

[0027] Figure 3 This is a schematic diagram of a multivariate long short-term memory network model structure provided in an embodiment of the present invention.

[0028] Figure 4 This is a schematic diagram of a multivariate time series model structure including an attention prototype network, provided in an embodiment of the present invention.

[0029] Figure 5 A comparison chart of the prediction performance of a multivariate long short-term memory network model based on three sets of variables and a full set of variables, and a multivariate time series model including an attention prototype network, provided in an embodiment of the present invention.

[0030] Figure 6 A comparison chart of training times for a multivariate long short-term memory network model based on three sets of variables and a full set of variables, and a multivariate time series model including an attention prototype network, provided in an embodiment of the present invention. Detailed Implementation

[0031] The present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. The purpose and effects of the present invention will become clearer. It should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0032] The phytoplankton monitoring scheme design method based on a multivariate time series model of the present invention first calculates the correlation between various environmental variables and phytoplankton density labels, and then ranks the variables according to the correlation. Next, based on the ranking results, the lowest-ranked variable values ​​are filtered out to reduce redundant monitoring indicators. Then, different time series models are trained under different variable sets to compare prediction performance and model training time, and the time series model that achieves the optimal balance between prediction performance and computational efficiency is selected. Finally, under the premise of budget constraints, existing monitoring foundation, and clear management priorities, the prediction results and uncertainty characterization of the optimal model are used to generate an optimized monitoring scheme for the next two years, achieving coordinated optimization of the monitoring indicator system, sampling frequency, spatiotemporal layout, and management needs.

[0033] As one implementation method, the phytoplankton monitoring scheme design method based on a multivariate time series model of this embodiment includes the following steps one through four.

[0034] Step 1: Dataset Construction and Sample Division: Divide all marine monitoring samples into training and test sets according to their chronological order.

[0035] In this embodiment, the training set consists of all historical samples up to the third-to-last year. These historical samples are merged into a single set and binary-classified. The test set consists independently of data from the two most recent years. During the evaluation phase, the data from each year is used as a separate test subset and subjected to individual binary classification. Therefore, the set D of all marine monitoring samples collected chronologically and the training set Di are... 训练集 and test set D 测试集 They are represented as follows:

[0036]

[0037]

[0038]

[0039] Among them, X n Y represents the variable vector of the nth sample. n This represents the binary label of the nth sample. Where 1 represents a high-density phytoplankton distribution area (greater than or equal to the median density of all phytoplankton samples), and 0 represents a low-density phytoplankton distribution area; η n Let M be the year corresponding to the nth sample, and M represent the total number of observation years for the historical samples. In this example, M=8.

[0040] Step 2: Calculate the correlation between each variable value and the phytoplankton density label using various variable filtering methods, sort all variable values ​​accordingly, and filter out the lower-ranked variable values ​​for each filtering method based on the sorting results, thereby obtaining various variable sets.

[0041] The variables used in this embodiment are silicate (Si), phosphate (PO4), nitrate (NO3), salinity (SSS), sea surface temperature (SST), ocean current (CV), and dissolved oxygen (DO).

[0042] In this embodiment, three variable filtering methods are used: mutual information method, Pearson correlation coefficient method, and variable importance ranking method based on random forest model. The specific calculation formulas are as follows:

[0043]

[0044]

[0045]

[0046] In the formula, I(X) is the importance score for the j-th variable value based on mutual information; j is the variable value index; j ;y) represents the variable value X j Mutual information between the phytoplankton density label y; For variable value X j The set of possible values; Let y be the set of possible values ​​for the phytoplankton density label. For variable value X j The value of the joint probability distribution of the label y at the point (x,y); For variable value X j The marginal probability at the value x; p y(y) represents the marginal probability of label y at the value y; N is the total number of samples; x ij y represents the value of the j-th variable in the i-th sample; i This represents the phytoplankton density label value corresponding to the i-th sample; Let be the mean of the j-th variable value across all samples; ρ(X) represents the mean of the phytoplankton density labels across all samples. j (y) represents the variable value X j The absolute value of the Pearson correlation coefficient with label y; Assign an importance score to the j-th variable value based on the Pearson correlation coefficient; Let be the variable importance score based on the j-th variable value in the random forest, be the average importance score across all decision trees, and K be the total number of decision trees in the random forest. In the k-th decision tree, the variable values... Gini impurity (a measure of the contribution of a variable to the tree's predictive performance) is introduced by the nodes involved in the partitioning.

[0047] Among them, the correlation ranking based on the Pearson correlation coefficient method is as follows: Figure 2 As shown in (a), the correlation strengths are as follows: silicate (0.200021), nitrate (0.19037), dissolved oxygen (0.183243), salinity (0.173815), sea surface temperature (0.132881), ocean current (0.02331), and phosphate (0.003421); the correlation ranking based on mutual information is as follows. Figure 2 As shown in (b), the correlation strengths are as follows: silicate (0.040221), salinity (0.030891), dissolved oxygen (0.028461), phosphate (0.016344), sea surface temperature (0.013002), nitrate (0.009018), and ocean current (0.00762); the results of the variable importance ranking method based on the random forest model are as follows. Figure 2 As shown in (c), the correlation levels are as follows: silicate (0.125439438), phosphate (0.103435165), nitrate (0.092411034), salinity (0.091157641), sea surface temperature (0.088522116), dissolved oxygen (0.084275428), and ocean current (0.069541295).

[0048] Step 3: Using the multiple variable sets and the full variable set obtained in Step 2, train a multivariate long short-term memory network model and a multivariate time series model containing an attention prototype network, respectively. Compare the prediction performance and training time of the two models on the multiple variable sets and the full variable set, and select the model with the best prediction performance and training time and its corresponding variable set.

[0049] A multivariate long short-term memory network model consists of an input layer, a hidden layer, and an output layer.

[0050] This embodiment increases the dimension of the input data to N (7 or 5) dimensions, as shown in the structural diagram below. Figure 3 As shown.

[0051] The input layer of a multivariate long short-term memory network model receives multi-source time series X. t and the hidden state h from the previous time step t-1 The previous cell state C t-1 The hidden layer selects, remembers, and updates information through forget gates, input gates, and output gates to obtain new unit states C. t and hidden state h t The output layer generates the prediction result (0,1) and sets the new cell state C. t and the new hidden state h t Pass it to the next time step t+1.

[0052] Among them, the multi-source time series are the following four types of multivariate time series in parallel:

[0053] (1) The total set of variables obtained are silicate (Si), phosphate (PO4), nitrate (NO3), salinity (SSS), sea surface temperature (SST), ocean current (CV) and dissolved oxygen (DO);

[0054] (2) Silicate (Si), nitrate (NO3), salinity (SSS), sea surface temperature (SST) and dissolved oxygen (DO) obtained after filtration based on Pearson correlation coefficient method;

[0055] (3) Silicate (Si), phosphate (PO4), salinity (SSS), sea surface temperature (SST) and dissolved oxygen (DO) obtained after filtration based on mutual information method;

[0056] (4) Silicate (Si), phosphate (PO4), nitrate (NO3), salinity (SSS) and sea surface temperature (SST) obtained after filtering based on the random forest model.

[0057] The output is Y= .

[0058] like Figure 4As shown, the multivariate time series model containing the attention prototype network includes branch one, which consists of a random dimension permutation module, a local variable extraction module, two high-order local modules with shared parameters and an average pooling layer, and branch two, which consists of an LSTM (Long Short-Term Memory) model and an average pooling layer. It also includes a splicing layer connecting the two branches, two fully connected layers, an embedding layer, an attention prototype network module, a gated multiplication module, a similarity transformation module, and a normalized exponential function module.

[0059] Among them, the four parallel multivariate time series are all used as inputs to the multivariate time series model containing the attention prototype network, and are respectively input to the random dimension permutation module and the LSTM model.

[0060] In branch one, the random dimension permutation module randomly permutes the variable / channel dimensions, changing only the variable order without altering the time series values ​​of each variable itself, and outputs multiple sets of multivariate time series after the random dimension permutation. The local variable extraction module includes multiple sub-modules consisting of one-dimensional convolutional layers, batch normalization layers, and rectified leakage functions, each extracting local variables from the corresponding multivariate time series. There are two local modules with shared parameters, which further extract higher-order local representations from the local variables extracted by the local variable extraction module, and then pass them through an average pooling layer to obtain the local variable vector.

[0061] In branch two, the four parallel multivariate time series are sequentially passed through an LSTM model and an average pooling layer two to obtain global feature representations.

[0062] The local variable vector obtained from branch one and the global feature representation obtained from branch two are concatenated in the feature dimension through a concatenation layer to obtain the fused feature representation. The fused feature representation is then input into two fully connected layers and an embedding layer to obtain the embedding vector of the sample to be classified.

[0063] The embedding vectors of the samples to be classified are input into the attention prototype network module to learn a prototype for each class. The class prototype is the feature representation (embedding) of each class, containing the same embedding size as the time series. Specifically, the class prototype is a weighted combination of training samples within the same class, where the weights of the training samples are trained by the attention layer. This learns a class prototype for each class that is close to data samples within the same class but far closer to data samples in different classes. The attention prototype network module outputs the distances between the embedding vectors.

[0064] The distance between the embedding vectors of the samples to be classified is processed by a gated multiplication module to obtain the gated multiplication feature representation. This feature representation is then processed sequentially by a similarity transformation module and a Softmax normalization function module to normalize the input into a class probability distribution, outputting the conditional class probability. .

[0065] The multivariate time series model incorporating an attention prototype network combines a random grouping permutation method with a multi-layer convolutional network, which can effectively learn low-dimensional variables from multi-source time series data and solve the problem of limited training labels.

[0066] In this embodiment, taking the multivariate time series obtained from the full variable set as an example, after the random dimension permutation module, a subset of multiple two-variable combinations is obtained. Based on expert experience, the four most important two-variable combinations for phytoplankton monitoring are selected from these two-variable combinations: nitric acid + dissolved oxygen, phosphoric acid + silica, ocean current + nitric acid, and salinity + sea surface temperature. These combinations are then input into the local variable extraction module.

[0067] In this embodiment, based on different combinations of feature variables S i With different model structures M j Obtain the corresponding prediction performance E i,j and model training time T i,j The baseline prediction performance E obtained using a full set of variables and a baseline model structure. 0,0 and baseline training time T 0,0 Based on the normalized baseline, the optimal variable combination and model structure are determined using the following weighted optimization formula:

[0068]

[0069] In the formula, i * The index of the variable set; j * λ is the index for the multivariate time series model; in this embodiment, λ is 1.

[0070] In this embodiment, the prediction performance and training time of the long short-term memory network model based on different variable filtering methods and the multivariate time series model containing an attention prototype network are respectively as follows: Figure 5 and Figure 6 As shown, where:

[0071] 1) The baseline prediction performance F1 score of the multivariate time series model with an attention prototype network based on the full variable set is 73.58% and the baseline training time is 120.81s; the baseline prediction performance F1 score of the multivariate long short-term memory network model based on the full variable set is 72.20% and the baseline training time is 315.12s.

[0072] 2) The results of the mutual information method are as follows: The baseline prediction performance F1 score of the multivariate time series model containing the attention prototype network is 70.18% and the baseline training time is 112.45s; the baseline prediction performance F1 score of the multivariate long short-term memory network model is 73.82% and the baseline training time is 316.17s.

[0073] 3) The results of the Pearson correlation coefficient method are as follows: the baseline prediction performance F1 score of the multivariate time series model containing the attention prototype network is 73.91% and the baseline training time is 112.12s; the baseline prediction performance F1 score of the multivariate long short-term memory network model is 70.85% and the baseline training time is 338.79s.

[0074] 4) The results of the variable importance ranking method based on the random forest model are as follows: the baseline prediction performance F1 score of the multivariate time series model containing the attention prototype network is 74.26% and the baseline training time is 218.03s; the baseline prediction performance F1 score of the multivariate long short-term memory network model is 71.17% and the baseline training time is 336.77s.

[0075] All F1 scores above are calculated based on positive samples (areas with high density of phytoplankton distribution).

[0076] In this embodiment, the optimal model selected is a multivariate time series model that includes an attention prototype network, and the corresponding set of variables is the variable set.

[0077] Step 4: Under the constraints of project budget, existing monitoring foundation, and management priorities, generate a phytoplankton monitoring plan for the future period based on the prediction results obtained from the optimal model and its corresponding variable set in Step 3.

[0078] In this embodiment, under the constraints of project budget, existing monitoring infrastructure, and defined management priorities, the prediction results for each monitoring station are based on the optimal time series model over a future planning period. A phytoplankton monitoring scheme for the future period is generated by solving the following optimization problem:

[0079]

[0080] In the formula, K represents the total number of monitoring stations, and T represents the total number of time segments in the planning period for the future; ω k The weight of the k-th monitoring station is determined based on management priorities; To map the forecast results to a single monotonic function that maps the forecast results to the monitoring benefits or monitoring demand intensity; As a decision variable, when The time indicates that a monitoring task is assigned to the k-th monitoring station in season t; otherwise, no task is assigned. k,t B represents the cost of performing one monitoring operation on the k-th monitoring station during season t; B represents the total monitoring expenditure under the project budget constraint. For the set of monitoring stations covered by the existing monitoring infrastructure, m k Z* represents the minimum number of monitoring sessions required for the k-th monitoring station to maintain the existing monitoring infrastructure during the planning period; Z* represents the optimal monitoring scheme that meets budget constraints, existing monitoring infrastructure constraints, and management priorities.

[0081] On the other hand, the present invention also provides an embodiment of a phytoplankton monitoring scheme design device based on a multivariate time series model.

[0082] The phytoplankton monitoring scheme design device based on a multivariate time series model provided in this embodiment includes one or more processors for implementing the phytoplankton monitoring scheme design method based on a multivariate time series model in the above embodiment.

[0083] The phytoplankton monitoring scheme design device based on a multivariate time series model of this invention can be applied to any device with data processing capabilities, such as a computer. The device can be implemented in software, hardware, or a combination of both. Taking software implementation as an example, as a logical device, it is formed by the processor of the device with data processing capabilities reading the corresponding computer program instructions from non-volatile memory into memory and executing them. From a hardware perspective, in addition to the processor, memory, network interface, and non-volatile memory, the device with data processing capabilities in the embodiment may also include other hardware depending on its actual functions; these will not be elaborated further.

[0084] The specific implementation process of the functions and roles of each unit in the above device can be found in the implementation process of the corresponding steps in the above method, and will not be repeated here.

[0085] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. 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 the present invention according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0086] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the phytoplankton monitoring scheme design method based on a multivariate time series model as described in the above embodiments.

[0087] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device, such as a plug-in hard disk, smart media card (SMC), SD card, flash card, etc., equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units of any data processing device and external storage devices. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0088] It will be understood by those skilled in the art that the above descriptions are merely preferred examples of the invention and are not intended to limit the invention. Although the invention has been described in detail with reference to the foregoing examples, those skilled in the art can still modify the technical solutions described in the foregoing examples or make equivalent substitutions for some of the technical features. All modifications and equivalent substitutions made within the spirit and principles of the invention should be included within the scope of protection of the invention.

Claims

1. A method for designing a phytoplankton monitoring scheme based on a multivariate time series model, characterized in that, Includes the following steps: S1: Divide all marine monitoring samples into training and testing sets according to their chronological order; S2: The correlation between each variable value and the phytoplankton density label is calculated using multiple different variable filtering methods. Based on this, all variable values ​​are sorted. Then, based on the sorting results, several variable values ​​that rank low in each variable filtering method are filtered out, thus obtaining multiple different variable sets. S3: Using the multiple variable sets and the full variable set obtained in S2, train a multivariate long short-term memory network model and a multivariate time series model containing an attention prototype network, respectively. Compare the prediction performance and training time of the two models on the multiple variable sets and the full variable set, and select the model with the best prediction performance and training time and its corresponding variable set. S4: Under the constraints of project budget, existing monitoring foundation and management priorities, the prediction results obtained from the optimal model obtained in S3 and its corresponding variable set are used to generate a phytoplankton monitoring plan for a future period of time. In S2, the multivariate time series model containing the attention prototype network includes branch one, which consists of a random dimension permutation module, a local variable extraction module, two high-order local modules with shared parameters, and an average pooling layer one; and branch two, which consists of an LSTM model and an average pooling layer two. It also includes a splicing layer connecting branch one and branch two, two fully connected layers, an embedding layer, an attention prototype network module, a gated multiplication module, a similarity transformation module, and a normalized exponential function module. The local variable extraction module includes multiple local variable extraction sub-modules consisting of one-dimensional convolutional layers, batch normalization layers, and rectified functions with leakage, which extract local variables from the corresponding multivariate time series.

2. The method for designing a phytoplankton monitoring scheme based on a multivariate time series model according to claim 1, characterized in that, In S2, the multivariate long short-term memory network model includes an input layer, a hidden layer, and an output layer.

3. The method for designing a phytoplankton monitoring scheme based on a multivariate time series model according to claim 1, characterized in that, The marine monitoring samples in S1 include variables such as silicate, phosphate, nitrate, salinity, sea surface temperature, ocean currents, and dissolved oxygen.

4. The method for designing a phytoplankton monitoring scheme based on a multivariate time series model according to claim 1, characterized in that, The variable filtering methods include mutual information method, Pearson correlation coefficient method, and variable importance ranking method based on random forest model.

5. The method for designing a phytoplankton monitoring scheme based on a multivariate time series model according to claim 4, characterized in that, In S2, the last two variable values ​​in each variable filtering method are filtered out.

6. A device for designing a phytoplankton monitoring scheme based on a multivariate time series model, characterized in that, One or more processors are used to implement the phytoplankton monitoring scheme design method based on a multivariate time series model as described in any one of claims 1 to 5.

7. An electronic device, characterized in that, include: One or more processors; A storage device for storing one or more programs, which, when executed by the electronic device, cause the electronic device to implement the phytoplankton monitoring scheme design method based on a multivariate time series model as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, It stores a program that, when executed by a processor, implements the phytoplankton monitoring scheme design method based on a multivariate time series model as described in any one of claims 1 to 5.

Images