Electric bulk carrier energy consumption prediction method and system based on multi-source data

Through multi-source data preprocessing and model training, an energy consumption prediction system was established, which solved the problems of energy consumption prediction and control of pure electric bulk carriers, and improved the accuracy and endurance of energy consumption prediction.

CN118982098AInactive Publication Date: 2024-11-19THREE GORGES ELECTRIC POWER JINZHOU ENERGY (HUBEI) CO LTD

Patent Information

Application Number
CN202410989444.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-19
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

How to predict and control energy consumption in pure electric bulk carriers during freight due to various factors is a problem that needs to be solved, especially when the installed capacity of ship energy batteries is limited, the expansion of continuous endurance is limited.

Method used

Using the energy consumption prediction method of electric bulk carriers based on multi-source data, the state data and environmental information are collected outside the ship are monitored and collected, and the data preprocessing and anomaly analysis are used for data preprocessing and anomaly analysis are established to establish a Stacking comprehensive model for energy consumption prediction. , combined with gradient enhancement decision tree algorithm, Ridge Ridge regression model and SVM algorithm for model training.

Benefits of technology

It realizes accurate prediction and analysis of the energy consumption of electric bulk carriers, improves the rationality and accuracy of the energy consumption prediction of pure electric bulk carriers, provides a foundation for transportation control, and enhances endurance and safety.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982098A_ABST
    Figure CN118982098A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of ship energy consumption prediction methods, and particularly relates to an electric bulk cargo ship energy consumption prediction method and system based on multi-source data. Comprising the steps of original data collection, original data preprocessing, data restoration processing, execution of multi-source data anomaly analysis and correction, energy consumption prediction composite model training, energy consumption prediction and the like. The scheme of predicting and analyzing the operation energy consumption of the electric bulk cargo ship based on the multi-source data in the operation process is realized, a technical basis is provided for constructing continuous energy consumption monitoring and management of the whole ship, and the reasonability and accuracy of energy consumption prediction of the pure electric bulk cargo ship are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of ship energy consumption prediction methods, and in particular relates to an electric bulk carrier energy consumption prediction method and system based on multi-source data. Background Art

[0002] With the advancement of sustainable energy development, China has launched a wide range of pure electric equipment, among which pure electric bulk carriers are a key development direction of inland cargo transportation. Pure electric bulk carriers have eliminated their dependence on traditional fossil energy and achieved zero emissions. They can effectively solve the water pollution and air pollution problems of inland rivers caused by ship navigation. Compared with traditional ships, they also have higher transportation efficiency under the same energy consumption. Relying on pure electric power and control systems, their control capabilities are also better than traditional ships. One of the current problems limiting the promotion and application of pure electric bulk carriers is the further expansion of their sustainable endurance. For pure electric bulk carriers, the energy consumption required for the ship's own navigation and various operations on board mainly comes from the limited storage capacity in the ship. In the case of limited installed capacity of ship energy batteries, how to predict and control the energy consumption of bulk carriers under the influence of various factors during cargo transportation is a problem that needs to be solved. Summary of the invention

[0003] The purpose of the present invention is to provide an electric bulk carrier energy consumption prediction method and system based on multi-source data based on the demand for energy consumption control of pure electric bulk carriers, which is used to construct and improve the energy consumption prediction task of pure electric bulk carriers, thereby providing a basis for the transportation control and energy consumption analysis of pure electric bulk carriers.

[0004] To achieve the above object, the present invention adopts the following technical solution.

[0005] A method for predicting energy consumption of an electric bulk carrier based on multi-source data comprises the following steps:

[0006] S1. Raw data collection

[0007] The status data and operating parameters of the bulk carrier during operation are collected based on the onboard monitoring and collection device; the hydrological environment and weather environment information inside and outside the bulk carrier collection period are obtained based on the off-board monitoring and collection component;

[0008] S2. Preprocessing of raw data: Preprocessing of the collected raw data includes:

[0009] S20, data repair processing, specifically: for empty data, data filling is performed by fitting or interpolation according to other data in the data group where the data is located; for abnormal data values ​​or data fragments that deviate from the reasonable range of the original data, the abnormal data should be deleted and regarded as empty data for filling;

[0010] S21, perform multi-source data anomaly analysis, for a group of multi-source raw data with an associated period t, select continuous data within a time period T=kt, k is an integer, to form a multi-source raw data set X T =x1,x2...x t ...x T};in represents the multi-source data value in the t-th data collection period within the time period T of the multi-source data set, where xtn represents the value of the n-th original data within xt; the sliding time window t' is used to convert the multi-source original data set X T Split to get the training set X A ={x t‘} and test set X B ={x t+1‘};in k' is a positive integer less than k, x t‘ Represents a set of measurement values ​​within the time period t'; constructs a DWGAT graph attention network using the aforementioned multi-source raw data as graph structure network node elements;

[0011] For two sets of original data with known correlation, the correlation embedding vectors between the corresponding nodes are created based on their prior information. For two sets of original data with unknown correlation, random embedding vectors are created. Each graph node is used as the initial node to perform random walks according to the preset number of steps. A node sequence pool is created based on the obtained node sequence to reflect the DWGAT graph attention network structure. The Word2Vec training model is used to create and optimize the missing embedding vectors between the original data to obtain the DWGAT graph attention network model.

[0012] The training set X consisting of multi-source data in time period t' A ={x t‘ Input the DWGAT graph attention network model obtained in step S215 to obtain the predicted value of the measured data in the next time period t+1', and use the test set X composed of the measured data in the next time period t+1' B ={x t+1‘ Verify the model’s predicted output, determine the model error, and adjust the DWGAT graph attention network model parameters until the model training results meet the standards;

[0013] S22. Error Correction

[0014] The trained DWGAT graph attention network model is used to verify the original data from multiple sources, and the measured values ​​whose errors between the measured values ​​and the model prediction values ​​exceed the threshold are verified and corrected, and the original data are integrated;

[0015] S3. Energy consumption prediction composite model training

[0016] Based on the Stacking comprehensive model, a comprehensive model for predicting energy consumption of bulk carriers is established. The historical data is collected and processed based on steps setp1 to S3 to create a prediction model training database to train the model and obtain the optimal comprehensive model for predicting energy consumption of bulk carriers.

[0017] S5. Energy consumption prediction

[0018] The integrated raw data is input into the comprehensive model for predicting energy consumption of bulk carriers to obtain the prediction results.

[0019] For the further improvement or preferred implementation of the above-mentioned electric bulk carrier energy consumption prediction method based on multi-source data, step S21 also includes time synchronization processing based on the acquisition cycle, taking the acquisition cycle of a specific original data as the standard cycle, and performing time series synchronization on other data, specifically:

[0020] For the original data of time series continuity whose acquisition cycle time scale does not match the standard cycle, splice and cut them to make their acquisition cycle consistent with the standard cycle;

[0021] For the intermittently collected periodic raw data, the periodic data is copied or completed to make it have time sequence continuity and processed based on step S20;

[0022] Homogeneous data with the same or similar time series change trends due to the correlation between the data themselves are screened and cleared, and the original data with better time series integrity are selected and other data are deleted.

[0023] A further improvement or preferred implementation method of the aforementioned electric bulk carrier energy consumption prediction method based on multi-source data, a comprehensive model for bulk carrier energy consumption prediction, is constructed with a gradient boosting decision tree algorithm as a meta-model, and a Ridge ridge regression model, a SVM algorithm model, and an extreme tree search model as basic models.

[0024] Further improvement or preferred implementation of the above-mentioned electric bulk carrier energy consumption prediction method based on multi-source data, training the model includes:

[0025] The model training database is divided into training data set and test data set in proportion; the original data in the training set is evenly divided into K basic data sets; the basic data sets are used to train the Ridge regression model, SVM algorithm model, and extreme tree search model. For each model, in each training cycle, one of the basic data sets is used as the validation set, and the remaining data sets are used as the training set to train each model. After obtaining the prediction results of each model, they are compared with the validation set to determine the optimal parameters of each model, and the optimal comprehensive model for bulk carrier energy consumption prediction is obtained.

[0026] An energy consumption prediction system for an electric bulk carrier includes a data acquisition monitoring module, a communication module, and a control center;

[0027] The data acquisition and monitoring module at least includes: an onboard monitoring and acquisition component for monitoring and collecting status data and operating parameters of the power supply, cooler, generator, propeller, and grid load in the electric bulk carrier during operation; an offboard monitoring and acquisition component for monitoring and collecting the operating environment of the electric bulk carrier, and the offboard monitoring component is used to obtain hydrological environment and weather environment information inside and outside the bulk carrier during the acquisition period;

[0028] The communication module includes an Ethernet or RS-485 network for realizing data transmission between the data acquisition monitoring module and the control center;

[0029] The control center includes a distributed processing device composed of at least two CPU units, at least one of which includes a data processing CPU unit for realizing data processing and diagnosis generation and a control output CPU unit for realizing process control and output.

[0030] For further improvement or preferred implementation scheme of the aforementioned electric bulk carrier energy consumption prediction system, the communication module specifically includes: an Ethernet transmission network based on the ModBus-TCP protocol for data transmission between power supplies, generator sets, coolers, excitation control equipment, thrusters, power grid loads and control centers, and an I / O acquisition control unit for connecting the data ports of each device, an RS-485 transmission network for data transmission with a hydrological environment information collection device and a weather environment information collection device, and an I / O acquisition control unit for connecting the data ports of each device; a number of MAC controllers, PHY controllers and MΠ interfaces are provided in the Ethernet transmission network.

[0031] A further improvement or preferred implementation scheme of the aforementioned electric bulk carrier energy consumption prediction system, the onboard monitoring and collection components at least include: a power supply information collection device for obtaining electrical parameters of each group of power supplies as well as output efficiency, temperature, and power supply effectiveness status parameters; a cooler information collection device for obtaining electrical parameters of the cooler controller as well as cooler water inlet and outlet, water temperature cooler operating parameters; a generator information collection device for obtaining electrical parameters of the generator controller and the supporting excitation control equipment as well as generator output efficiency and generator set effectiveness status parameters; a propeller information collection device for obtaining electrical parameters of the propeller controller as well as propeller power and speed status parameters; a generator information collection device for obtaining electrical parameters of the power grid load of the non-energy system in the electric bulk carrier as well as load real-time power and load access status parameters;

[0032] The off-board monitoring and collection components include: a hydrological environment information collection device for obtaining information on ocean currents, tides, internal waves, jump layers, and seawater density outside the electric bulk carrier that may affect the navigation of the ship; and a weather environment information collection device for obtaining wind speed, air pressure, visibility, air humidity, and temperature.

[0033] Its beneficial effects are:

[0034] The electric bulk carrier energy consumption prediction method and system based on multi-source data of the present application realizes a prediction and analysis scheme for the operating energy consumption of the electric bulk carrier based on multi-source data during the operation process, and provides a technical basis for building continuous energy consumption monitoring and management of the entire ship. By creating a data network model for data optimization and processing steps, it is possible to further compress the original data, effectively identify abnormal data based on the mathematical structure relationship between the multi-source original data, retain the correlation between the data, and improve the rationality and accuracy of the energy consumption prediction of pure electric bulk carriers. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is a schematic diagram of the data processing flow in the energy consumption prediction method of electric bulk carrier based on multi-source data;

[0036] Figure 2 Schematic diagram of the model training process in the electric bulk carrier energy consumption prediction method based on multi-source data. DETAILED DESCRIPTION

[0037] The present invention is described in detail below in conjunction with specific embodiments.

[0038] The present invention relates to an energy consumption prediction method for an electric bulk carrier based on multi-source data and a monitoring system thereof. The method is mainly used to realize continuous monitoring and prediction and early warning of energy consumption during continuous navigation of a ship, improve the perception of a pure electric bulk carrier of its own endurance and safety, and provide a basic means for the smooth progress of navigation planning and operation tasks. The method is characterized in that it can effectively process original data, compress and screen invalid data and homogeneous data, improve the validity of ship energy consumption data, and retain the correlation between data at the same time, so that the influence of continuous changes in the continuous navigation process on the overall energy consumption is further manifested, and an effective means is provided for long-term continuous monitoring and early warning.

[0039] The following is a detailed explanation with specific steps.

[0040] In order to achieve effective prediction of energy consumption data, it is necessary to obtain raw data related to ship energy consumption in real time, that is, step S1.

[0041] S1. Raw data collection, used to obtain complete and full energy consumption related continuous monitoring data;

[0042] In order to ensure the high efficiency and stability of data, the status data and operating parameters of the bulk carrier during operation should be collected based on the onboard monitoring and collection devices; the hydrological environment and weather environment information inside and outside the bulk carrier collection period should be obtained based on the off-board monitoring and collection components;

[0043] Accordingly, the monitoring system of the present invention includes a data acquisition monitoring module. Based on the ultimate purpose of the present invention, the necessary raw data is obtained. The data acquisition monitoring module should at least include the following parts:

[0044] A. On-board monitoring and collection components, used to monitor and collect status data and operating parameters of key parts such as power supply, cooler, generator, propeller, grid load, etc. in electric bulk carriers;

[0045] The on-board monitoring and collection components include at least: a power supply information collection device for obtaining electrical parameters of each group of power supplies and status parameters such as output efficiency, temperature, and power supply effectiveness; a cooler information collection device for obtaining electrical parameters of the cooler controller and cooler operating parameters such as cooler water inlet and outlet, water temperature, etc.; a generator information collection device for obtaining electrical parameters of the generator controller and supporting excitation control equipment and status parameters such as generator output efficiency and generator set effectiveness; a thruster information collection device for obtaining electrical parameters of the thruster controller and status parameters such as thruster power and speed; a generator information collection device for obtaining electrical parameters of the power grid load of the non-energy system in the electric bulk carrier and status parameters such as load real-time power and load access status;

[0046] B. Offboard monitoring and collection components, used to monitor and collect operating environment parameters of electric bulk carriers. Offboard monitoring components are used to obtain hydrological and weather environment information inside and outside the bulk carrier collection period;

[0047] The off-board monitoring and collection components include: a hydrological environment information collection device for obtaining information on the current, tide, internal wave, thermocline, seawater density, etc. outside the electric bulk carrier that may affect the navigation of the ship; a weather environment information collection device for obtaining information on wind speed, air pressure, visibility, air humidity, temperature, etc. that may affect the navigation of the ship

[0048] It should be noted that the off-board monitoring components and some on-board monitoring components are often not completely structures or devices belonging to the ship itself or the system itself. For example, some data are obtained based on existing ship monitoring stations, satellites, weather forecast systems, etc. located on land. The corresponding collection devices mentioned in the present invention are only functional statements that the system should include corresponding devices for obtaining the aforementioned data, but they can actually be called from related systems. These called devices are one of the components that constitute the functional structure of the complete monitoring system of the present invention.

[0049] Furthermore, to ensure that the energy consumption monitoring and early warning system can effectively acquire and transmit necessary data information, a complete communication module should be established, including: Ethernet or RS-485 network for data acquisition and monitoring module and control center to realize data transmission;

[0050] As a preferred solution of the present invention, the communication module specifically includes: an Ethernet transmission network based on the ModBus-TCP protocol for data transmission between power supplies, generator sets, coolers, excitation control equipment, thrusters, power grid loads and control centers, and an I / O acquisition control unit for connecting the data ports of each device, wherein the Ethernet transmission network is provided with a plurality of MAC controllers, PHY controllers and MΠ interfaces; and also includes an RS-485 transmission network for data transmission with a hydrological environment information collection device and a weather environment information collection device, and an I / O acquisition control unit for connecting the data ports of each device;

[0051] Since the raw data involves different types of data collected from a large number of different devices and systems, these data may lead to implementation errors due to various reasons. At the same time, since the collection devices corresponding to each raw data often have completely different collection cycles and data structures, in order to ensure the efficiency and stability of subsequent data utilization and processing, the collected raw data needs to be preprocessed as necessary, such as Figure 1 As shown, as necessary contents, the preprocessing in the present invention at least includes data repair processing and data synchronization processing.

[0052] S2. Raw data preprocessing

[0053] S20, data repair processing, refers to the processing of empty data and abnormal data generated in the original data due to various uncontrollable factors during collection and transmission. Specifically: for empty data, data should be filled by fitting, interpolation, etc. according to other data in the data group where the data is located to ensure data integrity; for abnormal data values ​​or data fragments that obviously deviate from the reasonable range of the original data, the abnormal data should be deleted and regarded as empty data for filling;

[0054] The data filling method mainly selects a suitable method to complete the data according to the type and characteristics of various types of original data. In order to realize continuous energy consumption monitoring and prediction, the method and system of the present invention need to consider complete data with time series continuity. In order to ensure the integrity of the data and improve the time series characteristics of the data, numerical fitting or interpolation processing is used to ensure effective data preprocessing. The specific fitting and interpolation algorithms or means can select existing algorithms or means according to the basic structural characteristics of the data or the graphic characteristics of the graphic data. Such algorithms or means are very mature, so the present invention will not describe them in detail.

[0055] On the basis of the aforementioned preliminary processing, we obtain relatively complete full-brightness original data, which is a very large amount of data. In the process of continuous monitoring, processing such a large amount of data will quickly increase the system computing resources and increase the burden on software and hardware. Taking into account the purpose of the present invention, while analyzing the characteristics of various energy consumption monitoring data, it can be clear that the original data has a large amount of time-series correlation characteristic data, including data that changes periodically, and correlation data with consistent data time-series change characteristics in the same period. The existence of these data can be used for streamlining and screening of original data, and at the same time, create an association structure between multi-source data to achieve clearer correlation characteristics on a relatively uniform time scale, which is conducive to quickly locating abnormal data and correcting it. At the same time, it can also provide a reference basis for expected data changes, and provide corresponding references in the subsequent further data optimization and utilization process. Therefore, the present invention further optimizes the data processing process, specifically based on the following steps:

[0056] S21, time synchronization processing based on the acquisition cycle, that is, taking the acquisition cycle of a specific raw data (generally a key raw data) as the standard cycle, performing time synchronization on other data;

[0057] S211. The original data of time series continuity whose acquisition cycle time scale does not match the standard cycle are spliced ​​and cut to make their acquisition cycle consistent with the standard cycle; the basic purpose of synchronous processing of the acquisition cycle is to make the collected original data relatively uniform in time series characteristics and improve the efficiency of subsequent database and model data processing.

[0058] S212. For intermittently collected periodic raw data, periodic data replication or completion is performed to make it have time continuity and process it based on step S20; data replication and completion can avoid the actual continuous data volume being missing due to equipment configuration limitations, retain continuous data change characteristics, and provide a basis for subsequent analysis and extraction of related features.

[0059] S213, based on steps 211 and 212, filter and remove homogeneous data with the same or similar time series change trends due to the correlation between the data themselves, select original data with better time series integrity and delete other data;

[0060] The existence of homogeneous data is mostly due to the structural adjacency of devices or parameters, or the direct or indirect control relationship between them, or the different manifestations of the same data in different configurations or structures. By screening and removing homogeneous data, duplicate data can be effectively compressed and the characteristic specificity of valid data can be improved.

[0061] Based on the above steps, we have obtained multi-source raw data based on the standard cycle. In the actual navigation process of the ship, the raw data are not independent of each other, but are interrelated and affected by each other. Therefore, the data judged as normal in a single raw data is inaccurate from the perspective of its associated data. Therefore, within the associated cycle of the multi-source data with correlation, multi-source data anomaly analysis should be further performed, such as Figure 1 As shown, it is implemented based on the following steps:

[0062] S214, for a group of multi-source original data with an associated period t, select continuous data within a time period T=kt, where k is an integer, to form a multi-source original data set X T ={x1,x2...x t ...x T};in represents the multi-source data value in the t-th data collection period within the time period T of the multi-source data set, where xtn represents the value of the n-th original data within xt; the sliding time window t' is used to convert the multi-source original data set X T Split to get the training set X A ={x t‘} and test set X B ={x t+1‘};in k' is a positive integer less than k, xt ‘ represents a set of measurements within a time period t';

[0063] Whether there is a correlation is mainly determined by whether the waveform graphs formed by continuous data within period t are similar. In actual implementation, the Euclidean distance, Manhattan distance, curve shape similarity, Pearson correlation coefficient and other parameters of the two waveform graphs are used for calculation. Based on the system computing power and accuracy requirements, a threshold is set and similar multi-source data is selected.

[0064] The DWGAT graph attention network is constructed with the aforementioned multi-source raw data as the graph structure network node elements; the DWGAT graph attention network model can be used to extract the mathematical structural relationship between data quantities from multi-source data, create a structural model based on its non-explicit internal relationship, and verify the measurement data based on this, revealing errors or abnormal data that are easily overlooked and difficult to detect between data, and improving the validity of the original data. Specifically:

[0065] S215. Based on the DWGAT graph attention network created in step S214, for two groups of original data with known correlation, create a correlation embedding vector between corresponding nodes according to their prior information, create a random embedding vector between two groups of original data with unknown correlation, use each graph node as the initial node to perform random walk according to a preset number of steps, create a node sequence pool reflecting the structure of the DWGAT graph attention network according to the obtained node sequence, and use the Word2Vec training model to create and optimize the missing embedding vectors between the original data to obtain the DWGAT graph attention network model; for two groups of original data with correlation, map the original data by constructing an embedding vector to characterize the potential correlation.

[0066] S216, the training set X composed of multi-source data in time period t' A ={x t‘ Input the DWGAT graph attention network model obtained in step S215 to obtain the predicted value of the measured data in the next time period t+1', and use the test set X composed of the measured data in the next time period t+1' B ={x t+1‘ Verify the model’s predicted output, determine the model error, and adjust the DWGAT graph attention network model parameters until the model training results meet the standards;

[0067] S22. Error Correction

[0068] The trained DWGAT graph attention network model is used to verify the original data from multiple sources, and the measured values ​​whose errors between the measured values ​​and the model prediction values ​​exceed the threshold are verified and corrected, and the original data are integrated;

[0069] In order to make fuller use of the time series characteristics of data and make effective predictions using the correlation characteristics between data, the present invention proposes a scheme for establishing a comprehensive model for predicting energy consumption of bulk carriers based on the Stacking comprehensive model, such as Figure 2 shown.

[0070] S3. Energy consumption prediction composite model training

[0071] Based on the Stacking comprehensive model, a comprehensive model for predicting energy consumption of bulk carriers is established. The gradient boosting decision tree algorithm is used as the original model, and the Ridge regression model, SVM algorithm model, and extreme tree search model are used for basic model training. The historical data are collected and processed based on steps setp1 to S3 to create a prediction model training database, which is divided into training data sets and test data sets in proportion; the original data in the training set is evenly divided into K (generalized into five parts, i.e., five-fold cross validation) basic data sets; the basic data sets are used to train the Ridge regression model, SVM algorithm model, and extreme tree search model. For each model, in each training cycle, one of the basic data sets is used as the verification set, and the remaining data sets are used as the training set to train each model, so as to obtain the optimal comprehensive model for predicting energy consumption of bulk carriers;

[0072] S4. Energy consumption prediction

[0073] The integrated raw data is input into the comprehensive model for predicting energy consumption of bulk carriers to obtain the prediction results.

[0074] In the process of executing the present invention and operating the system, continuous data processing and analysis are involved. In order to ensure the effective operation of the model, the operating system of the present invention should adopt a distributed processing system, that is, it includes a distributed processing device composed of at least two CPU units, and at least one of the distributed processing devices includes a data processing CPU unit for realizing data processing and diagnosis generation and a control output CPU unit for realizing process control and output.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for predicting energy consumption of electric bulk carriers based on multi-source data, characterized in that: The steps include: S1. Raw data collection The status data and operating parameters of the bulk carrier during operation are collected based on the onboard monitoring and collection device; the hydrological environment and weather environment information inside and outside the bulk carrier collection period are obtained based on the off-board monitoring and collection component; S2. Preprocessing of raw data: Preprocessing of the collected raw data includes: S20, data repair processing, specifically: for empty data, data filling is performed by fitting or interpolation according to other data in the data group where the data is located; for abnormal data values ​​or data fragments that deviate from the reasonable range of the original data, the abnormal data should be deleted and regarded as empty data for filling; S21, perform multi-source data anomaly analysis, for a group of multi-source raw data with an associated period t, select continuous data within a time period T=kt, k is an integer, to form a multi-source raw data set X T ={x1,x2...x t ...x T };in represents the multi-source data value in the tth data collection period within the time period T of the multi-source data group, where Represents x t The value of the nth original data in the time window t' is used to convert the multi-source original data set X T Split to get the training set X A ={x t‘ } and test set X B ={x t+1‘ };in k' is a positive integer less than k, x t ‘ Represents a set of measurement values ​​within the time period t'; constructs a DWGAT graph attention network using the aforementioned multi-source raw data as graph structure network node elements; For two sets of original data with known correlation, the correlation embedding vectors between the corresponding nodes are created based on their prior information. For two sets of original data with unknown correlation, random embedding vectors are created. Each graph node is used as the initial node to perform random walks according to the preset number of steps. A node sequence pool is created based on the obtained node sequence to reflect the DWGAT graph attention network structure. The Word2Vec training model is used to create and optimize the missing embedding vectors between the original data to obtain the DWGAT graph attention network model. The training set X consisting of multi-source data in time period t' A ={x t‘ Input the DWGAT graph attention network model obtained in step S215 to obtain the predicted value of the measured data in the next time period t+1', and use the test set X composed of the measured data in the next time period t+1' B ={x t+1‘ Verify the model’s predicted output, determine the model error, and adjust the DWGAT graph attention network model parameters until the model training results meet the standards; S22. Error Correction The trained DWGAT graph attention network model is used to verify the original data from multiple sources, and the measured values ​​whose errors between the measured values ​​and the model prediction values ​​exceed the threshold are verified and corrected, and the original data are integrated; S3. Energy consumption prediction composite model training Based on the Stacking comprehensive model, a comprehensive model for predicting energy consumption of bulk carriers is established. The historical data is collected and processed based on steps setp1 to S3 to create a prediction model training database to train the model and obtain the optimal comprehensive model for predicting energy consumption of bulk carriers. S4. Energy consumption prediction The integrated raw data is input into the comprehensive model for predicting energy consumption of bulk carriers to obtain the prediction results.

2. The method for predicting energy consumption of electric bulk carriers based on multi-source data according to claim 1 is characterized in that: The step S21 also includes time synchronization processing based on the acquisition cycle, taking the acquisition cycle of a specific original data as the standard cycle, and performing time synchronization on other data. Specifically: For the original data of time series continuity whose acquisition cycle time scale does not match the standard cycle, splice and cut them to make their acquisition cycle consistent with the standard cycle; For the intermittently collected periodic raw data, the periodic data is copied or completed to make it have time sequence continuity and processed based on step S20; Homogeneous data with the same or similar time series change trends due to the correlation between the data themselves are screened and cleared, and the original data with better time series integrity are selected and other data are deleted.

3. The method for predicting energy consumption of electric bulk carriers based on multi-source data according to claim 1 is characterized in that: The comprehensive model for predicting energy consumption of bulk carriers is constructed with a gradient boosting decision tree algorithm as a meta-model and a Ridge regression model, a SVM algorithm model and an extreme tree search model as basic models.

4. The method for predicting energy consumption of electric bulk carriers based on multi-source data according to claim 3 is characterized in that: The training of the model includes: The model training database is divided into training data set and test data set in proportion; the original data in the training set is evenly divided into K basic data sets; the basic data sets are used to train the Ridge regression model, SVM algorithm model, and extreme tree search model. For each model, in each training cycle, one of the basic data sets is used as the validation set, and the remaining data sets are used as the training set to train each model. After obtaining the prediction results of each model, they are compared with the validation set to determine the optimal parameters of each model, and the optimal comprehensive model for bulk carrier energy consumption prediction is obtained.

5. The electric bulk carrier energy consumption prediction system according to claim 1, characterized in that: Including data acquisition monitoring module, communication module, and control center; The data acquisition and monitoring module at least includes: an onboard monitoring and acquisition component for monitoring and collecting status data and operating parameters of the power supply, cooler, generator, propeller, and grid load in the electric bulk carrier during operation; An off-board monitoring and collection component for monitoring and collecting the operating environment of an electric bulk carrier, wherein the off-board monitoring component is used to obtain hydrological and weather environment information inside and outside the bulk carrier during a collection period; The communication module includes an Ethernet or RS-485 network for realizing data transmission between the data acquisition monitoring module and the control center; The control center includes a distributed processing device consisting of at least two CPU units, at least one of which includes a data processing CPU unit for implementing data processing and diagnosis generation and a control output CPU unit for implementing process control and output.

6. The electric bulk carrier energy consumption prediction system according to claim 5, characterized in that: The communication module specifically includes: an Ethernet transmission network based on the ModBus-TCP protocol for data transmission between power supplies, generator sets, coolers, excitation control equipment, thrusters, power grid loads and control centers, and an I / O acquisition control unit for connecting the data ports of each device; an RS-485 transmission network for data transmission with a hydrological environment information collection device and a weather environment information collection device, and an I / O acquisition control unit for connecting the data ports of each device; the Ethernet transmission network is provided with a number of MAC controllers, PHY controllers and MΠ interfaces.

7. The electric bulk carrier energy consumption prediction system according to claim 5, characterized in that: The on-board monitoring and collection component at least includes: a power supply information collection device for obtaining electrical parameters of each group of power supplies, as well as output efficiency, temperature, and power supply effectiveness status parameters; a cooler information collection device for obtaining electrical parameters of a cooler controller, as well as water inlet and outlet of the cooler, and water temperature cooler operation parameters; a generator information collection device for obtaining electrical parameters of a generator controller and supporting excitation control equipment, as well as generator output efficiency and generator set effectiveness status parameters; a thruster information collection device for obtaining electrical parameters of a thruster controller, as well as thruster power and speed status parameters; a generator information collection device for obtaining electrical parameters of a power grid load of a non-energy system in an electric bulk carrier, as well as real-time load power and load access status parameters; The off-board monitoring and collection component includes: a hydrological environment information collection device for obtaining information on ocean currents, tides, internal waves, thermocline, and seawater density outside the electric bulk carrier that may affect the navigation of the ship; and a weather environment information collection device for obtaining wind speed, air pressure, visibility, air humidity, and temperature.

Citation Information

Patent Citations

  • Multi-source time series data fault diagnosis method based on graph neural network and medium

    CN112783940A

  • Cross-scale time series data fusion method and system for Internet of Things

    CN116049764A

  • Ship traffic flow prediction method based on improved space-time diagram attention neural network

    CN116612664A

  • Ship energy consumption high-precision prediction system based on multi-model fusion

    CN117290673A

Cited By

  • Bulk cargo ship navigational speed prediction system based on feature engineering and integrated tree model

    CN121211408A

  • A robust optimization method for bulk carrier dispatching and carbon intensity compliance

    CN123093027A