A ship speed prediction and optimization method, device and readable storage medium

Abstract

The application discloses a ship speed prediction and optimization method and device and a readable storage medium, and relates to the field of ship speed prediction and optimization.The method comprises the following steps: establishing a ship speed prediction model by taking a main engine power, a ship heading, a flow speed, a relative flow direction, a wind speed, a relative wind direction, an effective wave height, a relative wave direction, and a combination period of wind waves and swell waves as input and taking a ship speed as output; obtaining ship route planning information and marine weather forecast information; dividing a ship planned route into a plurality of planned segments; obtaining relevant information of each planned segment in combination with the marine weather forecast information, and then predicting real ship speeds of each planned segment under corresponding weather sea conditions and different main engine powers in combination with the ship speed prediction model; and establishing a cost function according to the main engine power and the corresponding real ship speed, and constructing a ship speed optimization model in combination with constraint conditions, so that optimized speeds of each planned segment are obtained. The application can predict and optimize ship speeds.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of ship operation management, and particularly relates to a ship speed prediction and optimization method, device and readable storage medium. BACKGROUND

[0002] While the shipping industry is driving the development of the global economy, its greenhouse gas emissions are also rising year by year. Ship speed is a key parameter for calculating fuel consumption and emissions, and is an important factor for evaluating the effect of energy efficiency design index (EEDI) and ship energy efficiency management plan (SEEMP). In order to improve fuel efficiency and reduce greenhouse gas emissions, it is necessary to accurately predict the ship speed.

[0003] In actual navigation, due to the complexity of the ship itself and the influence of the ocean environment resistance, it is difficult for the ship to sail at the planned speed. Existing ship speed prediction technologies, such as ship model test and fluid dynamics calculation method, not only consume a lot of time and money, but also are difficult to accurately explore the relationship between multiple factors. This leads to the fact that they cannot achieve ideal results. At the same time, existing ship speed optimization technologies cannot accurately predict the true speed of the ship in a variable ocean environment, which also leads to the fact that they cannot achieve ideal results. The prediction and optimization of the speed of the planned voyage is one of the key factors for optimizing ship operation, and is crucial for improving ship operation efficiency.

[0004] In summary, existing ship speed prediction technologies are mostly based on physical models established by theoretical derivation and ship model test, which are difficult to accurately depict the influence of complex ocean environment on the state of the ship, and many physical models ignore related features such as ocean currents to simplify the model. Therefore, predicting ship speed based on physical models is not accurate. SUMMARY

[0005] In view of the above defects or improvement needs of the prior art, the present application provides a ship speed prediction and optimization method, device and readable storage medium, which is based on a data-driven method of machine learning to solve the problem of inaccurate prediction of existing ship speed. Compared with physical models, the data-driven method based on machine learning can avoid subjective assignment of parameters, and can consider more factors affecting ship speed and explore the nonlinear relationship between multiple factors, thereby establishing a more accurate nonlinear ship speed prediction model. The accuracy of the ship speed prediction model is further improved by optimizing the hyperparameters of the machine learning model through the particle swarm algorithm. The speed optimization model based on high-accuracy ship speed prediction makes the predicted ship speed more consistent with the true speed of the ship navigation, thereby reducing the influence of weather information on ship speed and improving the accuracy of the calculation of ship navigation cost, and better achieving ship speed optimization.

[0006] In a first aspect of the present application, a ship speed prediction and optimization method is provided, which comprises:

[0007] A ship speed prediction model is established by taking main engine power, ship heading, flow speed, relative flow direction to the ship heading, wind speed, relative wind direction to the ship heading, significant wave height, relative wave direction to the ship heading, and combined period of wind wave and swell as inputs, and taking ship speed as output;

[0008] Ship route planning information and marine weather forecast information are obtained, and the marine weather forecast information includes flow speed, flow direction, wind speed, wind direction, significant wave height, wave direction, and combined period of wind wave and swell;

[0009] A ship planned route is obtained according to the ship route planning information, and the ship planned route is divided into a plurality of planned segments;

[0010] Ship headings of the planned segments are obtained, and relevant information of the planned segments is obtained in combination with the marine weather forecast information, including ship heading, flow speed, relative flow direction to the ship heading, wind speed, relative wind direction to the ship heading, significant wave height, relative wave direction to the ship heading, and combined period of wind wave and swell;

[0011] Real ship speeds of the planned segments under corresponding weather sea conditions and different main engine powers are predicted according to different main engine powers and the relevant information of the planned segments in combination with the ship speed prediction model;

[0012] A cost function is established according to the main engine power and the corresponding real ship speed, and a ship speed optimization model is constructed in combination with constraint conditions;

[0013] Based on the ship speed prediction model and the ship speed optimization model, optimized speeds of the planned segments are obtained.

[0014] Further, a cost function is established according to the main engine power and the corresponding real ship speed, and a ship speed optimization model is constructed in combination with constraint conditions, which comprises:

[0015] The cost of the total distance is calculated based on the following formula:

[0016]

[0017] Wherein, cost is the cost of the i-th planned segment, n is the total number of planned segments; v0 is the design speed, v i is the ideal speed under the given main engine power, h0 represents the fuel consumption rate under the design speed; D i represents the distance of the i-th planned segment; v irThe real ship speed of the ith planned voyage section under the same main engine power is predicted by the ship speed prediction model; a1 and a2 are coefficients corresponding to the fuel cost and the time cost respectively;

[0018] Meanwhile, the following constraints are met:

[0019] T n ≤T max , to ensure that the ship arrives at the destination port on time and follows the specified latest arrival time; wherein, T i represents the total time of the first i planned voyage sections, and T0 is 0;

[0020] The minimum and maximum limits of the ideal speed of the ship in each planned voyage section are set; wherein, and are the minimum speed and the maximum speed respectively;

[0021] T i ≥0, requiring the start time of each planned voyage section to be non-negative to ensure the implementability of the voyage plan.

[0022] Further, based on the ship speed prediction model and the ship speed optimization model, the optimized speed of each planned voyage section is obtained, including:

[0023] For each planned voyage section, the real ship speed of the ship under different main engine powers is obtained, so as to obtain the cost of each planned voyage section under different real ship speeds;

[0024] Combined with the constraint conditions, the real ship speed of each planned voyage section corresponding to the lowest cost of the total voyage is obtained, and the ship is controlled to sail at the corresponding real ship speed in the corresponding planned voyage section.

[0025] Further, a ship speed prediction model based on a particle swarm optimization algorithm is established, including:

[0026] The ship speed prediction model is a Catboost model, and the Catboost model hyperparameters are optimized based on the particle swarm algorithm; wherein, the Catboost model hyperparameters include the maximum number of trees iterations, the tree depth depth and the learning rate learning_rate;

[0027] The dimensions are determined according to the number of parameters to be optimized, and the positions and velocities of the particles are randomly initialized;

[0028] The performance of the ship speed prediction model is taken as the fitness value, and the optimal Catboost model hyperparameters are obtained through iteration, so as to obtain the ship speed prediction model based on the particle swarm optimization algorithm.

[0029] Further, the ship planning route is divided into several planned segments, including:

[0030] According to the time resolution of the marine weather forecast information and the designed speed, the planned route is divided into several planned segments by inserting planned waypoints.

[0031] According to the planned segments and the changes of the ship heading and the marine weather forecast information in the adjacent planned segments, the planned segments are further segmented or combined to obtain the final planned segments.

[0032] Further, a ship speed prediction model is established, including:

[0033] Obtain ship navigation history information and marine weather history information;

[0034] Based on the single-source data cleaning and multi-source data fusion method, a multi-source data aggregation processing model is constructed, and the obtained ship navigation history information and marine weather history information are subjected to single-source data cleaning and multi-source data fusion to obtain a feature set.

[0035] Based on the Pearson correlation analysis method, the key feature set that has a significant impact on the ship speed is selected from the feature set, the key features are used as the input of the ship speed prediction model, and the key feature set is used to train the ship speed prediction model.

[0036] Wherein, obtaining ship navigation history information and marine weather history information includes:

[0037] Obtain ship AIS data, including time, position, and corresponding ship speed, main engine power and ship heading; wherein, the position includes longitude and latitude;

[0038] Obtain marine weather history information, including time, position, and corresponding flow rate, flow direction, wind speed, wind direction, significant wave height, wave direction, and combination period of wind wave and surge wave.

[0039] Further, the single-source data cleaning includes:

[0040] For each single-source data, identify the outliers and replace them with missing values, and then fill the single-source data based on the tensor decomposition method, and use the Z-score method to standardize the filled data;

[0041] Wherein, the identification method includes: for significantly abnormal data, use multi-dimensional clustering and fixed threshold method for identification; for non-significant abnormal data, use theoretical knowledge and experience model for identification;

[0042] For ship speed and main engine power, two-dimensional tensor filling is used, as follows:

[0043]

[0044] wherein q MT is the ship speed or main engine power of the Mth position of the ship at the Tth time; d M is the Mth position, t T is the Tth time; U and V are column vectors and row vectors of two dimensions of M and T, and u and v are element values;

[0045] For other navigation parameters other than the ship speed and the main engine power, a four-dimensional space is formed by the ship speed, the position, the time and the other navigation parameters, and the four-dimensional tensor decomposition method is used for filling.

[0046] Further, the multi-source data fusion includes:

[0047] Different data sets are matched and aligned based on the space-time attributes, and the same attribute features are judged and converted to realize consistent expression of the features; then, a data similarity measurement matrix is constructed by combining the hash function method and the Euclidean distance, redundant attributes are identified, and data deduplication is realized through attribute reduction.

[0048] If there are differences in the same attribute data records in the heterogeneous data sets, the rationality of the different source data is defined by using the distribution statistics method, the mean value of the reasonable data is taken for numerical merging, and the strong coupling integration of the multi-source data is realized.

[0049] According to a second aspect of the present application, a ship speed prediction and optimization device is provided, which includes a processor and a memory, the memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to realize the steps of the ship speed prediction and optimization method according to any one of the above.

[0050] According to a third aspect of the present application, a readable storage medium is provided, which stores programs or instructions, and the programs or instructions are executed by the processor to realize the steps of the ship speed prediction and optimization method according to any one of the above.

[0051] Overall, compared with the prior art, the above technical solutions conceived by the present application can achieve the following beneficial effects:

[0052] The application provides a ship speed prediction and optimization method, device and readable storage medium, constructs a ship speed prediction model and a ship speed optimization model, can quickly and at low cost predict a more accurate real ship speed by considering the influence of marine meteorological environment on the ship and the characteristics of the ship itself when the ship is sailing in the ocean, so as to solve the defects that the existing ship speed prediction technology not only consumes a large amount of time and funds, but also is difficult to accurately mine the relationship between multiple factors. In addition, the accurate and rapid ship speed prediction reduces the influence of inaccurate ship speed on cost calculation in the ship speed optimization, improves the accuracy of the calculation of the ship sailing cost, and establishes a cost function according to the main engine power and the real ship speed, and constructs a ship speed optimization model in combination with the constraint conditions, so as to solve the defect that the existing ship speed optimization cannot achieve ideal results.

[0053] In addition, the application constructs a multi-source data aggregation processing model based on a unique single-source data cleaning and multi-source data fusion strategy, trains a machine learning model based on particle swarm optimization through the fused data set, thereby constructing a ship speed prediction model, and improves the prediction performance of the ship speed prediction model. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 It is a flowchart of the ship speed prediction and optimization method;

[0055] Figure 2 It is a flowchart of the hyperparameter optimization of the machine learning model based on the particle swarm algorithm;

[0056] Figure 3 It is a flowchart of the optimization of the ship speed based on the ship speed prediction model and the ship speed optimization model;

[0057] Figure 4 It is a structural diagram of the ship speed prediction and optimization device;

[0058] Figure 5 It is a diagram of the ship speed optimization results under different wind and wave conditions;

[0059] Figure 6 It is a diagram of the cost of each voyage section using the optimized ship speed. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below in combination with the drawings and examples. It should be understood that the specific examples described herein are only used to explain the application and do not limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0061] The existing ship speed prediction technology is mostly based on physical models established by theoretical derivation and ship model experiment, which is difficult to accurately depict the influence of complex marine environment on the ship state, and many physical models ignore the related features such as ocean current to simplify the model. Therefore, the ship speed prediction based on the physical model is not accurate. Compared with the physical model, the data-driven method based on machine learning can avoid the subjective assignment of parameters, and can consider more factors affecting the ship speed and mine the nonlinear relationship between multiple factors, so as to establish a more accurate nonlinear ship speed prediction model. The ship speed prediction model is further optimized by the particle swarm algorithm to improve the accuracy of the ship speed prediction model. The speed optimization model based on the high-accuracy ship speed prediction makes the predicted ship speed more close to the real ship speed, thereby reducing the influence of weather information on the ship speed, and further improving the accuracy of the calculation of the ship navigation cost, and the cost function is established according to the main engine power and the real ship speed, and the ship speed optimization model is constructed combined with the constraint condition, which can better realize the optimization of the ship speed.

[0062] As Figure 1 shown, according to one embodiment of the application, a ship speed prediction and optimization method based on multi-source heterogeneous sensor data in a variable marine environment is proposed, comprising the following steps:

[0063] S101, obtaining ship navigation history information and marine weather history information;

[0064] S102, constructing a multi-source data aggregation processing model based on single-source data cleaning and multi-source data fusion, and processing the obtained ship navigation history information and marine weather history information;

[0065] S103, establishing a ship speed prediction model based on a particle swarm optimization algorithm according to the processed ship navigation history information and marine weather history information;

[0066] S104, obtaining ship route planning information and marine weather forecast information;

[0067] S105, constructing a ship speed optimization model according to a cost function and a constraint condition;

[0068] S106, establishing a ship speed prediction and optimization model based on the speed prediction model and the speed optimization model, and optimizing the speed of the planned route.

[0069] The ship navigation data and marine weather data are aggregated and processed, so that the predicted ship speed is more close to the real ship speed, thereby reducing the influence of weather information on the ship speed, and further improving the accuracy of the calculation of the ship navigation cost, and better realizing the optimization of the ship speed.

[0070] Specifically, first, the ship navigation history information and marine weather history information are obtained. For example, the ship navigation information such as the time, latitude and longitude coordinates, heading, speed, main engine power, draft, etc. when the ship actually sails, and the weather information around the ship. By establishing a multi-source data aggregation processing model, the ship navigation information and weather information are processed and fused. Then, a machine learning model based on particle swarm optimization is trained through the fused data set to build a ship speed prediction model. Secondly, the ship route planning information and marine weather forecast information are obtained, and the multi-source data aggregation processing model is used to process and fuse the ship planned route points and weather forecast information. Among them, the planned route point is set by the ship according to the designed speed under ideal conditions. The route points are divided according to the marine weather conditions of the planned route points, and the speed of each route segment is optimized according to the cost function and its limited conditions to achieve speed optimization.

[0071] In some embodiments, the ship navigation history information and marine weather history information are obtained, including:

[0072] By obtaining ship AIS data and marine weather history information from a weather forecast center, a data set M1 is established; wherein the ship AIS data includes ship speed, heading, time, longitude, latitude, main engine power, and ship static data (such as ship length, ship width, draft, etc.). Marine weather history information includes flow rate (m / s), flow direction (°), wind speed (m / s), wind direction (°), significant wave height (m), wave direction (°), and combined period of wind wave and swell wave (s).

[0073] In some embodiments, a multi-source data aggregation processing model is constructed based on a single-source data cleaning and multi-source data fusion method, including:

[0074] Based on the ship navigation history data and marine weather history information, the data is identified and processed;

[0075] For missing, outlier and other significant abnormal data, multi-dimensional clustering and fixed threshold method are used for identification; for non-significant abnormal data generated under special operation mode such as "ramming", theoretical knowledge and empirical model are used for identification; then the missing values are replaced with abnormal values, and then the single-source data set is filled based on tensor decomposition method, and the filled data is standardized by Z-score method, finally realizing high-quality cleaning of original data. Ship speed and main engine power data belong to two-dimensional time series, and a two-dimensional matrix is constructed; for other navigation parameters of the ship, a four-dimensional space is formed by speed, position (longitude and latitude), time and other navigation parameters. Taking two-dimensional tensor filling as an example, where d represents different position points of the ship, and t represents time series.

[0076] Therefore, the two-dimensional matrix (such as the speed of the ship at different time points) can be decomposed into:

[0077]

[0078] wherein q MT is the Mth point in time T ship attribute value (such as speed, main engine power); U and V are two-dimensional column vector and row vector of M and T. Through decomposition, the hidden patterns in time and space can be revealed, and used for prediction and filling of missing data points to achieve single-source data cleaning.

[0079] Assume that the two-dimensional matrix Q = UV is constructed, U is a matrix representing the spatial dimension (location point or parameter), and V is a matrix representing the time dimension. By decomposing the known data, the correlation in time and space is found. For example, the change pattern between adjacent time points and adjacent spatial positions can be reflected by the decomposed matrices U and V. Among them, the pattern reflects the trend of the change of the ship speed or other parameters under certain conditions during the voyage. The decomposed matrices U and V are used to fill in the missing values in the original matrix Q. For example, the ship speed data missing at some time points can be inferred by the pattern of the previous and next time points. Similarly, for the missing data in the spatial dimension (such as the missing main engine power at some position points), it can be reconstructed by the decomposed spatial pattern.

[0080] It should be noted that the navigation parameters refer to the ship heading, wind speed, wind direction and other characteristic variables related to the ship speed. Other navigation parameters of the ship refer to other characteristic data other than time, position, ship speed and main engine power. By integrating various navigation parameters at different times and positions, a complete navigation track can be formed, and the trends and correlations in time and space can be analyzed. For example, how does the wind speed and direction change with time and position, and how does it affect the ship speed. It also better reveals the abnormal points in the data, especially for some parameters with strong time and space correlation.

[0081] Based on the time and space attribute matching, different data sets are aligned, and the same attribute characteristics (such as wind direction and ship relative wind direction, and ship navigation history data may also contain ship relative wind direction data) are determined and converted (converted to the direction relative to the heading according to the direction of the meteorological data), to realize the consistent expression of the characteristics. Then, combined with the hash function method and the Euclidean distance, a data similarity measurement matrix is constructed to identify redundant attributes and realize data deduplication through attribute reduction. If there are differences in the same attribute data records in the heterogeneous data set, the distribution statistics method is used to define the rationality of the data from different sources, and the mean value of the reasonable data is taken for numerical merging (the mean value of multiple different values of the same attribute is processed), and finally a multi-source data aggregation processing model is constructed to realize the strong coupling integration of multi-source data.

[0082] The single-source data cleaning and multi-source data fusion are performed on the ship navigation data and the meteorological data to construct a multi-source data aggregation processing model. Based on the multi-source data aggregation processing model and the data set M1, the external marine meteorological data and the ship AIS data are cleaned and fused to establish a data set S1.

[0083] In some embodiments, a ship speed prediction model based on a particle swarm optimization algorithm is established according to ship navigation history information and marine meteorological history information, including:

[0084] Based on the Pearson correlation analysis method, a key feature set that has a significant influence on the ship speed is screened out;

[0085] The machine learning model hyperparameters are optimized based on the particle swarm algorithm;

[0086] A ship speed prediction model is established by training the optimized machine learning model using the data set S1.

[0087] Specifically, first, the Pearson correlation analysis method is used to calculate the correlation coefficients between different features to construct a correlation coefficient matrix, and a key feature set that has a significant influence on the ship speed is screened out. Finally, nine input variables are selected from the data set: main engine power, ship heading (°), flow rate (m), relative flow direction to the ship heading (°), wind speed (m / s), relative wind direction (°), significant wave height (m), relative wave direction (°), and combined period of wind and swell (s), and the ship speed (knots) is selected as the output variable, and the data set S1 is updated.

[0088] Second, as shown in Figure 2 The machine learning model hyperparameters are optimized based on the particle swarm algorithm, including:

[0089] The dimension N is determined according to the number n of parameters to be optimized, and the position and speed of the particles are randomly initialized;

[0090] The position attribute of each particle is an N-dimensional vector, and the range refers to the entire search space. The components of each dimension correspond to different parameters of the machine learning model, and the initialization range of each dimension is different;

[0091] The position vector of the i-th particle at time t can be represented as:

[0092]

[0093] Since all particles move in the same search space, when t = 0, the speed can be initialized to the range of (0, 1) in each dimension;

[0094] The velocity vector of the i-th particle at time t can be represented as:

[0095]

[0096] The position vector is assigned to the corresponding parameter of the model, and the performance on the training set is taken as the initial fitness value, and the mean square error (MSE) is set as the target optimization function of PSO, and the fitness value of the i-th particle at time t is:

[0097] F i(t) = (P i(t) → Model| traningset ) [metric＝MSE]

[0098] For the i-th particle, the individual optimal value at time t can be expressed as:

[0099] Pbest i(t) = max (F i(j) ), 0≤j≤t

[0100] Assuming there are m particles, the global optimal at time t can be expressed as:

[0101] Gbest (t) = max (Pbest k(t) ), 1≤k≤m

[0102] According to the target error value or the maximum number of iterations, the model training is terminated, and the optimized parameter value of the machine learning model is obtained.

[0103] In this embodiment, the model is a Catboost model. As shown in Figure 2 , the Catboost model hyperparameters are optimized based on the particle swarm algorithm, the dimension is 3 according to the number of parameters to be optimized (iterations, depth, learning_rate), and then the position and velocity of the particles are randomly initialized;

[0104] The position attribute of each particle is a 3-dimensional vector, and the range refers to the entire search space. The components of each dimension correspond to different parameters of the Catboost model, and the initialization range of each dimension is different;

[0105] The position vector of the i-th particle at time t can be expressed as:

[0106]

[0107] Since all particles move in the same search space, when t=0, the velocity can be initialized to the range of (0,1) in each dimension;

[0108] The velocity vector of the i-th particle at time t can be expressed as:

[0109]

[0110] The position vector is assigned to the corresponding parameter of the model, and the performance on the training set is taken as the initial fitness value, and the mean square error is set as the target optimization function of PSO, and the fitness value of the ith particle at time t is:

[0111] F i(t) = (P i(t) →Model| traningset ) [metric＝MSE]

[0112] For the ith particle, the individual optimal value at time t can be expressed as:

[0113] Pbest i(t) = max(F i(j) ), 0≤j≤t

[0114] Assuming there are m particles, the global optimal at time t can be expressed as:

[0115] Gbest (t) = max(Pbest k(t) ), 1≤k≤m

[0116] According to the target error value or the maximum number of iterations, the model training is terminated, and the optimized parameter value of the machine learning model is obtained, as shown in Table 1 below.

[0117] Table 1: Optimized parameter value table

[0118]

[0119] The optimized machine learning model is trained by using the data set S1 to establish a ship speed prediction model, and four standard regression performance indicators are used to evaluate the performance of the ship speed prediction model, including mean square error, root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).

[0120] In the reference example, 80% of the data (6952) is randomly selected as the training set to construct the ship speed prediction model, and the remaining 20% (1738) is used as the test set. The performance of the trained PSO-Catboost and Catboost speed prediction model is shown in Table 2 below.

[0121] Table 2: Ship speed prediction model performance table

[0122]

[0123] The PSO-Catboost speed prediction model performs better than the Catboost speed prediction model in four evaluation indicators, and it can be concluded that optimizing machine learning hyperparameters based on particle swarm algorithm can effectively improve the prediction performance of machine learning model.

[0124] In some embodiments, the ship route planning information and marine weather forecast information are obtained, including:

[0125] The ship route planning information including the origin port, destination port and estimated time of arrival, and the marine weather forecast information generated by the global ensemble forecast system from the National Oceanic and Atmospheric Administration, are obtained to establish a data set M2.

[0126] In some embodiments, a speed optimization model is constructed according to the cost function and the constraint condition, including:

[0127] The route cost function and the constraint condition are established;

[0128] Based on the formula:

[0129]

[0130] The cost of the voyage is calculated, where v0 is the designed speed (16 knots in this embodiment), v i is the ideal speed (no wind and no waves) under the given power (i.e. main engine power), v ir is the actual speed under the same power, D i represents the distance of the i-th leg, h0 represents the fuel consumption rate under the designed speed, and a1 and a2 correspond to the fuel cost and time cost respectively; divided by 24 represents conversion to days. The variable i represents the index of the segment on the planned route, i∈I represents that i is an element in the segment set I.

[0131] The constraint condition is satisfied at the same time: The time sequence is guaranteed to be complete, T n ≤T max Ensure that the ship arrives at the destination port on time and follows the specified latest arrival time. Set the minimum and maximum limits of the ship's speed on each leg. Require the start time of each leg to be non-negative to ensure the implementability of the navigation plan.

[0132] By calculating the cost of different speed options for the current leg and combining the optimal solution of the previous leg, the optimal decision for the current leg is found, and the recursion is performed until the last leg to solve the optimal decision for each leg.

[0133] It should be noted that there is a corresponding relationship between the main engine power and the ideal speed, which can be obtained through simulation or actual measurement.

[0134] In some embodiments, based on the speed prediction model and the speed optimization model, a ship speed prediction and optimization model is established, and the speed of the route is optimized, as shown in Figure 3 The ship speed prediction and optimization model comprises the following steps:

[0135] In step S1061, according to the time resolution of the marine weather forecast information M2, the ship speed is assumed to be sailing at the designed speed under ideal conditions, and a planned route point is inserted.

[0136] Based on the multi-source data aggregation processing model, the planned route point is aggregated with the marine weather forecast data to establish a data set S2.

[0137] For example, the marine weather forecast information is updated every hour, and the distance interval is obtained by multiplying the designed speed, and then a planned route point is inserted every distance interval along the planned route.

[0138] In step S1062, the marine weather conditions of each planned route point are determined according to the marine weather data in the data set S2, and the planned route is divided into i planned route points and i-1 planned route segments.

[0139] The planned route segment can be further divided or combined according to the ship heading change and the marine weather forecast information change in the planned route segment and the adjacent planned route segment to obtain the final planned route segment. For example, if the ship heading changes significantly in the planned route segment, a planned route point is inserted. If the conditions of multiple planned route segments are similar, the planned route segments can be combined.

[0140] In the reference example, the sea conditions and the Beaufort wind scale between the route points are divided into 19 segments, and the specific information is shown in Table 3.

[0141] Table 3 Route segment table

[0142]

[0143]

[0144] In step S1063, according to the marine weather conditions of each route segment, the real speed is predicted based on the speed prediction model, and the optimal speed is selected based on the route optimization model to minimize the cost and ensure arrival at the destination port within the predetermined time.

[0145] In step S1064, a ship speed prediction and optimization model is established to realize speed optimization to meet the requirement of minimizing the total operating cost of multi-segment voyage while following the latest arrival time.

[0146] By adopting the above technical solution, a ship speed prediction and optimization model is constructed.

[0147] According to the second aspect of the present application, as shown in Figure 4 A structural block diagram of a ship speed prediction device 200 is proposed. As shown in Figure 4 The ship speed prediction and optimization device 200 mainly includes:

[0148] The acquisition module one 201 is configured to acquire ship navigation information and marine weather history information.

[0149] The establishment module one 202 is configured to construct a multi-source data aggregation processing model based on a single-source data cleaning and multi-source data fusion method.

[0150] The establishment module two 203 is configured to establish a speed prediction model based on a particle swarm optimization algorithm according to the ship navigation information and the marine weather history information.

[0151] The acquisition module two 204 is configured to acquire ship route planning information and marine weather forecast information.

[0152] The establishment module three 205 is configured to construct a speed optimization model according to a cost function and a constraint condition.

[0153] The application module one 206 is configured to establish a ship speed prediction and optimization model based on the speed prediction model and the speed optimization model, and to perform speed optimization on a planned route.

[0154] As an optional implementation manner of the present embodiment, the application module one 202 includes:

[0155] The data processing module is configured to identify abnormal data of ship navigation data and marine weather information by using a multi-dimensional clustering and a fixed threshold method, and to fill in the abnormal data by interpolation.

[0156] The data fusion submodule is configured to align different data sets based on a spatio-temporal attribute matching, and to determine and convert the same attribute features to realize consistent expression of the features.

[0157] As an optional implementation manner of the present embodiment, the establishment module two 203 includes:

[0158] The feature screening submodule is configured to screen a key feature set having a significant influence on ship speed based on a Pearson correlation analysis method.

[0159] The hyperparameter optimization submodule is configured to optimize hyperparameters of a machine learning model based on a particle swarm optimization algorithm.

[0160] The model training submodule is configured to train an optimized machine learning model based on ship navigation data and marine weather information to establish a speed prediction model.

[0161] As an optional implementation manner of the present embodiment, the establishment module three 205 includes:

[0162] The cost of different speed options for the current leg is calculated by establishing a cost function and constraint conditions to solve the optimal decision for each leg, and a speed optimization model is constructed.

[0163] As an optional implementation of the embodiment, the application module 206 includes:

[0164] Application sub-module one: for aggregating the planned waypoints and marine weather forecast data based on a multi-source data aggregation processing model;

[0165] Application sub-module two: for predicting the real speed based on a speed prediction model according to the marine weather conditions of each leg;

[0166] Application sub-module three: for deciding to select the optimal speed based on a route optimization model according to the marine weather conditions of each leg;

[0167] Establishment sub-module: for establishing a ship speed prediction and optimization model, realizing speed optimization to maximize the reduction of the total operating cost of multi-leg voyage while complying with the latest arrival time.

[0168] According to a third aspect of the present application, a ship speed prediction and optimization device for ocean-going ships under variable marine environments based on multi-source heterogeneous sensor data is provided, which includes a processor and a memory, the memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to realize the steps of the ship speed prediction and optimization method for ocean-going ships under variable marine environments based on multi-source heterogeneous sensor data provided in the first aspect.

[0169] According to a fourth aspect of the present application, a readable storage medium is provided, which stores programs or instructions, and the programs or instructions are executed by a processor to realize the ship speed prediction and optimization method for ocean-going ships under variable marine environments based on multi-source heterogeneous sensor data according to any of the above technical solutions.

[0170] Figure 5 The optimized speed of the 6238TEU container ship under different wind and wave conditions is shown. As shown in Figure 6 Compared with the designed speed of the ship, the optimized speed significantly reduces the operating cost of each leg of the container ship. The strategic method saves about 81,011.16 dollars of cost, accounting for about 4.6% of the total operating cost.

[0171] In summary, the ship speed prediction and optimization model proposed by the present application performs better than existing methods in both ship speed prediction and speed optimization. The present application mainly includes a multi-source data aggregation processing model based on a unique single-source data cleaning and multi-source data fusion strategy and a machine learning ship speed prediction and optimization model based on a particle swarm algorithm. By considering the influence of marine meteorological environment and the characteristics of the ship itself when the ship is sailing in the ocean, the present application can quickly and at low cost predict a more accurate real speed for the ship sailing, to solve the defects of existing ship speed prediction technology that not only consumes a lot of time and money, but also is difficult to accurately mine the relationship between multiple factors. Through accurate and fast ship speed prediction, the influence of inaccurate speed on cost calculation in speed optimization is reduced, the accuracy of ship sailing cost calculation is improved, and the defects of existing ship speed optimization that cannot achieve ideal results are solved.

[0172] It should be noted that the size of the serial number of each step in the above embodiments does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0173] It should be noted that according to the needs of implementation, each step / component described in the present application can be split into more steps / components, or two or more steps / components or part of the operation of the steps / components can be combined into a new step / component to achieve the purpose of the present application.

[0174] Those skilled in the art will readily understand that the above is only a preferred embodiment of the present application and is not intended to limit the present application, and any modifications, equivalent replacements and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for predicting and optimizing ship speed, characterized in that, The method includes: A ship speed prediction model is established using the main engine power, ship heading, current speed, relative current direction to the ship heading, wind speed, relative wind direction to the ship heading, significant wave height, relative wave direction to the ship heading, and the combined period of wind waves and swells as inputs and ship speed as output. Obtain ship route planning information and marine meteorological forecast information, including current speed, current direction, wind speed, wind direction, significant wave height, wave direction, and the combination period of wind waves and swells; The planned routes of ships are obtained based on the ship route planning information, and the planned routes are divided into several planned segments; Obtain the ship's course for each planned segment, and combine marine meteorological forecast information to obtain relevant information for each planned segment, including ship's course, current speed, relative current direction to the ship's course, wind speed, relative wind direction to the ship's course, significant wave height, relative wave direction to the ship's course, and the combination period of wind waves and swells. Based on the relevant information of different main engine power and each planned voyage segment, and combined with the ship speed prediction model, the actual ship speed of each planned voyage segment under the corresponding weather and sea conditions and different main engine power is predicted; A cost function is established based on the main engine power and the corresponding actual ship speed, and a ship speed optimization model is constructed in conjunction with constraints, including: The cost of the total voyage is calculated based on the following formula: in, For the first The cost of the planned flight segment n Total number of planned flight segments; For design speed, For a given main engine power, the ideal speed is... This indicates the fuel consumption rate at the design speed; Indicates the first The distance of the planned flight segment; For the first The actual ship speed for the planned voyage under the same main engine power is predicted by the ship speed prediction model. and The coefficients are fuel cost and time cost, respectively. The following constraints must be met simultaneously: , To ensure that vessels arrive at their destination ports on time and in accordance with the prescribed latest arrival time; among which, Indicates the preceding Total time for each planned flight segment T 0 is 0; This sets the minimum and maximum limits for the ideal speed of the ship in each planned segment; among them, and These are the minimum speed and the maximum speed, respectively. The start time of each planned flight segment must be non-negative to ensure the feasibility of the flight plan. Based on the ship speed prediction model and the ship speed optimization model, the optimized speed for each planned voyage segment is obtained.

2. The ship speed prediction and optimization method according to claim 1, characterized in that, Based on the ship speed prediction model and the ship speed optimization model, the optimized speed for each planned segment is obtained, including: For each planned voyage segment, the actual ship speed at different main engine power is obtained, thereby obtaining the cost of each planned voyage segment at different actual ship speeds; By combining the constraints, the actual ship speeds for each planned segment are obtained when the total cost of the voyage is minimized, and then the ship is controlled to sail at the corresponding actual ship speeds in the corresponding planned segments.

3. The ship speed prediction and optimization method according to claim 1, characterized in that, A ship speed prediction model based on particle swarm optimization algorithm is established, including: The ship speed prediction model is the Catboost model, and the hyperparameters of the Catboost model are optimized based on the particle swarm optimization algorithm. The hyperparameters of the Catboost model include the maximum number of tree iterations, the tree depth, and the learning rate. The dimension is determined by the number of parameters to be optimized, and then the position and velocity of the particles are randomly initialized. The performance of the ship speed prediction model is used as the fitness value. The optimal hyperparameters of the Catboost model are obtained through iteration, thus obtaining a ship speed prediction model based on the particle swarm optimization algorithm.

4. The ship speed prediction and optimization method according to claim 1, characterized in that, The planned route for ships is divided into several planned segments, including: Based on the time resolution of marine meteorological forecast information and the design speed, planned waypoints are inserted into the planned route of the ship, dividing the planned route into multiple pre-planned segments; Based on the changes in ship course and marine meteorological forecast information within the pre-planned voyage segment and adjacent pre-planned voyage segments, the pre-planned voyage segment is further divided or combined to obtain the final planned voyage segment.

5. The ship speed prediction and optimization method according to claim 1, characterized in that, Establish a ship speed prediction model, including: Acquire historical ship navigation information and historical marine meteorological information; A multi-source data aggregation processing model is constructed based on the method of single-source data cleaning and multi-source data fusion. The acquired ship navigation history information and marine meteorological history information are cleaned from single sources and fused from multiple sources to obtain feature sets. Based on Pearson correlation analysis, a set of key features that have a significant impact on ship speed is selected from the feature set. These key features are used as input to the ship speed prediction model, and the model is trained using the set of key features. This includes acquiring ship navigation history information and marine meteorological history information, including: Acquire ship AIS data, including time, location, and corresponding ship speed, main engine power, and ship heading; the location includes longitude and latitude. Obtain historical marine meteorological information, including time, location, and corresponding current velocity, current direction, wind speed, wind direction, significant wave height, wave direction, and the combination period of wind waves and swells.

6. The ship speed prediction and optimization method according to claim 5, characterized in that, Single-source data cleaning includes: For each single-source data, outliers are identified and replaced with missing values. Then, the single-source data is filled using tensor decomposition, and the filled data is standardized using the Z-score method. The identification methods include: for significant anomalous data, multidimensional clustering and fixed threshold methods are used for identification; for non-significant anomalous data, theoretical knowledge and empirical models are used for identification. For ship speed and main engine power, a two-dimensional tensor is used for filling, as follows: in, q MT For the ship's first M The position at time T The ship's speed or main engine power; For the first M One location, For the first T A time period; U and V For two dimensions M and T Column vectors and row vectors, u and v For element values; For navigation parameters other than ship speed and main engine power, a four-dimensional space is formed by ship speed, position, time and other navigation parameters, and then filled in based on the four-dimensional tensor decomposition method.

7. The ship speed prediction and optimization method according to claim 5, characterized in that, Multi-source data fusion includes: Different datasets are aligned based on spatiotemporal attribute matching, and features with the same attribute are judged and transformed to achieve consistent feature representation. Then, a data similarity measurement matrix is ​​constructed by combining hash function method and Euclidean distance to identify redundant attributes, and data redundancy is achieved through attribute reduction. If there are differences in data records with the same attribute in heterogeneous datasets, the distribution statistics method is used to define the rationality of data from different sources, and the mean of the rational data is used for numerical merging to achieve strong coupling and integration of multi-source data.

8. A ship speed prediction and optimization device, characterized in that, include: A processor and a memory, wherein the memory stores a program or instructions that can run on the processor, and when the program or instructions are executed by the processor, implement the steps of the ship speed prediction and optimization method according to any one of claims 1 to 7.

9. A readable storage medium, characterized in that, It stores programs or instructions, which, when executed by a processor, implement the steps of the ship speed prediction and optimization method according to any one of claims 1 to 7.

Images