Ship energy consumption intelligent prediction method based on deep learning and navigation feature extraction

Abstract

A ship energy consumption intelligent prediction method based on deep learning and navigation feature extraction comprises the steps that firstly, navigation data and navigation domain environment data of a ship are obtained, data cleaning and analysis processing are carried out, abnormal values and missing values in the data are removed, and change features of the data are analyzed; carrying out time scale alignment on the navigation data, and carrying out space-time interpolation matching on the aligned navigation data and the navigation domain environment data; performing feature extraction based on a Pearson's correlation coefficient to simplify prediction model input; overlapping sampling is performed on the data, a training set and a test set are constructed, and meanwhile, a ship energy consumption prediction black box model is constructed based on a deep learning method and a neural network theory; then, training the prediction model by using the training set data, updating model parameters by adopting a gradient descent method, and storing the model parameters after the training is finished; and finally, verifying the performance of the trained prediction model based on the test set data. According to the invention, accurate prediction of ship energy consumption can be effectively realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of ship energy consumption prediction, and specifically relates to an intelligent prediction method for ship energy consumption based on deep learning and navigation feature extraction. Background Art

[0002] Ship energy consumption is not only directly related to the operating costs of shipping companies, but also closely related to national energy security and environmental protection. Ship energy consumption prediction is an important prerequisite for optimizing ship energy consumption, and it can provide further decision-making support for ship shipping mission arrangements. Ship energy consumption is mainly affected by a variety of factors, including ship type, sailing speed, load conditions, climate conditions, and marine environment.

[0003] At present, ship energy consumption prediction models can be mainly divided into the following three categories: white box model, black box model and gray box model ([1] Fan A, Yang J, Yang L, et al. A review of ship fuel consumption models [J]. Ocean engineering, 2022, 264: 112405.). Among them, the white box model is a relatively common ship energy consumption model. It mainly establishes the physical connection between the input and output of the ship system based on the physical relationship and empirical formula between the internal and external parameters of the ship. The white box model is transparent and interpretable, but it involves certain assumptions and simplifications in the modeling process, and the consideration of various factors is often not comprehensive ([2] Wang K, Xu H, Li J, et al. A novel dynamical collaborative optimization method of ship energy consumption based on aspatial and temporal distribution analysis of voyage data [J]. Applied Ocean Research, 2021, 112: 102657.), during a complete voyage, the energy consumption of a ship is affected by various internal and external factors, and the navigation conditions are complex and changeable. The use of a white box prediction model may result in large deviations. As a purely data-driven model, the black box model has high data requirements and is mainly constructed through statistical methods or machine learning techniques. It does not rely on the internal principles of the ship system. The black box model can well capture the nonlinear relationship between ship energy consumption and various navigation factors, and achieve accurate ship energy consumption prediction. Therefore, it has also received more and more attention and research ([3] Yuan Z, Liu J, Zhang Q, et al. Prediction and optimisation of fuel consumption for inland ships considering real-time status and environmental factors [J]. Ocean Engineering, 2021, 221: 108530.). The gray box model is a model between the white box model and the black box model. It establishes the model structure through prior knowledge in related fields and determines the model parameters by training existing data. Although it has a certain degree of interpretability, it is easily affected by noise and requires prior knowledge and training data at the same time, making it difficult to implement in practice.

[0004] Different ships often have different navigation status, navigation environment, operating tasks, navigation cycles, etc., so the dynamic characteristics of fuel consumption are also complex and diverse. Existing ship energy consumption prediction model methods mostly focus on merchant ships or inland ships, and the influencing factors considered are often not comprehensive or have poor versatility, making it difficult to achieve more accurate predictions in terms of ship energy consumption. Therefore, how to improve the prediction accuracy of ship energy consumption and achieve accurate prediction of ship energy consumption has become a key challenge. Summary of the invention

[0005] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to provide an intelligent prediction method for ship energy consumption based on deep learning and navigation feature extraction, so as to improve the prediction accuracy of ship energy consumption and realize accurate prediction of ship energy consumption.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] An intelligent prediction method for ship energy consumption based on deep learning and navigation feature extraction uses the characteristics of end-to-end data-driven neural networks and capturing potential patterns within the data to construct a black box model for ship energy consumption prediction. It also improves the prediction accuracy of the black box model for ship energy consumption prediction by analyzing the ship's navigation operation status and navigation environment conditions.

[0008] A method for intelligent prediction of ship energy consumption based on deep learning and navigation feature extraction, comprising the following steps:

[0009] Step 1: Obtain the ship's navigation data D ND and flight environment data D NED , where the navigation data D ND It includes internal ship data and external ship data. The internal ship data includes the speed, pitch, and azimuth of each propeller, and the external ship data includes the ship's speed over the ground, the ship's course over the ground, the longitude and latitude of the ship, etc. The navigation data D ND The ground speed data and ground heading data in can be expressed as: and Wherein, n1∈1,…,N1, n2∈1,…,N2, respectively represent the n1th data point of the speed over ground data and the n2th data point of the course over ground data, N1 represents the data length of the speed over ground data, and N2 represents the data length of the course over ground data; the navigation environment data D NED Including wind speed, wave height, sea speed, etc. in the navigation area, where wind speed and wave height data can be expressed as: and Among them, n3∈1,…,N3, n4∈1,…,N4, respectively represent the n3th data point of wind speed data and the n4th data point of wave height data, N3 represents the data length of wind speed data, and N4 represents the data length of wave height data;

[0010] Step 2: The navigation data D obtained in step 1 ND and flight environment data D NED Perform data cleaning and analysis to remove outliers and missing values ​​(NaN); draw the change curves of various types of data in the navigation data to determine whether some of the data have mismatches due to data transmission delays; draw the navigation environment data D NED The numerical change curves of various data in the navigation environment data D NED The changes in the spatial and temporal characteristics of the distribution;

[0011] Step 3: The ship's navigation data D ND The internal data of the ship and the external data of the ship are aligned on a time scale, and then the aligned navigation data D ND According to its time, longitude and latitude values ​​and flight environment data D NED Further interpolation matching is performed in time and space scales;

[0012] Step 4: Perform Pearson correlation analysis on the data processed and matched in step 3 to complete feature extraction. The Pearson correlation coefficient calculation formula is as follows:

[0013]

[0014] In the formula: X and Y represent different characteristic variables; and Refers to the mean of feature variables X and Y respectively; X i and Y i Respectively represent the i-th sample point of the feature variable X and the feature variable Y, i∈1,…,n, n represents the sample length of the feature variable; r(X,Y) represents the correlation coefficient between the feature variable X and the feature variable Y, which is used to measure the degree of correlation between the two feature variables; when r is greater than 0, it means that the feature variable X and the feature variable Y are positively correlated, and the closer the value is to 1, the greater the degree of positive correlation; conversely, when r is less than 0, it means that the feature variable X and the feature variable Y are negatively correlated, and the closer the value is to -1, the greater the degree of negative correlation;

[0015] Step 5: Use overlapping sampling to construct data samples for the data extracted from step 4 to create a data set. Overlapping sampling first uses a time window of length w to connect the data points obtained at w consecutive sampling moments into a high-dimensional vector as a sample point, and then the time window slides back a step length s for the next sampling. The data set S obtained after overlapping sampling k for: The number of samples obtained after overlapping sampling can be expressed as: Each sample can be represented as: Wherein, the subscript k refers to the feature variable k obtained after completing the feature extraction in step 4, s refers to the time window sliding step, w refers to the time window length, norm refers to the normalization operation, and L refers to the data length for overlapping sampling;

[0016] Step 6: Based on deep learning methods and neural network theory, a black box model for ship energy consumption prediction is constructed;

[0017] Step 7: The dataset S generated by overlapping sampling in step 5 k The training set and the test set are divided into 8:2, and the black box model of ship energy consumption prediction constructed in step 6 is used for training and testing; the RMSE loss function is used in the model training process, and the optimizer adopts the Adam optimizer; after obtaining the ship energy consumption prediction results output by the model, a series of performance evaluation indicators are determined to measure the model prediction performance. The series of performance evaluation indicators include RMSE, MAE, MSE and MRE, and pictures are drawn to analyze and show the consistency between the actual fuel consumption value and the predicted fuel consumption value of the prediction black box model, as well as the volatility of the predicted value; the calculation formulas of RMSE, MAE, MSE and MRE indicators are as follows:

[0018]

[0019]

[0020]

[0021]

[0022] Where: t represents the actual fuel consumption of the ship, represents the fuel consumption predicted by the prediction model; z∈1,…,N,t z and They represent the actual fuel consumption value of the ship at the zth sample point and the fuel consumption value predicted by the prediction model respectively, and N is the sample length of the test set of the black box model for ship energy consumption prediction.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] The present invention proposes an intelligent prediction method for ship energy consumption based on deep learning and navigation feature extraction. First, the ship's navigation data and navigation environment data are collected and analyzed, and a series of processing is performed, including removing data errors and abnormal points, vector synthesis, data matching, and feature analysis and extraction. Since the neural network can well capture the complex feature associations in ship navigation and the potential patterns between fuel consumption and navigation characteristics, these complex feature information can be effectively used to improve resource efficiency, enhance model performance, and achieve more accurate ship energy consumption prediction. The low volatility of the model prediction value and its high consistency with the actual value of the ship's fuel consumption also verify the effectiveness of the present invention, which effectively improves the accuracy of ship energy consumption prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of an embodiment of the present invention.

[0026] Figure 2 This is a comparison curve between the model prediction value and the actual fuel consumption value of the embodiment of the present invention.

[0027] Figure 3 This is a scatter plot comparing the predicted values ​​of the training set and test set samples of the model embodiment of the present invention with the actual fuel consumption values. DETAILED DESCRIPTION

[0028] The present invention is further described in detail below in conjunction with the embodiments and drawings.

[0029] like Figure 1 As shown, a method for intelligent prediction of ship energy consumption based on deep learning and navigation feature extraction includes the following steps:

[0030] Step 1: Obtain the ship's navigation data D ND and flight environment data D NED , where the navigation data D ND It includes internal ship data and external ship data. The internal ship data includes the speed, pitch, and azimuth of each propeller, and the external ship data includes the ship's speed over the ground, the ship's course over the ground, the longitude and latitude of the ship, etc. The navigation data D ND The ground speed data and ground heading data in can be expressed as: and Wherein, n1∈1,…,N1,n2∈1,…,N2,represent the n1th data point of the speed over ground data and the n2th data point of the course over ground data, respectively; N1 represents the data length of the speed over ground data, and N2 represents the data length of the course over ground data; the navigation environment data D NED Including wind speed, wave height, sea speed, etc. in the navigation area, where wind speed and wave height data can be expressed as: and Wherein, n3∈1,…,N3, n4∈1,…,N4, respectively represent the n3th data point of wind speed data and the n4th data point of wave height data, N3 represents the data length of wind speed data, and N4 represents the data length of wave height data;

[0031] Step 2: The navigation data D obtained in step 1 ND and flight environment data D NED Perform data cleaning and analysis to remove outliers and missing values ​​(NaN); draw the change curves of various types of data in the navigation data to determine whether some of the data have mismatches due to data transmission delays; draw the navigation environment data D NED The numerical change curves of various data in the navigation environment data D NED The changes in the spatial and temporal characteristics of the distribution;

[0032] Step 3: The ship's navigation data D ND The internal data of the ship and the external data of the ship are aligned on a time scale, and then the aligned navigation data D ND According to its time, longitude and latitude values ​​and flight environment data D NED Further interpolation matching is performed in time and space scales;

[0033] Step 4: Perform Pearson correlation analysis on the data processed and matched in step 3 to complete feature extraction. The Pearson correlation coefficient calculation formula is as follows:

[0034]

[0035] In the formula: X and Y represent different characteristic variables; and Refers to the mean of feature variables X and Y respectively; X i and Y i Respectively represent the i-th sample point of the feature variable X and the feature variable Y, i∈1,…,n, n represents the sample length of the feature variable; r(X,Y) represents the correlation coefficient between the feature variable X and the feature variable Y, which is used to measure the degree of correlation between the two feature variables; when r is greater than 0, it means that the feature variable X and the feature variable Y are positively correlated, and the closer the value is to 1, the greater the degree of positive correlation; conversely, when r is less than 0, it means that the feature variable X and the feature variable Y are negatively correlated, and the closer the value is to -1, the greater the degree of negative correlation;

[0036] Step 5: Use overlapping sampling to construct data samples and create a data set for the data extracted from the features in step 4. Overlapping sampling uses a sliding window mechanism defined by two key parameters: the time window length Window_size and the sliding step length Stride. Overlapping sampling first uses a time window of length w to connect the data points obtained at w consecutive sampling moments into a high-dimensional vector as a sample point, and then the time window slides backward by a step length s for the next sampling. The data set S obtained after overlapping sampling k for: The number of samples obtained after overlapping sampling can be expressed as: Each sample can be represented as: Wherein, the subscript k refers to the feature variable k obtained after completing the feature extraction in step 4, s refers to the time window sliding step, w refers to the time window length, norm refers to the normalization operation, and L refers to the data length for overlapping sampling;

[0037] Step 6: Based on deep learning methods and neural network theory, a black box model for ship energy consumption prediction is constructed; neural networks can automatically extract and learn complex features in data through a multi-layer structure. Especially when faced with high-dimensional, nonlinear, and noisy data, they can continuously optimize their internal parameters through training to capture the deep "patterns" in the data. This feature makes neural networks particularly good at dealing with nonlinear problems and is suitable for modeling and prediction tasks of complex systems;

[0038] Step 7: The dataset S generated by overlapping sampling in step 5 k The training set and the test set are divided into 8:2, and the black box model of ship energy consumption prediction constructed in step 6 is used for training and testing; the RMSE loss function is used in the model training process, and the optimizer adopts the Adam optimizer; after obtaining the ship energy consumption prediction results output by the model, a series of performance evaluation indicators are determined to measure the model prediction performance. The series of performance evaluation indicators include RMSE, MAE, MSE and MRE, and pictures are drawn to analyze and show the consistency between the actual fuel consumption value and the predicted fuel consumption value of the prediction black box model, as well as the volatility of the predicted value; the calculation formulas of RMSE, MAE, MSE and MRE indicators are as follows:

[0039]

[0040]

[0041]

[0042]

[0043] Where: t represents the actual fuel consumption of the ship, represents the fuel consumption predicted by the prediction model; z∈1,…,N,t z and They represent the actual fuel consumption value of the ship at the zth sample point and the fuel consumption value predicted by the prediction model respectively, and N is the sample length of the test set of the black box model for ship energy consumption prediction.

[0044] The embodiment verifies the effectiveness of the method of the present invention based on the actual navigation data of a voyage of a certain ship and the actual environmental data of the navigation area.

[0045] After data processing and feature extraction, a total of 7210 valid data are obtained; the valid data are first normalized and overlapped sampled, and then randomly divided into two groups: 80% for training and the remaining 20% ​​for testing; in order to enhance the stability and convergence of the training, the StepLR learning rate scheduler is used, and the learning rate is multiplied by the Gamma factor for decay every specified number of epochs (step_size). This method helps prevent overshoot because it can gradually reduce the learning rate during the training process, thereby encouraging the model to converge to the optimal solution. The model training parameters are shown in Table 1, the model structure is shown in Table 2, and the four performance evaluation index results of the prediction model are shown in Table 3:

[0046] Table 1 Prediction model hyperparameter settings

[0047]

[0048] Table 2 Prediction model structure parameters

[0049]

[0050] Table 3 Results of prediction model performance evaluation indicators

[0051]

[0052]

[0053] After the model training is completed, the test set data is input into the energy consumption prediction model to obtain the predicted fuel consumption value output by the energy consumption prediction model. Figure 2 and Figure 3 It can be seen that the method of the present invention has high accuracy in predicting ship energy consumption, the predicted fuel consumption value is close to the true value, and the predicted value fluctuates little and is relatively stable; even when the navigation environment and navigation status change and fluctuate, the method of the present invention can still effectively capture the potential pattern of fuel consumption, which verifies the superiority and applicability of the method of the present invention, indicating that it has potential utility in optimizing ship operation strategies, and is expected to improve the prediction accuracy of ships in complex navigation scenarios.

Claims

1. A method for intelligent prediction of ship energy consumption based on deep learning and navigation feature extraction, characterized in that: By utilizing the end-to-end data-driven nature of neural networks and the ability to capture potential patterns within data, a black box model for predicting ship energy consumption is constructed. The prediction accuracy of the black box model for predicting ship energy consumption is improved by analyzing the ship's navigation operation status and navigation environment conditions.

2. A method for intelligent prediction of ship energy consumption based on deep learning and navigation feature extraction, characterized in that: The following steps are involved: Step 1: Obtain the ship's navigation data D ND and flight environment data D NED , where the navigation data D ND It includes internal ship data and external ship data. The internal ship data includes the speed, pitch and direction of each propeller. The external ship data includes the ship's speed over the ground, the ship's course over the ground, the longitude and latitude of the ship, and the navigation data D ND The ground speed data and ground heading data in are expressed as: and Wherein, n1∈1,…,N1, n2∈1,…,N2, respectively represent the n1th data point of the speed over ground data and the n2th data point of the course over ground data, N1 represents the data length of the speed over ground data, and N2 represents the data length of the course over ground data; the navigation environment data D NED It includes wind speed, wave height and sea speed in the navigation area, where wind speed and wave height data are expressed as: and Among them, n3∈1,…,N3, n4∈1,…,N4, respectively represent the n3th data point of wind speed data and the n4th data point of wave height data, N3 represents the data length of wind speed data, and N4 represents the data length of wave height data; Step 2: The navigation data D obtained in step 1 ND and flight environment data D NED Perform data cleaning and analysis to remove outliers and missing values ​​(NaN); draw the change curves of various types of data in the navigation data to determine whether some of the data have mismatches due to data transmission delays; draw the navigation environment data D NED The numerical change curves of various data in the navigation environment data D NED The changes in the spatial and temporal characteristics of the distribution; Step 3: The ship's navigation data D ND The internal data of the ship and the external data of the ship are aligned on a time scale, and then the aligned navigation data D ND According to its time, longitude and latitude values ​​and flight environment data D NED Further interpolation matching is performed in time and space scales; Step 4: Perform Pearson correlation analysis on the data processed and matched in step 3 to complete feature extraction. The Pearson correlation coefficient calculation formula is as follows: In the formula: X and Y represent different characteristic variables; and Refers to the mean of feature variables X and Y respectively; X i and Y i Respectively represent the i-th sample point of the feature variable X and the feature variable Y, i∈1,…,n, n represents the sample length of the feature variable; r(X,Y) represents the correlation coefficient between the feature variable X and the feature variable Y, which is used to measure the degree of correlation between the two feature variables; when r is greater than 0, it means that the feature variable X and the feature variable Y are positively correlated, and the closer the value is to 1, the greater the degree of positive correlation; conversely, when r is less than 0, it means that the feature variable X and the feature variable Y are negatively correlated, and the closer the value is to -1, the greater the degree of negative correlation; Step 5: Use overlapping sampling to construct data samples for the data extracted from step 4 to create a data set. Overlapping sampling first uses a time window of length w to connect the data points obtained at w consecutive sampling moments into a high-dimensional vector as a sample point, and then the time window slides back a step length s for the next sampling. The data set S obtained after overlapping sampling k for: The number of samples obtained after overlapping sampling is expressed as: Each sample is represented as: Wherein, the subscript k refers to the feature variable k obtained after completing the feature extraction in step 4, s refers to the time window sliding step, w refers to the time window length, norm refers to the normalization operation, and L refers to the data length for overlapping sampling; Step 6: Based on deep learning methods and neural network theory, a black box model for ship energy consumption prediction is constructed; Step 7: The dataset S generated by overlapping sampling in step 5 k The training set and the test set are divided into 8:2, and the black box model for ship energy consumption prediction constructed in step 6 is used for training and testing; the RMSE loss function is used in the model training process, and the optimizer adopts the Adam optimizer; after obtaining the ship energy consumption prediction results output by the model, a series of performance evaluation indicators are determined to measure the model prediction performance. The series of performance evaluation indicators include RMSE, MAE, MSE and MRE, and pictures are drawn to analyze and show the consistency between the actual fuel consumption value and the predicted fuel consumption value of the prediction black box model, as well as the volatility of the predicted value; the calculation formulas of RMSE, MAE, MSE and MRE indicators are as follows: Where: t represents the actual fuel consumption of the ship, represents the fuel consumption predicted by the prediction model; z∈1,…,N,t z and They represent the actual fuel consumption value of the ship at the zth sample point and the fuel consumption value predicted by the prediction model respectively, and N is the sample length of the test set of the black box model for ship energy consumption prediction.

Images