CN106779137A - A kind of method that ship oil consumption is predicted according to sea situation and operating condition - Google Patents

Abstract

The invention discloses a method for predicting fuel consumption of a ship according to sea conditions and maneuvering conditions. The method for predicting fuel consumption of a ship first obtains a sample data set through data screening, data integration and normalization processing, and then establishes a multiple linear regression prediction model based on For the sample data set, the minimum shrinkage and selection operator (LASSO algorithm) is used to define a cost function, and the variable shrinkage and parameter selection are combined with the cross-validation and least angle regression (LARS) algorithm. Finally, the Osborne dual algorithm is used to solve the LASSO estimate, and the ship is calculated fuel consumption. A method for predicting fuel consumption of a ship according to sea conditions and maneuvering conditions proposed by the present invention can establish a functional relationship between the fuel consumption of a ship and various influencing factors, solve the problem of multicollinearity in the calculation of fuel consumption, and improve the calculation accuracy of fuel consumption of a ship. Energy saving and emission reduction in marine transportation is of great significance.

Description

A Method of Forecasting Fuel Consumption of Ships Based on Sea State and Maneuvering Conditions

technical field

The invention belongs to the field of ship shipping, and in particular relates to a method for predicting fuel consumption of a ship according to sea conditions and maneuvering conditions.

Background technique

Fuel consumption during ship navigation is affected by various uncertain factors such as meteorological and sea conditions, ship navigation status, and operating conditions, and fuel consumption prediction has important guiding significance for ship speed, course, and other operations. Accurate fuel consumption calculation can better guide the crew to steer the ship and ensure sailing with a low fuel consumption, low emission and high-efficiency maneuvering plan within the time required by the ship schedule. However, the uncertainty of influencing factors and the high correlation between some factors make it impossible to accurately determine the functional relationship between fuel consumption and various influencing factors, and it is also impossible to accurately calculate the fuel consumption of a ship during its voyage.

Statistical machine learning is based on the distribution of a large amount of historical data, combined with statistical theory to establish model assumptions, through the design algorithm process to estimate parameters from data and models, continuous learning to improve model performance, automatically find the laws that exist between data, and calculate a pair A research method that approximates the functional relationship of the natural model and uses the obtained functional relationship to predict unknown data, test the generalization ability of the model, and verify the model assumptions

Regression analysis is one of the commonly used methods in data mining. It reflects the characteristics of data attribute values, and expresses the relationship of data mapping in the form of functions, so as to discover the dependencies between attribute values. Regression analysis method is widely used in the prediction of data sequence and the study of correlation.

The LASSO algorithm is a biased estimation method in machine learning algorithms that can simplify the index set and deal with data with multicollinearity properties. _The algorithm introduces the L1 penalty function in the loss function, under the constraint that the sum of the absolute values of the regression coefficients is less than a constant, the residual sum of squares is minimized, so that the regression coefficients of some indicators are strictly equal to zero, that is to say , the LASSO algorithm has the characteristic of variable sparseness, which can eliminate redundant features, find only the variables needed by the problem, realize the simplification of the index set, and obtain a model with strong explanatory ability for the results.

The LARS algorithm is used to decide which variables are selected into the model and give parameter estimates for the corresponding variables. The algorithm does not directly add a variable in each step forward stepwise regression, but first finds the variable with the highest correlation with the dependent variable, and then adjusts the coefficient of the variable along the direction of the minimum square error. , the correlation coefficient between the variable and the residual item will gradually decrease until the correlation between the new variable and the residual is greater than the correlation between the variable and the residual at this time, and then readjust along the direction of the minimum square error, This process is repeated until all variables are selected into the set of independent variables, and finally the desired parameter estimates are obtained.

Contents of the invention

The invention provides a method for predicting fuel consumption of a ship according to sea conditions and maneuvering conditions, which can solve the problem of fuel consumption calculation during ship navigation, and comprehensively consider factors such as weather, sea conditions, navigation status, and maneuverability that affect ship navigation, by constructing a statistical model and Combined with the machine learning algorithm, the model is used for parameter learning and model training, and finally a fuel consumption prediction model is obtained through training to solve the problem of rapid and accurate prediction of ship fuel consumption under the influence of various factors.

The present invention provides a method for predicting fuel consumption of ships according to sea conditions and maneuvering conditions. The main invention is the model framework for predicting fuel consumption of ships based on the LASSO algorithm.

The ship fuel consumption prediction model framework based on the LASSO algorithm is based on the ship's historical voyage data and climate and sea state data. Firstly, a linear regression prediction model is established and assumptions are made, and then a cost function is defined by the LASSO algorithm, which is converted to solve the minimum square error _loss under the L1 constraint. Based on the optimized convex quadratic programming problem, combined with the LARS algorithm to solve the sparse solution of the LASSO problem, the coefficient shrinkage and variable selection are realized, and finally the ship fuel consumption prediction model is obtained through training.

In order to achieve the above object, the present invention mainly comprises the following steps:

Step (1) Raw data collection. Mainly use different equipment and ways to collect climate and sea state data, ship navigation status data and ship maneuvering data.

Step (2) data preprocessing. Perform data screening on the raw data collected in (1), perform data denoising, cropping, and synchronization processing, then integrate data from multiple data sources while removing redundancy, and finally perform unified normalization processing on the integrated data , to get the model training data set.

Step (3) Establish a fuel consumption prediction model. According to the distribution characteristics of the training data set in step (2) and empirical assumptions that the input feature variables and the corresponding predicted output variables satisfy a linear relationship, a multiple linear regression prediction model is established.

Step (4) parameter learning and model training. Based on the training data obtained in step (2), the LASSO algorithm is used to define a cost function, which is transformed into a convex quadratic programming problem for the minimization of the square error loss under the L ₁ constraint, and the model parameters are finally combined with cross-validation and LARS algorithm Learn, and use the Osborne dual algorithm to solve the LASSO estimate. Finally, the estimated value of LASSO is substituted into the fuel consumption prediction model established in step (3) to obtain the trained fuel consumption prediction model.

Step (5) model application. Inputting the unknown meteorological and sea state data and ship maneuvering data into the fuel consumption prediction model obtained in step (5), can output the final fuel consumption prediction result.

Compared with existing artificial neural network (BP-ANN), support vector regression (SVR) predictive model, the present invention has following characteristics:

1. The present invention can realize variable sparseness and solve problems such as multicollinearity and overfitting.

The present invention mainly uses the LASSO algorithm combined with the least squares method to define a cost function, and based on the training data, combines the cross-validation and the LARS algorithm to solve the sparse solution, realizes parameter contraction and selection, and solves the problem of multicollinearity while realizing variable selection. In addition, the LASSO algorithm introduces _an L1 regularization term in the objective function, and weighs the proportion between the regularization term and the square error term through the regularization coefficient, which prevents the model from overfitting to a certain extent.

2. The present invention has strong generalization ability and result interpretation ability.

The present invention comprehensively considers a variety of possible factors affecting the fuel consumption prediction of ships, can obtain accurate fuel consumption prediction results on unknown test data sets, and can better predict fuel consumption mutations caused by mutations of influencing factors, and has strong generalization ability and ability to interpret predictions. .

Description of drawings

Fig. 1 is a block diagram of a method for predicting fuel consumption of ships according to sea conditions and maneuvering conditions in the present invention;

Figure 2 is a comparison chart of prediction results of LASSO, BP-ANN and SVR prediction models;

Figure 3 is a comparison chart of cumulative scores of prediction results of LASSO, BP-ANN and SVR prediction models;

Figure 4 shows the average absolute error values of the prediction results of LASSO, BP-ANN and SVR prediction models.

detailed description

A method for predicting fuel consumption of ships according to sea conditions and maneuvering conditions provided by the present invention will be described in detail below in conjunction with the accompanying drawings. Figure 1 is a block diagram of a method for predicting fuel consumption of ships based on sea conditions and handling conditions.

As shown in Fig. 1, the construction of the model of the present invention mainly includes five steps: step (1) raw data collection. Mainly use different equipment and ways to collect climate and sea state data, ship navigation status data and ship maneuvering data; step (2) data preprocessing. Perform data screening on the raw data collected in (1), perform data denoising, cropping, and synchronization processing, then integrate data from multiple data sources while removing redundancy, and finally perform unified normalization processing on the integrated data , to obtain a model training data set; step (3) to establish a fuel consumption prediction model. According to the distribution characteristics of the training data set in step (2) and the empirical assumption that the input feature variable and the corresponding predicted output variable satisfy a linear relationship, a multiple linear regression prediction model is established; step (4) parameter learning and model training. Based on the training data obtained in step (2), the LASSO algorithm is used to define a cost function, which is transformed into a convex quadratic programming problem for the minimization of the square error loss under the L ₁ constraint, and the model parameters are finally combined with cross-validation and LARS algorithm Learn, and use the Osborne dual algorithm to solve the LASSO estimate. Finally, the LASSO estimated value is substituted into the fuel consumption prediction model established in step (3) to obtain the trained fuel consumption prediction model; step (5) model application. Inputting the unknown meteorological and sea state data and ship maneuvering data into the fuel consumption prediction model obtained in step (4), can output the final fuel consumption prediction result.

Step 1: Raw Data Collection

The original data mainly include meteorological and sea state data, ship navigation history data and ship maneuvering data. Meteorological data mainly comes from weather forecast information released by the Meteorological Bureau, sea state data is indirectly measured by some sensors, ship navigation history data and maneuvering data are obtained through the noon report of the sailing log and high-speed, high-precision sensor components installed on the ship, and finally the data is transmitted to the database server for storage and analysis. In order to predict the fuel consumption of ships, the signals that need to be collected include fuel consumption, fuel density, fuel temperature, trim angle, heel angle, ship height, pitch, rudder angle, true heading, yaw angle, draft, wind angle, wind speed, and water speed , ground speed, longitude, and latitude, in which fuel consumption is used as the output response, and other variables are used as input variables of the model.

Step 2: Data preprocessing

Data preprocessing is the premise to ensure high-quality model prediction results. There are generally a large number of incomplete, inconsistent, repetitive, high-dimensional and noisy data in the massive raw data collected in practice, which seriously affects the execution efficiency of machine learning algorithms and the complexity of the model. In the present invention, the original data collected in step 1 is processed through data screening, data integration and normalization to obtain a sample data set.

Step 3: Establish a fuel consumption prediction model

Since the fuel consumption of a ship is the result of multiple factors acting together, it is assumed that the fuel consumption and each influencing factor satisfy a linear relationship, so multiple linear regression is used to establish a prediction model, and the fuel consumption prediction model can be expressed by the following formula:

y _i = β ^T x _i +∈ _i (1)

In the formula, x _i represents the i-th sample, and y _i is the response variable corresponding to the i-th sample, that is, fuel consumption. ∈ _i is a random error that obeys normal distribution, β=(β ₁ ,β ₂ ,…,β _p ) ^T is the regression coefficient variable, where β _j represents the jth regression coefficient.

Step 4: Parameter learning and model training

Because there may be a high degree of correlation between factors that affect fuel consumption, such as wind speed and wave height, air pressure and wind strength, cargo weight and draft, etc., in order to ensure that each input variable in step 4 is independent of each other, the present invention adopts the LASSO algorithm. Parameter shrinkage and variable selection. The LASSO algorithm can be expressed as solving the problem of minimizing the square error _loss under the L1 constraint as follows:

In the formula, N represents the sample size, p represents the number of input variables contained in each sample, x _ij represents the jth input variable corresponding to the i-th sample, β _j represents the regression coefficient corresponding to the j-th input variable, and t represents the prior The set free parameter that determines the degree of regularization, t≥0.

Express (2) in matrix form:

st||β _j || ₁ ≤ t (3)

because so,

Equation (3) can be rewritten as:

st||β _j || ₁ ≤ t (5)

The Lagrange multiplier method is expressed as:

In the formula, λ≥0 is an adjustment parameter, which weighs the proportion of the error square term and the L ₁ regularization term. In the present invention, the combination of the cross-validation method and the LARS algorithm is adopted to calculate the parameter regularization path, and simultaneously solve the parameter λ in the formula (6), and λ corresponds to the λ value corresponding to the smallest mean square error in the mean value of all cross-validation results . Then the Osborne dual algorithm is used to solve the LASSO estimate. Finally, the estimated value of LASSO is substituted into formula (1) to obtain the fuel consumption prediction model.

Step Five: Model Application

Inputting the meteorological and sea state conditions and ship maneuvering condition data into the fuel consumption prediction model trained in step 4 can predict the fuel consumption of the ship under the weather and sea state conditions and maneuvering conditions. The predicted fuel consumption value is used to calculate the future ship's sailing speed according to the cubic function relationship between fuel consumption and sailing speed, so as to guide the crew to use low fuel consumption, low emission and high efficiency within the specified sailing period on the premise of ensuring safe sailing. Steering plan sailing.

Figure 2 is a comparison chart of prediction results of LASSO, BP-ANN and SVR prediction models. The figure shows that 20 sets of unknown meteorological and sea state data and ship maneuvering data are input into three fuel consumption prediction models, and the final fuel consumption prediction results and real fuel consumption values are output. It can be seen from the figure that under the same navigation conditions, the LASSO prediction model is more accurate in predicting fuel consumption, more sensitive to changes in the navigation environment, and the prediction results are more stable.

Figure 3 is a comparison chart of cumulative scores of prediction results of LASSO, BP-ANN and SVR prediction models. The mean absolute error (MAE) refers to the result of averaging the absolute value of the difference between the predicted value and the true value. The cumulative score is defined as:

Among them, M is the size of the unknown data set, and _Me<δ means that the absolute error of the prediction result is smaller than the size of the data set of δ tons/day. It can be seen from the figure that the error between the prediction result of the LASSO model and the real value is smaller than that of the BP-ANN model and the SVR model, and the prediction accuracy is higher.

Figure 4 shows the comparison results of the average absolute error values of the prediction results of LASSO, BP-ANN and SVR prediction models, which are obtained on the basis of 5000 test sets. It can be seen that the average absolute error of the prediction results of the LASSO model is the smallest, and the variation range of the average absolute error is also the smallest.

Various modifications and changes can be made to the present invention by those skilled in the art. Thus, the present invention covers the modifications and changes that come within the scope of the appended claims and their equivalents.

Claims (1)

1. a kind of method that ship oil consumption is predicted according to sea situation and operating condition, it is characterised in that it is described according to sea situation with manipulate The method of conditional forecasting ship oil consumption includes following five steps：

Raw data acquisition step：The Weather Forecast Information from weather bureau's issue is collected as meteorological data；By sensor Sea state data is measured indirectly；Ship's navigation historical data is obtained by log noon report；By installing biography aboard ship Sensor component obtains Ship Controling data；Data above is finally sent to database server storage；

Data prediction step：Data screening is carried out for the initial data gathered in raw data acquisition step, data are gone Make an uproar, cut and synchronization process, then the integrated data from multiple data sources remove redundancy simultaneously, and finally integrated data are entered The unified normalized of row, obtains model training data set；

Oil consumption forecast model establishment step：According to the characteristic distributions of training dataset and empirical hypothesis in data prediction step Meet linear relationship between input feature vector variable and corresponding prediction output variable, set up Multiple Linear Regression Forecasting Models of Chinese；

Parameter learning and model training step：It is fixed using LASSO algorithms based on the training data obtained in data prediction step An adopted cost function, and be converted to solution L₁The convex quadratic programming problem of the lower square error minimization of loss of constraint, most terminates Close cross validation to learn model parameter with LARS algorithms, and LASSO is solved using Osborne Conjugate Search Algorithms and estimate；Most In LASSO estimates substitution oil consumption forecast model establishment step is set up oil consumption forecast model afterwards, the oil consumption after being trained Forecast model；

Model applying step：Unknown meteorological sea state data and Ship Controling data input oil consumption forecast model establishment step are obtained To oil consumption forecast model in, export final oil consumption and predict the outcome.