Ship navigation oil consumption estimation system based on deep learning

Abstract

The invention discloses a ship navigation oil consumption estimation system based on deep learning, and particularly relates to the field of ship navigation oil consumption estimation, which comprises a data acquisition terminal, a data preprocessing terminal, a deep learning terminal, a real-time monitoring terminal and a control center, the system specifically comprises a multi-source data acquisition module, a data preprocessing module, a model construction module and an oil consumption estimation module. The multi-source data acquisition module is used for acquiring ship navigation influence factor data from multiple angles through a data acquisition terminal; the data preprocessing module is used for preprocessing the ship navigation influence factor data through a data preprocessing terminal; according to the method, multi-source data acquisition is deployed, and a composite feature system is constructed, so that the problems that a traditional method only depends on a single variable and a multivariable coupling effect is difficult to capture are effectively solved, more comprehensive characterization of complex influence factors of ship navigation is realized, and a data foundation is laid for high-precision oil consumption estimation.

Description

Technical Field

[0001] The present invention relates to the technical field of fuel consumption prediction for ship navigation. More specifically, the present invention relates to a ship navigation fuel consumption prediction system based on deep learning. Background Art

[0002] In the field of ship navigation, accurate prediction of fuel consumption is crucial for reducing operating costs, reducing environmental pollution, and enhancing navigation economy. Traditional ship fuel consumption prediction methods mainly rely on empirical formulas or simplified physical models. Such methods usually only consider a few single variables such as ship speed and cargo volume, and it is difficult to effectively capture the complex dynamic characteristics under the coupling effect of multiple variables during the ship navigation process. The comprehensive influence of multi-dimensional data such as real-time engine condition parameters, such as engine speed, propeller torque, ocean environment parameters, and ship's own state parameters, is not fully considered, resulting in a significant increase in fuel consumption prediction error in complex sea conditions or when the ship's condition suddenly changes. In addition, traditional methods lack the ability to deeply mine historical data and real-time learning, and it is difficult to adapt to the changing external environment and internal conditions during the ship navigation process, making the real-time performance and accuracy of the prediction results difficult to meet the development needs of modern ship intelligence and energy conservation.

[0003] Therefore, there is an urgent need for a ship navigation fuel consumption prediction system based on deep learning. By integrating multi-source heterogeneous sensor data and constructing a deep learning model with dynamic feature learning ability, it can achieve high-precision and real-time prediction of fuel consumption during ship navigation, solve the problems of insufficient prediction accuracy and poor real-time adaptability of traditional methods in multi-variable coupling scenarios, and provide a scientific decision-making basis for energy-saving ship navigation. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the present invention provides a ship navigation fuel consumption prediction system based on deep learning, through the following solutions to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solution: A ship navigation fuel consumption prediction system based on deep learning, including a data acquisition terminal, a data preprocessing terminal, a deep learning terminal, a real-time monitoring terminal, and a control center, specifically including: Multi-source data acquisition module: Obtain data on ship navigation influencing factors from multiple angles through the data acquisition terminal; Data preprocessing module: Perform preprocessing operations on the ship navigation influencing factor data through the data preprocessing terminal; Model construction module: Construct and train an improved spatio-temporal feature fusion neural network model ST-FNN through the deep learning terminal to capture the time-series dependence and spatial coupling relationship of fuel consumption influencing factors; Fuel consumption prediction module: The control center realizes the fuel consumption prediction feedback by outputting the fuel consumption required for the planned route based on the improved spatio-temporal feature fusion neural network model ST-FNN.

[0006] Preferably, the data of ship navigation influencing factors includes ship motion parameters, engine working condition parameters, marine environment parameters, and ship self-parameters; the ship motion parameters specifically include: real-time speed V, course angle and acceleration a; the engine working condition parameters specifically include engine speed n, propeller torque T, and fuel injection quantity Q; the marine environment parameters specifically include seawater density , ocean current velocity U, wind direction , wind speed and significant wave height ; the ship self-parameters include cargo weight M, draft depth d, and hull fouling roughness R.

[0007] Preferably, the preprocessing operation includes outlier detection and repair, normalization processing, and composite feature construction; the outlier detection and repair: using the Local Outlier Factor (LOF) algorithm to detect outliers in each parameter sequence, and for the detected abnormal data points, using the weighted average of the data at the previous and next moments for repair, and the weighting coefficient is determined according to the reciprocal of the time interval; the normalization processing: uniformly mapping all parameters to the interval [0, 1], using the min-max normalization method; the composite feature construction: includes constructing a relative wind speed feature model, a wave interference feature model, and a propulsion efficiency feature model.

[0008] Preferably, the relative wind speed feature model is used to input the real-time speed V, wind speed , wind direction and course angle into the relative wind speed feature model to obtain the relative wind speed: ; the wave interference feature model is used to input the significant wave height , draft depth d, relative wind speed and real-time speed V into the wave interference feature model to obtain the wave interference factor: , where represents the reference speed parameter, which is used to correct the hydrodynamic nonlinear effect during low-speed ship navigation; the propulsion efficiency feature model is used to input the propeller torque T, engine speed n, and fuel injection quantity Q into the propulsion efficiency feature model to obtain the propulsion efficiency coefficient: , where represents the fuel density, and HV represents the fuel calorific value.

[0009] Preferably, the improved spatio-temporal feature fusion neural network model ST-FNN includes an input layer, a spatio-temporal feature extraction layer, a spatial feature interaction layer, and an output layer; the input layer: receives the preprocessed data of ship navigation influencing factors and composite features, and sets the input dimension N = 17; the spatio-temporal feature extraction layer: consists of two layers of bidirectional long short-term memory networks Bi-LSTM, each layer containing 128 memory units, which are used to capture long-distance dependencies in the time series and spatio-temporal evolution features of ship motion; the spatial feature interaction layer: adopts a graph neural network GNN structure, regards the preprocessed data of ship navigation influencing factors and composite features as graph nodes, constructs a parameter correlation graph, and captures the spatial coupling relationship between parameters through graph convolution operations; the output layer: adopts a fully connected layer, and outputs the estimated fuel consumption rate f = , where, and represent the weights and biases of the output layer, represents the output vector after spatio-temporal feature fusion.

[0010] Preferably, the graph convolution operation is specifically expressed as: , where v represents a node and l represents the number of layers; represents the feature vector of the v node in the l-th layer; represents the feature vector of the adjacent node u of the v node in the l-th layer, represents the set of neighborhood nodes of node v; |N(v)| represents the number of elements in the neighborhood node set; represents the weight matrix of the l-th layer, which is used to linearly transform the features of neighborhood nodes; represents the bias vector of the l-th layer, which is used to preserve the independence of the node's own features; represents the activation function, which is used to introduce non-linear transformation.

[0011] Preferably, the method for training the improved spatio-temporal feature fusion neural network model ST-FNN is: adopting the mean squared error MSE loss function , where P is the number of training samples, and are the estimated fuel consumption and actual fuel consumption of the i-th sample respectively; the optimization algorithm adopts the adaptive moment estimation Adam algorithm, sets the initial learning rate to 1e-4, and the learning rate decays to 0.8 times the original every 50 training cycles. During the training process, an early stopping strategy is adopted to prevent overfitting, and the training stops when the validation set loss no longer decreases within 20 training cycles of the early stopping patience value.

[0012] Preferably, the fuel consumption prediction feedback method is as follows: deploy the trained ST-FNN model to the ship control center. During the ship's voyage, based on the output of the ST-FNN model, obtain the predicted fuel consumption rate at the current moment. , the control center compares the prediction result with the historical average fuel consumption data and displays it through the real-time monitoring terminal. When the predicted fuel consumption rate deviates from the historical average fuel consumption data by more than 20%, automatically identify the parameters that have a significant impact on fuel consumption. At the same time, the system automatically triggers an alarm to prompt the crew to check the current navigation status.

[0013] Technical effects and advantages of the present invention: 1. By deploying multi-source data acquisition terminals, the present invention can obtain ship motion parameters, engine operating conditions parameters, marine environment parameters, and ship self-parameters in real time, and construct a composite feature system including relative wind speed, wave interference factor, and propulsion efficiency coefficient, effectively solving the problem that traditional methods rely only on single variables and are difficult to capture the coupling effect of multiple variables, realizing a more comprehensive characterization of complex influencing factors in ship navigation, and laying a data foundation for high-precision fuel consumption prediction; 2. With the help of the improved spatio-temporal feature fusion neural network model ST-FNN, the present invention uses the bidirectional long short-term memory network Bi-LSTM to capture long-distance dependencies in time series and the graph neural network GNN to mine coupling associations in parameter space, solving the problems of insufficient dynamic feature learning ability of traditional models in the background technology and significant prediction errors under complex sea conditions, achieving in-depth modeling of the spatio-temporal dynamic characteristics of ship navigation and improving the accuracy of fuel consumption prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a schematic diagram of the terminal connection of the present invention; Figure 2 is a schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0016] As shown in the attached Figure 1 drawings, a ship navigation fuel consumption prediction system based on deep learning includes a data acquisition terminal, a data preprocessing terminal, a deep learning terminal, a real-time monitoring terminal, and a control center; The data acquisition terminal: responsible for collecting multi-dimensional and multi-modal raw data during the ship's navigation in real time, providing basic information for subsequent analysis; The data preprocessing terminal: denoises, normalizes, and enhances the features of the raw data to improve data quality and mine implicit information; The deep learning terminal: mines the spatio-temporal correlation laws in the data through a deep learning model to achieve high-precision prediction of the fuel consumption rate; The real-time monitoring terminal: presents real-time data, prediction results, and anomaly warnings in a visual manner to assist the crew in making quick decisions; The control center: responsible for coordinating the collaborative work of each terminal, integrating the data processing results, and generating intelligent decision-making suggestions.

[0017] As shown in the appendix Figure 2 A ship navigation fuel consumption prediction system based on deep learning also includes: The multi-source data acquisition module: obtains data on factors affecting ship navigation from multiple angles through the data acquisition terminal; In this embodiment, it should be specifically noted that: the data on factors affecting ship navigation includes ship motion parameters, engine operating parameters, marine environment parameters, and ship self-parameters; the ship motion parameters specifically include: real-time speed V, course angle course angle and acceleration a; the engine operating parameters specifically include engine speed n, propeller torque T, and fuel injection quantity Q; the marine environment parameters specifically include seawater density , ocean current velocity U, wind direction , wind speed and significant wave height ; the ship self-parameters include cargo weight M, draft depth d, and hull fouling roughness R; the real-time speed V directly affects the ship's propulsion power requirement, and the higher the speed, the non-linear growth of fuel consumption per unit distance may be; the course angle reflects the relative relationship between the ship's navigation direction and environmental factors and affects the navigation resistance; the acceleration a characterizes the ship's acceleration and deceleration frequency, and frequent speed changes will increase fuel consumption; the engine speed n and propeller torque T reflect the power output intensity and are directly related to the fuel injection quantity Q; the greater the seawater density , the change in hull buoyancy affects the draft depth, indirectly changing the navigation resistance; the ocean current velocity U, wind direction , wind speed and significant wave height : following the current or wind can reduce energy consumption, while going against the current or wind or high waves will significantly increase the propulsion resistance; the cargo weight M and draft depth d: the greater the load, the deeper the draft, and the greater the frictional resistance between the hull and seawater; the hull fouling roughness R: fouling will increase the surface resistance of the hull.

[0018] Data preprocessing module: preprocess the data of the influencing factors of ship navigation through a data preprocessing terminal; In this embodiment, it should be specifically noted that: the preprocessing operations include outlier detection and repair, normalization processing, and composite feature construction; the outlier detection and repair: use the Local Outlier Factor (LOF) algorithm to detect outliers in each parameter sequence. For the detected abnormal data points, use the weighted average of the data at the previous and next moments for repair, and the weighting coefficient is determined according to the reciprocal of the time interval; the normalization processing: map all parameters to the interval [0, 1] uniformly, using the min-max normalization method; the composite feature construction: includes constructing a relative wind speed feature model, a wave interference feature model, and a propulsion efficiency feature model; it should be understood that: the purpose of data preprocessing is to clean the original data, unify the data scale, and construct composite features to reveal the physical associations between parameters; the relative wind speed feature model is used to input the real-time ship speed V, wind speed , wind direction and course angle into the relative wind speed feature model to obtain the relative wind speed: ; the wave interference feature model is used to input the significant wave height , draft depth d, relative wind speed and real-time ship speed V into the wave interference feature model to obtain the wave interference factor: , where represents the reference speed parameter, which is used to correct the hydrodynamic nonlinear effect during low-speed ship navigation; it should be understood that: the ratio of wave height to draft depth reflects the impact intensity of waves on the hull. Combining the ratio of relative wind speed to ship speed, the navigation resistance interference under the combined action of waves and wind is quantified; the propulsion efficiency feature model is used to input the propeller torque T, engine speed n, and fuel injection quantity Q into the propulsion efficiency feature model to obtain the propulsion efficiency coefficient: , where represents the fuel density, and HV represents the calorific value of the fuel.

[0019] Model construction module: construct and train an improved spatio-temporal feature fusion neural network model ST-FNN through a deep learning terminal to capture the time series dependence and spatial coupling relationship of fuel consumption influencing factors; In this embodiment, it should be specifically noted that: the improved spatio-temporal feature fusion neural network model ST-FNN includes an input layer, a spatio-temporal feature extraction layer, a spatial feature interaction layer, and an output layer; the input layer: receives the preprocessed original ship navigation influencing factor data and composite features, and sets the input dimension N = 17; it should be understood that: the dimension N = 17 specifically includes: 14 ship navigation influencing factor data and 3 composite features, and converts the original data and artificially designed key features into digital signals recognizable by the model to form a 17-dimensional input vector; the spatio-temporal feature extraction layer consists of two layers of bidirectional long short-term memory networks Bi-LSTM, and each layer contains L = 128 memory units, which are used to capture long-distance dependencies in the time series and spatio-temporal evolution features of ship motion; it should be understood that: Bi-LSTM processes the sequence through forward and backward neurons simultaneously, extracts context information of past and future moments, and the input data at each moment is input from both the "past → future" and "future → past" directions simultaneously to capture the dependencies of previous and subsequent moments. The memory unit simulates the human memory mechanism, and controls the retention and forgetting of information through the "forget gate", "input gate", and "output gate" to avoid the "memory decay" of long sequence data. After being processed by two layers of Bi-LSTM, a hidden state vector containing time series features is generated ; the spatial feature interaction layer: adopts the graph neural network GNN structure, regards each parameter as a graph node, constructs a parameter association graph, and captures the spatial coupling relationship between parameters through graph convolution operations to generate a hidden state vector containing the spatial coupling relationship of parameters ; the specific expression of the graph convolution operation is: , where v represents the node and l represents the layer number; represents the feature vector of the v node in the l-th layer; represents the feature vector of the adjacent node u of the v node in the l-th layer, represents the set of neighborhood nodes of node v; |N(v)| represents the number of elements in the neighborhood node set; represents the weight matrix of the l-th layer, which is used for linear transformation of the neighborhood node features; represents the bias vector of the l-th layer, which is used to retain the independence of the node's own features; represents the activation function, which is used to introduce non-linear transformation; the output layer: adopts a fully connected layer, and outputs the estimated fuel consumption rate f = through a linear activation function. Among them, the fuel consumption is a continuous physical quantity, and there is no need for non-linear transformation, and the result is directly output through linear combination, and represent the weights and biases of the output layer, which are used to "map" the feature vector to the fuel consumption rate, represents the output vector after spatio-temporal feature fusion, which is the time feature and spatial features are concatenated into a long vector; the method for training the improved spatio-temporal feature fusion neural network model ST-FNN is as follows: the mean squared error MSE loss function is adopted , where P is the number of training samples, and are the estimated fuel consumption and the actual fuel consumption of the i-th sample respectively; calculate the "mean squared difference" between the estimated fuel consumption and the actual fuel consumption of all training samples. The smaller the difference, the more accurate the model; the optimization algorithm adopts the adaptive moment estimation Adam algorithm, the initial learning rate is set to 1e-4, and the learning rate decays to 0.8 times the original every 50 training cycles. The early stopping strategy is adopted during the training process to prevent overfitting, and the training stops when the loss of the validation set no longer decreases within 20 training cycles of the early stopping patience value.

[0020] Fuel consumption estimation module: The control center outputs the fuel consumption required for the planned route based on the improved spatio-temporal feature fusion neural network model ST-FNN to achieve fuel consumption estimation feedback.

[0021] In this embodiment, specifically, it should be noted that the fuel consumption estimation feedback method is as follows: deploy the trained ST-FNN model to the ship control center. During the ship's navigation, based on the output of the ST-FNN model, obtain the estimated fuel consumption rate at the current moment , and the control center compares the estimated result with the historical average fuel consumption data and displays it through the real-time monitoring terminal. When the estimated fuel consumption rate deviates from the historical average fuel consumption data by more than 20%, automatically identify the parameters that have a significant impact on fuel consumption, and at the same time, the system automatically triggers an alarm to prompt the crew to check the current navigation status.

[0022] Secondly: In the accompanying drawings of the disclosed embodiments of the present invention, only the structures related to the disclosed embodiments of the present disclosure are involved. Other structures can refer to the general design. Without conflict, the same embodiment and different embodiments of the present invention can be combined with each other; Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A ship navigation fuel consumption prediction system based on deep learning, characterized in that It includes a data acquisition terminal, a data preprocessing terminal, a deep learning terminal, a real-time monitoring terminal, and a control center, specifically including: Multi-source data acquisition module: Obtain data on factors affecting ship navigation from multiple angles through the data acquisition terminal; Data preprocessing module: Perform preprocessing operations on the data on factors affecting ship navigation through the data preprocessing terminal; Model construction module: Construct and train an improved spatio-temporal feature fusion neural network model ST-FNN through the deep learning terminal to capture the time series dependence and spatial coupling relationship of fuel consumption influencing factors; Fuel consumption estimation module: The control center outputs the fuel consumption required for the planned route based on the improved spatio-temporal feature fusion neural network model ST-FNN to achieve fuel consumption estimation feedback.

2. The ship navigation fuel consumption prediction system based on deep learning according to claim 1, wherein: The data on factors affecting ship navigation includes ship motion parameters, engine operating condition parameters, marine environment parameters, and ship own parameters; the ship motion parameters specifically include: real-time speed V, course angle and acceleration a; the engine operating condition parameters specifically include engine speed n, propeller torque T, and fuel injection quantity Q; the marine environment parameters specifically include seawater density , ocean current velocity U, wind direction , wind speed and significant wave height ; the ship own parameters include cargo weight M, draft d, and hull fouling roughness R.

3. The ship navigation fuel consumption prediction system based on deep learning according to claim 1, characterized in that: The preprocessing operations include outlier detection and repair, normalization processing, and composite feature construction; The outlier detection and repair: Use the Local Outlier Factor (LOF) algorithm to detect outliers in each parameter sequence. For the detected abnormal data points, use the weighted average of the data at the previous and next moments for repair, and the weighting coefficient is determined according to the reciprocal of the time interval; The normalization processing: Map all parameters to the [0, 1] interval uniformly, using the min-max normalization method; The composite feature construction: Include constructing a relative wind speed feature model, a wave interference feature model, and a propulsion efficiency feature model.

4. The ship navigation fuel consumption prediction system based on deep learning according to claim 3, characterized in that: The relative wind speed characteristic model is used to input the real-time ship speed V, wind speed , wind direction and course angle into the relative wind speed characteristic model to obtain the relative wind speed: ; the wave interference characteristic model is used to input the significant wave height , draft d, relative wind speed and real-time ship speed V into the wave interference characteristic model to obtain the wave interference factor: , where represents the reference speed parameter, which is used to correct the hydrodynamic nonlinear effect during low-speed navigation of the ship; The propulsion efficiency characteristic model is used to input the propeller torque T, the engine speed n, and the fuel injection quantity Q into the propulsion efficiency characteristic model to obtain a propulsion efficiency coefficient: , where represents the fuel density, and HV represents the fuel calorific value.

5. The ship navigation fuel consumption prediction system based on deep learning according to claim 1, characterized in that: The improved spatio-temporal feature fusion neural network model ST-FNN includes an input layer, a spatio-temporal feature extraction layer, a spatial feature interaction layer, and an output layer; the input layer: receives the preprocessed data of ship navigation influencing factors and composite features, and sets the input dimension to 17; the spatio-temporal feature extraction layer: consists of two layers of bidirectional long short-term memory networks Bi-LSTM, each layer containing 128 memory units, which are used to capture the long-distance dependencies in the time series and the spatio-temporal evolution characteristics of ship motion; the spatial feature interaction layer: adopts a graph neural network GNN structure, regards the preprocessed data of ship navigation influencing factors and composite features as graph nodes, constructs a parameter correlation graph, and captures the spatial coupling relationship between parameters through graph convolution operations; the output layer: adopts a fully connected layer, and outputs the estimated fuel consumption rate f = , where and represent the weights and biases of the output layer, represents the output vector after spatio-temporal feature fusion.

6. The ship navigation fuel consumption prediction system based on deep learning according to claim 5, characterized in that: The specific representation of the graph convolution operation is as follows: , where v represents a node and l represents the layer number; represents the feature vector of node v in the l-th layer; represents the feature vector of the adjacent node u of node v in the l-th layer, represents the set of neighborhood nodes of node v; |N(v)| represents the number of elements in the neighborhood node set; represents the weight matrix in the l-th layer, which is used to perform a linear transformation on the neighborhood node features; represents the bias vector in the l-th layer, which is used to retain the independence of the node's own features; represents the activation function, which is used to introduce a non-linear transformation.

7. The ship navigation fuel consumption prediction system based on deep learning according to claim 1, characterized in that: The method for training the improved spatio-temporal feature fusion neural network model ST-FNN is as follows: the mean squared error MSE loss function is adopted , where P is the number of training samples, and are the predicted fuel consumption and the actual fuel consumption of the i-th sample respectively; The optimization algorithm uses the Adaptive Moment Estimation (Adam) algorithm, sets the initial learning rate to 1e-4, and the learning rate decays to 0.8 times the original value every 50 training cycles. The early stopping strategy is adopted during the training process to prevent overfitting, and the training stops when the validation set loss no longer decreases within 20 training cycles of the early stopping patience value.

8. The ship navigation fuel consumption prediction system based on deep learning according to claim 1, characterized in that: The fuel consumption prediction feedback method is as follows: Deploy the trained ST-FNN model to the ship control center. During the ship's voyage, based on the output of the ST-FNN model, obtain the predicted fuel consumption rate at the current moment. , The control center compares the prediction result with the historical average fuel consumption data and displays it through the real-time monitoring terminal. When the predicted fuel consumption rate deviates from the historical average fuel consumption data by more than 20%, automatically identify the parameters that have a significant impact on fuel consumption, and at the same time, the system automatically triggers an alarm to prompt the crew to check the current navigation status.

Images