Prediction system and method for partial pressure of carbon dioxide in sea area

Abstract

The invention relates to a sea area carbon dioxide partial pressure prediction system and method. The system comprises a historical data acquisition module used for acquiring target carbon dioxide partial pressure data of a target sea area; the annual time sequence prediction module is used for inputting the target carbon dioxide partial pressure data into a pre-trained prediction model to obtain predicted annual carbon dioxide partial pressure data of the target sea area, the prediction model is obtained through training of a training set, and the training set comprises historical carbon dioxide partial pressure data of the target sea area in several years; the prediction model is constructed by adopting a back propagation neural network; the monthly spatial distribution prediction module is used for extracting monthly spatial distribution of the carbon dioxide partial pressure data of the target sea area in the prediction year based on the carbon dioxide partial pressure data in the prediction year; and the visualization module is used for visually displaying the prediction year carbon dioxide partial pressure data and the monthly space distribution condition of the prediction year carbon dioxide partial pressure data. According to the invention, accurate prediction of carbon dioxide partial pressure can be realized.

Description

Technical Field

[0001] The present invention relates to the technical field of carbon dioxide partial pressure prediction, and particularly to a prediction system and method for carbon dioxide partial pressure in sea areas. Background Art

[0002] The partial pressure of carbon dioxide is a key parameter describing the concentration of carbon dioxide in seawater, which directly reflects the ability of seawater to absorb carbon dioxide in the atmosphere and its changing trend. Existing prediction systems often rely on a large amount of historical data, and the complexity of steps such as data cleaning, preprocessing, and feature extraction is extremely high, resulting in low prediction efficiency; moreover, traditional prediction methods often fail to achieve ideal prediction accuracy; in addition, the prediction results of the partial pressure of carbon dioxide are usually presented in the form of time series, lacking an intuitive display of spatial distribution. Therefore, the present invention provides a prediction system and method for carbon dioxide partial pressure in sea areas. Summary of the Invention

[0003] The purpose of the present invention is to provide a prediction system and method for carbon dioxide partial pressure in sea areas. Through the fitting of a backpropagation neural network and external calls, the non-linear characteristics of data can be accurately captured, the complexity of data processing can be reduced, and the prediction efficiency can be improved.

[0004] To achieve the above purpose, the present invention provides the following solutions:

[0005] A prediction system for carbon dioxide partial pressure in sea areas, comprising: a historical data acquisition module, an annual time series prediction module, and a monthly spatial distribution prediction module;

[0006] The historical data acquisition module is used to acquire target carbon dioxide partial pressure data of a target sea area, where the target carbon dioxide partial pressure data is the carbon dioxide partial pressure data of the year before the prediction year;

[0007] The annual time series prediction module is used to input the target carbon dioxide partial pressure data into a pre-trained prediction model to obtain the predicted carbon dioxide partial pressure data of the target sea area in the prediction year. The prediction model is obtained by training with a training set, and the training set includes the historical carbon dioxide partial pressure data of the target sea area for several years. The prediction model is constructed using a backpropagation neural network;

[0008] The monthly spatial distribution prediction module is used to extract the monthly spatial distribution of the carbon dioxide partial pressure data of the target sea area in the prediction year based on the predicted carbon dioxide partial pressure data of the prediction year;

[0009] The visualization module is used to visually display the predicted carbon dioxide partial pressure data of the prediction year and the monthly spatial distribution of the predicted carbon dioxide partial pressure data of the prediction year.

[0010] Optionally, the training set includes an input set and an output set. Among them, the input set includes carbon dioxide partial pressure data from year n to year m, and the output set includes carbon dioxide partial pressure data from year n + 1 to year m + 1. Both the input set and the output set are in the form of space × time.

[0011] Optionally, before training the prediction model with the training set, it further includes preprocessing the training set. The preprocessing includes replacing the NaN values in the training set with 0.

[0012] Optionally, after obtaining the predicted-year carbon dioxide partial pressure data of the target sea area, it further includes integrating the predicted-year carbon dioxide partial pressure data into the size of longitude × latitude × 12 months.

[0013] Optionally, the system is constructed using the MATLAB Appdesigner tool, and the prediction model is in the form of external call.

[0014] To further achieve the above object, the present invention also provides a prediction method for the carbon dioxide partial pressure in a sea area, including:

[0015] Obtain the target carbon dioxide partial pressure data of the target sea area, where the target carbon dioxide partial pressure data is the carbon dioxide partial pressure data of the year before the predicted year;

[0016] Input the target carbon dioxide partial pressure data into a pre-trained prediction model to obtain the predicted-year carbon dioxide partial pressure data of the target sea area. Among them, the prediction model is obtained by training with a training set, and the training set includes the historical carbon dioxide partial pressure data of the target sea area for several years. The prediction model is constructed using a backpropagation neural network;

[0017] Extract the monthly spatial distribution of the carbon dioxide partial pressure data of the predicted year of the target sea area based on the predicted-year carbon dioxide partial pressure data.

[0018] Optionally, the training set includes an input set and an output set. Among them, the input set includes carbon dioxide partial pressure data from year n to year m, and the output set includes carbon dioxide partial pressure data from year n + 1 to year m + 1. Both the input set and the output set are in the form of space × time.

[0019] Optionally, before training the prediction model with the training set, it further includes preprocessing the training set. The preprocessing includes replacing the NaN values in the training set with 0.

[0020] The beneficial effects of the present invention are:

[0021] Through the fitting method of the backpropagation neural network, the present invention can accurately capture the non-linear characteristics of data; and by externally calling the neural network, it reduces the data processing complexity, optimizes the data processing process, improves the prediction efficiency, and reduces the R & D and maintenance costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0023] Figure 1 Schematic diagram of the interface structure of a prediction system for the partial pressure of carbon dioxide in the sea area according to an embodiment of the present invention;

[0024] Figure 2 Time series prediction of the partial pressure of carbon dioxide in Zhejiang Province in 2014 according to an embodiment of the present invention;

[0025] Figure 3 Spatial distribution of the predicted partial pressure of carbon dioxide in Zhejiang Province in June 2014 and the actual spatial distribution according to an embodiment of the present invention;

[0026] Figure 4 Spatial distribution error of the partial pressure of carbon monoxide in Zhejiang Province in June 2014 according to an embodiment of the present invention;

[0027] Figure 5 Distribution of the partial pressure of carbon dioxide in Zhejiang Province in July 2022 according to an embodiment of the present invention;

[0028] Figure 6 Schematic diagram of the working process of a prediction system for the partial pressure of carbon dioxide in the sea area according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0030] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0031] This embodiment provides a prediction system for the partial pressure of carbon dioxide in the sea area, including: a historical data acquisition module, an annual time series prediction module, and a monthly spatial distribution prediction module;

[0032] The historical data acquisition module is used to acquire the target partial pressure of carbon dioxide data of the target sea area, where the target partial pressure of carbon dioxide data is the partial pressure of carbon dioxide data of the year before the prediction year;

[0033] The annual time series prediction module is used to input the target partial pressure of carbon dioxide data into a pre-trained prediction model to predict the future annual partial pressure of carbon dioxide data of the target sea area. The prediction model is obtained by training with a training set, and the training set includes the historical partial pressure of carbon dioxide data of the target sea area for several years. The prediction model is constructed using a backpropagation neural network;

[0034] The monthly spatial distribution prediction module is used to extract the monthly spatial distribution of the partial pressure of carbon dioxide data of the prediction year of the target sea area based on the future annual partial pressure of carbon dioxide data;

[0035] The visualization module is used to visually display the partial pressure of carbon dioxide data of the prediction year and the monthly spatial distribution of the partial pressure of carbon dioxide data of the prediction year.

[0036] Specifically, this embodiment can accurately capture the non-linear characteristics of the data through the fitting method of the backpropagation neural network; and by externally calling the neural network, it reduces the data processing complexity, optimizes the data processing process, improves the prediction efficiency, and reduces the R & D and maintenance costs.

[0037] Further, the training set includes an input set and an output set. Among them, the input set includes the partial pressure of carbon dioxide data from year n to year m, and the output set includes the partial pressure of carbon dioxide data from year n + 1 to year m + 1. Both the input set and the output set are in the form of space × time.

[0038] Further, before training the prediction model with the training set, it also includes preprocessing the training set. The preprocessing includes replacing the NaN values in the training set with 0.

[0039] Further, after obtaining the partial pressure of carbon dioxide data of the prediction year of the target sea area, it also includes integrating the partial pressure of carbon dioxide data of the prediction year into the size of longitude × latitude × 12 months.

[0040] Further, the system is constructed using the MATLAB Appdesigner tool, and the prediction model is in the form of external call.

[0041] Taking the prediction of the partial pressure of carbon dioxide in the sea area of Zhejiang Province as an example, a prediction system for the partial pressure of carbon dioxide in the sea area proposed in this embodiment is described in detail as follows:

[0042] (1) Construct a dataset and train a backpropagation neural network;

[0043] In recent years, the rapid development of machine learning and neural networks has enabled scientific researchers to better analyze the dynamic changes of marine ecosystems. Due to its powerful non-linear fitting ability and self-learning ability, the neural network method has been widely used in various non-linear predictions. Based on the backpropagation neural network method, this embodiment trains a dataset for 23 years, uses the carbon dioxide partial pressure situation of the previous year to predict the carbon dioxide partial pressure situation of the next year, and accurately captures the laws and spatial distribution characteristics of the carbon dioxide partial pressure in the waters of Zhejiang Province.

[0044] 1. Data description: The data comes from the monthly satellite observation data of carbon dioxide partial pressure of Copernicus (https: / / data.marine.copernicus.eu / products). The selected time range is from 1998 to 2022; the spatial range is longitude [119.625, 124.875] and latitude [25.875, 31.125], and the grid size is 22×22.

[0045] 2. Process the dataset: This embodiment mainly predicts the distribution of carbon dioxide partial pressure in Zhejiang Province from January to December 2022. Therefore, the carbon dioxide partial pressure data from 1998 to 2020 is used as the input dataset, and the size of this input set is 22×22×12×23. The carbon dioxide partial pressure data from 1999 to 2021 is used as the output dataset, and the size of this output set is 22×22×12×23. Training is carried out with this dataset. Among them, 22×22 is the spatial distribution, and 12×23 is the time dimension of monthly data and annual data. For the convenience of training, it is integrated into a spatial×time dataset, and the dataset size is 484×276.

[0046] 3. Train the neural network: The dataset size is 484×276, including 484 feature points and 276 samples. It should be noted that there should be no NaN values in the training of the backpropagation neural network. Therefore, the NaN values corresponding to land areas should be excluded. In this embodiment, all NaN values are replaced with 0. In Matlab, use the neural network fitting tool 14.4 for training and output the trained neural network.

[0047] (2) Build a system using the MATLAB Appdesigner tool;

[0048] This embodiment comprehensively and accurately grasps the dynamic changes of the carbon dioxide partial pressure in the waters of Zhejiang Province through two parts: annual time series prediction and monthly spatial distribution prediction, providing a scientific basis for marine environmental protection and response to climate change.

[0049] The annual time series prediction of the partial pressure of carbon dioxide in the sea area of Zhejiang Province refers to an overview of the prediction situation for the whole year after the selected year. This prediction is based on the trained backpropagation neural network, which inputs the data of the previous year to predict the situation of this year. For example, if the predicted year is 1999, the background dataset will select the data of 1998 and input it into the trained backpropagation neural network. The output data is 1999. After being processed in the form of mean and root mean square error, it is expressed in the form of a time series. By combining and viewing, we can more comprehensively understand the overall prediction situation of that year, evaluate the accuracy and stability of the prediction, and provide support for decision-making. The monthly spatial distribution prediction of the partial pressure of carbon dioxide in the sea area of Zhejiang Province refers to visualizing and comparing the predicted situation of the monthly average spatial distribution with the actual situation after selecting the year and month, and calculating the error between the two as an evaluation index, including RMSE (root mean square error), MAE (mean absolute error), MAPE (mean absolute percentage error), and R 2 (coefficient of determination), and realizing a monthly performance evaluation through the prediction situation combining specific values and spatial distribution. The specific platform construction is as Figure 1 shown.

[0050] According to the platform construction, for the annual time series prediction module and the monthly spatial distribution prediction module in this embodiment, one operation logic is to first perform annual prediction, that is, read the data on the "year regulator", such as 1999. Then, in the dataset loaded by the code, the data of 1998 will be selected and input into the backpropagation neural network for prediction. The predicted data will be divided into 12 months, and the mean and root mean square error of each month will be calculated as the content of the annual time series prediction module. Then, read the data on the "month regulator", such as January. Then, the first group of data will be selected from the just-predicted 12 groups of data, that is, the predicted data for January, and then read the actual data for January for plotting and calculation.

[0051] After pressing the "Predict" button, each module will start running. Therefore, only a callback function needs to be added to the code part of the button, call the name of a certain coordinate axis or a certain value, and plot or update the value under this coordinate axis to realize the prediction or simulation process.

[0052] (1) Annual time series prediction module:

[0053] The annual time series prediction module only has two parts of code logic. One is the mean value, and the other is the root mean square error. As long as the year selected for prediction in each run is read in the "Year Regulator", first drag in two year labels and place them at the titles of two panels respectively. After each selection, the selected year will be updated to achieve an intuitive effect. Then, input the data of the previous year of the selected year into the external neural network, and calculate the average and root mean square error of the predicted data and the actual situation for 12 months respectively. Note that the predicted data and the actual data except for NaN are required. Plot the line chart of the mean value on the corresponding coordinate axis, and plot the bar chart of the root mean square error on the corresponding coordinate axis. This embodiment predicts the partial pressure of carbon dioxide in the sea area of Zhejiang Province in 2014. As Figure 2 shown, it is the time series prediction situation in 2014.

[0054] (2) Monthly spatial distribution prediction module:

[0055] The monthly spatial distribution prediction module also has two parts of code logic. One is the spatial distribution, and the other is the error situation. Similar to the annual prediction, the month selected for prediction in each run needs to be read in the "Month Regulator". Drag in two month labels and place them at the titles of two panels respectively. After each selection, the selected month will be updated. Input the whole-year dataset for prediction outside the neural network, extract the spatial distribution of the corresponding month in the whole year after prediction, change the output predicted data to a spatial dimension matrix, and plot it on the corresponding coordinate axis. Read the actual data of the corresponding year and month and plot it on the corresponding coordinate axis. As Figure 3 shown, it is the predicted spatial distribution situation and the actual spatial distribution situation of the partial pressure of carbon dioxide in June 2014.

[0056] For the error situation, the error value is calculated by comparing the predicted monthly and annual spatial distribution data with the actual monthly and annual spatial distribution data, including RMSE (root mean square error), MAE (mean absolute error), MAPE (mean absolute percentage error), and R 2 (coefficient of determination). Input the four data into their respective numerical texts respectively. As Figure 4 shown, it is the error situation of the spatial distribution of the partial pressure of carbon dioxide in June 2014.

[0057] Generally speaking, the production of this system is mainly based on the MATLAB Appdesigner tool, and a more strict dataset division method is adopted for training. It not only realizes the independence between the training set and the validation set, but also can make full use of historical data, avoiding data leakage and overfitting problems. The R of the overall predicted data 2The values all reached above 0.9 in 2022, improving the accuracy of carbon dioxide partial pressure prediction; and calling the neural network externally to build a simple prediction system, reducing the prediction cost and improving the prediction efficiency.

[0058] As an implementable mode, in addition to the backpropagation neural network for constructing the prediction model, such as neural network models like radial basis neural network, long short-term memory network, etc., principal component analysis, support vector machine, random forest, etc. can also be adopted. Although they are different in structure, they can all be trained to fit historical data for predicting the carbon dioxide partial pressure.

[0059] This embodiment predicts the distribution of carbon dioxide partial pressure in Zhejiang Province from January to December in 2022. Taking the prediction situation in July 2022 as an example, see Figure 5 , the overall annual trend of the predicted carbon dioxide partial pressure in the sea area of Zhejiang Province in 2022 is generally consistent with the actual annual trend, but there is a phenomenon of being on the high side in the average values of individual months and years. Taking July as an example for the spatial distribution, its predicted distribution situation is also basically consistent with the actual spatial distribution situation. The root mean square error of the prediction for this month is 11.3935, the mean absolute error is 6.8278, the mean absolute percentage error is 3.2518%, and the coefficient of determination reaches 0.9946, which is in good agreement with the true distribution situation.

[0060] As Figure 1 and Figure 6 shown, the system interface structure and working process are described as follows: In the left side, the predicted years that can be selected are from 1999 to 2021 (training set) and 2022 (test set), the prediction range is 12 months, and after prediction, the root mean square error, mean absolute error, mean absolute percentage error, and coefficient of determination of the predicted month and year will be displayed; on the right side is the visualization interface of the carbon dioxide partial pressure in the sea area of Zhejiang Province, where the prediction situation can be visually compared with the real situation.

[0061] To further optimize the technical solution, this embodiment also provides a method for predicting the carbon dioxide partial pressure in the sea area, including:

[0062] Obtain the target carbon dioxide partial pressure data of the target sea area, where the target carbon dioxide partial pressure data is the carbon dioxide partial pressure data of the year before the predicted year;

[0063] Input the target carbon dioxide partial pressure data into the pre-trained prediction model to obtain the predicted carbon dioxide partial pressure data of the target sea area in the predicted year, where the prediction model is obtained through training with a training set, the training set includes the historical carbon dioxide partial pressure data of the target sea area for several years, and the prediction model is constructed using a backpropagation neural network;

[0064] Extract the monthly spatial distribution of the predicted annual carbon dioxide partial pressure data in the target sea area based on the predicted annual carbon dioxide partial pressure data.

[0065] Further, the training set includes an input set and an output set. Among them, the input set includes the carbon dioxide partial pressure data from year n to year m, and the output set includes the carbon dioxide partial pressure data from year n + 1 to year m + 1. Both the input set and the output set are in the form of space × time.

[0066] Further, before training the prediction model with the training set, it also includes preprocessing the training set. The preprocessing includes replacing the NaN values in the training set with 0.

[0067] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention should fall within the protection scope determined by the claims of the present invention.

Claims

1. A prediction system for the partial pressure of carbon dioxide in the sea, characterized in that: include: Historical data acquisition module, annual time series prediction module, monthly spatial distribution prediction module, and visualization module; The historical data acquisition module is used to acquire target carbon dioxide partial pressure data of the target sea area, wherein the target carbon dioxide partial pressure data is the carbon dioxide partial pressure data of the year before the forecast year; The annual time series prediction module is used to input the target carbon dioxide partial pressure data into a pre-trained prediction model to obtain the predicted annual carbon dioxide partial pressure data of the target sea area, wherein the prediction model is obtained by training a training set, the training set includes the historical carbon dioxide partial pressure data of the target sea area for several years, and the prediction model is constructed using a back propagation neural network; The monthly spatial distribution prediction module is used to extract the monthly spatial distribution of the predicted annual carbon dioxide partial pressure data of the target sea area based on the predicted annual carbon dioxide partial pressure data; The visualization module is used to visualize the predicted annual carbon dioxide partial pressure data and the monthly spatial distribution of the predicted annual carbon dioxide partial pressure data.

2. The prediction system for the partial pressure of carbon dioxide in the sea area according to claim 1, characterized in that: The training set includes an input set and an output set, wherein the input set includes carbon dioxide partial pressure data from n to m years, and the output set includes carbon dioxide partial pressure data from n+1 to m+1 years, and both the input set and the output set are in the form of space×time.

3. The prediction system for the partial pressure of carbon dioxide in the sea area according to claim 2, characterized in that: Before training the prediction model through the training set, the training set is also preprocessed, and the preprocessing includes: replacing NaN values ​​in the training set with 0.

4. The prediction system for the partial pressure of carbon dioxide in the sea area according to claim 1, characterized in that: After obtaining the predicted annual carbon dioxide partial pressure data of the target sea area, the method further includes integrating the predicted annual carbon dioxide partial pressure data into a size of longitude×latitude×12 months.

5. The prediction system for the partial pressure of carbon dioxide in the sea area according to claim 1, characterized in that: The system is constructed using MATLAB Appdesigner tools, and the prediction model is in the form of external calls.

6. A method for predicting the partial pressure of carbon dioxide in sea areas, characterized in that: include: Obtaining target carbon dioxide partial pressure data of the target sea area, wherein the target carbon dioxide partial pressure data is the carbon dioxide partial pressure data of the year before the forecast year; Inputting the target carbon dioxide partial pressure data into a pre-trained prediction model to obtain the predicted annual carbon dioxide partial pressure data of the target sea area, wherein the prediction model is obtained by training a training set, the training set includes several years of historical carbon dioxide partial pressure data of the target sea area, and the prediction model is constructed using a back propagation neural network; The monthly spatial distribution of the carbon dioxide partial pressure data for the predicted year of the target sea area is extracted based on the carbon dioxide partial pressure data for the predicted year.

7. The method for predicting the partial pressure of carbon dioxide in sea areas according to claim 6, characterized in that: The training set includes an input set and an output set, wherein the input set includes carbon dioxide partial pressure data from n to m years, and the output set includes carbon dioxide partial pressure data from n+1 to m+1 years, and both the input set and the output set are in the form of space×time.

8. The method for predicting the partial pressure of carbon dioxide in sea areas according to claim 7, characterized in that: Before training the prediction model through the training set, the training set is also preprocessed, and the preprocessing includes: replacing NaN values ​​in the training set with 0.

Images