A ship carbon emission feature prediction method based on a dynamic method and an attention mechanism

Abstract

The application discloses a ship carbon emission feature prediction method based on a dynamic method and an attention mechanism, adopts an AIS data-based dynamic method and a TFT time sequence prediction model based on the attention mechanism, utilizes ship AIS data, ship Lloyd's file data, pollutant emission factor parameters and geographic information, establishes a high-resolution ship carbon emission list, and realizes multi-period pollutant emission feature prediction of the ship. Through data cleaning, AIS data trajectory integration, data fusion and data completion and other operations, the AIS data is preprocessed, and the accuracy of the AIS data is ensured. According to the ship type and the ship running state, the dynamic method is used to estimate and analyze the pollutant emission amount, and the dynamic ship behavior is considered in real time when analyzing the carbon emission law and predicting the carbon emission feature.

Description

TECHNICAL FIELD

[0001] The present application relates to a ship pollutant emission estimation technology and a time series prediction technology, in particular to a ship carbon emission multi-feature prediction method based on a power method and an attention mechanism. BACKGROUND

[0002] With the shipping industry playing an increasingly important role in international trade activities and the global economy, the greenhouse gas emissions of the shipping industry have also increased significantly, especially in ports with heavy ship traffic, which has a negative impact on the marine environment and human health. The most effective way to reduce environmental pressure is to reduce pollutant emissions, improve treatment rates, and reduce energy consumption. Therefore, a detailed ship emission inventory should be developed for the waters near the port where ships are frequently active, and the emission characteristics in the future period should be predicted based on the past emission characteristics to master the characteristics of ship carbon emissions so that relevant departments can take effective management strategies to reduce ship emissions.

[0003] The power method based on AIS data is used by most scholars to estimate ship pollutant emissions, and the method is practical. However, this method requires a ship AIS system. Due to factors such as incomplete coverage of inland river AIS base stations, blocked ship uplink signals, and atmospheric radio interference of the AIS system, AIS data has problems such as missing, errors, and duplication. The estimation of the pollutant emission inventory of inland ships still has limitations. The ship atmospheric emission inventory is an important basis for preventing and controlling ship atmospheric pollution, and it is necessary to conduct localized research to make it suitable for the actual situation of ship emissions, and then establish an accurate ship emission inventory.

[0004] In addition, the ship emission inventory contains rich time series dynamic information, and time series can be used to predict dynamic evolution processes. In the past few decades, scientists have been working to develop specialized models that can capture the underlying patterns of time series to effectively infer the future. The development of attention mechanisms has improved long-term dependency learning, and the transformer architecture has achieved state-of-the-art performance in a variety of natural language processing applications. Recent work has also demonstrated the benefits of using attention mechanisms in time series prediction applications, which outperform comparable recurrent networks.

[0005] Based on the above problems and status, the present application adopts a power method based on AIS data and a TFT timing prediction model based on an attention mechanism, uses ship AIS data, ship Lloyd's profile data, pollutant emission factor parameters and geographic information, establishes a high-resolution ship carbon emission inventory, and realizes multi-period pollutant emission characteristic prediction of the ship. The present application pre-processes the AIS data through data cleaning, AIS data trajectory integration, data fusion and data completion, etc., to ensure the accuracy of the AIS data. The power method is used to estimate and analyze the pollutant emission according to the ship type and the ship running state, to ensure that the analysis of the carbon emission law and the prediction of the carbon emission characteristics consider the dynamic ship behavior in real time. The nationality, type and other attributes of the ship are associated to study the characteristics of the pollutant emission source in the region and the factors affecting the carbon emission from the aspects of the ship registration place, the ship type and other ship attributes and space-time, to master the origin and development of the ship carbon emission, thereby providing data support for the accurate carbon emission reduction policy and carbon trading of China. SUMMARY

[0006] In view of the current pressure faced by ship pollution control and the problems existing in the current method, the present application provides a ship carbon emission characteristic prediction method based on a power method and an attention mechanism, realizes the estimation of the ship carbon emission and the prediction of the future emission characteristics. The present application solves the calculation error problem caused by information missing, error, repetition, abnormality, etc. in the process of device use and data transmission and collection of the AIS data due to reasons such as sensor characteristics, signal interference, transmission channel congestion, etc., through data cleaning, AIS data trajectory integration, data fusion and data completion, etc., to ensure the accuracy of the subsequent calculation. Based on the processed data, a high-resolution ship carbon emission inventory of the ship is established by using the power method, and the data in the emission inventory is sorted into a timing sequence to realize the multi-period characteristic prediction of the ship pollutant emission.

[0007] The ship carbon emission feature prediction method based on the power method and the attention mechanism comprises four modules. The first module is an AIS data preprocessing model: in view of the problems of faults, errors, discontinuities and the like that may exist in the storage and management of AIS data and the problems of data loss, abnormality, brevity, unstable rate and the like that may exist in the process of multi-path incremental AIS data access, according to the demand of ship carbon emission tracing, in combination with ship Lloyd data and geographic information, data cleaning, AIS data trajectory integration, data fusion and data completion operations are performed on the data, thereby providing a high-quality data basis for subsequent calculation. The second module is the establishment of a ship high-resolution emission list: by using the processed ship AIS data and geographic information, the ship power method is adopted, and according to the pollutant emission factor, the ship operating state and other ship attributes, the carbon emissions of the ship main engine, auxiliary engine and boiler are calculated respectively, and a high-resolution carbon emission list that is specific to a single ship, an emission device and an emission time is established. The third module is the prediction of ship carbon emission features: the module extracts the data items related to the prediction target in the ship carbon emission list, establishes a ship carbon emission time sequence, and uses the TFT time series prediction model based on the attention mechanism to learn the relationship between the features, thereby realizing multi-period feature prediction of ship carbon emissions. The fourth module is the analysis of ship carbon emission feature rules: according to the established high-resolution ship carbon emission list and the prediction results of future carbon emission features, the ship carbon emission feature rules are analyzed from different dimensions such as time and space, ship ownership, ship type and navigation state, and the future emission features are mastered according to the prediction results, thereby providing data support for relevant departments to develop emission reduction measures.

[0008] To achieve the above object, the technical scheme adopted by the present application is as follows:

[0009] Step S1: Ship AIS data preprocessing

[0010] The present application uses ship AIS data to complete the calculation and prediction of ship carbon emissions. Due to the possible information loss, errors, repetition, abnormalities and the like of AIS data in the process of equipment use and data transmission and collection, interference is caused to subsequent data mining analysis. Therefore, the AIS data needs to be preprocessed.

[0011] Step S11: Data cleaning of AIS data

[0012] Abnormal data includes data anomalies, behavior anomalies and position anomalies. Among them, data anomalies include data items exceeding thresholds and missing, behavior anomalies include navigation state exceeding thresholds and track not conforming to common sense, and position anomalies include ships deviating from channels and exceeding the research scope. For abnormal data, the data needs to be removed or corrected in combination with application scenarios and theoretical basis.

[0013] Step S12: AIS data trajectory integration

[0014] Since AIS devices send signals every 3 seconds to several minutes, the data volume is very large, which is very inconvenient for subsequent calculation process, so it is necessary to calculate and integrate the massive AIS data. The track points of a ship in a complete activity in a device (channel / anchorage / berth) are integrated into a track, corresponding to a data, which is convenient for subsequent processing and calculation.

[0015] Step S13: AIS data fusion and completion

[0016] Combined with the ship's Lloyd's archive data, the AIS data is matched and fused through the MMSI code of the ship, so that the dynamic trajectory information in the AIS data and the static attributes of the ship are combined. For data not included in the Lloyd's database, the relationship formula calculation, statistics and average value calculation method are used to determine the final complete ship pollutant emission database in combination with the actual application scene and theoretical knowledge.

[0017] Step S2: Establishing a high-resolution ship carbon emission inventory

[0018] The application adopts a ship power method, 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 states and different port devices, thereby establishing a high-resolution ship carbon emission inventory.

[0019] Step S3: Ship carbon emission feature prediction model based on attention mechanism

[0020] Based on the ship carbon emission inventory established in step S2, the attributes related to the prediction target are extracted and arranged into time series data as the input of the prediction model. The ship carbon emission feature prediction model based on attention mechanism learns the relationship between different features in the time series data and the time dependence of the time series sequence, and makes a prediction of the future ship carbon emission based on the past ship carbon emission features.

[0021] Step S4: Ship carbon emission feature rule analysis

[0022] According to the established ship carbon emission inventory and the prediction of the ship carbon emission features, the ship nationality and type are associated and researched, and the ship carbon emission features are analyzed from different angles such as spatio-temporal emission rule, ship navigation state, emission component, etc. to master the characteristics of ship carbon emission, thereby providing strong data support and theoretical reference for relevant departments to formulate targeted emission reduction policies.

[0023] Compared with the prior art, the application has the beneficial effects that:

[0024] The present application guarantees the data quality of AIS data through data cleaning, trajectory integration, data fusion and data completion operations on ship AIS data, avoids the influence of subsequent calculation results caused by data quality problems. And the emission inventory established guarantees high granularity, high precision in time, space and emission sources, further improves the precision of the ship emission inventory, and masters the origin and development of ship carbon emissions and the driving factors affecting ship carbon emissions. In addition, the characteristic prediction of ship carbon emission situation also achieves relatively accurate effect, and the trend change of ship carbon emission is grasped in advance, which provides scientific decision basis for relevant departments to formulate targeted emission reduction measures. BRIEF DESCRIPTION OF DRAWINGS

[0025] The present application will be better understood from the following detailed description of the embodiments of the present application, taken in conjunction with the accompanying drawings, in which like reference numerals refer to like parts, wherein:

[0026] Figure 1 is a flowchart of the inventive method;

[0027] Figure 2 is a flowchart of AIS data preprocessing;

[0028] Figure 3 is a Temporal fusion transformer time sequence prediction model based on attention mechanism used in the present application. DETAILED DESCRIPTION

[0029] In view of the current pressure faced by ship pollution control, the present application provides a ship carbon emission characteristic prediction method based on dynamic method and attention mechanism, realizes the establishment of high-resolution ship carbon emission inventory and the prediction of future multi-period ship emission characteristics, analyzes the ship carbon emission characteristic law from different dimensions such as time and space, ship belonging place, ship type and navigation state, and judges the future emission trend according to the prediction result, so as to provide data support for relevant departments to develop emission reduction measures. The ship carbon emission characteristic prediction method based on dynamic method and attention mechanism provided by the present application is composed of four modules, which are ship AIS data preprocessing module, establishment of high-resolution ship carbon emission inventory module, ship carbon emission characteristic prediction model based on attention mechanism and ship carbon emission characteristic law analysis module.

[0030] The first step is the preprocessing of AIS data: Addressing potential issues such as faults, errors, and interruptions during AIS data storage and management, as well as data gaps, anomalies, incompleteness, and unstable rates during multi-channel incremental AIS data access, this module performs data cleaning, AIS data trajectory integration, data fusion, and data completion operations based on the needs of ship carbon emission tracing and practical application scenarios, combined with ship Lloyd's Register data and geographic information. This provides a high-quality data foundation for subsequent calculations. The second module is the establishment of a high-resolution ship emission inventory: Combining geographic information and employing the ship dynamics method, this module calculates the carbon emissions of the ship's main engine, auxiliary engines, and boilers based on pollutant emission factors, ship operating status, and other ship attributes, establishing a high-resolution ship carbon emission inventory specific to each ship, its equipment, and emission time. The third module is the prediction of ship carbon emission characteristics: This module extracts data items related to the prediction target from the ship carbon emission inventory, establishes a ship carbon emission time series, and uses a TFT time series prediction model based on an attention mechanism to learn the relationships between various features, achieving multi-time-period feature prediction of ship carbon emissions. The fourth module is the analysis of the characteristics and patterns of ship carbon emissions: Based on the established high-resolution ship carbon emission inventory and the prediction of future ship carbon emission characteristics, the module analyzes the characteristics and patterns of ship carbon emissions from different dimensions such as time and space, ship location, ship type, and navigation status, and grasps future emission trends based on the prediction results, providing data support for relevant departments to formulate emission reduction measures.

[0031] In this invention, such as Figure 1 As shown, a method for predicting ship carbon emission characteristics based on the dynamic method and attention mechanism includes the following steps:

[0032] Step S1: Pollen data preprocessing

[0033] This invention uses the NanoZoomer-SQ Slice Scanner to convert physical glass slides into digital whole-slide images (WSI). Because WSI images have extremely large pixels, they are not conducive to direct training and prediction of the model. Furthermore, pollen grains are very sparse and small in size, significantly disproportionate to the overall image. Therefore, it is necessary to perform layered cropping of the WSI images to obtain images of optimal size suitable for model input.

[0034] Step S1: Ship AIS data preprocessing

[0035] The present application uses ship AIS data to complete the calculation and prediction of carbon emissions. Due to the possible information missing, errors, repetition, abnormalities and other conditions in the equipment use and data transmission collection process of AIS data, it will interfere with the subsequent data mining analysis. Therefore, the AIS data needs to be preprocessed to ensure the accuracy of the data and the accuracy of the subsequent calculation. The data processing process is as shown in Figure 2

[0036] Step S11: Data cleaning of AIS data

[0037] Using big data analysis technology, the problems of faults, errors, interruptions and the like that may exist in the storage and management process of the massive inventory AIS data, and the problems of data missing, abnormalities, brevity, unstable rate and the like that may exist in the process of accessing multiple incremental AIS data, are combined with the actual application scene to achieve satisfactory analysis results.

[0038] According to the comparison between the real-time speed of the ship and the maximum speed of the ship, the abnormal speed of the ship is judged, if the real-time speed of the ship is greater than the maximum speed of the ship, the track point is removed; the ship heading angle range is 0 to 360 degrees, which is considered as abnormal heading angle and the track point is removed; the track point is removed if the MMSI number of the ship is not nine digits; the track point is removed if the ship type, ship name and other data fields exceed the threshold value. The track point with missing dynamic information data items such as time, track point position, running speed and heading angle is removed. If the ship track point position falls outside the Tianjin port range, it is considered as abnormal track point position and should be removed. In addition, according to the ship IMO number, ship name and reporting time, the AIS information of repeated reporting is removed.

[0039] Step S12: AIS data track integration

[0040] Since AIS sends a signal every 3 seconds to several minutes, the data volume is very large, which is very inconvenient for the subsequent calculation process, so the massive AIS data needs to be calculated and integrated first. All AIS track points of the same ship in the same device (channel / anchorage / berth) are integrated to calculate the average running speed, navigation time and the like in the port device. The AIS data after integration of the track includes ship MMSI, ship type, ship country, ship length, ship width, entry time, exit time, device id and name, average running speed, navigation time and the like. All track points of a ship in a device (channel / anchorage / berth) for a complete activity are integrated into a track, which corresponds to a data, which can be directly used for the calculation of the ship pollutant emission inventory.

[0041] Step S13: AIS data fusion and completion

[0042] ​The ship Lloyd's file includes basic information of the ship, such as ship MMSI, ship building year, ship country, main engine maximum rated continuous power, main engine speed (high speed machine above 1000 rpm, medium speed machine 300 < n < 1000 rpm, low speed machine < 300 rpm), main engine manufacturer information, auxiliary engine maximum rated power, engine number, design maximum speed, fuel type, auxiliary engine maximum rated power, ship type, etc. Through the MMSI code provided by the ship Lloyd's database, the static attribute information of the corresponding ship in the Lloyd's database is searched, and is matched and fused with the AIS data after the trajectory is integrated.

[0043] For data not contained in the Lloyd's database, a fitting relationship or statistical average value method is used to determine, and a complete ship pollutant emission data database is finally established. For ship building year, ship manufacturer information, the mode of the corresponding data in the Lloyd's database is used. For ship tonnage, ship main engine speed, the average value of the corresponding data in the Lloyd's database is calculated.

[0044] The engine power of most ships can be extracted from the Lloyd's database, however, for some ships that cannot be found in the database, in order to provide more accurate estimation results, it is assumed that the engine power of the ship is related to the type of the ship and its total tonnage, then the missing engine power information can be complemented by the amount of missing information. For ships lacking engine power information but having total tonnage information, a nonlinear regression method is used to predict the engine power of these ships; for ships lacking engine power information and total tonnage information, the engine power of other ships of similar size is used as the engine power of these ships; for ships with only MMSI number and ship type information, the average power of all ships of the same type in the port is used as the engine power of these ships; for ships with only MMSI number, the average power of all ships in the port is used as the engine power of these ships. The missing AE power can be predicted by the ratio of AE power to ME power of a specific ship type. In order to ensure the accuracy of the ship emission estimation results in this study, the ratio of main engine power to auxiliary engine power of different types of ships is determined by referring to the China Ship Atmospheric Pollutant Emission Inventory 2016.

[0045] Step S2: Establishing a high-resolution ship carbon emission inventory

[0046] For the estimation of ship greenhouse gas CO2 emissions, this study adopts the "top-down" dynamic method, which is based on the functional relationship between the energy output (unit: kW·h) of the ship's main engine, auxiliary machinery and boiler and the emission factor of various emissions. The emission factor used in the calculation is measured in g / kW·h, and is finally corrected by combining the corresponding emission factor correction coefficient. Combined with geographic information, the CO2 emissions of a single ship in a single channel, anchorage and berth are calculated, and the CO2 emissions of different ships at different times in the entire port network are integrated to obtain the real-time emissions of different facilities in the entire port, and then a high-resolution ship CO2 emission inventory is established.

[0047] Step S21: Estimation of carbon emissions of ship main engine

[0048] The estimation formula of ship main engine pollutant emissions is as follows:

[0049] Ei = MCR x LF x Act x EFix FCF x CF x 10 -6 (1)

[0050] Where Ei is the estimated amount of CO2 emissions of the ship's main engine, measured in tons; MCR is the maximum continuous rated power of the ship's main engine, measured in KW; LF is the load factor of the ship's main engine, dimensionless, and the calculation formula is:

[0051]

[0052] Act is the sailing time of the ship, measured in h; EFi is the emission factor of CO2, measured in g / kW·h, and the EFi of low-speed diesel engine is 620, and the EFi of medium-speed diesel engine is 683; FCF is the fuel correction factor, dimensionless, and for different fuel types such as RO (2.7% S), HFO (1.5% S), MGO (0.5% S), MDO (1.5% S), MGO (0.1% S), the FCF of CO2 is 1; CF is the low load adjustment coefficient of the ship's main engine, dimensionless, and for load factor LF of 1%, 2%, 3%, 4%, 5%, 6%, 7%, 8%, 9%, 10%, 11%, 12%, 13%, 14%, 15%, 16%, 17%, 18%, 19%, and >=20%, the CF is 5.82, 3.28, 2.44, 2.01, 1.76, 1.59, 1.47, 1.38, 1.31, 1.25, 1.21, 1.17, 1.14, 1.11, 1.08, 1.06, 1.04, 1.03, 1.01, and 1, respectively.

[0053] Step S22: Estimation of carbon emissions of ship auxiliary machinery

[0054] The estimation formula of ship main engine pollutant emissions is as follows:

[0055] Eai = A_MCR x LF_a x Act x EFai x FCF_a x 10 -6 (3)

[0056] Wherein Eai is the estimated amount of CO2 emissions of marine auxiliary machinery, unit is ton; A_MCR is the maximum continuous rated power of marine auxiliary machinery, unit is KW, which is supplemented by the data in the Lloyd's database of ship, for the missing, it is calculated by the ratio of the power of marine auxiliary machinery and the maximum continuous rated power of marine auxiliary machinery, the calculation formula is as follows:

[0057] A_MCR = AMR x MCR (4)

[0058] AMR is the ratio of the power of marine auxiliary machinery and the maximum continuous rated power of marine auxiliary machinery, for the ship type of car carrier, bulk carrier, container ship, cruise ship, general cargo ship, ocean tugboat, reefer ship, ro-ro ship, oil tanker and other types of ship, AMR is 0.266, 0.222, 0.22, 0.278, 0.191, 0.222, 0.406, 0.259, 0.211 and 0.222 respectively; MCR is the maximum continuous rated power of marine main engine; LF_a is the load factor of marine auxiliary machinery, dimensionless unit, which is determined according to the type of ship and navigation state, as shown in table 1; Act is the running time of marine auxiliary machinery, unit is h; EFai is the emission factor of CO2, unit is g / kW h, the value is 683; FCF_a is the fuel correction coefficient, dimensionless unit, for different fuel types of RO(2.7% S), HFO(1.5% S), MGO(0.5% S), MDO(1.5% S), MGO(0.1% S), the FCF_a of CO2 is 1.

[0059] Table 1. Load factor of marine auxiliary machinery LF_A

[0060]

[0061] Step S23: estimation of marine boiler emissions

[0062] The estimation formula of marine boiler pollutant emissions is as follows:

[0063] Ebi = B Energy x Act x EFbi x 10 -6 (5)

[0064] Wherein Ebi is the CO2 emission of the ship boiler, the unit is ton; B_Energy is the load power of the ship boiler, the unit is kW, the boiler is usually only started when the main load of the ship is less than or equal to 20%, and the boiler is in the closed state when the ship is normally sailing at sea, so the boiler load power B_Energy is 0 in the cruising state and the deceleration zone sailing state, and in other sailing states, for the ship type of car carrier, bulk carrier, container ship, cruise ship, general cargo ship, ocean tugboat, refrigerated ship, ro-ro ship, oil tanker and other types of ship, B_Energy is 371, 109, 506, 1393, 137, 0, 109, 464, 3000 and 137 respectively; Act is the time of the ship boiler running, the unit is h; EFbi is the CO2 emission factor of the ship boiler, which is 970g / kW·h.

[0065] Step S3: ship carbon emission feature prediction model based on attention mechanism

[0066] The application adopts a novel DNN architecture based on attention mechanism, a time fusion Transformer (TFT) model, as shown in the figure. Figure 3 The model learns the relationship between different features in the time series data and the time dependence of the time series sequence, and makes a prediction on the future carbon emission of the ship based on the past ship carbon emission features. The main components of the model are: (1) gating mechanism: controls whether to skip some components in the architecture to adapt to the data set and application scenario; (2) variable selection network: TFT performs variable selection through the variable selection network applied to static covariates and time-dependent covariates, selects relevant input variables at each time step, and removes unnecessary noise input; (3) static covariate encoder: adjusts the time dynamic variable by encoding the context vector, thereby integrating static features into the network; (4) time processing module: learns the time relationship from observed and known time-varying inputs, uses a seqtoseq layer for local processing, and uses a multi-head attention to capture long-term dependencies; (5) prediction interval: determines the possible target value range in each prediction range through quantile prediction.

[0067] The application takes the high-resolution CO2 emission inventory established in step S2 as the input of the model, realizes the prediction of the multi-period emission characteristics of the ship carbon in Tianjin Port. The input includes known inputs related to target values and static metadata, the equipment id in the emission inventory is extracted as a static variable, the time information year, month, day, week, distance from the start time days and the pollutant emission of each device with a granularity of days as time-varying variables, and the model is used to learn the feature relationship between them and the target prediction value (ship carbon emission) to generate the possible target value range of the ship in the prediction time range.

[0068] The present application divides the data set into training set, validation set and test set according to the ratio of 7:2:1 with time step of day. The training set is used for feature learning, the test set is used for hyperparameter optimization, and the test set is used for performance evaluation of the model. Hyperparameter optimization is performed by random search, and the optimal model parameters are selected. The appropriate number of iterations and epoch size can be set. The complete search range of all hyperparameters is as follows:

[0069] ·Hidden Layer size—10,20,40,80,160,240,320

[0070] ·Dropout rate—0.1,0.2,0.3,0.4,0.5,0.7,0.9

[0071] ·Minibatch size—64,128,256

[0072] ·Learning rate—0.0001,0.001,0.01

[0073] ·Max gradient norm—0.01,1.0,100.0

[0074] ·Num heads—1,4

[0075] The present application uses the data of the past 30 days to predict the carbon emission of the ship in the future 1 day. For model training and hyperparameter optimization, the joint minimization quantile loss is used, and the sum of all quantile outputs is:

[0076]

[0077]

[0078] where Ω is the domain of training data containing M samples, W represents the weight of TFT, Q is the set of output quantiles (the present application uses Q={0.1, 0.5, 0.9}, and (.)+=max(0,.). For the test set, the standardized quantile loss in the whole prediction range, the present application focuses on P50 and P90 risk:

[0079]

[0080] where is the domain of test samples.

[0081] In addition, the RMSE and MAE are used as the evaluation indexes of the model. The RMSE is the standard deviation of the prediction error, the error represents the distance between the actual point and the predicted regression line, the RMSE represents the distribution of the actual point around the predicted regression line, and the range is [0, +∞], the greater the error, the greater the value.

[0082] The calculation formula is:

[0083]

[0084] The range of MAE is [0, +∞], the greater the error, the greater the value. The calculation formula is:

[0085]

[0086] These methods are widely used to evaluate the accuracy of regression problems, and for all indexes, the lower the value, the better.

[0087] Step S4: Ship carbon emission characteristic rule analysis

[0088] According to the established ship carbon emission inventory and the prediction of the characteristics of ship carbon emission, the ship nationality and type are associated, the characteristics of ship carbon emission are mastered from different angles such as space-time emission rule, ship navigation state, ship type, ship attribution, and ship equipment, multi-dimensional and fine-grained ship carbon emission characteristics in multiple time periods are realized, future ship carbon emission trend is understood, and strong data support and theoretical reference are provided for relevant departments to formulate targeted emission reduction policies. Through the analysis of the space-time dimension of ship carbon emission, the time period and port facility with more emission are mastered, so that the relevant departments can timely adjust the available human resources and supervision work; through the analysis of the navigation state dimension of ship carbon emission, the emission of ships under different navigation states is quantified, so that the port management personnel can further optimize the loading and unloading scheme and reduce unnecessary emission; through the analysis of the attribution dimension of ship carbon emission, the relevant departments can control the carbon emission transaction tax and increase the control intensity of the ship of the country / region with large carbon emission; through the analysis of the type dimension of ship carbon emission, the port cargo loading and unloading structure is mastered, so that the local government can utilize capital, technology and human resources to promote the implementation of relevant clean energy projects and further optimize the port environmental protection structure.

Claims

1. A ship carbon emission feature prediction method based on a power method and an attention mechanism, characterized in that, The ship carbon emission feature prediction method consists of four modules; the first module is an AIS data preprocessing model: according to the demand of ship carbon emission tracing, combined with ship Lloyd's data and geographic information, data cleaning, AIS data trajectory integration, data fusion and data completion operations are carried out on the data to provide a high-quality data basis for subsequent calculation; the second module is the establishment of a high-resolution ship carbon emission list: using the processed ship AIS data and geographic information, using the ship power method, according to the pollutant emission factor, ship operating state and other ship attributes, the carbon emissions of the ship main engine, auxiliary engine and boiler are calculated respectively, and a high-resolution ship carbon emission list is established, which is specific to a single ship, the equipment where the emission is located and the emission time; the third module is the prediction of ship carbon emission features: extracting the data items related to the prediction target from the high-resolution ship carbon emission list, establishing a ship carbon emission time series, using the TFT time series prediction model based on the attention mechanism, learning the relationship between each feature, and realizing the multi-period feature prediction of ship carbon emission; the fourth module is the analysis of ship carbon emission feature law: according to the established high-resolution ship carbon emission list and the prediction results of future carbon emission features, the ship carbon emission feature law is analyzed from different dimensions of time and space, ship ownership, ship type and navigation state, and the future emission features are mastered according to the prediction results, providing data support for relevant departments to develop emission reduction measures; The high-resolution ship carbon emission list established in the ship carbon emission feature prediction model based on the attention mechanism extracts the attributes related to the prediction target and arranges them into time series data as the input of the prediction model; the ship carbon emission feature prediction model based on the attention mechanism learns the relationship between different features in the time series data and the time dependence of the time series, and makes predictions about the future carbon emission of the ship based on the past ship carbon emission features.

2. The ship carbon emission feature prediction method based on the power method and attention mechanism according to claim 1, characterized in that, In the preprocessing of ship AIS data, ship AIS data is used to complete the calculation and prediction of ship carbon emission; Step S11: Data cleaning of AIS data; Abnormal data includes data anomaly, behavior anomaly and position anomaly; Among them, data anomaly includes data exceeding threshold and missing, behavior anomaly includes navigation state exceeding threshold and track inconsistent with common sense, and position anomaly includes ship deviating from the channel and exceeding the research scope; the abnormal data is eliminated or corrected; Step S12: AIS data trajectory integration; First, the massive AIS data is calculated and integrated; the track points of a complete activity of a ship in a device are integrated into a track, corresponding to a data; Step S13: AIS data fusion and completion; Combined with the ship's Lloyd's file data, through the ship's MMSI code, the dynamic trajectory information in the AIS data and the static attributes of the ship are combined. For data not contained in the Lloyd's database, the fitting relationship formula calculation or statistical average value method is used to determine the final complete ship pollutant emission data database combined with the actual application scene and theoretical knowledge.

3. The ship carbon emission feature prediction method based on the power method and attention mechanism according to claim 1, characterized in that, In the establishment of high-resolution ship carbon emission inventory, the ship power method is used, AIS data, ship Lloyd's data, emission factor parameters and geographic information are used to calculate the carbon emissions of the main engine, auxiliary engine and boiler of the ship in different navigation states and different port equipment, thereby establishing a high-resolution ship carbon emission inventory.

4. The ship carbon emission feature prediction method based on the power method and attention mechanism according to claim 1, characterized in that, In the analysis of the characteristics of ship carbon emissions, according to the established high-resolution ship carbon emission inventory and the prediction of the characteristics of ship carbon emissions, the correlation research of ship nationality and type is carried out, and the characteristics of ship carbon emissions are analyzed from the time and space emission law, ship navigation state and emission component, so as to master the characteristics of ship carbon emissions.

Images