Optimal navigational speed evaluation method based on neural network

Abstract

The invention discloses an optimal navigational speed evaluation method based on a neural network, which is a method for obtaining optimal navigational speed evaluation by combining an oil consumption prediction model oriented to inland river luxurious tourism ships with a navigational speed optimization model based on a genetic algorithm. A back propagation neural network-based fuel consumption prediction model is established by using actual ship navigation data, a nonlinear planning navigational speed optimization model is established on the basis of a fuel consumption model and leg division, and a genetic algorithm is used as a solving algorithm on the basis of fuel consumption model establishment and navigational speed optimization model establishment. And carrying out navigational speed optimization on one section of navigational distance of the target ship from the course to obtain a high-precision prediction result.

Description

Technical Field

[0001] This invention relates to the field of energy harvesting and electric propulsion vessels, specifically to an optimal speed evaluation method based on neural networks. Background Technology

[0002] In recent years, ship energy efficiency optimization has become a focus of academic attention, with researchers studying and proposing various energy efficiency optimization methods. Improving ship energy efficiency can be achieved through various technological optimization methods and operational management optimization methods. Technological optimization methods include propeller redesign, hull antifouling measures, engine improvements, hull optimization design, and the use of clean energy. Investing in these new technologies requires additional costs, while improving ship energy efficiency through various operational measures during operation is less costly or even requires no investment. Therefore, ship owners prefer to improve ship energy efficiency through operational management. Operational management-based energy efficiency optimization methods mainly include speed optimization, route optimization, trim optimization, and fleet deployment optimization. Although these measures can improve ship energy efficiency and reduce fuel consumption to some extent, since the power required to maintain the current speed is proportional to the cube of the speed, even a slight decrease in ship speed can significantly improve ship energy efficiency. A 10% decrease in speed is equivalent to a reduction of approximately 27% in fuel consumption per unit time. According to relevant research, reducing ship speed by 2-3 knots from the design speed can significantly reduce ship fuel consumption and lower operating costs. Therefore, speed optimization is a more critical optimization method.

[0003] However, reduced speed also has some adverse effects. For example, due to the decrease in speed, a larger fleet is needed to handle the same volume of trade, ordering new vessels incurs additional costs, and it also affects the shipping experience for customers. Furthermore, excessively low speeds inevitably increase sailing time and reduce the combustion efficiency of the main engine, thus increasing fuel consumption per unit distance. Considering these other adverse effects of reduced speed, determining the optimal economic speed for a vessel is of significant research importance. Summary of the Invention

[0004] This invention aims to provide a neural network-based optimal speed assessment method. It combines a fuel consumption prediction model for inland luxury tourist vessels with a speed optimization model based on a genetic algorithm to obtain the optimal speed assessment. A fuel consumption prediction model based on a backpropagation neural network (BPNN) is established using real-ship navigation data. Based on the fuel consumption model and segment division, a nonlinear programming speed optimization model is constructed. Building upon the fuel consumption model and speed optimization model, a genetic algorithm is used as the solution algorithm to optimize the speed of the target vessel for a segment of the route. This method can achieve high-accuracy prediction results.

[0005] To solve the above problems, the present invention adopts the following technical solution: A neural network-based optimal speed assessment method combines a fuel consumption prediction model for inland luxury tourist vessels with a speed optimization model based on a genetic algorithm to obtain the optimal speed assessment. A fuel consumption prediction model based on a back propagation neural network (BPNN) is established using actual ship navigation data. Based on the fuel consumption model and segment division, a nonlinear programming speed optimization model is constructed. Using a genetic algorithm as the solution algorithm, the speed of the target vessel is optimized for a segment of the route, yielding high-precision prediction results.

[0006] The specific steps of this method are as follows: S1. Data Acquisition and Preprocessing: The data acquisition and preprocessing system is designed with a two-layer network structure, namely the perception layer and the decision layer. The perception layer is completed by distributed data acquisition units to collect data such as cabin equipment and ship navigation parameters; the decision layer is completed by the system server to realize real-time processing of the collected data, as well as the collection of deep learning training data in the early stage and the storage of algorithm optimization data in the later stage. The perception layer and the decision layer achieve high-speed data communication through an industrial Ethernet star connection; S2. Fuel Consumption Prediction Model Construction: The fuel consumption prediction model is divided into three categories: white-box model, black-box model and gray-box model of ship fuel consumption; among them, the structure and parameters of the white-box model of ship fuel consumption are known, and it is established based on the relevant knowledge in the ship field and the interaction relationship between ship-engine-propeller; the structure and parameters of the black-box model of ship fuel consumption are unknown, and it is established based on training with massive historical operating data; the structure and parameters of the black-box model of ship fuel consumption are unknown, and it is established based on training with massive historical operating data; The gray box model is between the white box model and the black box model of ship fuel consumption. Some of its physical feature data can be directly measured, while other features that are difficult to obtain are obtained through data training. S3, Target ship fuel consumption model construction: Combine the massive historical monitoring data of the real ship operation monitoring platform, and use the backpropagation neural network to construct the black box model of ship fuel consumption for the target ship. S4, Speed ​​optimization: (1) Speed ​​optimization assumptions and segmented speed optimization model construction: By performing speed segmentation and rapid optimization on different segments within the target ship's fixed route, the total fuel consumption of the route objective function is minimized. Based on the above fuel consumption model and segment division, a nonlinear programming optimization model is constructed. (2) Speed ​​optimization based on genetic algorithm: Based on the fuel consumption model construction and speed optimization model construction, the genetic algorithm is used as the solution algorithm to optimize the speed of one segment of the navigation distance of each sub-segment of the target ship's route from Chongqing to Yichang.

[0007] In step S1, the sensing layer consists of a signal acquisition box, fuel flow meters arranged on the generator inlet and return oil pipelines, a shaft power tester installed on the shafting, an energy metering device embedded in the ship's main switchboard, a ship engine room monitoring and alarm system (not part of this system), the main switchboard system, a ship draft sensor, and ship communication and navigation equipment. The signal acquisition box contains a PLC module to acquire data from Ethernet signals, serial port signals, AI signals, and DI signals, meeting the system's signal acquisition requirements. Fuel flow meters are arranged on the inlet and return oil pipelines of fuel-consuming equipment. Combined with fuel refueling information and fuel level data from the ship's fuel tanks, they drive the fuel statistics algorithm module to obtain accurate fuel consumption data for ship navigation and docking. The magnetic strip sensed by the shaft power meter is attached to the ship's shafting. Combined with the high-precision sensing sensor of the shaft power meter, static verification and ship navigation process data are used to determine the fuel consumption. The system uses dynamic verification and other methods to acquire power, speed, and torque signals of the ship's shafting. An energy metering device embedded in the main switchboard, combined with current transformers mounted on the main switchboard's copper busbars, statistically analyzes input voltage and current signals, converting them into active power, reactive power, and power factor parameters. This data is used to calculate the total power generation of the ship's generator sets and the specific power consumption of each branch circuit. The engine room monitoring and alarm system provides the operating status and alarm signals of all engine room equipment during navigation, as well as information on fuel refueling and specific fuel level data in the fuel tanks during navigation. Combined with a tank capacity model, it calculates the specific fuel quantity for each time period, assisting flow meters in accurate fuel consumption statistics. The main switchboard system is used for the distribution of electric propulsion power and monitoring the operating status of propulsion equipment during navigation. Combined with the power station management system, it provides a safe, reliable, high-quality, and economical intelligent power supply management system.

[0008] Furthermore, the power station management system also has functions of power generation management, load management, and power distribution management; the ship's draft sensor reflects the ship's draft and trim status under the current river conditions; the wind speed and direction instrument, log, depth sounder, and GPS mounted on the ship's end reflect signals such as wind speed and direction, ship speed, current water depth of the Yangtze River, ship's navigation position, and ship speed during the ship's navigation process.

[0009] Furthermore, the serial port module, AI module, and DI module inside the signal acquisition box interact with sensors within the system and sensors / systems outside the system. The data interface of the acquisition module has a built-in noise filter combined with a Kalman filter algorithm module to ensure the authenticity of the data. The logic control program of the acquisition module is designed with a network disconnection reconnection mechanism to ensure the reliability of data communication. The data interaction cycle between each system and device reaches the level of hundreds of milliseconds, which improves the response efficiency of the entire system from the perception level.

[0010] Furthermore, in step S2, the white-box model structure and parameters of ship fuel consumption include the ship's main engine obtaining power through burning fuel, the power being transmitted through the gearbox and main engine shafting equipment to the propeller, and the propeller, receiving the transmitted power, interacting with the water flow to propel the ship forward; and considering the influence of wind speed, wind direction, current speed, current direction, and ship draft on the ship's state, as well as the influence of downstream and upstream currents on the main engine fuel consumption; the black-box model training algorithm for ship fuel consumption adopts the extraneous tree regression algorithm, Naive Bayes algorithm, human neural network, support vector machine supervised algorithm, linear regression learning algorithm, or random forest regression; when constructing the gray-box model of ship fuel consumption, the black part is segmented out through mathematical methods, and then the mathematical model of this unobservable part is inferred using partial data.

[0011] Furthermore, in step S3, the backpropagation neural network is a multi-layer feedforward network, including an input layer, a hidden layer, and an output layer. This backpropagation neural network can be supervised and has the characteristics of signal forward propagation and error backpropagation. The computational learning process of the backpropagation neural network is divided into two stages, including signal forward propagation and error backpropagation. During the backpropagation process, the weights and biases from the hidden layer to the output layer and from the input layer to the hidden layer are optimized sequentially.

[0012] Furthermore, the backpropagation neural network computation includes the input layer's common... There are 10 node variables, and the hidden layer has a total of 10 nodes. There are [number] nodes in the output layer. indivual, For the first The input value of each node, For the first hidden layer The node and the first input layer The weights between nodes For the hidden layer The threshold of each node, Let be the activation function of the hidden layer. For the output layer The node and the hidden layer The weights between nodes Then it is the output layer. The threshold of each node, The activation function of the output layer. The output of the final output layer nodes; during the forward propagation of the neural network signal, the output value of the output layer... The calculation formula is as follows: (1) Backpropagation of neural network errors will adjust and optimize the weights and thresholds of each layer. The error correction of the prediction result is as follows: (2) The corrected formulas for calculating the weights and thresholds of each node are as follows: (3) (4) (5) (6) Because black-box models are highly dependent on datasets, when selecting target vessels, vessels with smaller errors between fuel consumption flow meter measurement data and manual measurement data are chosen to ensure the validity of the flow meter sampling data.

[0013] Furthermore, in step S4, a nonlinear programming optimization model is constructed, with the objective function as follows: (9) Constraints: (10) In the formula: For the first Ground speed for each segment of the flight path; For Japanese standard vessels The initial draft of each segment of the navigation route; For the target vessel The tail draft of each segment; For the target vessel The heading angle for each segment; For the first Wind that divides the navigation segment Quantity; For the first Wind that divides the navigation segment Quantity; For the first The meaningful wave height for dividing a flight segment; For the first The effective wave period for each segment; For the first The main wave angle for each segment of the navigation route; For the first Flow divided into segments Quantity; For the first Flow divided into segments Quantity; For the first The distance of each segment; To optimize ship departure times; The latest time for the ship to depart; This represents the expected duration of the ship's total sailing time. , For the minimum and maximum ground speed limits, in Equation (10), constraint 1 means that the sum of the segmented voyage distances must be equal to the total voyage distance; constraint 2 means that the voyage time must be less than the planned time; constraint 3 means that the departure time of the ship must be earlier than the planned time; constraint 4 means that the speed of the ship during the voyage cannot be greater than or less than the maximum or minimum speed of the ship. Furthermore, a specific precision constraint is added to the original objective function to perform global optimization of the flight path to one decimal place. The modified objective rainfall number is as follows: (11) (12) in, The constraint is integer, and ∈[60, 1301, variable substitution in the iterative operation The values ​​were adjusted to the required precision for this paper, and convergence was achieved after 895 iterations.

[0014] The beneficial effects of this invention are: This invention proposes an optimal speed assessment method based on neural networks. It combines a fuel consumption prediction model for inland luxury tourist vessels with a speed optimization model based on genetic algorithms to obtain the optimal speed assessment method. Based on the monitoring data and information of the vessel's navigation status and power status, the method assesses the vessel's power status, navigation and loading status, etc., and provides the vessel with assessment results and speed optimization solutions. This enables real-time monitoring, assessment and optimization of the vessel's power consumption, thereby continuously improving the level of vessel power management. Attached Figure Description

[0015] Figure 1 Block diagram of a data acquisition system for luxury cruise ships; Figure 2 Block diagram of the data acquisition system for luxury cruise ships; Perception layer system block diagram; Figure 3 This is a schematic diagram of a backpropagation neural network; Figure 4 This is a flight route map from Chongqing to Yichang. Figure 5 This is a flowchart of the genetic algorithm. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0017] like Figures 1 to 5 As shown in the figure, the present invention provides an optimal speed evaluation method based on neural networks, which combines a fuel consumption prediction model with a speed optimization model based on a genetic algorithm to obtain the optimal speed evaluation result. The method includes the following steps: S1. Data Acquisition and Preprocessing S1.1 Data Acquisition and Preprocessing System Design like Figure 1 As shown in Figure 2, this system adopts a two-layer network structure: a perception layer and a decision layer. The perception layer consists of distributed data acquisition units that collect data on cabin equipment and ship navigation parameters. The decision layer, handled by the system server, performs real-time processing of the collected data, as well as the collection of training data for deep learning and the storage of data for algorithm optimization. High-speed data communication between devices is achieved through a star-shaped connection using industrial Ethernet.

[0018] The sensing layer of this system consists of a signal acquisition box, fuel flow meters arranged on the generator inlet and outlet oil pipelines, a shaft power tester installed on the shaft system, an energy metering device embedded in the ship's main switchboard, a ship engine room monitoring and alarm system that is not part of this system, the main switchboard system, a ship draft sensor, and ship communication and navigation equipment.

[0019] The signal acquisition box houses a PLC module, capable of acquiring data from various signal types including Ethernet, serial, AI, and DI signals, meeting the system's signal acquisition requirements. Fuel flow meters are installed on the inlet and outlet lines of major fuel-consuming equipment. Combined with fuel refueling information and fuel level data from the ship's fuel tanks, they drive a fuel statistics algorithm module to obtain relatively accurate fuel consumption data during ship navigation and docking. A shaft power meter, mounted on the ship's shafting, uses a magnetic strip and a high-precision sensor to acquire power, speed, and torque signals from the shafting through static and dynamic verification during navigation. This provides accurate, real-time power data for subsequent speed optimization model training. An energy metering device embedded in the main distribution board, combined with a current transformer mounted on the main distribution board's copper busbar, statistically analyzes input voltage and current signals. The signal is then converted into parameters such as active power, reactive power, and power factor, used to statistically analyze the total power generation of the ship's generator sets and the specific power consumption of each branch equipment. This is the core statistical device of this energy-saving system. The engine room monitoring and alarm system provides signals such as the operating status and alarm status of the entire ship's engine room equipment during navigation. It also provides information on ship fuel refueling and specific fuel level data in the fuel tanks during navigation. Combined with the tank capacity model, it can calculate the specific fuel quantity of the ship at different time periods, assisting the flow meter in accurate fuel consumption statistics. The main switchboard system is used for the distribution of electric propulsion power and monitoring the operating status of propulsion equipment during ship navigation. Combined with the power station management system, it provides a safe, reliable, high-quality, and economical intelligent power supply management system. The power station management system also has power generation management, load management, and power distribution management functions. The ship's draft sensor reflects the ship's draft and trim status under the current river conditions. The anemometer, log, depth sounder, and GPS mounted on the ship's end reflect signals such as wind speed and direction, ship speed, current Yangtze River depth, ship position, and ship speed during navigation.

[0020] The serial port module, AI module, and DI module inside the signal acquisition box interact with sensors within the system and with sensors / systems outside the system. The data interface of the acquisition module has a built-in noise filter combined with a Kalman filter algorithm module to ensure the authenticity of the data. The logic control program of the acquisition module is designed with a network disconnection reconnection mechanism to ensure the reliability of data communication. The data interaction cycle between various systems and devices can reach the level of hundreds of milliseconds, which improves the response efficiency of the entire system from the perception level.

[0021] S2, Fuel Consumption Prediction Model Construction The construction of a fuel consumption model is fundamental to ship speed optimization, and the accuracy of the fuel consumption model's predictions significantly impacts the accuracy of speed optimization. When navigating inland waterways, a ship's fuel consumption is the result of multiple influencing factors, including wind speed, wind direction, current speed, current direction, and ship draft. Even when maintaining the same speed or RPM while sailing steadily on the Yangtze River, the ship's main engine fuel consumption will vary considerably under different wind, wave, and current conditions. Furthermore, measurement errors in ship fuel consumption can also affect the accuracy of the fuel consumption model. Fuel consumption prediction models can be categorized into three types: white-box models, black-box models, and gray-box models.

[0022] S2.1 White Box Model The structure and parameters of the white-box model are known. The white-box model of ship fuel consumption is based on relevant knowledge in the field of shipbuilding and is established on the interaction relationship between ship, engine, and propeller. The ship's main engine obtains power by burning fuel. The power is transmitted through the gearbox, main engine shaft system, and other equipment to the propeller. The propeller, which receives the transmitted power, interacts with the water flow to propel the ship forward. Factors such as wind speed, wind direction, current speed, current direction, and ship draft can greatly affect the state of the ship, especially with and against the current, which have a huge impact on the main engine fuel consumption.

[0023] S2.2 Black Box Model Unlike traditional models, the structural parameters of a ship fuel consumption black-box model are unknown. While the prediction accuracy of these models is satisfactory, they require massive amounts of historical operational data for training. Common training algorithms for ship fuel consumption black-box models include Extraneous Tree Regression, Naive Bayes, Human Neural Networks, Support Vector Machine supervised learning, Linear Regression, and Random Forest Regression. Black-box models offer better prediction accuracy and data fitting performance, but they also have several drawbacks, such as the reliance on massive amounts of historical operational data and the inability of the model structure and parameters to directly reflect the professional knowledge of the shipping industry.

[0024] S2.3 Gray Box Model The gray box model falls between the white box and black box models. Some of its physical characteristics can be directly measured, while others, which are difficult to obtain, are derived through data training. When constructing a gray box model for ship fuel consumption, it is necessary to use mathematical methods to segment out the "black box" portion, and then use partial data to infer the mathematical model of this obscure part. The gray box model is one of the options for constructing fuel consumption models. S3. Construction of the target ship fuel consumption model While the inherent logical relationships and parameter structures of white-box models are known, and they offer high visualization capabilities when constructing ship fuel consumption models, their predictive accuracy is limited. Conversely, black-box models offer satisfactory predictive accuracy and data fitting performance, but require massive amounts of historical operational data for training, and their parameter structures are unknown. Taking advantage of our ability to obtain massive amounts of historical monitoring data from real-ship operation monitoring platforms, this paper utilizes a backpropagation neural network to construct a black-box fuel consumption model for the target ship.

[0025] A backpropagation neural network consists of an input layer, hidden layers, and an output layer; it is a multi-layer feedforward network. This network can be supervised and features both signal forward propagation and error backward propagation. The computational learning process of a backpropagation neural network is divided into two stages: signal forward propagation and error backward propagation. During backpropagation, the weights and biases from the hidden layer to the output layer and from the input layer to the hidden layer are optimized sequentially. Figure 3 As shown, the input layer has a total of There are 10 node variables, and the hidden layer has a total of 10 nodes. There are [number] nodes in the output layer. indivual. For the first The input value of each node, For the first hidden layer The node and the first input layer The weights between nodes For the hidden layer The threshold of each node, Let be the activation function of the hidden layer. For the output layer The node and the hidden layer The weights between nodes Then it is the output layer. The threshold of each node, The activation function of the output layer. This is the output for the nodes in the final output layer.

[0026] Therefore, during the forward propagation of the backpropagation neural network signal, the output value of the output layer... The calculation formula is as follows: (1) Backpropagation of neural network errors will adjust and optimize the weights and thresholds of each layer. The error correction of the prediction result is as follows: (2) The corrected formulas for calculating the weights and thresholds of each node are as follows: (3) (4) (5) (6) Because black-box models are highly dependent on datasets, this paper selects vessels with minimal errors in fuel consumption flow meter measurements compared to manual measurements to ensure the validity of the flow meter sampling data. The target vessel is equipped with several sensors to collect real-time data on various meteorological conditions encountered during navigation, as well as the operation of onboard main engines and other equipment. The sampling frequency of the main engine fuel consumption flow meter is 1 Hz. Since instantaneous equipment operation data cannot accurately reflect the impact on main engine fuel consumption, the data acquisition system converts the collected data into corresponding tables, statistically analyzing main engine fuel consumption values ​​at 10-minute, 1-hour, and 1-day intervals, summing the collected data values, and taking the average of the data collected at 10-minute intervals for the ship's draft at both ends. Fuel consumption monitoring data for the Yangtze River luxury cruise ship's route from Chongqing to Yichang was obtained through an intelligent energy efficiency management system deployed on the ship's bow. The main parameters of the ship are shown in Table 1.

[0027] Table 1. Main parameters of Yangtze River luxury cruise ships

[0028] The intelligent energy efficiency management system acquired a dataset recording fuel consumption of the main engine at 10-minute intervals, totaling approximately 10,000 records. After acquiring the monitoring data, the ship operation monitoring data was cleaned to improve data quality and ensure that the training data for the fuel consumption model met data quality requirements. After data cleaning, 6,180 fuel consumption data records were selected. Since the more feature parameters used to build the backpropagation neural network computational model, the more complex the calculation becomes, it is necessary to select appropriate feature parameters. That is, the selected feature parameters should have a significant correlation between the input feature parameters and the output, while having a low correlation among the input feature parameters themselves. The correlation coefficient is a statistical indicator reflecting the degree of linear correlation between variables. Its value range is [1, 1], with positive values ​​indicating positive correlation and negative values ​​indicating negative correlation. The closer the value is to 0, the weaker the linear relationship. The formula for calculating the correlation coefficient is as follows: (7) in, Let 1 be the variable. For variable 2, , The mean values ​​of variables 1 and 2 are used. By studying the correlation coefficients among multiple random variables, the input layer feature parameters with high correlation are merged to reduce the dimensionality of the input layer, thereby improving the accuracy and efficiency of network computation. Through calculation and screening, the selected model input parameters include flow direction, flow velocity, main wave direction angle, meaningful wave height, significant wave period, wind speed, wind direction, heading angle, ship draft, and speed. After selecting feature parameters, preprocessing, and standardizing the ship navigation data, the dataset is randomly shuffled and divided into training and test sets in a 7:3 ratio. 4326 data samples are used as the training set, and 1854 samples are used as the test set. The training parameters of the fuel consumption model are shown in Table 2. The input layer includes the 9 parameters shown in Table 2, and the output layer outputs the ship's main engine fuel consumption. The model has 5 hidden layers, each with 9 nodes.

[0029] Table 2 Input parameters for fuel consumption model training

[0030] After training is complete, the mean absolute error of the training results is calculated using the following formula: (8) in, For each group of predicted values, For each group of actual values.

[0031] After training, the model's average absolute error was 2.3%, and the accuracy and generalization ability of the fuel consumption model basically met the requirements of engineering applications.

[0032] S4, Speed ​​Optimization S4.1 Speed ​​Optimization Assumptions and Construction of Segmented Speed ​​Optimization Model Due to the long flight distance from Chongqing to Yichang, see Figure 4 Fuel consumption in the early and late stages of a voyage, or passenger disembarkation midway, can cause changes in the ship's draft. Therefore, speed optimization is only performed on the distance immediately after departure or before arrival at port. When optimizing speed segments for ships on a fixed route, the total voyage time, speed limits between adjacent segments, and the limitations imposed by the captain and navigator must be considered. Furthermore, meteorological data recorded in the meteorological database is spaced at intervals of 1 degree of longitude and 1 degree of latitude, approximately 60 nautical miles. Therefore, based on meteorological data in the target ship's track map and inflection points in the route during navigation, segments are clustered. Each segment is divided into smaller segments of 50–80 nautical miles to ensure high commonality in meteorological data and heading among different segments, while lower commonality exists between different segments. The aim of this method is to quickly optimize speed segments within the target ship's fixed route to minimize the total fuel consumption of the route objective function. Based on the above fuel consumption model and segment division, a nonlinear programming optimization model is constructed.

[0033] Objective function: (9) Constraints: (10) In the formula: For the first Ground speed for each segment of the flight path; For Japanese standard vessels The initial draft of each segment of the navigation route; For the target vessel The tail draft of each segment; For the target vessel The heading angle for each segment; For the first Wind that divides the navigation segment Quantity; For the first Wind that divides the navigation segment Quantity; For the first The meaningful wave height for dividing a flight segment; For the first The effective wave period for each segment; For the first The main wave angle for each segment of the navigation route; For the first Flow divided into segments Quantity; For the first Flow divided into segments Quantity; For the first The distance of each segment; To optimize ship departure times; The latest time for the ship to depart; This represents the expected duration of the ship's total sailing time. , The minimum and maximum ground speed limits are set. In Equation (10), constraint 1 means that the sum of the segmented voyage distances must be equal to the total voyage distance; constraint 2 means that the voyage time must be less than the planned time; constraint 3 means that the departure time of the ship must be earlier than the planned time; constraint 4 means that the ship's speed during the voyage cannot be greater than or less than the ship's maximum or minimum speed, so as to ensure that the target ship maintains a good voyage state during the voyage; constraint 5 means that the speed change of each segment cannot exceed 20%, because the speed change of the segment must conform to the driving habits of the captain and other operators, and excessive speed change will lead to a significant increase in main engine fuel consumption. In the constraint optimization calculation, it is assumed that the speed of each segment remains unchanged, the encountered meteorological data remains unchanged, and since the heading angle of the target ship does not change much within the segment, it is assumed that the heading angle of the ship's voyage within the segment remains unchanged in the optimization calculation. In addition, the historical meteorological database measures and records ocean currents at the same latitude and longitude at 24-hour intervals, and wind and waves at 6-hour intervals. Therefore, when the target ship is in Arrive at the The location point needs to be obtained. When collecting meteorological data for a flight segment, use Distance after time The most recent meteorological data as the first Meteorological data for the flight segment.

[0034] S4.2 Speed ​​Optimization Based on Genetic Algorithm Based on the construction of fuel consumption and speed optimization models, a genetic algorithm is used as the solution algorithm to optimize the speed of a segment of each sub-segment of the target ship's route from Chongqing to Yichang. The genetic algorithm is a computational model designed based on the laws of biological evolution, simulating natural selection and genetic evolution in the process of biological evolution. It is an adaptive stochastic search method widely used in optimization calculations across various industries. The genetic algorithm process is as follows: Figure 5 As shown.

[0035] To accelerate the algorithm's search for the optimal solution during population initialization, the speed range was narrowed based on the maximum and minimum speed limits for each segment, using 9–10 kN as the initial population value range. Global speed optimization was performed with a population size of 50, a maximum of 1000 iterations, a crossover rate of 0.8, a mutation rate of 0.05, and an elite count of 0.05 times the population size. To overcome the large calculation errors of integer models and the slow optimization due to the high precision of simple genetic algorithms, and considering practical engineering applications, a specific precision constraint was added to the original objective function to perform global optimization of the flight path to one decimal place. The modified target rainfall number is as follows: (11) (12) in, The constraint is integer, and ∈[60, 1301, variable substitution in the iterative operation The values ​​were adjusted to the precision required in this paper. Convergence was calculated after 895 iterations.

[0036] As shown in Table 3, the total fuel consumption after iterative optimization is 15.86t, while the actual fuel consumption of the target ship before optimization is 16.24t. The optimization results indicate that when sailing at the optimized speed, the target ship can save 0.38t of fuel, resulting in a total fuel consumption reduction of 2.34%.

[0037] Table 3 Comparison of data before and after optimization

[0038] This method utilizes a backpropagation neural network to establish a fuel consumption model for the target vessel. The model's average absolute error is 2.3%, and its accuracy and generalization ability basically meet the requirements for engineering applications. Secondly, an optimization model with a minimum total fuel consumption along the route and fixed-precision nonlinear constraints is constructed. A genetic algorithm is then used, and historical meteorological data is used to optimize the target vessel's route speed in segments. The calculation results show that the target vessel's fuel consumption is reduced by 2.34% after speed optimization, indicating that segmented speed optimization for navigation vessels is a feasible solution.

[0039] Based on the principle of energy saving through speed reduction, determining the optimal energy-saving speed for ship navigation—that is, making a dynamic balance between the expected profit and expected loss of the objective function—has significant economic and environmental implications. This paper studies and optimizes the speed of a target ship under fixed-precision nonlinear constraints. The results show that the method of segmenting the ship and optimizing its speed is feasible, which can not only reduce carbon emissions but also, to some extent, reduce the total voyage time.

Claims

1. A method for evaluating optimal speed based on neural networks, characterized in that, A method for obtaining optimal speed assessment is adopted by combining a fuel consumption prediction model for luxury inland waterway tourist vessels with a speed optimization model based on a genetic algorithm. A fuel consumption prediction model based on a backpropagation neural network is established using actual ship navigation data. Based on the fuel consumption model and the segment division, a nonlinear programming speed optimization model is constructed. Based on the construction of the fuel consumption model and the speed optimization model, a genetic algorithm is used as the solution algorithm to optimize the speed of the target vessel from a certain segment of the route, and obtain high-precision prediction results.

2. The optimal speed evaluation method based on neural networks according to claim 1, characterized in that, The specific steps of this method are as follows: S1. Data Acquisition and Preprocessing: The data acquisition and preprocessing system is designed with a two-layer network structure, namely the perception layer and the decision layer. The perception layer is completed by distributed data acquisition units to collect data such as cabin equipment and ship navigation parameters; the decision layer is completed by the system server to realize real-time processing of the collected data, as well as the collection of deep learning training data in the early stage and the storage of algorithm optimization data in the later stage. The perception layer and the decision layer achieve high-speed data communication through an industrial Ethernet star connection; S2. Fuel Consumption Prediction Model Construction: The fuel consumption prediction model is divided into three categories: white-box model, black-box model and gray-box model of ship fuel consumption; among them, the structure and parameters of the white-box model of ship fuel consumption are known, and it is established based on the relevant knowledge in the ship field and the interaction relationship between ship-engine-propeller; the structure and parameters of the black-box model of ship fuel consumption are unknown, and it is established based on training with massive historical operating data; the structure and parameters of the black-box model of ship fuel consumption are unknown, and it is established based on training with massive historical operating data; The gray box model is between the white box model and the black box model of ship fuel consumption. Some of its physical feature data can be directly measured, while other features that are difficult to obtain are obtained through data training. S3, Target ship fuel consumption model construction: Combine the massive historical monitoring data of the real ship operation monitoring platform, and use the backpropagation neural network to construct the black box model of ship fuel consumption for the target ship. S4, Speed ​​optimization: (1) Speed ​​optimization assumptions and segmented speed optimization model construction: By performing speed segmentation and rapid optimization on different segments within the target ship's fixed route, the total fuel consumption of the route objective function is minimized. Based on the above fuel consumption model and segment division, a nonlinear programming optimization model is constructed. (2) Speed ​​optimization based on genetic algorithm: Based on the fuel consumption model construction and speed optimization model construction, the genetic algorithm is used as the solution algorithm to optimize the speed of one segment of the navigation distance of each sub-segment of the target ship's route from Chongqing to Yichang.

3. The optimal speed evaluation method based on neural networks according to claim 2, characterized in that, In step S1, the sensing layer consists of a signal acquisition box, fuel flow meters arranged on the generator inlet and return oil pipelines, a shaft power tester installed on the shafting, an energy metering device embedded in the ship's main switchboard, a ship engine room monitoring and alarm system (not part of this system), the main switchboard system, a ship draft sensor, and ship communication and navigation equipment. The signal acquisition box contains a PLC module to acquire data from Ethernet signals, serial port signals, AI signals, and DI signals, meeting the system's signal acquisition requirements. Fuel flow meters are arranged on the inlet and return oil pipelines of fuel-consuming equipment. Combined with fuel refueling information and fuel level data from the ship's fuel tanks, they drive the fuel statistics algorithm module to obtain accurate fuel consumption data for ship navigation and docking. The magnetic strip sensed by the shaft power meter is attached to the ship's shafting. Combined with the high-precision sensing sensor of the shaft power meter, the power, speed, and torque signals of the ship's shafting are obtained through static verification and dynamic verification during ship navigation. The power metering device embedded in the ship's main switchboard, combined with the current transformer mounted on the copper busbar of the main switchboard, statistically analyzes the input voltage and current signals, and then converts them into active power, reactive power, and power factor parameters. This is used to calculate the total power generation of the ship's generator sets and the specific power consumption of each branch circuit equipment. The engine room monitoring and alarm system provides the operating status and alarm signals of the ship's engine room equipment during navigation, as well as information on ship refueling and specific fuel level data in the fuel tanks during navigation. Combined with the tank capacity model, it calculates the specific fuel quantity of the ship at different times, assisting the flow meter in accurately calculating fuel consumption. The main switchboard system is used for the distribution of electric propulsion power and monitoring the operating status of propulsion equipment during navigation. Combined with the power station management system, it provides a safe, reliable, high-quality, and economical intelligent power supply management system.

4. The optimal speed evaluation method based on neural networks according to claim 3, characterized in that, The power station management system also has functions of power generation management, load management, and power distribution management; the ship's draft sensor reflects the ship's draft and trim status under the current river conditions; the wind speed and direction instrument, log, depth sounder, and GPS mounted on the ship's end reflect the wind speed and direction, ship speed, current water depth of the Yangtze River, ship's position, and ship speed signals during the ship's navigation process, respectively.

5. The optimal speed evaluation method based on neural networks according to claim 3, characterized in that, The serial port module, AI module, and DI module inside the signal acquisition box interact with sensors within the system and sensors outside the system. The data interface of the acquisition module has a built-in noise filter combined with a Kalman filter algorithm module to ensure the authenticity of the data. The logic control program of the acquisition module is designed with a network disconnection reconnection mechanism to ensure the reliability of data communication. The data interaction cycle between various systems and devices reaches the level of hundreds of milliseconds, which improves the response efficiency of the entire system from the perception level.

6. The optimal speed evaluation method based on neural networks according to claim 2, characterized in that, In step S2, the white-box model structure and parameters of ship fuel consumption include the ship's main engine obtaining power through burning fuel, the power being transmitted through the gearbox and main engine shafting equipment to the propeller, and the propeller interacting with the water flow to propel the ship forward; it also needs to consider the influence of wind speed, wind direction, current speed, current direction, and ship draft on the ship's state, as well as the influence of downstream and upstream currents on the main engine fuel consumption; the black-box model training algorithm for ship fuel consumption adopts the extra-tree regression algorithm, Naive Bayes algorithm, human neural network, support vector machine supervised algorithm, linear regression learning algorithm, or random forest regression; when constructing the gray-box model of ship fuel consumption, the black part is segmented out by mathematical methods, and then the mathematical model of this unobservable part is inferred using partial data.

7. The optimal speed evaluation method based on neural networks according to claim 2, characterized in that, In step S3, the backpropagation neural network is a multi-layer feedforward network, including an input layer, a hidden layer, and an output layer. This backpropagation neural network can be supervised and has the characteristics of signal forward propagation and error backpropagation. The computational learning process of the backpropagation neural network is divided into two stages, including signal forward propagation and error backpropagation. During the backpropagation process, the weights and biases from the hidden layer to the output layer and from the input layer to the hidden layer are optimized sequentially.

8. The optimal speed evaluation method based on neural networks according to claim 7, characterized in that, The specific calculations of a backpropagation neural network include: the total number of input layers. There are 10 node variables, and the hidden layer has a total of 10 nodes. There are [number] nodes in the output layer. indivual, For the first The input value of each node, For the first hidden layer The node and the first input layer The weights between nodes For the hidden layer The threshold of each node, Let be the activation function of the hidden layer. For the output layer The node and the hidden layer The weights between nodes Then it is the output layer. The threshold of each node, The activation function of the output layer. The output of the final output layer nodes; during the forward propagation of the neural network signal, the output value of the output layer... The calculation formula is as follows: （1） Backpropagation of neural network errors will adjust and optimize the weights and thresholds of each layer. The error correction of the prediction result is as follows: （2） The corrected formulas for calculating the weights and thresholds of each node are as follows: （3） （4） （5） （6） Because black-box models are highly dependent on datasets, when selecting target vessels, vessels with smaller errors between fuel consumption flow meter measurement data and manual measurement data are chosen to ensure the validity of the flow meter sampling data.

9. The optimal speed evaluation method based on neural networks according to claim 2, characterized in that, In step S4, a nonlinear programming optimization model is constructed, with the objective function as follows: （9） Constraints: （10） In the formula: For the first Ground speed for each segment of flight For Japanese standard vessels The first draft of each segment of the navigation route. For the target vessel The tail draft of each segment of the navigation route. For the target vessel The heading angle for each segment of the flight path, For the first Wind that divides the navigation segment Quantity, For the first Wind that divides the navigation segment Quantity, For the first The meaningful wave height for dividing the navigation segment For the first The effective wave period for dividing flight segments For the first The main wave angle for dividing the navigation segment For the first Flow divided into segments Quantity, For the first Flow divided into segments Quantity, For the first The distance of each segment, To optimize ship departure time The latest time for the ship to depart. This represents the expected duration of the ship's total voyage time. , For the minimum and maximum ground speed limits, in Equation (10), constraint 1 means that the sum of the segmented voyages must be equal to the total voyage distance, constraint 2 means that the voyage time must be less than the planned time, constraint 3 means that the departure time of the ship must be earlier than the planned time, and constraint 4 means that the speed of the ship during the voyage cannot be greater than or less than the maximum or minimum speed of the ship.

10. The optimal speed evaluation method based on neural networks according to claim 9, characterized in that, By adding a specific precision constraint to the original objective function to perform global optimization of the flight path to one decimal place, the modified objective rainfall number is as follows: （11） （12） in, The constraint is integer, and ∈[60, 1301, variable substitution in the iterative operation The values ​​were adjusted to the required precision for this paper, and convergence was achieved after 895 iterations.

Images