A ship structure safety prediction method based on artificial neural network

Abstract

The application discloses a ship structure safety prediction method based on an artificial neural network, historical data collection; the collected historical data are processed, time sequence alignment is performed on multi-source heterogeneous historical data, and the processed historical data are normalized; an artificial neural network model is constructed; the artificial neural network model is trained by using the historical data; wave, speed and draft data collected in real time are input into the trained artificial neural network model, the artificial neural network model obtains stress results of a ship structure, rain flow counting is used to extract stress cycles for fatigue cumulative damage calculation, and safety state evaluation is performed according to the stress and fatigue cumulative damage. The application realizes real-time dynamic prediction of ship structure stress and fatigue life.

Description

TECHNICAL FIELD

[0001] The application relates to a safety prediction method, in particular to a ship structure safety prediction method based on an artificial neural network, and belongs to the technical field of intelligent ship navigation safety. BACKGROUND

[0002] In recent years, the shipping industry has developed rapidly, and the trend of large-scale and high-speed of ships is becoming more and more obvious. In the whole process of ship operation, the safety of ship structure is always the core element to ensure the safety of navigation, smooth transportation of goods and safety of personnel and property. The ocean environment is complex and changeable, and various factors such as sea waves, sea currents and weather conditions are intertwined, which have a significant impact on the ship structure. The traditional ship structure stress monitoring and fatigue life evaluation method mainly relies on finite element analysis, empirical formula and periodic inspection, and cannot realize real-time dynamic prediction and early warning. When facing the complex and changeable ocean environment, many limitations are exposed, and these methods cannot fully consider various complex factors and their dynamic changes, cannot fully dig and utilize a large amount of historical data for deep analysis and learning, and lead to a relatively simple understanding of the safety state of the ship structure. In addition, there are obvious deficiencies in the integration and use of real-time sea state data, future weather and sea state prediction data, so that the evaluation results lack both timeliness and foresight. With the development of artificial intelligence technology, artificial neural networks have shown great ability in nonlinear data processing. Therefore, combining neural network algorithm to realize real-time dynamic prediction of ship structure stress and fatigue life has important engineering application value. Developing a ship structure safety dynamic prediction system integrating multi-source data and neural network technology has become a key problem to be solved in the shipping field, and has extremely important practical significance and application value. SUMMARY

[0003] The technical problem to be solved by the application is to provide a ship structure safety prediction method based on an artificial neural network, and to realize real-time dynamic prediction of ship structure stress and fatigue life.

[0004] To solve the above technical problems, the technical scheme adopted by the application is:

[0005] A ship structure safety prediction method based on an artificial neural network comprises the following steps:

[0006] S1, historical data acquisition;

[0007] S2, processing the collected historical data, time sequence alignment of multi-source heterogeneous historical data, and normalization of the processed historical data;

[0008] S3, constructing an artificial neural network model;

[0009] S4, training the artificial neural network model by using historical data;

[0010] S5, inputting the real-time collected wave, speed, and draft data into the trained artificial neural network model, obtaining the stress result of the ship structure by the artificial neural network model, extracting stress cycles by using the rainflow counting method to perform fatigue cumulative damage calculation, and performing safety state evaluation according to the stress and fatigue cumulative damage.

[0011] Further, the step S1 is specifically collecting the stress of the key area of the ship body through the stress sensor of the key part of the ship, collecting the sailing speed and heading through the ship navigation system, collecting the ship loading information through the loading instrument, and collecting the wave information through the wave radar, wherein the wave information includes the wave height, the wave period, and the wave direction.

[0012] Further, the processing of the collected historical data in the step S2 is specifically preliminary screening of the collected historical data to eliminate obviously erroneous or abnormal data points, data cleaning of the wave information, the ship loading information, and the stress collected by the wave radar, the loading instrument, and the stress sensor to remove outliers and noises, and supplementing the data with missing values by using the interpolation method.

[0013] Further, the time sequence alignment of the multi-source heterogeneous historical data in the step S2 is specifically that the collected historical data is multi-source heterogeneous data, the multi-source heterogeneous data is counted to the time dimension of seconds, the sailing speed and heading collected by the navigation system of the ship and the ship loading information collected by the loading instrument are second-level data and do not need to be processed, the stress collected by the stress sensor is millisecond-level, the redundant data is removed and the second-level data is retained, and the wave information collected by the wave radar is minute-level data, and the processing method is,

[0014] the wave height obeys Rayleigh distribution ,

[0015] wherein h is the wave height, is the effective wave height per minute output by the wave radar, and e is a natural constant,

[0016] by inverse transform sampling method, the second-level wave height generation formula is wherein ;

[0017] the fluctuation of the wave direction in a short time obeys uniform distribution , , is the wave direction angle at a moment, is a random disturbance angle;

[0018] the wave period changes little in 1 minute, and the second-level wave period is set as T is the minute scale wave period.

[0019] Further, the step S2 is specifically to normalize the processed historical data by using the formula , wherein X is the historical data, X min is the minimum value in the historical data, and X max is the maximum value in the historical data.

[0020] Further, the step S3 is specifically to construct a BP neural network as a basic prediction model, the basic prediction model comprising an input layer, a hidden layer and an output layer, the number of neurons in the input layer being consistent with the number of features of the historical data, the hidden layer being provided with one or more layers, the number of neurons in the hidden layer being determined through experimental debugging, the activation function being ReLU or sigmoid function, and the number of neurons in the output layer being 1 and outputting the ship structure stress.

[0021] Further, the step S4 is specifically to train the artificial neural network model by using the collected historical data, divide the historical data into 80% training set and 20% validation set, monitor the loss function and accuracy of the artificial neural network model in real time during the training process, dynamically adjust the parameters of the artificial neural network model according to the training result, and gradually reduce the loss function of the artificial neural network model until it reaches the minimum.

[0022] Further, the step S5 is specifically to input the real-time collected and processed wave information, speed, heading and ship loading information into the trained artificial neural network model, rapidly obtain the ship structure stress by the artificial neural network model, extract the stress cycle by using the rainflow counting method, perform the fatigue cumulative damage calculation, and evaluate the safety state according to the predicted ship structure stress and fatigue cumulative damage in real time, or predict the stress and fatigue life of the key area of the ship in the next 24 hours in combination with the future sea wave information of the weather forecast, judge the prediction result of the safety state according to the preset safety threshold, continue to maintain real-time monitoring if the prediction result shows that the safety state is safe, and immediately issue an alarm to the crew through sound and light alarm to remind the crew to take effective measures in time if the prediction result reaches the warning or dangerous state.

[0023] Compared with the prior art, the present application has the following advantages and effects: the present application provides a ship structure safety prediction method based on artificial neural network, comprehensively considers various complex sea condition factors and their dynamic changes in the ship operation process, uses advanced artificial neural network algorithm to fully utilize and mine a large amount of historical data, performs deep analysis and learning, and integrates sea condition data and ship data in real time, so that the ship structure safety evaluation result has timeliness, in addition, future weather and sea condition prediction data are accessed, so that the ship structure safety evaluation result has foresight, the safety of the ship structure in actual sea condition operation is ensured, and the service life of the ship can be greatly prolonged; the ship historical data is processed, and the unified multi-source heterogeneous data is realized, so that the multi-source heterogeneous ship data can be trained and predicted by the neural network model. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of a ship structure safety prediction method based on artificial neural network of the present application.

[0025] Figure 2 is a schematic diagram of the present application for statistical multi-source heterogeneous data to time dimension of seconds.

[0026] Figure 3 is a structure schematic diagram of the artificial neural network model of the present application.

[0027] Figure 4 is a flowchart of real-time stress and fatigue damage prediction of the present application.

[0028] Figure 5 is a flowchart of 24-hour future stress and fatigue damage prediction of the present application. DETAILED DESCRIPTION

[0029] In order to clearly and completely describe the technical solutions adopted by the present application to achieve the predetermined technical purposes, the technical solutions in the embodiments of the present application will be described in detail 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, not all embodiments, and the technical means or technical features in the embodiments of the present application can be replaced without creative labor, which will be described in detail below with reference to the drawings and in combination with the embodiments.

[0030] As shown in Figure 1 , a ship structure safety prediction method based on artificial neural network of the present application comprises the following steps:

[0031] S1, historical data acquisition.

[0032] The multi-source historical data is collected by sensors on the ship for neural network training or for real-time prediction. Specifically, the stress of the key area of the ship body is collected by the stress sensor on the key part of the ship, the sailing speed and heading are collected by the ship navigation system, the ship loading information is collected by the loading instrument, and the wave information including wave height, wave period and wave direction is collected by the wave radar. These information can be used for the training of the neural system and real-time prediction.

[0033] Meanwhile, the wave information of the future 24 hours weather forecast provided by the professional meteorological forecasting agency is also accessed, including wave height, wave period, wave direction and other data. These data are the basis for subsequent neural network model analysis and prediction, and their accuracy and integrity directly affect the reliability of the safety prediction.

[0034] S2, processing the collected historical data, time aligning the multi-source heterogeneous historical data, and normalizing the processed historical data.

[0035] The collected historical data is preliminarily screened to eliminate obviously erroneous or abnormal data points, and the wave information, ship loading information and stress collected by the wave radar, loading instrument and stress sensor are cleaned to remove outliers and noise, for example, by setting reasonable wave height, draft and stress range, the data points obviously exceeding the normal range are filtered out. For wave height data, if the measured wave height at a certain time is far beyond the historical wave height data of the sea area, and the difference with the wave height data of the surrounding time is too large, it can be determined as an outlier and eliminated. For data with missing values, interpolation method is used for supplement, so as to ensure the integrity and accuracy of the data. Then, advanced filtering algorithm is used to remove noise interference.

[0036] The collected historical data is multi-source heterogeneous data, and the frequency is not synchronized. In order to train the model in the same time dimension, the multi-source heterogeneous ship data needs to be time aligned. As shown in Figure 2 The multi-source heterogeneous data is counted in the time dimension of seconds. The sailing speed, heading collected by the ship navigation system and the ship loading information collected by the loading instrument are second-level data, which need not be processed. The stress collected by the stress sensor is millisecond-level, and the redundant data is removed to retain the second-level data. The wave information collected by the wave radar is minute-level data, and its processing method is

[0037] Wave height obeys Rayleigh distribution ,

[0038] Wherein, h is the wave height, is the effective wave height output by the wave radar every minute, e is the natural constant,

[0039] The formula for generating the wave height in seconds is where The significant wave height is defined as the average of the largest one-third of the wave heights.

[0040] The fluctuation of the wave direction in a short time is subject to a uniform distribution is the wave direction angle at a certain time, is a random disturbance angle.

[0041] The wave period changes very little in one minute, and the wave period in seconds is set as , and T is the wave period in minutes.

[0042] The historical data is normalized by the formula where X is the historical data, X min is the minimum value in the historical data, and X max is the maximum value in the historical data.

[0043] S3, constructing an artificial neural network model.

[0044] As shown in Figure 3 , a BP neural network is constructed as a basic prediction model, which includes an input layer, a hidden layer, and an output layer. The number of input layer neurons is consistent with the number of historical data features. The input data features include wave data (wave height, wave period, wave direction), ship loading information (i.e., draft), speed, and stress. One or more hidden layers are set, and the number of hidden layer neurons is determined through experimental debugging. For example, the preset number of input layers is , is the number of input nodes, is the number of output nodes, . The activation function uses ReLU or sigmoid function to realize nonlinear transformation of the input data. The number of output layer neurons is 1 and outputs the ship structure stress.

[0045] S4, training the artificial neural network model using historical data.

[0046] The collected historical data is used to train the artificial neural network model, and the historical data is divided into 80% training set and 20% validation set. In the training process, the loss function and accuracy of the artificial neural network model are monitored in real time, and the parameters of the artificial neural network model are dynamically adjusted according to the training results, so that the loss function of the artificial neural network model gradually decreases until it reaches the minimum.

[0047] ​​S5, input the real-time collected wave, speed, draft data into the trained artificial neural network model, the artificial neural network model obtains the stress result of the ship structure, and uses the rain flow counting method to extract the stress cycle to perform fatigue cumulative damage calculation, and the safety state is evaluated according to the stress and fatigue cumulative damage.

[0048] As shown in Figure 4 , the wave information, speed, heading and ship loading information processed in real time are input into the trained artificial neural network model, the artificial neural network model quickly calculates the ship structure stress and uses the rain flow counting method to extract the stress cycle, performs fatigue cumulative damage calculation, and the system evaluates the safety state in real time according to the predicted ship structure stress and fatigue cumulative damage, and the system judges the prediction result of the safety state according to the preset safety threshold, if the prediction result shows that the safety state is safe, the real-time monitoring is continued, if the prediction result reaches the warning or dangerous state, the system immediately alarms the crew through sound and light to remind the crew to take effective measures in time.

[0049] As shown in Figure 5 , combined with future sea wave information, the artificial neural network model can predict the ship key area stress and fatigue life in the next 24 hours, and provide trend analysis of ship navigation safety risk in the future. The structure safety prediction of the future 24-hour scheduled route is evaluated, if the weather has a great impact on the structure strength, the route optimization can be performed to select the optimal route scheme with the minimum stress.

[0050] The application provides a ship structure safety prediction method based on an artificial neural network, comprehensively considers various complex sea state factors and their dynamic changes in the ship operation process, uses an advanced artificial neural network algorithm to fully utilize and mine a large amount of historical data, performs deep analysis and learning, and integrates sea state data and ship data in real time, so that the ship structure safety evaluation result is timely, and in addition, the future meteorological and sea state prediction data are accessed, so that the ship structure safety evaluation result is forward-looking, the safety of the ship structure in the actual sea state operation is ensured, and the service life of the ship can be greatly prolonged; the ship historical data are processed, and the unified multi-source heterogeneous data are realized, so that the multi-source heterogeneous ship data can be trained and predicted by the neural network model.

[0051] The above merely describes preferred embodiments of the present application, and is not intended to limit the present application in any form. Although the present application has been disclosed with preferred embodiments as above, it is not intended to limit the present application. Any person skilled in the art, without departing from the technical solution of the present application, can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes. However, as long as it does not deviate from the technical solution of the present application, and is within the spirit and principle of the present application, any simple modification, equivalent replacement and improvement of the above embodiments are still within the protection scope of the technical solution of the present application.

Claims

1. A method for predicting ship structural safety based on artificial neural networks, characterized in that... Includes the following steps: S1. Historical data collection; Specifically, step S1 involves collecting stress in key areas of the hull using stress sensors on key parts of the ship, collecting navigation speed and heading through the ship navigation system, collecting ship loading information through the loading instrument, and collecting wave information through wave radar, including wave height, wave period, and wave direction. S2. Process the collected historical data, perform time-series alignment on multi-source heterogeneous historical data, and normalize the processed historical data. The time-series alignment of the multi-source heterogeneous historical data in step S2 specifically involves: The collected historical data is multi-source heterogeneous data; this data is statistically analyzed to a time dimension measured in seconds. The navigation system's collected speed and heading data, and the loader's collected loading information, are second-level data and require no processing. The stress sensor's collected stress data is in milliseconds; redundant data is removed, retaining only second-level data. The wave radar's collected wave information is in minutes-level data, and its processing method is as follows... Wave height follows a Rayleigh distribution , Where h is the wave height. Let be the effective wave height per minute output by the wave radar, and e be the natural constant. The formula for generating second-level wave height using the inverse transform sampling method is as follows: ,in ; The wave direction fluctuates uniformly over a short period of time. , , The wave angle at a certain moment, For random perturbation angles; The wave cycle changes very little within 1 minute; the wave cycle is set to a second-level value. T represents the wave cycle on a minute scale; S3. Construct an artificial neural network model; Specifically, step S3 involves constructing a BP neural network as a basic forecast model. The basic forecast model includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is consistent with the number of features in the historical data. One or more hidden layers are set, and the number of neurons in the hidden layer is determined through experimental debugging. The activation function is ReLU or sigmoid. The number of neurons in the output layer is 1, and it outputs the ship structure stress. S4. Train the artificial neural network model using historical data; S5. Input the real-time collected wave, speed, and draft data into the trained artificial neural network model. The artificial neural network model obtains the stress results of the ship structure and uses the rainflow counting method to extract stress cycles for fatigue cumulative damage calculation. Based on the stress and fatigue cumulative damage, a safety status assessment is performed.

2. The ship structural safety prediction method based on artificial neural networks according to claim 1, characterized in that: The specific steps for processing the collected historical data in step S2 are as follows: the collected historical data is initially screened to remove obviously erroneous or abnormal data points; the wave information, ship loading information and stress collected by wave radar, loading instrument and stress sensor are cleaned to remove outliers and noise; and for data with missing values, interpolation is used to supplement them.

3. The ship structural safety prediction method based on artificial neural networks according to claim 1, characterized in that: The normalization of the processed historical data in step S2 specifically involves applying a formula to the historical data. Normalization, where X represents historical data, X min X is the minimum value in the historical data. max This represents the maximum value from historical data.

4. The ship structural safety prediction method based on artificial neural networks according to claim 1, characterized in that: Specifically, step S4 involves training the artificial neural network model using collected historical data, dividing the historical data into an 80% training set and a 20% validation set. During the training process, the loss function and accuracy of the artificial neural network model are monitored in real time, and the parameters of the artificial neural network model are dynamically adjusted based on the training results, so that the loss function of the artificial neural network model is gradually reduced until it is minimized.

5. The ship structural safety prediction method based on artificial neural networks according to claim 1, characterized in that: Specifically, step S5 involves inputting the real-time collected and processed wave information, speed, heading, and ship loading information into a trained artificial neural network model. The artificial neural network model quickly calculates the ship's structural stress and uses the rainflow counting method to extract stress cycles and perform fatigue cumulative damage calculation. The system performs a real-time safety status assessment based on the predicted ship structural stress and fatigue cumulative damage, or combines the future wave information from the weather forecast to predict the stress and fatigue life of key areas of the ship in the next 24 hours. The system judges the prediction results of the safety status based on the preset safety threshold. If the prediction results show that the ship is in a safe state, real-time monitoring continues. If the prediction results reach the warning or dangerous state, the system immediately issues an alarm to the crew through audible and visual alarms to remind the crew to take effective measures in a timely manner.

Images