Ship CO2 emission prediction method

Abstract

The invention belongs to the technical field of machine learning, and discloses a ship CO2 emission prediction method, which comprises the following seven steps: data acquisition and recording, data cleaning and preprocessing, data alignment, machine learning model construction, transfer learning and model adaptation, model verification and evaluation, and model iteration and optimization. According to the method, the CO2 emission prediction model of the single ship is constructed through machine learning according to the actually measured data, and the model is used as the pre-training model, so that the accuracy of CO2 emission prediction can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine learning, and in particular relates to a method for predicting ship CO2 emissions by using a machine learning algorithm and modeling with measured data. Background Art

[0002] In the context of economic globalization, the shipping industry plays a vital role. As one of the main means of transportation for trade and communication between countries in the world, ships not only carry goods, but also serve as a link between different countries and regions. However, with the rapid development of the shipping industry, the issue of carbon emissions from ships has gradually become a focus of global attention. The transportation activities of ships are inevitably accompanied by a large amount of carbon emissions. As one of the main greenhouse gases, CO2 not only has a direct impact on climate change, but also poses a threat to marine ecosystems and human health. However, there are still many challenges to achieve effective control and reduction of ship carbon emissions. The current monitoring methods of ship carbon emissions have limitations. Conventional statistical models often have difficulty in accurately predicting the emission level of ships, and are limited by data quality and timeliness. As a powerful data analysis tool, machine learning algorithms have shown amazing application prospects in various fields. With its adaptability, processing power and intelligence, machine learning algorithms have brought new possibilities for ship carbon emission prediction. It can effectively process complex data structures, automatically discover patterns, and achieve accurate predictions, which is expected to provide important support for environmental protection and sustainable development in the shipping industry. Therefore, it is necessary to propose a new method to establish an efficient and accurate ship carbon emission prediction model by combining machine learning algorithms and measured data to meet the urgent needs of carbon emission control in the current shipping industry.

[0003] Patent CN202310645870.3 discloses a method for predicting ship carbon emission characteristics based on the dynamic method and attention mechanism. The patent adopts the ship dynamic method, and uses AIS data, ship Lloyd's data, emission factor parameters and geographic information to calculate the carbon emissions of the main engine, auxiliary engine and boiler of the ship in different navigation conditions and different port equipment. The patent uses empirical emission factors rather than measured values ​​to calculate carbon emissions, and the accurate value of the calculated carbon emissions cannot be guaranteed. The carbon emission characteristics analyzed in the patent are a macro analysis of the carbon emission laws between ships, rather than the carbon emission laws of a single ship under different working conditions.

[0004] Patent CN202311412061.4 discloses a method and device for calculating the ship pollutant emission flux for a specific time and space. The patent obtains the pollutant emission factor, STSD and AIS data of the ship pollutants, and calculates the ship emission flux within a given time and space range of specific pollutants. However, the patent uses empirical emission factors rather than measured values ​​to calculate ship pollutants, and only refers to the time interval and some ship parameters for emission calculation, so the calculated value of pollutant emissions cannot be guaranteed to be accurate.

[0005] Patent CN202310874620.7 discloses a ship emission inventory prediction method and system based on AIS data. The patent establishes a ship activity level database based on the ship's AIS data and calculates the ship emission inventory in combination with the emission factor. The patent also calculates the emission inventory without measured emission factors. The formula used to calculate the emission inventory is an empirical formula, which limits the accuracy of the calculated emission values.

[0006] The ship emission value calculation method of the above patent takes into account the ship's emission factor and calculates the ship's emission value through time intervals and ship parameters. However, there is no measured ship operation data, and the accuracy of the calculation results obtained using empirical emission factors cannot be guaranteed. At the same time, the ship's emissions are calculated only based on the ship's emission factor, time, and ship parameters. This method ignores many details in the operation of the ship, resulting in further reduction in accuracy. The accuracy of emission calculations can be improved by modeling the measured data through machine learning. At the same time, the transfer learning method can effectively predict the target variable under different feature variables. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a modeling flow chart of a method for predicting CO2 emissions from ships.

[0008] Figure 2 It is a model diagram of the CO2 emission prediction method based on the artificial neural network algorithm of the present invention.

[0009] Figure 3 It is a model diagram of the CO2 emission prediction method based on artificial neural network algorithm and measured data of the present invention. Summary of the invention

[0010] In order to overcome the technical problems in the background technology, the present invention provides a method for predicting CO2 emissions of a ship, comprising the following steps:

[0011] S1: Data collection and recording: The vehicle-mounted gaseous emission tester is used to measure and record the instantaneous value of the ship's CO2 emissions in real time as the target variable, and the engine speed, engine power, exhaust flow, fuel consumption and ship speed at the same time point are simultaneously recorded as characteristic variables.

[0012] S2: Data cleaning and preprocessing: Detect and process missing values ​​in the collected data, identify and process outliers, and ensure the integrity and accuracy of the data; at the same time, format the data, including standardization of date and time formats and data type conversion, to meet the needs of subsequent analysis.

[0013] S3: Data alignment: Use Python to accurately align the feature variables and target variables at the same time, ensuring that each set of data contains complete and corresponding feature variables and target variable values, providing a basis for building machine learning models.

[0014] S4: Machine learning model construction: Based on the artificial neural network algorithm, the data processed in step S3 is used as training data, where engine speed, engine power, exhaust flow, fuel consumption, and ship speed are used as input feature variables, and CO2 emissions are used as output target variables. By adjusting the model parameters, the model's explained variance score, root mean square error, and determination coefficient are optimized until the best prediction performance is achieved, thereby constructing an efficient CO2 emissions prediction model.

[0015] S5: Transfer learning and model adaptation: Use the prediction model constructed in step S4 as a pre-trained model, import a new data set containing only two characteristic variables: ship speed and fuel consumption, and use transfer learning technology to fine-tune the new model while keeping most of the model structure unchanged to adapt to the new combination of characteristic variables, thereby improving the prediction accuracy and efficiency of the model on the new data set.

[0016] S6: Model verification and evaluation: The constructed prediction model is comprehensively verified and evaluated using a variety of indicators such as mean square error, linear correlation coefficient, explained variance, etc. to ensure the accuracy and effectiveness of the model in predicting CO2 emissions; at the same time, a cross-validation method is used to improve the robustness and reliability of the evaluation results.

[0017] S7: Model iteration and optimization: Based on the verification and evaluation results, continuously analyze the prediction error, adjust the model parameters, and continuously optimize the model through an iterative process to further improve the prediction accuracy and reliability.

[0018] Furthermore, in the step S2 of data cleaning and preprocessing, the processing of outliers includes but is not limited to deletion, replacement with normal values, or estimation using interpolation method.

[0019] Furthermore, the S4 machine learning model construction step also includes optimization and adjustment of the neural network structure and training parameters to improve the learning and generalization capabilities of the model.

[0020] Furthermore, the S5 transfer learning and model adaptation step also includes necessary preprocessing of the new dataset to ensure that its format is compatible with the pre-trained model, and monitors performance changes during the transfer learning process to adjust the learning strategy and parameters in a timely manner.

[0021] Furthermore, in the S6 model validation and evaluation step, the cross-validation method used includes but is not limited to K-fold cross-validation to ensure that the evaluation results can fully reflect the performance of the model on different data subsets.

[0022] Compared with the prior art, the present invention has the following advantages:

[0023] The present invention is driven by a large amount of measured data and machine learning algorithms. First, the measured CO2 emissions rather than the empirical CO2 emission factor value ensures the accuracy of the calculation. Secondly, the machine learning algorithm can automatically learn and optimize the prediction model, so that the prediction results are more intelligent and reliable, thereby achieving more accurate predictions. The present invention additionally considers the impact of engine-related parameters and environmental conditions on emissions, and predicts through measured input characteristic variables. The machine learning algorithm can better capture the impact of various engine parameters on carbon emissions during ship operation, thereby making the prediction of emissions more accurate. DETAILED DESCRIPTION

[0024] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0025] like Figure 1 As shown, a method for predicting CO2 emissions of a ship comprises the following steps:

[0026] S1: Data collection and recording: The vehicle-mounted gaseous emission tester is used to measure and record the instantaneous value of the ship's CO2 emissions in real time as the target variable, and the engine speed, engine power, exhaust flow, fuel consumption and ship speed at the same time point are simultaneously recorded as characteristic variables. The relationship between the characteristic variable and the target variable is as shown in formula (1):

[0027] E(CO2)=f(p,n,v,F) (1)

[0028] Among them, E(CO2) is CO2 emission, p is engine power, n is engine speed, v is ship speed, and F is ship fuel consumption.

[0029] S2: Data cleaning and preprocessing: Detect and process missing values ​​in the collected data, identify and process outliers, and ensure the integrity and accuracy of the data; at the same time, format the data, including standardization of date and time formats and data type conversion, to meet the needs of subsequent analysis.

[0030] S3: Data alignment: Use Python to accurately align the feature variables and target variables at the same time, ensuring that each set of data contains complete and corresponding feature variables and target variable values, providing a basis for building machine learning models.

[0031] S4: Machine learning model construction: Based on the artificial neural network algorithm, the data processed in step S3 is used as training data, where engine speed, engine power, exhaust flow, fuel consumption, and ship speed are used as input feature variables, and CO2 emissions are used as output target variables. By adjusting the model parameters, the explained variance score, root mean square error, and determination coefficient of the model are optimized until the best prediction performance is achieved, thereby constructing an efficient CO2 emissions prediction model. Appropriately adjust the relevant parameters and evaluate the model to obtain:

[0032] Explained variance score: 0.9999061185981677;

[0033] Root mean square error (RMSE): 2.6092191854044278;

[0034] Mean square error (MSE): 6.808024757482545;

[0035] Coefficient of determination (R^2): 0.9999060978828336.

[0036] S5: Transfer learning and model adaptation: The prediction model constructed in step S4 is used as a pre-trained model, and a new data set containing only two characteristic variables, ship speed and fuel consumption, is imported. Through transfer learning technology, the new model is fine-tuned on the basis of keeping most of the model structure unchanged to adapt to the new combination of characteristic variables, thereby improving the prediction accuracy and efficiency of the model on the new data set. The relationship between the characteristic variables and the target variables after transfer learning is as shown in formula (2):

[0037] E(CO2)=g(v,F) (2)

[0038] Fine-tune the pre-trained model to adapt it to a new model with only two feature variables, and evaluate the model to get:

[0039] Explained variance score: 0.9997816553810736;

[0040] Root mean square error (RMSE): 3.35276825230011;

[0041] Mean square error (MSE): 11.241054953631535;

[0042] Coefficient of determination (R^2): 0.9997805469861097.

[0043] S6: Model verification and evaluation: The constructed prediction model is comprehensively verified and evaluated using a variety of indicators such as mean square error, linear correlation coefficient, explained variance, etc. to ensure the accuracy and effectiveness of the model in predicting CO2 emissions; at the same time, a cross-validation method is used to improve the robustness and reliability of the evaluation results.

[0044] S7: Model iteration and optimization: Based on the verification and evaluation results, continuously analyze the prediction error, adjust the model parameters, and continuously optimize the model through an iterative process to further improve the prediction accuracy and reliability.

[0045] Preferably, in the step S2 of data cleaning and preprocessing, the processing of outliers includes but is not limited to deletion, replacement with normal values, or estimation using interpolation method.

[0046] Preferably, the S4 machine learning model construction step also includes optimizing and adjusting the neural network structure and training parameters to improve the learning and generalization capabilities of the model.

[0047] Preferably, the S5 transfer learning and model adaptation step also includes necessary preprocessing of the new data set to ensure that its format is compatible with the pre-trained model, and monitors performance changes during the transfer learning process to adjust the learning strategy and parameters in a timely manner.

[0048] Preferably, in the S6 model validation and evaluation step, the cross-validation method used includes but is not limited to K-fold cross-validation to ensure that the evaluation results can fully reflect the performance of the model on different data subsets.

[0049] Although embodiments of the present invention have been described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting CO2 emissions from a ship, characterized by: The following steps are involved: S1: Data collection and recording: The vehicle-mounted gaseous emission tester is used to measure and record the instantaneous value of the ship's CO2 emissions in real time as the target variable, and the engine speed, engine power, exhaust flow, fuel consumption and ship speed at the same time point are simultaneously recorded as characteristic variables; S2: Data cleaning and preprocessing: Detect and process missing values ​​in the collected data, identify and process outliers, and ensure the integrity and accuracy of the data; at the same time, format the data, including standardization of date and time formats and data type conversion, to meet the needs of subsequent analysis; S3: Data alignment: Use Python to accurately align feature variables and target variables at the same time, ensuring that each set of data contains complete and corresponding feature variables and target variable values, providing a basis for machine learning model construction; S4: Machine learning model construction: Based on the artificial neural network algorithm, the data processed in step S3 is used as training data, where engine speed, engine power, exhaust flow, fuel consumption, and ship speed are used as input feature variables, and CO2 emissions are used as output target variables. By adjusting the model parameters, the explained variance score, root mean square error, and determination coefficient of the model are optimized until the best prediction performance is achieved, thereby constructing an efficient CO2 emissions prediction model; S5: Transfer learning and model adaptation: The prediction model constructed in step S4 is used as a pre-trained model, and a new data set containing only two characteristic variables, namely, ship speed and fuel consumption, is imported. Through transfer learning technology, the new model is fine-tuned to adapt to the new combination of characteristic variables while keeping most of the model structure unchanged, thereby improving the prediction accuracy and efficiency of the model on the new data set; S6: Model verification and evaluation: Use multiple indicators such as mean square error, linear correlation coefficient, explained variance, etc. to comprehensively verify and evaluate the constructed prediction model to ensure the accuracy and effectiveness of the model in predicting CO2 emissions; at the same time, use cross-validation methods to improve the robustness and reliability of the evaluation results; S7: Model iteration and optimization: Based on the verification and evaluation results, continuously analyze the prediction error, adjust the model parameters, and continuously optimize the model through an iterative process to further improve the prediction accuracy and reliability.

2. The ship CO2 emission prediction method according to claim 1, characterized in that: In the S2 data cleaning and preprocessing step, the processing of outliers includes but is not limited to deletion, replacement with normal values, or estimation using interpolation method.

3. The ship CO2 emission prediction method according to claim 1 or 2, characterized in that: The S4 machine learning model construction step also includes optimizing and adjusting the neural network structure and training parameters to improve the learning and generalization capabilities of the model.

4. The method for predicting CO2 emissions from ships according to claim 1, characterized in that: The S5 transfer learning and model adaptation step also includes necessary preprocessing of the new dataset to ensure that its format is compatible with the pre-trained model, and monitoring performance changes during the transfer learning process to adjust the learning strategy and parameters in a timely manner.

5. The ship CO2 emission prediction method according to claim 1, characterized in that: In the S6 model validation and evaluation step, the cross-validation method used includes but is not limited to K-fold cross-validation to ensure that the evaluation results can fully reflect the performance of the model on different data subsets.

Images