Ship oil consumption and carbon intensity grade prediction method based on SA-LSTM model

Through the SA-LSTM model combined with self-attention mechanism and LSTM network, multi-source heterogeneous spatiotemporal data predicts ship fuel consumption and carbon intensity levels, solving the problem of low prediction accuracy in the existing technology, and achieving more efficient and accurate Ship energy consumption forecast and carbon intensity rating.

CN119066331APending Publication Date: 2024-12-03SHANGHAI MARITIME UNIVERSITY

Patent Information

Application Number
CN202411042876.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing ship carbon intensity rating prediction method based on machine learning algorithms has poor prediction results and needs to improve prediction accuracy.

Method used

The ship fuel consumption and carbon intensity level prediction method based on the SA-LSTM model is adopted, and the multi-source heterogeneous spatiotemporal data is collected in real time for preprocessing and feature extraction, and the fuel consumption data is predicted using the self-attention mechanism and the LSTM network, and the carbon intensity level is calculated.

Benefits of technology

It improves the accuracy of ship fuel consumption and carbon intensity level prediction, enhances the timeliness and dynamic adjustment capabilities of prediction, can process complex time series data more effectively, and has a good tolerance for incomplete data sequences.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119066331A_ABST
    Figure CN119066331A_ABST
Patent Text Reader

Abstract

The invention relates to a ship fuel consumption and carbon intensity grade prediction method based on an SA-LSTM model, and the method comprises the following steps: collecting multi-source heterogeneous spatio-temporal data of a ship, and carrying out the preprocessing, and obtaining multi-source data which is matched and fused in time and space; performing feature extraction based on the multi-source data, calculating correlation coefficients between different features and ship fuel consumption, and selecting key features as an input sequence according to the correlation coefficients; inputting the input sequence into a pre-constructed SA-LSTM model, and predicting ship fuel consumption data; based on the ship fuel consumption data, the ship carbon strength is calculated, and the carbon strength grade is obtained through further calculation. Compared with the prior art, the method has the advantages that the accuracy of ship carbon strength and rating prediction is improved, and the like.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of ship energy conservation and emission reduction, and in particular to a method for predicting ship fuel consumption and carbon intensity level based on a SA-LSTM model. Background Art

[0002] In order to promote carbon emission reduction in the shipping industry, in June 2021, the 76th session of the Marine Environment Protection Committee (MEPC) of the International Maritime Organization adopted short-term greenhouse gas emission reduction measures for ships, forming an international shipping carbon intensity rules mechanism. Starting from 2023, international ships with a gross tonnage of 5,000 and above will be required to carry out ship operation carbon intensity index (CII) rating. Ships will be rated as A, B, C, D or E, representing excellent, slightly excellent, medium, slightly poor or low performance levels respectively. Ships with unsatisfactory ratings will be required to make rectifications.

[0003] Accurately predicting the carbon intensity of ships and their ratings is an important basis for the future control of ship CII ratings. In this context, effective monitoring of ship fuel consumption and carbon intensity levels has become an important issue, which is closely related to the current and future operations of shipping companies. Patent application CN117150886A discloses a ship speed optimization method based on machine learning algorithms for carbon intensity ratings. This method uses a random forest regression algorithm to predict ship fuel consumption and obtain a carbon intensity rating. However, experimental verification shows that the prediction results of this method are not particularly ideal, and further optimization is still needed to improve the prediction accuracy. Summary of the invention

[0004] The purpose of the present invention is to provide a method for predicting ship fuel consumption and carbon intensity level based on the SA-LSTM model, which can effectively improve the prediction accuracy of ship fuel consumption and carbon intensity level.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model includes the following steps:

[0007] Collect multi-source heterogeneous spatiotemporal data of ships in real time and perform preprocessing to obtain multi-source data that is matched and fused in time and space;

[0008] Extracting features based on the multi-source data, calculating correlation coefficients between different features and ship fuel consumption, and selecting key features as input sequences according to the correlation coefficients;

[0009] Inputting the input sequence into a pre-built SA-LSTM-based ship fuel consumption prediction model to predict ship fuel consumption data;

[0010] Based on the ship fuel consumption data, the ship carbon intensity is calculated and further calculated to obtain the carbon intensity grade.

[0011] Furthermore, the multi-source heterogeneous spatiotemporal data includes AIS data, ship-related real-time data collected by sensors, and meteorological and sea condition data.

[0012] Furthermore, the pre-processing step includes:

[0013] Cleaning and screening the AIS data, and performing aggregation processing according to a set frequency;

[0014] Converting data units of the ship-related real-time data;

[0015] Grid division is performed according to longitude and latitude based on the aggregated AIS data, the real-time ship-related data after unit conversion, and the meteorological and sea condition data;

[0016] The gridded meteorological and sea condition data are matched with the geographical location and time of the ship to achieve the correspondence in time and space, and the various directions in the meteorological and sea condition data are converted into the relative directions of the ship. Finally, they are fused with the gridded ship-related real-time data and AIS data according to time to obtain an input sequence that is matched and fused in time and space.

[0017] Furthermore, the set frequency of the AIS data during aggregation processing is consistent with the collection frequency of meteorological and sea condition data.

[0018] Furthermore, the AIS data includes dynamic data and static data, and the ship-related real-time data includes real-time fuel consumption data of the ship's main engine, auxiliary engine, and boiler, and the ship's main engine speed and trim data.

[0019] Furthermore, the step of selecting key features includes:

[0020] All correlation coefficients are arranged in descending order by taking the absolute value method, and the features corresponding to the first k correlation coefficients are taken as key features.

[0021] Furthermore, the SA-LSTM-based ship fuel consumption prediction model weights the elements at different positions in the input sequence through the self-attention mechanism to obtain the importance of each element, and uses the LSTM network to capture the long-term dependencies in the input sequence to predict the ship fuel consumption data.

[0022] Furthermore, the step of predicting ship fuel consumption data includes:

[0023] Input sequence representation: The input sequence is represented as X = x1, x2..., x n , where t is the sequence length;

[0024] Set the query, key and value: For any given input X = [x1,...,x n ], first set the weights W of query, key and value respectively q , W k , W v , perform linear transformation on the input and obtain the corresponding results Q, K, V, the expression is: Q = W q X, K = W k X, V = W v X;

[0025] Calculate the attention score matrix S: Evaluate the correlation between input and input. The attention score design method adopts the additive attention method, which is expressed as:

[0026] S(q,k)=v*tanh(W q Q+W k K)

[0027] In the formula, v represents the learnable vector, tanh represents the activation function, and W q , W k represents a learnable parameter;

[0028] Attention weight calculation: Perform a softmax operation on each row of the attention score matrix S to obtain the attention weight matrix A. The calculation expression of the attention weight matrix is:

[0029] A ij =softmax(S(q,k))

[0030] In the formula, A ij is the attention weight of position i to position j;

[0031] Attention weighted representation: The attention weight matrix is ​​used to perform weighted summation on the embedded vector to obtain the attention weighted representation vector of each position. The expression of the attention weighted representation vector is:

[0032] C=A(Q,K,V)=softmax(S(q,k))V

[0033] Where C is the attention weighted representation vector;

[0034] The attention weighted representation vector is input into the LSTM network. The LSTM network uses a recursive operation of time step t to capture the long-term dependencies in the sequence and finally outputs the predicted ship fuel consumption data, where the expression of the recursive operation is:

[0035]

[0036] In the formula, i t 、f t , g t , o t are the activation values ​​of the input gate, forget gate, memory cell state and output gate respectively, σ is the sigmoid activation function, W ii For x t The weight matrix to the input gate, x t is the current input, b ii For x t Bias to the input gate, W hi h t-1 The weight matrix to the input gate, h t-1 is the hidden state at the previous moment, b hi h t-1 Bias to the input gate, W if For x t To the weight matrix of the forget gate, b if For x t The bias to the forget gate, W hf h t-1 To the weight matrix of the forget gate, b hf h t-1 to the bias of the forget gate, tanh is the tanh activation function, W ig For x t to the weight matrix of the candidate memory unit, b ig For x t The bias to the candidate memory cell, W hg h t-1 to the weight matrix of the candidate memory unit, b hg h t-1 The bias to the candidate memory cell, W io For x t The weight matrix to the output gate, b io For x t Bias to the output gate, W ho h t-1 The weight matrix to the output gate, b ho h t-1 Bias to the output gate, c t is the memory cell state at the current moment, c t-1 is the memory unit state at the previous moment, h t In hidden state.

[0037] Furthermore, the calculation expression of the ship carbon intensity is:

[0038]

[0039] Where, CII tis the carbon intensity up to time t, CF is the carbon conversion coefficient of fuel oil, is the sum of fuel consumption up to time t, C represents the deadweight tonnage of the target ship; Represents the total sailing distance of the target ship up to time t.

[0040] Furthermore, the calculation expression of the carbon intensity grade is:

[0041]

[0042] Where, CII Req is the annual carbon intensity grade of the ship, Z is the reduction factor of carbon intensity CII in different years, C is the deadweight tonnage of the ship, and a and c are parameters of different ship types.

[0043] Compared with the prior art, the present invention has the following beneficial effects:

[0044] (1) The present invention can provide a comprehensive and detailed energy consumption prediction basis for the model by selecting multi-source heterogeneous spatiotemporal data, and perform fuel consumption prediction based on the SA-LSTM-based ship fuel consumption prediction model. Since the SA-LSTM model is integrated with the self-attention mechanism, it will pay more attention to the data that contributes more to fuel consumption, thereby improving the accuracy of the fuel consumption prediction results. Through further grade calculation, a more accurate carbon intensity grade result can be obtained.

[0045] (2) The present invention collects and processes multi-source heterogeneous data such as ship AIS data, fuel consumption data, weather and sea conditions in real time. The model can accurately capture the operating status and energy consumption of the ship. In combination with real-time meteorological data, the prediction is more timely and has dynamic adjustment capabilities.

[0046] (3) The AIS data, real-time ship-related data collected by sensors, and weather and sea condition data used in the present invention are all highly sequential. The LSTM part can effectively handle the long-term and short-term dependency problems of time series data, memorize and utilize important information from the past, and forget information that is no longer important, thereby providing accurate predictions for the ship's real-time and future fuel consumption. The self-attention mechanism enables the model to automatically identify and pay more attention to the most important information or time points for the current prediction task when analyzing AIS, fuel consumption, and weather and sea condition data, thereby improving the model's sensitivity to key information. By combining LSTM and the self-attention mechanism, the SA-LSTM model can process complex time series data more efficiently, especially when faced with long time series or data with complex patterns, while still maintaining good performance.

[0047] (4) After implementing and experimenting with a variety of typical machine learning and deep learning algorithms, the present invention selected the SA-LSTM model with the highest average accuracy for ship fuel consumption prediction data. The fuel consumption data predicted by this model is further processed and converted into carbon intensity up to a certain moment, which can ensure the effectiveness and accuracy of the carbon intensity rating prediction.

[0048] (5) The SA-LSTM model of the present invention combines the self-attention mechanism and the long short-term memory network (LSTM), and has certain advantages in processing missing values ​​in data. It has a tolerance for breakpoints or missing data in the sequence data, and improves the processing ability of incomplete data sequences. By learning from the effective information in the sequence, SA-LSTM can still give a more reliable prediction result in the face of missing data. While achieving a higher precision prediction, there is no need to fill or delete the missing values ​​first like some other algorithms. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0050] Figure 2 It is a schematic diagram of the SA-LSTM model structure of the present invention;

[0051] Figure 3 is a schematic diagram of correlation coefficients between different features of the present invention and fuel consumption;

[0052] Figure 4 This is a comparison chart of carbon intensity results obtained by calculating the fuel consumption output of different models of the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0054] This embodiment provides a method for predicting ship fuel consumption and carbon intensity level based on the SA-LSTM model. The method integrates multi-source heterogeneous data such as ship fuel mass flow meter sensor data, ship archive data, ship AIS data, ship meteorological and sea condition data, and performs processing and spatiotemporal fusion; then a variety of feature selection methods are used to construct feature subsets, and the correlation between each feature is analyzed; secondly, the data is divided into training sets and test sets in proportion according to the time series, and a deep learning prediction model is established to dynamically predict the ship's fuel consumption, carbon intensity, and annual CII level, and compare it with the traditional machine learning model, and optimize the hyperparameters of different models through Bayesian search and cross-validation methods. This method can help effectively improve ship energy efficiency, control ship CII ratings, and scientifically respond to relevant carbon reduction policies and regulations. Specifically, such as Figure 1 As shown, the method comprises the following steps:

[0055] Step 1: Collect and process multi-source heterogeneous spatiotemporal data of ships.

[0056] The multi-source heterogeneous spatiotemporal data of ships in this embodiment mainly include three categories: AIS data, ship-related real-time data, and meteorological and sea condition data.

[0057] 1. Ship AIS data collection and processing: Ship AIS data includes dynamic and static data. Due to the influence of weather, location, equipment and other factors, the update frequency of AIS data varies, and large areas of data may be missing in some areas. The longitude and latitude changes of adjacent AIS points in the interpolated data can be used to calculate the ship's sailing distance. The AIS data needs to be cleaned and screened. Considering that the collection frequency of meteorological and sea condition data is once an hour, in order to maintain a consistent time resolution, the present invention aggregates the ship's AIS data on an hourly basis.

[0058] 2. Onboard sensor data and processing: Obtain real-time data related to the ship by installing various sensors. The real-time fuel consumption data of the ship's main engine, auxiliary engine and boiler is obtained through mass flow meters. The ship's main engine speed and trim data are obtained through corresponding sensor data. The original unit of the heavy oil and light oil data of the main engine, auxiliary engine and boiler is kg / h, and the data unit needs to be converted into kg according to the time interval of each data.

[0059] 3. Meteorological and sea state data and processing: The present invention obtains meteorological and sea state data through the European Centre for Medium-Range Weather Forecasts (ECMWF) and Copernicus Marine Service (Copernicus). The data set includes wind components, temperature, humidity, characteristic wave height and periodic frequency at 10 meters above sea level. Among them, Copernicus provides sea state data sets for multiple times a year, including sea water temperature and flow velocity components at different sea water depths. The collection frequency of the meteorological data and sea state data used is once an hour.

[0060] 4. These data are downloaded in the form of a grid divided by longitude and latitude. The size of the grid can be selected. The weather and sea state data need to be matched with the ship's geographic location and time to achieve temporal and spatial correspondence. Finally, considering the ship's heading, the various directions in the weather and sea state data need to be converted into the relative direction of the ship. Finally, the pre-processed AIS data, ship-related real-time data, weather data and sea state data need to be fused according to time to obtain multi-source data that is matched and fused in time and space.

[0061] Step 2: Extract features and select key features as input sequences.

[0062] A variety of feature selection methods are used to select multiple features from multi-source data, and the correlation coefficients between different features and ship fuel consumption are calculated. All feature correlation coefficients are sorted from high to low by taking the absolute value method, and the features corresponding to the first k correlation coefficients are selected as key features.

[0063] Step 3: Establishment and prediction of ship fuel consumption prediction model based on SA-LSTM.

[0064] 1. After implementing and experimenting with multiple typical machine learning and deep learning algorithms, including gradient boosting, random forest, artificial neural network, long short-term memory network, long short-term memory neural network with self-attention mechanism, regression algorithm, etc., the embodiment of the present invention takes the long short-term memory neural network with self-attention mechanism with the highest average accuracy for ship fuel consumption prediction data as the system modeling algorithm, further processes the fuel consumption data predicted by the above model, and converts it into carbon intensity up to a certain moment, which can ensure the effectiveness and accuracy of carbon intensity rating prediction.

[0065] The present invention embeds the self-attention mechanism into the LSTM model to form a new deep learning architecture, called SA-LSTM (Self-Attention Long Short-Term Memory, a long short-term memory neural network based on the self-attention mechanism). The self-attention mechanism is derived from the Transformer model, which is a mechanism that obtains the importance of each element by weighting the elements at different positions in the sequence. The SA-LSTM model structure established by the present invention is shown in the attached Figure 2 .

[0066] 2. Specifically for the present invention, the introduction of the self-attention mechanism is achieved through the following steps:

[0067] a) Input sequence representation: The sequence of ship operating parameters is represented as X = x1, x2..., x n , where t is the sequence length.

[0068] b) Set the query, key and value: For any given input X = [x1,...,x n ], first set the weights W of query, key and value respectively q , W k , W v , perform linear transformation on the input and obtain the corresponding results Q, K, V, the expression is: Q = W q X, K = W k X, V = W v X.

[0069] c) Calculate the attention score matrix S: evaluate the correlation between the input (by query) and the input (by key). The attention score design method adopts the additive attention method, which is expressed as:

[0070] S(q,k)=v*tanh(W q Q+W k K)

[0071] In the formula, v represents the learnable vector, tanh represents the activation function, and W q , W k represents a learnable parameter;

[0072] d) Attention weight calculation: Perform a softmax operation on each row of the attention score matrix S to obtain the attention weight matrix A. The calculation expression of the attention weight matrix is:

[0073] A ij =softmax(S(q,k))

[0074] Among them, A is the attention matrix, Aij Represents the attention weight of position i to position j;

[0075] e) Attention weighted representation: The attention weight matrix is ​​used to perform weighted summation on the embedded vector to obtain the attention weighted representation vector of each position. The expression of the attention weighted representation vector is:

[0076] C=A(Q,K,V)=softmax(S(q,k))V

[0077] Where C is the attention weighted representation vector;

[0078] f) Input the obtained attention weighted representation vector C into the LSTM model as its input sequence. SA-LSTM can capture the long-term dependencies in the sequence through recursive operations at time step t, and finally output the predicted ship fuel consumption data from the LSTM network, where the recursive formula is as follows:

[0079]

[0080] Among them, i t 、f t , g t , o t are the activation values ​​of the input gate, forget gate, memory cell state and output gate respectively, σ is the sigmoid activation function, W ii For x t The weight matrix to the input gate, x t is the current input, b ii For x t Bias to the input gate, W hi h t-1 The weight matrix to the input gate, h t-1 is the hidden state at the previous moment, b hi h t-1 Bias to the input gate, W if For x t To the weight matrix of the forget gate, b if For x t The bias to the forget gate, W hf h t-1 To the weight matrix of the forget gate, b hf h t-1 to the bias of the forget gate, tanh is the tanh activation function, W ig For x t to the weight matrix of the candidate memory unit, b ig For x t The bias to the candidate memory cell, W hg h t-1 to the weight matrix of the candidate memory unit, bhg h t-1 The bias to the candidate memory cell, W io For x t The weight matrix to the output gate, b io For x t Bias to the output gate, W ho h t-1 The weight matrix to the output gate, b ho h t-1 Bias to the output gate, c t is the memory cell state at the current moment, c t-1 is the memory unit state at the previous moment, h t In hidden state.

[0081] Step 4: Carbon intensity rating.

[0082] 1. The present invention further processes the fuel consumption data predicted by the above model and converts it into the carbon intensity CII up to a certain moment. t , the specific calculation method is as follows:

[0083]

[0084] Where CF represents the carbon conversion coefficient of fuel oil; represents the sum of fuel consumption up to time t; C represents the deadweight tonnage of the target ship; Represents the total sailing distance of the target ship up to time t.

[0085] 2. The above calculation method can be used to obtain the carbon intensity of the target ship at each moment, thereby determining the change in the carbon intensity of the ship within a year. In addition, according to the definition of carbon intensity, the tail number calculated by this method is the annual carbon intensity value of the target ship.

[0086] 3. Based on the above calculated ship carbon intensity CII t The calculation results will be used to assign specific ratings to ships, determine the annual operational carbon intensity index (RequiredCII) required for the ship based on the CII reference baseline for a specific ship type in 2019, and then determine the rating boundary based on the boundary parameters of the ship type, thereby determining the annual carbon intensity level of the ship. The specific calculation formula is as follows:

[0087]

[0088] Where: C is the deadweight tonnage (DWT) of the ship; a and c are parameters of different ship types; Z is the reduction factor of CII in different years.

[0089] 4. In the last step, the real-time ship fuel consumption data (including the fuel consumption value predicted by the model) and the ship carbon intensity CII are t and the ship’s annual carbon intensity operational rating (CII) Req Stored in the ship or shore database.

[0090] This example uses a 2400TEU container ship as an experimental object to verify the effectiveness of the above method, specifically:

[0091] 1. Data sample collection.

[0092] Some data samples are shown in Table 1 below.

[0093] Table 1 Partial data samples

[0094]

[0095] 2. Feature Selection

[0096] The correlation coefficients of all features are arranged by taking the absolute value method, and the drawn bar chart is as follows Figure 3 shown.

[0097] 3. Fuel consumption prediction results

[0098] The performance of each model on different data sets was tested, and the specific results of the evaluation indicators of each model were calculated and shown in Table 2 below.

[0099] Table 2 Comparison of evaluation results of each model

[0100]

[0101] 4. Carbon intensity forecast

[0102] The output of each fuel consumption prediction model is further converted into carbon intensity, such as Figure 4 As shown in the figure, the carbon intensity distribution corresponding to the output of the SA-LSTM model is closer to the actual carbon intensity distribution of the target ship, while the gap between LSTM and ANN and the actual carbon intensity distribution is large.

[0103] 5. Rating prediction results

[0104] The ship type in this case study is a container ship, so a = 1984, c = 0.489. The rating boundary of the target ship is calculated according to the formula, as shown in Table 3 below.

[0105] Table 3 Rating results

[0106]

[0107] The carbon intensity prediction results of each model year are compared with the boundary parameters to determine the carbon intensity level achieved in the year, as shown in Table 4 below.

[0108] Table 4 Carbon intensity rating results of each model

[0109]

[0110] In summary, the key to predicting the carbon intensity of a ship in the embodiment of the present invention is to predict the fuel consumption of the ship. By converting the fuel consumption into the carbon emissions of the ship, the carbon emission intensity of the ship can be calculated according to the sailing distance of the ship. The research on ship fuel consumption prediction mainly adopts a bottom-up dynamic method based on ship AIS data, as well as a method based on traditional machine learning. It mainly relies on noon reported fuel consumption data, engine speed, speed and other data for learning and prediction. It is very rare to use a method with higher precision mass flow meter and large model related deep learning technology. Combining the real-time fuel consumption data of the mass flow meter and the large model technology will greatly improve the accuracy of ship fuel consumption prediction. On this basis, a more accurate ship carbon intensity and CII annual rating forecast can be obtained, which is of great significance to ship carbon management.

[0111] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0112] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0113] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0114] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0115] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0116] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0117] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model, characterized in that: The following steps are involved: Collect multi-source heterogeneous spatiotemporal data of ships in real time and perform preprocessing to obtain multi-source data that is matched and fused in time and space; Extracting features based on the multi-source data, calculating correlation coefficients between different features and ship fuel consumption, and selecting key features as input sequences according to the correlation coefficients; Inputting the input sequence into a pre-built SA-LSTM-based ship fuel consumption prediction model to predict ship fuel consumption data; Based on the ship fuel consumption data, the ship carbon intensity is calculated and further calculated to obtain the carbon intensity grade.

2. According to claim 1, a method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model is characterized in that: The multi-source heterogeneous spatiotemporal data include AIS data, ship-related real-time data collected by sensors, and meteorological and sea condition data.

3. According to claim 2, a method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model is characterized in that: The pre-processing steps include: Cleaning and screening the AIS data, and performing aggregation processing according to a set frequency; Converting data units of the ship-related real-time data; Grid division is performed according to longitude and latitude based on the aggregated AIS data, the real-time ship-related data after unit conversion, and the meteorological and sea condition data; The gridded meteorological and sea condition data are matched with the geographical location and time of the ship to achieve the correspondence in time and space, and the various directions in the meteorological and sea condition data are converted into the relative directions of the ship. Finally, they are fused with the gridded ship-related real-time data and AIS data according to time to obtain an input sequence that is matched and fused in time and space.

4. According to claim 2, a method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model is characterized in that: The set frequency of the AIS data during aggregation processing is consistent with the collection frequency of meteorological and sea condition data.

5. According to claim 2, a method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model is characterized in that: The AIS data includes dynamic data and static data, and the ship-related real-time data includes real-time fuel consumption data of the ship's main engine, auxiliary engine, and boiler, and the ship's main engine speed and trim data.

6. A method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model according to claim 1, characterized in that: The step of selecting key features comprises: All correlation coefficients are arranged in descending order by taking the absolute value method, and the features corresponding to the first k correlation coefficients are taken as key features.

7. The method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model according to claim 1 is characterized in that: The SA-LSTM-based ship fuel consumption prediction model weights the elements at different positions in the input sequence through the self-attention mechanism to obtain the importance of each element, and uses the LSTM network to capture the long-term dependencies in the input sequence to predict the ship fuel consumption data.

8. The method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model according to claim 7 is characterized in that: The step of predicting ship fuel consumption data comprises: Input sequence representation: The input sequence is represented as X = x1, x2..., x n , where t is the sequence length; Set query, key and value: For any given input X = [x1, ..., x n ], first set the weights W of query, key and value respectively q , W k , W v , perform linear transformation on the input and obtain the corresponding results Q, K, V, the expression is: Q = W q X, K = W k X, V = W v X; Calculate the attention score matrix S: Evaluate the correlation between input and input, where the attention score design method adopts the additive attention method, expressed as: S(q,k)=v*tanh(W q Q+W k K) In the formula, v represents the learnable vector, tanh represents the activation function, and W q , W k represents a learnable parameter; Attention weight calculation: Perform a softmax operation on each row of the attention score matrix S to obtain the attention weight matrix A. The calculation expression of the attention weight matrix is: A ij =softmax(S(q,k)) In the formula, A ij is the attention weight of position i to position j; Attention weighted representation: The attention weight matrix is ​​used to perform weighted summation on the embedded vector to obtain the attention weighted representation vector of each position. The expression of the attention weighted representation vector is: C=A(Q,K,V)=softmax(S(q,k))V Where C is the attention weighted representation vector; The attention weighted representation vector is input into the LSTM network. The LSTM network uses a recursive operation of time step t to capture the long-term dependencies in the sequence and finally outputs the predicted ship fuel consumption data, where the expression of the recursive operation is: In the formula, i t 、f t , g t , o t are the activation values ​​of the input gate, forget gate, memory cell state and output gate respectively, σ is the sigmoid activation function, W ii For x t The weight matrix to the input gate, x t is the current input, b ii For x t Bias to the input gate, W hi h t-1 The weight matrix to the input gate, h t-1 is the hidden state at the previous moment, b hi h t-1 Bias to the input gate, W if For x t To the weight matrix of the forget gate, b if For x t The bias to the forget gate, W hf h t-1 To the weight matrix of the forget gate, b hf h t-1 to the bias of the forget gate, tanh is the tanh activation function, W ig For x t to the weight matrix of the candidate memory unit, b ig For x t The bias to the candidate memory cell, W hg h t-1 to the weight matrix of the candidate memory unit, b hg h t-1 The bias to the candidate memory cell, W io For x t The weight matrix to the output gate, b io For x t Bias to the output gate, W ho h t-1 The weight matrix to the output gate, b ho h t-1 Bias to the output gate, c t is the memory cell state at the current moment, c t-1 is the memory unit state at the previous moment, h t In hidden state.

9. The method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model according to claim 1, characterized in that: The calculation expression of the ship carbon intensity is: Where, CII t is the carbon intensity up to time t, CF is the carbon conversion coefficient of fuel oil, is the sum of fuel consumption up to time t, C represents the deadweight tonnage of the target ship; Represents the total sailing distance of the target ship up to time t.

10. The method for predicting ship fuel consumption and carbon intensity level based on SA-LSTM model according to claim 1, characterized in that: The calculation expression of the carbon intensity grade is: Where, CII Req is the annual carbon intensity grade of the ship, Z is the reduction factor of carbon intensity CII in different years, C is the deadweight tonnage of the ship, and a and c are parameters of different ship types.

Citation Information

Patent Citations

  • A method and system for ship tracking

    CN109102528A

  • GCN-LSTM ship traffic flow prediction method based on attention mechanism

    CN114881295A

Cited By

  • Ship carbon emission space-time prediction method and carbon emission supervision platform using same

    CN121233661A