Method and device for predicting fuel consumption and shaft power of ship

Abstract

The invention provides a ship fuel consumption and shaft power prediction method and device, and the method comprises the steps: carrying out the preprocessing of ship operation parameter data collected by a sensor system, and obtaining the preprocessed data; extracting features from the preprocessed data to obtain feature data, wherein the feature data comprises operation features related to ship fuel consumption and shaft power; and inputting the feature data into a pre-trained feedforward neural network model to obtain predicted values of ship fuel consumption and shaft power. Through systematic data preprocessing, physical driving feature engineering and the feedforward neural network of the optimization architecture, high-precision prediction of the fuel consumption of the ship main engine and the power of the propeller shaft is realized, and the navigational speed loss is reliably estimated. Through the method, the data quality is improved, the problem of data distortion in the prior art is solved, and through deep fusion of physical characteristics and environmental factors, the prediction result better conforms to the actual operation condition of the ship.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ship power consumption control, and particularly relates to a ship fuel consumption and shaft power prediction method and device. BACKGROUND

[0002] Ship energy efficiency and carbon emissions have become an important issue in global climate governance, especially with the International Maritime Organization's net-zero emission framework requirements, the shipping industry is facing more stringent emission control requirements. The shipping industry consumes a large amount of fossil fuels and emits a large amount of carbon dioxide every year, accounting for about 13% of global transportation industry carbon emissions. In response to emission reduction regulations, the improvement of ship energy efficiency and carbon emission control has become the core direction of research. At present, the existing ship energy efficiency prediction technology mainly includes empirical method, physical model method and traditional machine learning method. However, these methods generally have the problems of being unable to adapt to dynamic sea conditions in real time, low precision, and lack of effective data quality control of the model. With the development of ship Internet of Things technology, a large amount of high-frequency data has been generated, but the existing technology has not effectively solved the problems of "poor data quality", "weak feature extraction" and "low dynamic adaptability". Therefore, a high-precision fuel consumption and shaft power prediction technology based on feedforward neural network (FNN) emerges as the times require, which provides more refined energy efficiency management solutions for ship operation by fusing multi-source data and machine learning algorithm. SUMMARY

[0003] The application provides a ship fuel consumption and shaft power prediction method and device to at least solve the problems of low prediction accuracy, insufficient data processing, unreasonable feature selection and poor model adaptability of the ship fuel consumption and shaft power prediction in the prior art.

[0004] According to a first aspect of the application, a ship fuel consumption and shaft power prediction method is provided, comprising: preprocessing ship operation parameter data collected by a sensor system to obtain preprocessed data; extracting features from the preprocessed data to obtain feature data, the feature data including operation features related to ship fuel consumption and shaft power; inputting the feature data into a pre-trained feedforward neural network model to obtain predicted values of ship fuel consumption and shaft power.

[0005] In an embodiment, preprocessing the ship operation parameter data collected by the sensor system comprises: identifying and correcting abnormal values in the ship operation parameter data, including using an interpolation algorithm to correct the jump caused by the abnormal values, or correcting the abnormal values based on a statistical method; The ship operation parameter data is smoothed, including smoothing the data within a predetermined time window by applying a moving average method to remove high-frequency noise and improve the stability of the data. Missing values in the ship operation parameter data are filled.

[0006] In an embodiment, feature data is obtained by extracting features from the preprocessed data, including: Derivative features related to ship fuel consumption and shaft power are generated by mathematical transformation or combination of the preprocessed data; From the derivative features related to ship fuel consumption and shaft power, derivative features closely related to the predicted values of ship fuel consumption and shaft power are selected, and redundant derivative features are screened out to obtain derivative feature data.

[0007] In an embodiment, derivative features related to ship fuel consumption and shaft power are generated by mathematical transformation or combination of the preprocessed data, including: Wind effect features are calculated according to wind speed and direction to generate derivative features for quantifying the influence of wind on the ship propulsion system; Hull fouling features are generated according to the ship operation time to generate derivative features reflecting the influence of hull fouling on propulsion efficiency; Current features are calculated according to the difference between the ground speed and the water speed to generate derivative features for quantifying the influence of sea current on ship speed; The standard deviation of the key parameters is calculated to generate parameter standard deviation features, which are derivative features reflecting the stability of the ship propulsion system working condition.

[0008] In an embodiment, from the derivative features related to ship fuel consumption and shaft power, features closely related to the predicted values of ship fuel consumption and shaft power are selected, and redundant features are screened out to obtain derivative feature data, including: By evaluating the correlation of each derivative feature with ship fuel consumption and shaft power, derivative features with correlation to ship fuel consumption and shaft power are selected; Redundant derivative features are screened out by feature selection algorithm to retain derivative features that have independent contribution to the prediction of ship fuel consumption and shaft power; Based on the selected derivative features, a final derivative feature data set is constructed.

[0009] In an embodiment, the training method of the feedforward neural network model includes: The feature training data is preprocessed, including data normalization and data shuffling; The preprocessed feature training data is input into the feedforward neural network model for forward propagation calculation to obtain the predicted value; The loss between the predicted value and the true value is calculated, and the trained feedforward neural network model is obtained by adjusting the network weight according to the loss value by using a back propagation algorithm.

[0010] According to a second aspect of the present application, a ship fuel consumption and shaft power prediction device is also provided, comprising: a preprocessing unit configured to preprocess the ship operation parameter data collected by the sensor system to obtain preprocessed data; a feature extraction unit configured to extract features from the preprocessed data to obtain feature data, the feature data comprising operation features related to the ship fuel consumption and shaft power; a prediction unit configured to input the feature data into the pre-trained feedforward neural network model to obtain the predicted values of the ship fuel consumption and shaft power.

[0011] In an embodiment, the preprocessing unit comprises: an outlier identification and correction module configured to identify and correct outliers in the ship operation parameter data, including correcting the jumps caused by outliers using an interpolation algorithm, or correcting outliers based on statistical methods; a smoothing processing module configured to smooth the ship operation parameter data, including smoothing the data within a predetermined time window by applying a moving average method to remove high-frequency noise and improve data stability; a missing value filling module configured to fill missing values in the ship operation parameter data.

[0012] In an embodiment, the feature extraction unit comprises: a derived feature generation module configured to generate derived features related to the ship fuel consumption and shaft power by mathematically transforming or combining the preprocessed data; a screening module configured to select derived features closely related to the predicted values of the ship fuel consumption and shaft power from the derived features related to the ship fuel consumption and shaft power, while screening out redundant derived features to obtain derived feature data.

[0013] In an embodiment, the derived feature generation module comprises: a first calculation module configured to calculate wind effect features based on wind speed and wind direction to generate derived features for quantifying the influence of wind on the ship propulsion system; a ship hull fouling feature generated based on the ship operation time to generate derived features for reflecting the influence of ship hull fouling on propulsion efficiency; a second calculation module configured to calculate current features based on the difference between the ground speed and the water speed to generate derived features for quantifying the influence of current on the ship speed; The third calculation module is configured to calculate the standard deviation of the key parameters to generate a parameter standard deviation feature, which is a derived feature reflecting the stability of the working condition of the ship propulsion system.

[0014] In an embodiment, the screening module comprises: The evaluation selection module is configured to select the derived features that have correlations with the ship fuel consumption and shaft power by evaluating the correlations of the derived features with the ship fuel consumption and shaft power; The redundancy removal module is configured to remove the redundant derived features by the feature selection algorithm to retain the derived features that have independent contributions to the prediction of the ship fuel consumption and shaft power; The dataset generation module is configured to construct a final derived feature dataset based on the selected derived features.

[0015] In an embodiment, the training method of the feedforward neural network model comprises: The feature training data is preprocessed, including data normalization and data shuffling; The preprocessed feature training data is input into the feedforward neural network model to perform forward propagation calculation to obtain a predicted value; The loss between the predicted value and the true value is calculated, and the trained feedforward neural network model is obtained by adjusting the network weight according to the loss value using the back propagation algorithm.

[0016] The present application realizes high-precision prediction of the ship main engine fuel consumption and propeller shaft power, and reliable estimation of the speed loss through systematic data preprocessing, physically driven feature engineering and optimized architecture of the feedforward neural network. Through the above method, not only the data quality is improved and the data distortion problem in the prior art is overcome, but also the prediction result is more consistent with the actual operation condition of the ship through the deep fusion of physical properties and environmental factors. In addition, the present application has stronger adaptability and generalization ability, and can support ship energy efficiency optimization, carbon emission control and speed optimization applications. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0018] Figure 1 A flowchart of a ship fuel consumption and shaft power prediction method provided by the present application.

[0019] Figure 2A method flowchart for preprocessing ship operation parameter data collected by a sensor system in an embodiment of the present application.

[0020] Figure 3 A method flowchart for extracting features from preprocessed data to obtain feature data in an embodiment of the present application.

[0021] Figure 4 A method flowchart for generating derived features related to ship fuel consumption and shaft power in an embodiment of the present application.

[0022] Figure 5 A method flowchart for obtaining derived feature data in an embodiment of the present application.

[0023] Figure 6 A method flowchart for training a feedforward neural network model in an embodiment of the present application.

[0024] Figure 7 A structural block diagram of a ship fuel consumption and shaft power prediction device provided by the present application.

[0025] Figure 8 A structural block diagram of a preprocessing unit in an embodiment of the present application.

[0026] Figure 9 A structural block diagram of a feature extraction unit in an embodiment of the present application.

[0027] Figure 10 A structural block diagram of a derived feature generation module in an embodiment of the present application.

[0028] Figure 11 A structural block diagram of a screening module in an embodiment of the present application.

[0029] Figure 12 A specific embodiment of an electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0031] Ship energy efficiency and carbon emission has become an important issue in global climate governance, especially with the International Maritime Organization's net-zero emission framework requirement, the shipping industry faces more stringent emission control requirements. The shipping industry consumes a large amount of fossil fuels and emits a large amount of carbon dioxide every year, accounting for about 13% of global transportation industry carbon emissions. In response to emission reduction regulations, the improvement of ship energy efficiency and carbon emission control has become the core direction of research. At present, the existing ship energy efficiency prediction technology mainly includes empirical method, physical model method and traditional machine learning method. However, these methods generally have the problems of being unable to adapt to dynamic sea conditions in real time, low precision and lack of effective data quality control in the model. With the development of ship Internet of Things technology, a large amount of high-frequency data has been generated, but the existing technology has not effectively solved the problems of "poor data quality", "weak feature extraction" and "low dynamic adaptability".

[0032] In order to solve the problems in the prior art, the present application provides a ship fuel consumption and shaft power prediction method, as shown in Figure 1 The method comprises the following steps: S101: Preprocessing the ship operation parameter data collected by the sensor system to obtain preprocessed data; Real-time operation parameters (such as ship speed, draft, propeller shaft speed, fuel consumption, etc.) are collected through the sensor system of the ship, and these raw data are processed to correct abnormal values, fill in missing values and smooth high-frequency noise. This preprocessing process can ensure that the data used in subsequent analysis has high quality, avoiding the adverse effects of sensor failure, data loss or noise on model training.

[0033] S102: Extracting features from the preprocessed data to obtain feature data, the feature data including operation features related to ship fuel consumption and shaft power; On the basis of data preprocessing, further feature extraction and derivation are performed on the data to generate features closely related to ship fuel consumption and shaft power. These features include wind effect features, hull fouling features, sea current features, and other derived features reflecting the operation state of the ship. Through the extraction of these features, the model can mine more potential information from the data and capture complex nonlinear relationships that affect ship performance.

[0034] S103: Inputting the feature data into a pre-trained feedforward neural network model to obtain the predicted values of ship fuel consumption and shaft power.

[0035] The extracted feature data is input into a trained feedforward neural network (FNN) model for prediction. Feedforward neural network is a machine learning model with strong nonlinear modeling capability, which can learn the complex relationships in ship operation data through multiple layers of neurons and activation functions, providing high-precision prediction of fuel consumption and shaft power.

[0036] The method for predicting ship fuel consumption and shaft power provided in this application aims to address the problem of low accuracy in monitoring ship energy efficiency and performance in existing technologies. Traditional methods for predicting ship fuel consumption and shaft power, such as empirical methods, physical model methods, and traditional machine learning methods (such as support vector machines and random forests), all have certain limitations. They cannot effectively capture complex dynamic factors in ship operation, such as wind effects, ocean current influences, and hull fouling, resulting in large prediction errors. Compared with traditional physical modeling or empirical models, this data-driven modeling method can better adapt to the complex environmental changes and real-time data changes in ship operation, making the prediction results more reliable and having higher generalization ability.

[0037] In one embodiment, the ship operation parameter data collected by the sensor system is preprocessed, such as... Figure 2 As shown, it includes: S201: Identify and correct outliers in ship operation parameter data, including using interpolation algorithms to correct jumps caused by outliers, or correcting outliers based on statistical methods. S202: Smooth the ship operation parameter data, including smoothing the data within a predetermined time window by applying the moving average method to remove high-frequency noise and improve the stability of the data; S203: Fill in missing values ​​in the ship operation parameter data.

[0038] In one specific embodiment, outliers are first corrected using an interpolation algorithm. The interpolation algorithm calculates the difference between adjacent data points in the time series and sets a reasonable threshold (usually based on the ship's inertia assumption) to identify outliers. If the change in a data point exceeds this threshold, it is considered an outlier and corrected. The algorithm steps are as follows: A. Calculate the difference between the current data point and the previous time point:

[0039] This value reflects the magnitude of the parameter change between two consecutive moments; if the change is too large, it may be an outlier. No. i Parameter measurements at a given time (such as time series data for latitude, longitude, etc.) B. Calculate the time interval Calculate the first i The time and the previous time ( i -1) Time difference:

[0040] Used to determine whether the sampling interval is reasonable (if the interval is too long, the large change in parameters may be caused by the actual working conditions rather than measurement error).

[0041] C. Determine the dynamic threshold value Dynamic threshold value D i Take the previous moment i -1 and the next moment i The maximum value of the parameter change amplitude: Its physical meaning is: taking the normal change amplitude of adjacent moments as the benchmark, to judge whether the current change is beyond the reasonable range.

[0042] D. Abnormal value judgment and correction If the following two conditions are met at the same time, it is determined that is an abnormal value, and is corrected: Condition 1: The parameter change amplitude exceeds the dynamic threshold value sf times, that is (indicates that the current change is much larger than the normal change of adjacent moments); Condition 2: The time interval does not exceed the maximum allowed value, that is ( ) minutes, to ensure that the working condition is stable within the interval.

[0043] The abnormal value correction formula is: That is, replace the abnormal value with the average of the previous moment and the next moment, which not only preserves the continuity of the time series, but also avoids introducing new bias.

[0044] By combining the dynamic change characteristics of the parameters and the time interval constraint , this algorithm can accurately identify and correct abnormal values caused by sensor errors or data transmission failures, while preserving the information of the real running state of the ship.

[0045] Abnormal value identification The core operating parameters of the ship propulsion system, such as the propeller shaft speed, shaft torque, and main engine fuel consumption, have strict physical coupling relationships. This relationship is the basis for ship propulsion efficiency evaluation and modeling. When the propeller shaft speed remains stable, the shaft torque needs to be maintained within a reasonable range matching the speed - if the shaft speed is constant but the torque abnormally increases or drops, such data is mostly caused by sensor instantaneous failure, data transmission interference, or extreme non-operating conditions. If such abnormal data is directly used for model training, it will seriously distort the normal correlation between parameters, causing the data-driven model prediction results to deviate from the physical reality, and unable to meet the precision requirements of ship performance monitoring.

[0046] Based on this, this study proposes a statistical outlier removal algorithm (Cleaning Algorithm). The core logic utilizes the strong statistical correlation between primary and secondary parameters to identify and remove outlier data points that violate physical laws within a defined operating condition range. Unlike traditional single-parameter threshold detection methods, this algorithm isolates operating conditions through primary parameters, avoiding misjudgments caused by differences in the normal fluctuation range of parameters under different operating conditions. Simultaneously, it maximizes the retention of effective data from regular operating conditions, providing a high-quality sample foundation for subsequent data-driven modeling.

[0047] Input parameters: preprocessed dataset, main parameters p 1. Secondary parameter p 2 A. Main Parameter Grouping (Operating Condition Isolation): Based on the measured range of the main parameter (propeller shaft speed), the system divides the parameters into several continuous, non-overlapping operating condition groups with a step size of s=1 rpm. .

[0048] B. Calculate within-group statistics and establish anomaly detection criteria for each working condition group. Independently calculate secondary parameters mean with standard deviation , .

[0049] The mean reflects the statistical center of the next parameter under this operating condition, and the formula is: , The standard deviation reflects the normal fluctuation range of the parameter under this operating condition. Since each group of data is sample data, an unbiased estimate is used, and the formula is:

[0050] C. Outlier identification and removal: For each data point, based on its group... Statistic and Through the following formula Determine if a data point is an outlier: If a data point satisfies the formula, it is considered an outlier, and all timestamp parameter records corresponding to that data point are deleted (to ensure data consistency).

[0051] When the deviation of a secondary parameter under a certain operating condition from the statistical center of that operating condition exceeds k=3 times the standard deviation, the deviation has exceeded the normal fluctuation range and violates the physical correlation law of the ship propulsion system.

[0052] 3) Data smoothing processing To reduce high-frequency noise and improve model stability, a simple moving average (SMA) was used with a window size of 5 minutes. The calculation formula is as follows: in, For a certain parameter in the first...i minute value is the first t minute smoothed value.

[0053] In an embodiment, features are extracted from the pre-processed data to obtain feature data, as shown in Figure 3 , including: S301: generating derived features related to ship fuel consumption and shaft power by performing mathematical transformation or combination on the pre-processed data; S302: selecting derived features closely related to the predicted values of ship fuel consumption and shaft power from the derived features related to ship fuel consumption and shaft power, while screening out redundant derived features, to obtain derived feature data.

[0054] After data preprocessing, feature extraction is performed next, and derived features related to ship fuel consumption and shaft power are generated. The derived features include wind effect, hull fouling, sea current, and standard deviation features, which help the model more accurately capture the relationship between ship operating conditions and fuel consumption, shaft power.

[0055] In an embodiment, derived features related to ship fuel consumption and shaft power are generated by performing mathematical transformation or combination on the pre-processed data, as shown in Figure 4 , including: S401: calculating wind effect features according to wind speed and wind direction to generate derived features for quantifying the impact of wind on the ship propulsion system; In a specific embodiment, the wind resistance experienced by the ship during navigation is directly related to the size of the wind speed and the heading angle of the wind relative to the ship: when the wind blows along the ship's cross direction (90° or 270°), the wind resistance is the largest; when the wind blows along the same direction (0°) or opposite direction (180°) of the ship's heading, the wind resistance is the smallest. To quantify this synergistic effect, the "wind effect" feature is introduced, which converts the two-dimensional wind parameters (wind speed, wind direction) into a one-dimensional representation through the product of wind speed and wind direction cosine value, as follows:

[0056] is the measured relative wind speed is the relative wind direction The physical meaning of the cosine term is the "projection component of wind speed in the ship's heading direction" - a positive value represents a tailwind (helps navigation, wind effect is positive), a negative value represents a headwind (hinders navigation, wind effect is negative), and the larger the absolute value, the stronger the impact of wind on ship resistance / propulsion power.

[0057] This feature solves the problem of "difficulty for the model to automatically learn the synergistic effect of wind speed and direction when they are input separately" in raw data, making the impact of wind on the propulsion system more directly captured by the model.

[0058] S402: Generate a hull fouling feature according to the ship operation time, and generate a derived feature for reflecting the impact of hull fouling on the propulsion efficiency; In a specific embodiment, hull fouling (such as marine organism attachment) will cause the roughness of the ship surface to increase, thereby increasing the water resistance of the ship, ultimately resulting in a decrease in speed at the same shaft power or an increase in fuel consumption at the same speed. Therefore, the "operation time difference" is transformed by logarithm to generate a fouling feature, and the formula is as follows: Fouling = ln(TimeElapsed).

[0059] TimeElapsed is the "difference between the current sampling timestamp and the first sampling timestamp" (unit: day), reflecting the operation time of the ship since the last hull cleaning; Logarithmic transformation can not only represent the growth trend of fouling over time, but also avoid the problem of "excessive amplification of late fouling" caused by linear transformation, so that the feature value maintains a reasonable dynamic range throughout the operation cycle.

[0060] S403: Calculate the sea current feature according to the difference between the ground speed and the water speed, and generate a derived feature for quantifying the impact of sea current on the ship speed; S404: Calculate the standard deviation of the key parameters, generate the parameter standard deviation feature, and use it as a derived feature to reflect the stability of the ship propulsion system working condition.

[0061] In a specific embodiment, the impact of sea current on ship speed is represented by the "difference between ground speed (SOG) and water speed (STW)"—when the sea current and the ship heading are in the same direction, SOG> STW, the difference is positive (sea current assistance); when the sea current is opposite, SOG< STW, the difference is negative. To directly quantify the impact of sea current on the propulsion system, generate a "sea current feature", and the formula is as follows:

[0062] SOG is the ground speed (unit: kn, GPS acquisition); STW is the water speed (unit: kn, log acquisition); the physical meaning of the feature value is "the component of the sea current in the direction of the ship's heading" (unit: kn).

[0063] This feature converts the "implicit environmental interference" of sea current into an explicit parameter, enabling the model to more accurately establish the relationship between "shaft power, speed, and fuel consumption" under different sea current conditions.

[0064] In another specific embodiment, the features further include a parameter standard deviation feature (STDofKeyParameters) and a mean draft feature (DraftTot): The instantaneous fluctuations of the ship propulsion system belong to "noise interference" rather than real performance changes - if the original high-frequency data is directly input, the model is easy to misjudge such fluctuations as performance features, resulting in a decrease in prediction stability. Therefore, based on the 5-minute time window in the "data smoothing" step, the standard deviation of the core parameters is calculated to generate the "parameter standard deviation feature" to quantify the stability of the working condition, as follows:

[0065] x For the core original parameters (in this study, the selected parameters are the water speed STW and the propeller shaft speed, as shown in Table 2 of the reference document); is the sampling value of the parameter at the t minute in the 5-minute window; is the mean value of the parameter in the 5-minute window; reflects the fluctuation degree of the parameter in a short time.

[0066] The draft of the ship directly affects the underwater wet surface area of the ship body, and then affects the water resistance - under different draft states, even if the shaft speed and the speed are the same, the main engine fuel consumption and the shaft power will have significant differences. In the original data, "fore and aft draft" is two independent parameters, if input into the model at the same time, it is easy to cause information redundancy. Therefore, the "mean draft feature" is generated by the arithmetic mean of the fore and aft draft, to simplify the representation of the influence of the draft on the propulsion system, as follows:

[0067] DraftAft is the aft draft (unit: m), DraftFore is the fore draft (unit: m), both of which are collected by pressure sensors; the mean draft can better represent the overall draft state of the ship, avoiding misjudgment of the model due to the small difference between the fore and aft draft.

[0068] In an embodiment, from the derived features related to the ship fuel consumption and the shaft power, features closely related to the predicted values of the ship fuel consumption and the shaft power are selected, and redundant features are removed, to obtain derived feature data, as shown in Figure 5 , including: S501: By evaluating the correlation of each derived feature with the ship fuel consumption and the shaft power, the derived features having correlation with the ship fuel consumption and the shaft power are selected; S502: By a feature selection algorithm, redundant derived features are removed, and derived features having independent contribution to the prediction of the ship fuel consumption and the shaft power are retained; S503: Construct a final derived feature dataset based on the selected derived features.

[0069] The core of feature selection is to screen input variables from the "original parameters + derived features" that meet the following three criteria: there is a clear physical causal relationship with the target variable; low redundancy with other input features; and the input dimension matches the model complexity.

[0070]

[0071] In an embodiment, as shown in Figure 6 The training method of the feedforward neural network model includes: S601: Preprocessing of feature training data, including data normalization and data shuffling; S602: Input the preprocessed feature training data into the feedforward neural network model, and perform forward propagation calculation to obtain the predicted value; S603: Calculate the loss between the predicted value and the true value, and adjust the network weight according to the loss value to obtain the trained feedforward neural network model.

[0072] In the data-driven modeling of ship propulsion system, the selection of model architecture needs to adapt to two core requirements: one is to accurately capture the strong nonlinear coupling relationship of the propulsion system (such as the nonlinear relationship between shaft speed, torque and MEFOC, and the dynamic interaction between environmental factors and ship resistance); The second is compatible with high-frequency, multi-dimensional preprocessed data sets. Traditional linear models cannot fit such complex nonlinear relationships, while recurrent neural networks (RNN) are good at time series modeling, but the parameter correlation of the ship propulsion system depends more on "current working condition characteristics" rather than "historical time series dependence", and RNN is prone to gradient explosion problem, not suitable for large-scale sample training.

[0073] In this application, a feedforward neural network (FNN) is used to construct a prediction model, and the precision is improved through parameter optimization and training strategy.

[0074] 1) Model structure parameter determination

[0075] 2) FNN model training process and key strategies Model training is the core step to achieve accurate mapping of "features to target", and this method designs a three-step process of "data preprocessing, training execution, and generalization capability optimization", each step is designed with specific strategies for ship data characteristics and FNN model characteristics: A. Data preprocessing before training To avoid training failure caused by data distribution bias and dimension difference, three key processing of input data are needed: Data Shuffling. Ship operation data has time series characteristics. If the training / validation set is directly divided according to the time sequence, the model will prefer to learn the "working condition characteristics of a specific time window", which will lead to insufficient generalization ability. Therefore, the "full data set random shuffling" strategy is adopted. The specific operation is: based on the timestamp, all samples are shuffled to ensure that the training set contains samples of different time periods, different working conditions, low speed / high speed, and wind / against wind; the shuffled data uses a fixed random seed (seed = 42) to ensure that the experiment is repeatable.

[0076] Data set division (Train-Validation-Test Split). According to the sample size and model verification requirements, two division schemes are adopted: Conventional training division: 80% training set + 20% validation set, used for model parameter debugging; Final test division: 70% training set + 20% validation set + 10% test set, used to evaluate the generalization ability of the model.

[0077] Data normalization (Normalization). The dimensions of the input features are significantly different. If directly input, the FNN will pay too much attention to the large dimension features and ignore the small dimension features. Therefore, Z-score normalization is adopted, the formula is:

[0078] Where: x is the original feature value; is the mean of the feature in the training set; is the standard deviation of the feature in the training set.

[0079] 3) Model training execution and monitoring The training process is implemented based on the Keras framework. The parameter initialization is: weight matrix Adopt He normal initialization Activation function, bias vector initialized to 0; Training iterations: a total of 50 rounds of epochs, each epoch includes two major steps: Forward propagation: input the input features of 64 samples through the input layer→hidden layer→output layer to calculate the predicted value ; Back propagation: calculate the gradient of the loss to each parameter through the Adam optimizer Update the model.

[0080] Optimization strategy: adopt 5-fold cross-validation (k-folds) to improve the generalization ability of the model; Integrated optimization: For the MEFOC prediction task, 10 FNN models with the same structure are trained, and the output mean is taken as the final prediction result.

[0081] The data quality of the present application is high and the distortion risk is low: through the three-step preprocessing of "interpolation-cleaning-smoothing", the abnormal time series such as GPS position can be accurately repaired, and the abnormal values of related parameters can be removed, which only affects about 7% of data points (interpolation), deletes 3.35% of data (cleaning), and the maximum difference of parameter mean after processing is only 0.74%, which ensures the reliability of data while avoiding distortion.

[0082] The prediction accuracy is significantly better than traditional methods: the FNN model with optimized architecture has the lowest RMSE of 0.78t / 24hr (R 2 =0.998) for main engine fuel consumption (MEFOC) prediction and the lowest RMSE of 132.07kW (R 2 =0.999) for shaft power prediction, which is much more accurate than empirical models and some existing data-driven models.

[0083] The speed loss estimation is more reliable and applicable to a wide range of scenarios: compared with ISO19030, the present application can retain 22000 data points and estimate the speed loss (5.12%) that conforms to the actual law of loss increase caused by hull fouling, reducing the uncertainty of the results. Breakthrough the limitations of existing technologies in "data distortion, weak feature correlation, and poor model generalization", realize the high-precision prediction of ship main engine fuel consumption (MEFOC) and propeller shaft power, and reliable estimation of speed loss, ultimately support ship energy efficiency optimization and carbon emission control.

[0084] The present application realizes high-precision prediction of ship main engine fuel consumption and propeller shaft power, and reliable estimation of speed loss through systematic data preprocessing, physically driven feature engineering, and optimized architecture of feedforward neural network. Through the above methods, not only the data quality is improved and the data distortion problem in existing technologies is overcome, but also the prediction results are more consistent with the actual operation of the ship through the deep integration of physical properties and environmental factors. In addition, the present application has stronger adaptability and generalization ability, which can support ship energy efficiency optimization, carbon emission control and speed optimization and other applications.

[0085] Based on the same inventive concept, the embodiment of the present application also provides a ship fuel consumption and shaft power prediction device, which can be used to implement the method described in the above embodiment, as described in the following embodiment. Since the principle of solving the problem of the ship fuel consumption and shaft power prediction device is similar to that of the ship fuel consumption and shaft power prediction method, the implementation of the ship fuel consumption and shaft power prediction device can be referred to the implementation of the ship fuel consumption and shaft power prediction method, and the repeated parts will not be described here. The term "unit" or "module" used below can be a combination of software and / or hardware that implements a predetermined function. Although the system described in the following embodiment is preferably implemented in software, hardware, or a combination of software and hardware is also possible and is conceived.

[0086] According to a second aspect of the present application, a ship fuel consumption and shaft power prediction device is also provided, as shown in Figure 7 comprising: a preprocessing unit 701 configured to preprocess ship operation parameter data collected by a sensor system to obtain preprocessed data; a feature extraction unit 702 configured to extract features from the preprocessed data to obtain feature data, the feature data including operation features related to ship fuel consumption and shaft power; a prediction unit 703 configured to input the feature data into a pre-trained feedforward neural network model to obtain a predicted value of ship fuel consumption and shaft power.

[0087] In an embodiment, as shown in Figure 8 the preprocessing unit 701 comprises: an outlier identification and correction module 801 configured to identify and correct outliers in the ship operation parameter data, including using an interpolation algorithm to correct the jump caused by the outliers, or correcting the outliers based on a statistical method; a smoothing processing module 802 configured to smooth the ship operation parameter data, including smoothing the data within a predetermined time window by applying a moving average method to remove high-frequency noise and improve the stability of the data; a missing value filling module 803 configured to fill missing values in the ship operation parameter data.

[0088] In an embodiment, as shown in Figure 9 the feature extraction unit 702 comprises: a derived feature generation module 901 configured to generate derived features related to ship fuel consumption and shaft power by performing mathematical conversion or combination on the preprocessed data; The screening module 902 is configured to select the derived features closely related to the ship fuel consumption and shaft power prediction value from the derived features related to the ship fuel consumption and shaft power, and remove redundant derived features to obtain derived feature data.

[0089] In an embodiment, as shown in Figure 10 The derived feature generation module 901 includes: The first calculation module 1001 is configured to calculate the wind effect feature according to the wind speed and the wind direction, and generate a derived feature for quantifying the influence of the wind on the ship propulsion system; The second calculation module 1002 is configured to generate the hull fouling feature according to the ship operation time, and generate a derived feature for reflecting the influence of the hull fouling on the propulsion efficiency; The third calculation module 1003 is configured to calculate the sea current feature according to the difference between the ground speed and the water speed, and generate a derived feature for quantifying the influence of the sea current on the ship speed; The fourth calculation module 1004 is configured to calculate the standard deviation of the key parameters, and generate a parameter standard deviation feature for reflecting the stability of the working condition of the ship propulsion system.

[0090] In an embodiment, as shown in Figure 11 The screening module 902 includes: The evaluation selection module 1101 is configured to select the derived features having correlation with the ship fuel consumption and shaft power by evaluating the correlation of each derived feature with the ship fuel consumption and shaft power; The redundancy removal module 1102 is configured to remove the redundant derived features by a feature selection algorithm, and retain the derived features having independent contribution to the prediction of the ship fuel consumption and shaft power; The data set generation module 1103 is configured to construct a final derived feature data set based on the selected derived features.

[0091] In an embodiment, the training method of the feedforward neural network model includes: The feature training data is preprocessed, including data normalization and data shuffling; The preprocessed feature training data is input into the feedforward neural network model to perform forward propagation calculation to obtain a prediction value; The loss between the prediction value and the true value is calculated, and the backward propagation algorithm is adopted to adjust the network weight according to the loss value to obtain the trained feedforward neural network model.

[0092] Those skilled in the art will appreciate that embodiments of the present application can be readily used as software, hardware, or a combination of software and hardware. In a software embodiment, various software modules are stored in memory units (e.g., DRAM, SRAM, ROM) and executed by one or more general purpose or special purpose processors. Various embodiments of the present application can be implemented in software and / or firmware applications.

[0093] The present application is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagrams, as well as combinations of flows and / or blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagrams block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified by the flow or flows and / or block or blocks.

[0094] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instructions means which implement the function specified in the flowchart and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified by the flow or flows and / or block or blocks.

[0095] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the function specified in the flowchart and / or block diagrams flow or flows and / or block or blocks. Figure 1 one or more flows and / or blocks Figure 1 means for carrying out the function specified by the flow or flows and / or block or blocks.

[0096] The present application is described in reference to the flowchart and / or block diagrams of the method, apparatus (system) and computer program product according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagrams, as well as combinations of flows and / or blocks in the flowchart and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart and / or block diagrams block or blocks.

[0097] The embodiment of the present application further provides a specific implementation of an electronic device capable of implementing all steps in the method in the above embodiment, referring to Figure 12 , and the electronic device specifically includes the following contents: Figure 12 An entity structure schematic diagram of the electronic device provided by the embodiment of the present application is provided, and the electronic device includes a processor 1201, a memory 1202 and a bus 1203.

[0098] The processor 1201, the memory 1202 and the bus 1203 are in communication with each other.

[0099] The processor 1201 is configured to invoke a computer program in the memory 1202, and the processor implements all steps in the method in the above embodiment when executing the computer program, for example, the processor implements the following steps when executing the computer program: S101: preprocessing ship operation parameter data collected by a sensor system to obtain preprocessed data; S102: extracting features from the preprocessed data to obtain feature data, the feature data including operation features related to ship fuel consumption and shaft power; S103: inputting the feature data into a pre-trained feedforward neural network model to obtain predicted values of the ship fuel consumption and the shaft power.

[0100] The embodiment of the present application further provides a computer readable storage medium capable of implementing all steps in the method in the above embodiment, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement all steps in the method in the above embodiment, for example, the processor implements the following steps when executing the computer program: S101: preprocessing ship operation parameter data collected by a sensor system to obtain preprocessed data; S102: extracting features from the preprocessed data to obtain feature data, the feature data including operation features related to ship fuel consumption and shaft power; S103: inputting the feature data into a pre-trained feedforward neural network model to obtain predicted values of the ship fuel consumption and the shaft power.

[0101] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on its differences from other embodiments. In particular, for hardware + program embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. Although the embodiments in this specification provide the method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive means. The order of steps listed in the embodiments is merely one possible execution order among many steps and does not represent the only execution order. In actual device or terminal product execution, the methods can be executed in the order shown in the embodiments or drawings or in parallel (e.g., in a parallel processor or multi-threaded processing environment, or even a distributed data processing environment). The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, product, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, product, or apparatus. Without further limitations, the presence of other identical or equivalent elements in the process, method, product, or apparatus that includes said elements is not excluded. For ease of description, the above devices are described in terms of function, divided into various modules. Of course, in implementing the embodiments of this specification, the functions of each module can be implemented in one or more software and / or hardware, or the module implementing the same function can be implemented by a combination of multiple sub-modules or sub-units, etc. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings or direct couplings or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, such that the instructions, which are executable by the processor of the computer or other programmable data processing device, produce instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0102] Those skilled in the art will appreciate that embodiments of the present specification can be devised for a method, a system, or a computer program product. Accordingly, embodiments of the present specification can be embodied in the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, embodiments of the present specification can be in the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk memory, CD-ROM, optical storage medium, etc.) embodying computer usable program code. Each of the various embodiments of the present specification is described in a progressive manner, and reference can be made to other embodiments for the same or similar parts. Each embodiment focuses on the differences from other embodiments. In particular, for system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and reference can be made to the description of the method embodiments. In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the embodiments of the present specification.

[0103] In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction. The above is only an embodiment of the embodiments of the present specification and is not intended to limit the embodiments of the present specification. The embodiments of the present specification can have various modifications and changes for those skilled in the art. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the embodiments of the present specification shall be included in the scope of the claims of the embodiments of the present specification.

Claims

1. A method for predicting ship fuel consumption and shaft power, characterized in that, include: Preprocess the ship operation parameter data collected by the sensor system to obtain preprocessed data; Feature data is obtained by extracting features from the preprocessed data, and the feature data includes operational features related to ship fuel consumption and shaft power. The feature data is input into a pre-trained feedforward neural network model to obtain predicted values ​​for ship fuel consumption and shaft power.

2. The method for predicting ship fuel consumption and shaft power according to claim 1, characterized in that, The preprocessing of ship operation parameter data collected by the sensor system includes: Identify and correct outliers in the ship operation parameter data, including using interpolation algorithms to correct jumps caused by the outliers, or correcting the outliers based on statistical methods; The ship operation parameter data is smoothed, including smoothing the data within a predetermined time window by applying a moving average method to remove high-frequency noise and improve the stability of the data. Missing values ​​in the ship operation parameter data are filled in.

3. The method for predicting ship fuel consumption and shaft power according to claim 1, characterized in that, The step of extracting features from the preprocessed data to obtain feature data includes: By performing mathematical transformations or combinations on the preprocessed data, derived features related to ship fuel consumption and shaft power are generated. From the derived features related to ship fuel consumption and shaft power, we select derived features that are closely related to the predicted values ​​of ship fuel consumption and shaft power, and at the same time filter out redundant derived features to obtain the derived feature data.

4. The method for predicting ship fuel consumption and shaft power according to claim 3, characterized in that, The process of generating derived features related to ship fuel consumption and shaft power by mathematically transforming or combining the preprocessed data includes: Based on wind speed and wind direction, wind effect characteristics are calculated, and derived characteristics are generated to quantify the impact of wind on ship propulsion systems. Based on the ship's operating time, hull fouling characteristics are generated, and derived characteristics are generated to reflect the impact of hull fouling on propulsion efficiency. Ocean current characteristics are calculated based on the difference between the speed over land and the speed over water, and derived characteristics are generated to quantify the impact of ocean currents on ship speed. Calculate the standard deviation of key parameters and generate parameter standard deviation features, which are derived features that reflect the stability of the ship propulsion system under operating conditions.

5. The method for predicting ship fuel consumption and shaft power according to claim 3, characterized in that, The process involves selecting features closely related to the predicted values ​​of ship fuel consumption and shaft power from the derived features related to ship fuel consumption and shaft power, while simultaneously filtering out redundant features to obtain the derived feature data, including: By evaluating the correlation between each derived feature and ship fuel consumption and shaft power, derived features that are correlated with ship fuel consumption and shaft power are selected. Redundant derived features are removed by using a feature selection algorithm, while retaining derived features that make independent contributions to the prediction of ship fuel consumption and shaft power. Based on the selected derived features, the final derived feature dataset is constructed.

6. The method for predicting ship fuel consumption and shaft power according to claim 1, characterized in that, The training method for the feedforward neural network model includes: Preprocess the feature training data, including data normalization and data shuffling; The preprocessed feature training data is input into the feedforward neural network model for forward propagation calculation to obtain the predicted value. The loss between the predicted value and the true value is calculated, and the backpropagation algorithm is used to adjust the network weights according to the loss value to obtain the trained feedforward neural network model.

7. A device for predicting ship fuel consumption and shaft power, characterized in that, include: The preprocessing unit is used to preprocess the ship operation parameter data collected by the sensor system to obtain preprocessed data; The feature extraction unit is used to extract features from the preprocessed data to obtain feature data, the feature data including operational features related to ship fuel consumption and shaft power; The prediction unit is used to input the feature data into a pre-trained feedforward neural network model to obtain predicted values ​​of ship fuel consumption and shaft power.

8. The ship fuel consumption and shaft power prediction device according to claim 7, characterized in that, The preprocessing unit includes: The outlier identification and correction module is used to identify and correct outliers in the ship operation parameter data, including correcting the jumps caused by the outliers using an interpolation algorithm, or correcting the outliers based on statistical methods. The smoothing module is used to smooth the ship operation parameter data, including smoothing the data within a predetermined time window by applying the moving average method to remove high-frequency noise and improve the stability of the data. The missing value filling module is used to fill in the missing values ​​in the ship operation parameter data.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the ship fuel consumption and shaft power prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the ship fuel consumption and shaft power prediction method according to any one of claims 1 to 7.

Images