Marine fuel oil sulfur content dynamic identification method and device

Abstract

The invention discloses a ship fuel oil sulfur content dynamic identification method and device, and the method comprises the steps: S1, deploying distributed sensor nodes at key positions of a ship fuel oil system, and collecting the states and environmental parameters of fuel oil at different positions; s2, constructing a multi-modal sensor network based on distributed sensor nodes, and uploading state parameters and environmental parameters; s3, deploying edge computing nodes on the ship to carry out data preprocessing and feature extraction; s4, using an improved multi-layer perceptron MLP-LSTM hybrid model, introducing an attention mechanism and a regularization technology to train a large amount of historical data, and establishing a fuel oil sulfur content prediction model; s5, according to a model prediction result, giving an alarm or suggesting a navigation strategy to sailors in real time; according to the method, data are collected and processed in real time based on the multi-modal sensor network and the edge calculation node, dynamic identification of the sulfur content of the fuel oil is achieved through the improved MLP-LSTM hybrid model, and the identification accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of monitoring sulfur content in ship fuel, and more particularly to a method and device for dynamically identifying sulfur content in ship fuel. Background Art

[0002] The sulfur content in ship fuel is one of the environmental protection indicators that the International Maritime Organization (IMO) focuses on. The sulfur dioxide emissions generated by the combustion of high-sulfur fuel oil have a serious impact on the atmospheric environment. Therefore, controlling the sulfur content in ship fuel is crucial to reducing the impact of the marine transportation industry on the environment. In particular, the regulation that the sulfur content of ship fuel shall not exceed 0.5% implemented globally in 2020 has made it particularly important to monitor the sulfur content in ship fuel.

[0003] At present, the technologies used to detect the sulfur content of ship fuel mainly include laboratory chemical analysis, online sensor measurement and spectral analysis. For example: Laboratory chemical analysis is a traditional detection method. After collecting samples, chemical reagents are used in the laboratory for analysis to determine the sulfur content. Although this method is highly accurate, it is time-consuming and cannot achieve real-time monitoring; online sensor measurement, specific sensors can be installed on board to monitor changes in sulfur content in fuel in real time, such as the sniffing method; however, these sensors may be affected by environmental factors, resulting in unstable or inaccurate measurement results; Spectral analysis: Using infrared or ultraviolet spectroscopy technology to quickly analyze the sulfur content in fuel, this method can provide more accurate results, but the equipment cost is high and requires regular calibration.

[0004] Both laboratory chemical analysis and some online measurement technologies have the problem of long response time and cannot meet the needs of instant decision-making; especially when using advanced spectral analysis technology, the initial investment and subsequent maintenance costs are high, which is a considerable expense for small and medium-sized shipping companies; online sensors are easily affected by external conditions such as temperature and humidity, resulting in data deviation; most existing detection methods are static or semi-static, making it difficult to achieve continuous tracking of changes in fuel sulfur content during ship operation.

[0005] Therefore, how to achieve real-time and continuous online monitoring of the sulfur content of ship fuel and improve the stability and accuracy of sulfur content detection are problems that technical personnel in this field urgently need to solve. Summary of the invention

[0006] In view of this, the present invention provides a method and device for dynamically identifying the sulfur content in ship fuel to solve some of the technical problems mentioned in the background technology.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A method for dynamically identifying sulfur content in ship fuel comprises the following steps:

[0009] S1. Deploy distributed sensor nodes at key locations of the ship's fuel system to collect fuel state parameters and environmental parameters at different locations;

[0010] S2. Build a multimodal sensor network based on distributed sensor nodes and upload state parameters and environmental parameters;

[0011] S3. Deploy edge computing nodes on the ship to perform data preprocessing and feature extraction;

[0012] S4. Using the improved multi-layer perceptron MLP-LSTM hybrid model, and introducing the attention mechanism and regularization technology, a fuel sulfur content prediction model is established by training a large amount of historical data;

[0013] S5. Issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

[0014] Preferably, key locations of the ship's fuel system include the fuel tank outlet, before and after the filter, the heat exchanger outlet and the main engine inlet; each sensor node is equipped with a variety of sensors, including electrochemical sensors, optical sensors, temperature sensors, pressure sensors, flow sensors, humidity sensors and vibration sensors.

[0015] Preferably, the state parameters and environmental parameters of the fuel at different locations are specifically: the concentration of hydrogen sulfide or sulfur dioxide in the fuel, the absorption spectrum, temperature, pressure, flow rate, ambient humidity and ambient vibration data of the fuel.

[0016] Preferably, the edge computing node performs data preprocessing as follows: using a moving average filter to remove high-frequency noise in the data; and using a Z-Score standardization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0017] Preferably, the features extracted by the edge computing node include time series features, frequency domain features, and combined features of temperature and pressure.

[0018] Preferably, the time series feature is to calculate the data mean and data standard deviation of the past n time points, the frequency domain feature is to perform Fourier transform on the time series data and extract the frequency domain feature; the combined feature is to multiply the temperature and pressure data to form a new feature.

[0019] Preferably, in step S4, the fuel sulfur content prediction model is established by using an improved multi-layer perceptron MLP-LSTM hybrid model, and introducing an attention mechanism and regularization technology to train a large amount of historical data. The specific content is:

[0020] S41. receiving multidimensional time series data through an input layer, including original data, time series features, frequency domain features, and combined features;

[0021] S42. Capturing the temporal dependency of the time series data by a first LSTM layer including 128 neurons and a second LSTM layer including 64 neurons;

[0022] S43. The introduction of attention mechanism enables the model to focus on the important parts of the time series;

[0023] S44. Use the Dropout layer to randomly discard some neurons to prevent overfitting;

[0024] S45. Perform nonlinear mapping, integrate the features extracted by the LSTM layer, and use multiple Dense fully connected layers to further extract high-level features;

[0025] S46. Update the model parameters through the Adam optimizer using the mean square error as the loss function, and introduce regularization terms during the model training process to prevent overfitting;

[0026] S47. Output the predicted sulfur content value through the output layer, in terms of mean square error and R 2 The performance of the model on the validation set was evaluated by score and K-fold cross validation was performed;

[0027] S48. Use the trained fuel sulfur content prediction model to predict the sulfur content of the preprocessed data through forward propagation.

[0028] Preferably, the calculation method for capturing the time dependency of time series data by the LSTM layer in S42 is:

[0029] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0030] i t =σ(W i ·[h t-1 , x t ]+b i )

[0031]

[0032]

[0033] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0034] h t =o t ☉tanh(C t )

[0035] Among them, f t is the output of the forget gate, i t is the input gate output, is the candidate cell state, C t is the cell state, o t is the output gate output, h t is the hidden state, W f , W i , W C , W o is the weight matrix, b f , b i , b C , b o is the bias vector, x t is the input data at time t, h t-1 is the hidden state at the t-1th moment, σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the element-by-element multiplication;

[0036] In step S44, the Dropout layer is:

[0037]

[0038] Among them, y is the output after Dropout, x is the input data, and p is the drop probability;

[0039] In step S45, the fully connected layer is:

[0040] y=σ(W·x+b)

[0041] Among them, y is the output, x is the input data, W is the weight matrix, b is the bias vector, and σ is the activation function;

[0042] The loss function of the regularization term introduced in step S46 is:

[0043]

[0044] Among them, MSE is the mean square error, λ is the regularization coefficient, and W i is the weight matrix;

[0045] The loss function mean square error MSE is:

[0046]

[0047] Where N is the number of samples, y iis the actual sulfur content of the i-th sample, is the predicted sulfur content of the i-th sample.

[0048] Preferably, in step S48, the process of forward propagation is specifically as follows:

[0049] h1=σ(W1x+b1)

[0050] h2=σ(W2h1+b2)

[0051]

[0052] Among them, x is the input feature vector, W1, W2 and W3 are weight matrices, b1, b2 and b3 are bias vectors, and σ is the activation function.

[0053] A device for dynamically identifying sulfur content in ship fuel, based on the method for dynamically identifying sulfur content in ship fuel, comprises: a multimodal sensor network, an edge computing node, a model building and training module, and a fuel sulfur content prediction model;

[0054] The multimodal sensor network includes distributed sensor nodes and wireless transmission modules. The distributed sensor nodes are deployed at key locations of the ship's fuel system. Each sensor node includes multiple sensors for collecting state parameters and environmental parameters of fuel at different locations and transmitting them to edge computing nodes through wireless transmission modules.

[0055] Edge computing nodes are used for data preprocessing and feature extraction of state parameters and environmental parameters;

[0056] Model building and training module, which is used to establish a fuel sulfur content prediction model by using an improved multi-layer perceptron MLP-LSTM hybrid model and introducing attention mechanism and regularization technology to train a large amount of historical data;

[0057] The fuel sulfur content prediction model is used to predict the sulfur content of ship fuel and issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

[0058] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and device for dynamically identifying sulfur content in ship fuel, which has the following advantages:

[0059] Based on the multimodal sensor network, the edge computing node can process the collected data in real time, extract features and perform preliminary processing to ensure the real-time and accuracy of the data; the improved MLP-LSTM hybrid model combines time series features, frequency domain features and combined features to realize dynamic identification of fuel sulfur content, and the model can capture the characteristics of fuel data from multiple angles to improve the richness and representativeness of features; the introduction of regularization terms in the model training process can further prevent overfitting and ensure that the model performs more stably on new data; through the user-friendly monitoring interface, the crew can view the predicted results of fuel sulfur content and other related parameters in real time, and understand the fuel status in a timely manner. When the sulfur content exceeds the set threshold, the system automatically issues an alarm to the crew and provides corresponding treatment suggestions to help the crew take timely measures to avoid illegal emissions. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0061] Figure 1 A schematic diagram of a method for dynamically identifying sulfur content in ship fuel provided by the present invention;

[0062] Figure 2 A schematic diagram of the construction and training of the fuel sulfur content prediction model provided by the present invention;

[0063] Figure 3 This is a schematic diagram of a device for dynamically identifying sulfur content in ship fuel provided by the present invention. DETAILED DESCRIPTION

[0064] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0065] The embodiment of the present invention discloses a method for dynamically identifying sulfur content in ship fuel, such as Figure 1 , including the following steps:

[0066] S1. Deploy distributed sensor nodes at key locations of the ship's fuel system to collect fuel state parameters and environmental parameters at different locations;

[0067] S2. Build a multimodal sensor network based on distributed sensor nodes and upload state parameters and environmental parameters;

[0068] S3. Deploy edge computing nodes on the ship to perform data preprocessing and feature extraction;

[0069] S4. Using the improved multi-layer perceptron MLP-LSTM hybrid model, and introducing the attention mechanism and regularization technology, a fuel sulfur content prediction model is established by training a large amount of historical data;

[0070] S5. Issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

[0071] In order to further implement the above technical solutions, the key positions of the ship's fuel system include the fuel tank outlet, before and after the filter, the heat exchanger outlet and the main engine inlet; each sensor node is equipped with a variety of sensors, including electrochemical sensors, optical sensors, temperature sensors, pressure sensors, flow sensors, humidity sensors and vibration sensors.

[0072] In order to further implement the above technical solution, the state parameters of the fuel at different locations and the environmental parameters are specifically: the concentration of hydrogen sulfide or sulfur dioxide in the fuel, the absorption spectrum of the fuel, temperature, pressure, flow, ambient humidity and ambient vibration data.

[0073] In order to further implement the above technical solution, the edge computing node performs data preprocessing as follows: using a moving average filter to remove high-frequency noise in the data; using the Z-Score standardization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

[0074] The moving average filter is:

[0075]

[0076] Z-Score is normalized to:

[0077]

[0078] Among them, x is a certain eigenvalue, μ is the mean of the feature, and σ is the standard deviation of the feature.

[0079] In order to further implement the above technical solution, the features extracted by the edge computing node include time series features, frequency domain features, and combined features of temperature and pressure.

[0080] In order to further implement the above technical solution, the time series feature is to calculate the data mean and data standard deviation of the past n time points, the frequency domain feature is to perform Fourier transform on the time series data and extract the frequency domain features; the combined feature is to multiply the temperature and pressure data to form a new feature.

[0081] The data mean is:

[0082]

[0083] The standard deviation of the data is:

[0084]

[0085] Among them, x i is the eigenvalue at the i-th time point, and n is the number of time points;

[0086] The frequency domain features are: frequency component, extract the main frequency component of the signal; power spectrum density, calculate the power spectrum density of the signal; peak frequency, extract the peak frequency of the signal; total energy, calculate the total energy of the signal;

[0087] Combined features:

[0088] Combined Feature=Temperature×Pressure

[0089] Among them, Temperature is the data of the temperature sensor, and Pressure is the data of the pressure sensor.

[0090] In order to further implement the above technical solution, Figure 2 In step S4, the improved multi-layer perceptron MLP-LSTM hybrid model is used, and the attention mechanism and regularization technology are introduced to train a large amount of historical data. The specific content of establishing the fuel sulfur content prediction model is as follows:

[0091] S41. receiving multidimensional time series data through an input layer, including original data, time series features, frequency domain features, and combined features;

[0092] S42. Capturing the temporal dependency of the time series data by a first LSTM layer including 128 neurons and a second LSTM layer including 64 neurons;

[0093] The first LSTM layer captures the temporal dependencies of the time series data and returns the sequence for subsequent LSTM layers to process. Parameters: units = 128 indicates the number of LSTM units, input_shape = (100, 10) specifies the shape of the input data, and return_sequences = True indicates that the complete output sequence is returned.

[0094] The function of the second LSTM layer is to further extract time series features and not return the sequence; parameter: units = 64 indicates the number of LSTM units;

[0095] S43. The introduction of attention mechanism enables the model to focus on the important parts of the time series;

[0096] S44. Use the Dropout layer to randomly discard some neurons to prevent overfitting;

[0097] S45. Perform nonlinear mapping, integrate the features extracted by the LSTM layer, and use multiple Dense fully connected layers to further extract high-level features;

[0098] S46. Update the model parameters through the Adam optimizer using the mean square error as the loss function, and introduce regularization terms during the model training process to prevent overfitting;

[0099] S47. Output the predicted sulfur content value through the linear activation function Linear(x)=x output layer, and use the mean square error and R 2 Score the performance of the model on the validation set and perform K-fold cross validation (divide the data into K parts, use one part as the validation set each time, and the rest as the training set, repeat K times, and take the average performance indicator);

[0100] S48. Use the trained fuel sulfur content prediction model to predict the sulfur content of the preprocessed data through forward propagation.

[0101] In order to further implement the above technical solution, the calculation method of the LSTM layer in S42 to capture the time dependency of time series data is:

[0102] f t =σ(W f ·[h t-1 ,x t ]+b f )

[0103] i t =σ(W i ·[h t-1 ,x t ]+b i )

[0104]

[0105]

[0106] o t =σ(W o ·[h t-1 ,x t ]+b o )

[0107] h t =o t ☉tanh(C t )

[0108] Among them, f t is the output of the forget gate, i t is the input gate output, is the candidate cell state, C t is the cell state, o t is the output gate output, h t is the hidden state, W f , W i , W C , W o is the weight matrix, b f , b i , b C , b o is the bias vector, x t is the input data at time t, h t-1 is the hidden state at the t-1th moment, σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the element-by-element multiplication;

[0109] In step S44, the Dropout layer is:

[0110]

[0111] Among them, y is the output after Dropout, x is the input data, and p is the drop probability;

[0112] In step S45, the fully connected layer is:

[0113] y=σ(W·x+b)

[0114] Among them, y is the output, x is the input data, W is the weight matrix, b is the bias vector, and σ is the activation function;

[0115] The loss function of the regularization term introduced in step S46 is:

[0116]

[0117] Among them, MSE is the mean square error, λ is the regularization coefficient, and W i is the weight matrix;

[0118] The loss function L is the mean square error MSE, specifically:

[0119]

[0120] Where N is the number of samples, y i is the actual sulfur content of the i-th sample, is the predicted sulfur content of the i-th sample.

[0121] R in metrics used to evaluate models 2 The score is used to measure the explanatory power of the model, specifically:

[0122]

[0123] Among them, y i is the actual sulfur content of the i-th sample, is the predicted sulfur content of the i-th sample, is the mean value of the actual sulfur content;

[0124] The Adam optimizer used to update the model parameters is:

[0125]

[0126] Among them, θ t is the current parameter, α is the learning rate, and L is the loss function.

[0127] In order to further implement the above technical solution, in step S48, the process of forward propagation is specifically as follows:

[0128] h1=σ(W1x+b1)

[0129] h2=σ(W2h1+b2)

[0130]

[0131] Among them, x is the input feature vector, W1, W2 and W3 are weight matrices, b1, b2 and b3 are bias vectors, and σ is the activation function.

[0132] A device for dynamically identifying sulfur content in ship fuel, based on a method for dynamically identifying sulfur content in ship fuel, such as Figure 3 , including: multimodal sensor network, edge computing nodes, model building and training modules, fuel sulfur content prediction model;

[0133] The multimodal sensor network includes distributed sensor nodes and wireless transmission modules. The distributed sensor nodes are deployed at key locations of the ship's fuel system. Each sensor node includes multiple sensors for collecting state parameters and environmental parameters of fuel at different locations and transmitting them to edge computing nodes through wireless transmission modules.

[0134] Edge computing nodes are used for data preprocessing and feature extraction of state parameters and environmental parameters;

[0135] Model building and training module, which is used to establish a fuel sulfur content prediction model by using an improved multi-layer perceptron MLP-LSTM hybrid model and introducing attention mechanism and regularization technology to train a large amount of historical data;

[0136] The fuel sulfur content prediction model is used to predict the sulfur content of ship fuel and issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

[0137] A computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, a method for dynamically identifying the sulfur content of ship fuel is implemented.

[0138] A processing terminal includes a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and is characterized in that when the processor executes the computer program, a method for dynamically identifying the sulfur content of ship fuel is implemented.

[0139] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0140] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for dynamically identifying sulfur content in ship fuel, characterized in that: The following steps are involved: S1. Deploy distributed sensor nodes at key locations of the ship's fuel system to collect fuel state parameters and environmental parameters at different locations; S2. Build a multimodal sensor network based on distributed sensor nodes and upload state parameters and environmental parameters; S3. Deploy edge computing nodes on the ship to perform data preprocessing and feature extraction; S4. Using the improved multi-layer perceptron MLP-LSTM hybrid model, and introducing the attention mechanism and regularization technology, a fuel sulfur content prediction model is established by training a large amount of historical data; S5. Issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

2. A method for dynamically identifying sulfur content in ship fuel according to claim 1, characterized in that: The key locations of the ship's fuel system include the fuel tank outlet, before and after the filter, the heat exchanger outlet and the main engine inlet; each sensor node is equipped with a variety of sensors, including electrochemical sensors, optical sensors, temperature sensors, pressure sensors, flow sensors, humidity sensors and vibration sensors.

3. A method for dynamically identifying sulfur content in ship fuel according to claim 1, characterized in that: The state parameters and environmental parameters of the fuel at different locations are specifically: the concentration of hydrogen sulfide or sulfur dioxide in the fuel, the absorption spectrum of the fuel, temperature, pressure, flow, ambient humidity and ambient vibration data.

4. A method for dynamically identifying sulfur content in ship fuel according to claim 1, characterized in that: The data preprocessing performed by the edge computing node is as follows: using a moving average filter to remove high-frequency noise in the data; using the Z-Score standardization method to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1.

5. A method for dynamically identifying sulfur content in ship fuel according to claim 4, characterized in that: The features extracted by edge computing nodes include time series features, frequency domain features, and combined features of temperature and pressure.

6. A method for dynamically identifying sulfur content in ship fuel according to claim 5, characterized in that: The time series feature is to calculate the data mean and standard deviation of the past n time points. The frequency domain feature is to perform Fourier transform on the time series data and extract the frequency domain features. The combined feature is to multiply the temperature and pressure data to form a new feature.

7. A method for dynamically identifying sulfur content in ship fuel according to claim 5, characterized in that: In step S4, the improved multi-layer perceptron MLP-LSTM hybrid model is used, and the attention mechanism and regularization technology are introduced to train a large amount of historical data to establish a fuel sulfur content prediction model. The specific content is: S41. receiving multidimensional time series data through an input layer, including original data, time series features, frequency domain features, and combined features; S42. Capturing the temporal dependency of the time series data by a first LSTM layer including 128 neurons and a second LSTM layer including 64 neurons; S43. The introduction of attention mechanism enables the model to focus on the important parts of the time series; S44. Use the Dropout layer to randomly discard some neurons to prevent overfitting; S45. Perform nonlinear mapping, integrate the features extracted by the LSTM layer, and use multiple Dense fully connected layers to further extract high-level features; S46. Update the model parameters through the Adam optimizer using the mean square error as the loss function, and introduce regularization terms during the model training process to prevent overfitting; S47. Output the predicted sulfur content value through the output layer, in terms of mean square error and R 2 The performance of the model on the validation set was evaluated by score and K-fold cross validation was performed; S48. Use the trained fuel sulfur content prediction model to predict the sulfur content of the preprocessed data through forward propagation.

8. A method for dynamically identifying sulfur content in ship fuel according to claim 7, characterized in that: The calculation method for capturing the time dependency of time series data by the LSTM layer in S42 is: f t =σ(W f ·[h t-1 ,x t ]+b f ) i t =σ(W i ·[h t-1 ,x t ]+b i ) the t =σ(W o ·[h t-1 ,x t ]+b o ) h t =o t ☉tanh(C t ) Among them, f t is the output of the forget gate, i t is the input gate output, is the candidate cell state, C t is the cell state, o t is the output gate output, h t is the hidden state, W f , W i , W C , W o is the weight matrix, b f , b i , b C , b o is the bias vector, x t is the input data at time t, h t-1 is the hidden state at the t-1th moment, σ is the Sigmoid activation function, tanh is the hyperbolic tangent activation function, and ⊙ is the element-by-element multiplication; In step S44, the Dropout layer is: Among them, y is the output after Dropout, x is the input data, and p is the drop probability; In step S45, the fully connected layer is: y=σ(W·x+b) Among them, y is the output, x is the input data, W is the weight matrix, b is the bias vector, and σ is the activation function; The loss function of the regularization term introduced in step S46 is: Among them, MSE is the mean square error, λ is the regularization coefficient, and W i is the weight matrix; The loss function mean square error MSE is: Where N is the number of samples, yi is the actual sulfur content of the i-th sample, is the predicted sulfur content of the i-th sample.

9. A method for dynamically identifying sulfur content in ship fuel according to claim 7, characterized in that: In step S48, the process of forward propagation is specifically as follows: h1=σ(W1x+b1) h2=σ(W2h1+b2) Among them, x is the input feature vector, W1, W2 and W3 are weight matrices, b1, b2 and b3 are bias vectors, and σ is the activation function.

10. A device for dynamically identifying sulfur content in ship fuel, characterized in that: A method for dynamically identifying sulfur content in ship fuel oil based on any one of claims 1 to 9, comprising: a multimodal sensor network, an edge computing node, a model building and training module, and a fuel oil sulfur content prediction model; The multimodal sensor network includes distributed sensor nodes and wireless transmission modules. The distributed sensor nodes are deployed at key locations of the ship's fuel system. Each sensor node includes multiple sensors for collecting fuel state parameters and environmental parameters at different locations and transmitting them to edge computing nodes through wireless transmission modules. Edge computing nodes are used for data preprocessing and feature extraction of state parameters and environmental parameters; Model building and training module, which is used to establish a fuel sulfur content prediction model by using an improved multi-layer perceptron MLP-LSTM hybrid model and introducing attention mechanism and regularization technology to train a large amount of historical data; The fuel sulfur content prediction model is used to predict the sulfur content of ship fuel and issue warnings or suggest navigation strategies to the crew in real time based on the model prediction results.

Images