LSTM-based operation optimization method for marine low-speed engine waste heat ORC system

Abstract

This invention discloses an LSTM-based method for optimizing the operation of a marine low-speed engine waste heat ORC system. The method involves collecting power generation and temperature control data from the ORC system during ocean voyages across multiple monitoring cycles and performing standardized preprocessing. Each monitoring cycle is divided into multiple time nodes, and comprehensive power change and comprehensive heat transfer change values ​​are generated. A single-cycle two-dimensional feature vector set is constructed, and K-medoids clustering is used to extract vector center points. The similarity of center points is used to determine the consistency of cycle features, accurately distinguishing between long-term and short-term waste heat fluctuation cycles. An LSTM neural network is introduced to perform time-series prediction of the two types of fluctuation cycles, outputting long-term and short-term waste heat prediction features. Based on the prediction features, dynamic adjustment schemes for the ORC system adapted to long-term trends and short-term disturbances are generated, optimizing the working fluid pump speed, circulation rate, and evaporation and condensation conditions.

Description

Technical Field

[0001] This invention relates to the field of marine energy analysis, and more specifically, to an LSTM-based method for optimizing the operation of marine low-speed engine waste heat ORC systems. Background Technology

[0002] As the global shipping industry continues its green and low-carbon transformation, energy conservation and emission reduction have become mandatory development indicators for the ocean-going vessel industry. Very Large Ore Carriers (VLOCs) and other ocean-going dry bulk carriers are characterized by long sailing cycles and high equipment loads. During long-term operation, the main engines and auxiliary engines continuously generate a large amount of medium- and low-grade waste heat, mainly including waste heat from cylinder liner cooling water, low-temperature exhaust gas from diesel engines, and heat dissipation from electromechanical equipment. The Organic Rankine Cycle (ORC), with its advantages of strong low-temperature adaptability, compact equipment structure, and convenient operation and maintenance, has become the core technology solution for the resource recovery of low-temperature waste heat from ships. It can convert waste low-temperature heat into clean electricity, effectively reducing ship fuel consumption and emissions of sulfur oxides (SOx), nitrogen oxides (NOx), and carbon dioxide (CO2), achieving effective control of emissions from low-carbon and zero-carbon fuel low-speed engines on ships.

[0003] However, the current mainstream ORC system operation control technology in China is still mainly based on traditional fixed parameter PID control and static operating condition matching control. The core control logic relies on the fixed operating parameters set by the ship's rated navigation conditions. It can only make minor passive corrections for instantaneous operating data. It lacks the dual superposition characteristics analysis of long-term trend gradual fluctuations and short-term sudden disturbance fluctuations in the waste heat state, lacks accurate trend prediction of power generation and temperature control states, cannot distinguish between long-term trend changes and short-term invalid disturbances, and has lag in regulation and low operating condition adaptation accuracy. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and proposes an operation optimization method for marine low-speed engine waste heat ORC system based on LSTM.

[0005] The first aspect of this invention provides an LSTM-based method for optimizing the operation of a marine low-speed engine waste heat ORC system, comprising: S1. During multiple monitoring cycles, power generation status data and temperature control status data are collected through the ORC system of the marine low-speed engine, and the collected data is preprocessed. S2. In each monitoring cycle, multiple time nodes are set, and power characteristics and temperature control characteristics are analyzed based on power generation status data and temperature control status data, and comprehensive power change value and comprehensive heat exchange change value are generated for each time node. S3. Construct a feature vector set based on a monitoring cycle according to the comprehensive power change value and the comprehensive heat exchange change value. Select the vector center point of the feature vector set by clustering. Based on the vector center point, identify the cycle with feature consistency in multiple monitoring cycles and mark the long-term fluctuation cycle and the short-term fluctuation cycle. S4. Based on the collected data, an LSTM prediction model is introduced to perform serialized prediction of the corresponding power generation and temperature control states in the long-term and short-term fluctuation cycles, and to obtain long-term and short-term prediction features based on the waste heat state. S5. Generate the first and second adjustment schemes of the ORC system based on the long-term and short-term prediction characteristics, respectively.

[0006] In this solution, S1 specifically includes: The power generation status data includes real-time power generation, multi-level power generation condition labels, real-time working fluid flow rate, and working fluid inlet and outlet pressure parameters. The temperature control status data includes waste heat exchange temperature difference, heat exchange temperature, heat exchanger inlet and outlet temperatures, temperature control adjustment range, and heat exchange flow rate parameters. During the data collection process, multiple monitoring cycles are continuous and uninterrupted time-series monitoring cycles, while the duration of a single monitoring cycle is a fixed time period.

[0007] In this solution, S1 includes: The preprocessing specifically includes wavelet threshold denoising, missing value repair, outlier removal, and data normalization. Wavelet threshold denoising uses the db3 wavelet basis function, with a decomposition level of 3 layers and a threshold value of 0.02-0.05 to denoise the collected data for high-frequency noise. The missing value repair adopts piecewise linear interpolation to filter and fill short-term missing data with a duration of less than 2 seconds within a single period. The outlier removal adopts the 3σ criterion to remove abnormal data points that exceed the normal operating condition parameter range. The normalization process adopts the maximum-minimum normalization algorithm to uniformly map all parameter data to a preset standard range.

[0008] In this solution, S2 includes: In each monitoring cycle, N time points are set; At each time point, multi-dimensional power generation parameters and multi-dimensional temperature control parameters are obtained from the collected data. The parameter change rate of adjacent nodes is weighted and calculated for the power generation status to obtain the comprehensive power change value. The combined temperature control change value is obtained by weighting the rate of change of parameters of adjacent nodes for the temperature control status.

[0009] In this solution, S3 specifically refers to: At a given time point, a two-dimensional feature vector is generated based on the changes in comprehensive power and comprehensive heat transfer. Multiple feature vectors are calculated for each monitoring cycle, and a feature vector set is formed. In a feature vector set, the feature vector set of a single monitoring period is used as an independent clustering sample. K initial cluster centers are set, and the number of cluster centers, the iteration convergence threshold and the maximum number of iterations are preset. The distance between vectors is calculated using cosine distance. Multiple optimal vector center points of the feature vector set of this period are selected through iterative clustering. The mean value of multiple vector center points obtained by clustering for a single monitoring period is calculated to obtain the mean vector center point that uniquely represents the overall fluctuation characteristics of that period.

[0010] In this solution, S3 further includes: The feature similarity between the center points of the mean vectors of different monitoring periods is calculated using cosine similarity. Feature similarity intervals are set based on the maximum distance, the shortest distance and the preset similarity threshold. If two consecutive monitoring periods are selected, and the distance between the center points of the corresponding two mean vectors is within a similar interval, then the two consecutive monitoring periods are marked, and all monitoring periods are identified to obtain the long-term fluctuation period. The remaining unmarked periods are taken as short-term fluctuation periods.

[0011] In this solution, S4 includes: For long-term and short-term fluctuation cycles, the corresponding power generation parameters and temperature control parameters are obtained from the collected data, resulting in the first and second collected data, respectively. Construct an LSTM time series prediction model, specifically a two-layer stacked deep neural network model. The prediction model includes an input layer, a two-layer LSTM hidden layer, a fully connected mapping layer, and an output layer. The datasets whose parameter fluctuation range is within a preset threshold compared to the first and second collected data are selected from the historical collected data and labeled as the first training set and the second training set. The first and second training sets are imported into the prediction model to perform long-term and short-term fluctuation prediction training of power generation status, respectively, until the preset number of iterations is reached. The first and second collected data are serialized based on time nodes to obtain the power generation parameter sequence and the temperature control parameter sequence. The sequences are then imported into the prediction model. Long-term and short-term prediction periods are set to predict the power generation status and temperature control status. Long-term and short-term prediction features are generated based on the prediction data.

[0012] In this solution, S5 specifically refers to: Based on the long-term prediction characteristics, the long-term increasing or decreasing trend of waste heat is analyzed, and the operating speed of the working fluid pump, the total amount of working fluid circulation, the heat exchange temperature of the evaporator chamber, and the operating pressure of the condenser chamber of the ORC system are adjusted to generate the first adjustment scheme. Based on the short-term prediction characteristics, the instantaneous change disturbance of waste heat is analyzed, and the instantaneous circulation rate of the working fluid, the opening degree of the temperature control valve, and the flow rate of the heat exchange medium are monitored and adjusted to generate a second adjustment scheme.

[0013] A second aspect of the present invention also provides an LSTM-based operational optimization system for a marine low-speed engine waste heat ORC system. The system includes a memory, a processor, and an interface. The memory includes an LSTM-based operational optimization program for the marine low-speed engine waste heat ORC system. When executed by the processor, the LSTM-based operational optimization program for the marine low-speed engine waste heat ORC system performs the following steps: S1. During multiple monitoring cycles, power generation status data and temperature control status data are collected through the ORC system of the marine low-speed engine, and the collected data is preprocessed. S2. In each monitoring cycle, multiple time nodes are set, and power characteristics and temperature control characteristics are analyzed based on power generation status data and temperature control status data, and comprehensive power change value and comprehensive heat exchange change value are generated for each time node. S3. Construct a feature vector set based on a monitoring cycle according to the comprehensive power change value and the comprehensive heat exchange change value. Select the vector center point of the feature vector set by clustering. Based on the vector center point, identify the cycle with feature consistency in multiple monitoring cycles and mark the long-term fluctuation cycle and the short-term fluctuation cycle. S4. Based on the collected data, an LSTM prediction model is introduced to perform serialized prediction of the corresponding power generation and temperature control states in the long-term and short-term fluctuation cycles, and to obtain long-term and short-term prediction features based on the waste heat state. S5. Generate the first and second adjustment schemes of the ORC system based on the long-term and short-term prediction characteristics, respectively.

[0014] A third aspect of the present invention also provides a computer-readable storage medium comprising an LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system. When the LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system is executed by a processor, it implements the steps of the LSTM-based operation optimization method for a marine low-speed engine waste heat ORC system as described in any of the preceding claims.

[0015] This invention discloses an LSTM-based method for optimizing the operation of a marine low-speed engine waste heat ORC system. The method involves collecting power generation and temperature control data from the ORC system across multiple monitoring cycles and performing standardized preprocessing. Each monitoring cycle is divided into multiple time nodes, and comprehensive power change and comprehensive heat transfer change values ​​are generated. A single-cycle two-dimensional feature vector set is constructed, and K-medoids clustering is used to extract vector center points. The similarity of center points is used to determine the consistency of cycle features, accurately distinguishing between long-term and short-term waste heat fluctuation cycles. An LSTM neural network is introduced to perform time-series prediction of the two types of fluctuation cycles, outputting long-term and short-term waste heat prediction features. Based on the prediction features, dynamic adjustment schemes for the ORC system adapted to long-term trends and short-term disturbances are generated, optimizing the working fluid pump speed, circulation rate, and evaporation and condensation conditions.

[0016] This invention solves the problems of large fluctuations in low-temperature heat sources, poor adaptability of traditional control systems, unstable system operation, and low waste heat utilization rate in large ocean-going vessels, and significantly improves the efficiency of low-temperature waste heat recovery in ships. Attached Figure Description

[0017] Figure 1 A flowchart of an operation optimization method for a marine low-speed engine waste heat ORC system based on LSTM is shown. Figure 2 A schematic diagram of the training of the LSTM prediction model of the present invention is shown; Figure 3 A block diagram of an LSTM-based marine low-speed engine waste heat ORC system operation optimization system is shown. Detailed Implementation

[0018] To better understand the above-mentioned objects, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It is understood that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0019] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0020] Figure 1 A flowchart of an operation optimization method for a marine low-speed engine waste heat ORC system based on LSTM is shown.

[0021] like Figure 1 As shown, the first aspect of the present invention provides an operation optimization method for a marine low-speed engine waste heat ORC system based on LSTM, comprising: S1. During multiple monitoring cycles, power generation status data and temperature control status data are collected through the ORC system of the marine low-speed engine, and the collected data is preprocessed. S2. In each monitoring cycle, multiple time nodes are set, and power characteristics and temperature control characteristics are analyzed based on power generation status data and temperature control status data, and comprehensive power change value and comprehensive heat exchange change value are generated for each time node. S3. Construct a feature vector set based on a monitoring cycle according to the comprehensive power change value and the comprehensive heat exchange change value. Select the vector center point of the feature vector set by clustering. Based on the vector center point, identify the cycle with feature consistency in multiple monitoring cycles and mark the long-term fluctuation cycle and the short-term fluctuation cycle. S4. Based on the collected data, an LSTM prediction model is introduced to perform serialized prediction of the corresponding power generation and temperature control states in the long-term and short-term fluctuation cycles, and to obtain long-term and short-term prediction features based on the waste heat state. S5. Generate the first and second adjustment schemes of the ORC system based on the long-term and short-term prediction characteristics, respectively.

[0022] According to an embodiment of the present invention, step S1 specifically includes: The power generation status data includes real-time power generation, multi-level power generation condition labels, real-time working fluid flow rate, and working fluid inlet and outlet pressure parameters. The temperature control status data includes waste heat exchange temperature difference, heat exchange temperature, heat exchanger inlet and outlet temperatures, temperature control adjustment range, and heat exchange flow rate parameters. During the data collection process, multiple monitoring cycles are continuous and uninterrupted time-series monitoring cycles, while the duration of a single monitoring cycle is a fixed time period.

[0023] It is understood that the analysis process in this embodiment is applicable to predictive analysis of all navigation conditions for low-speed engines used in ocean-going vessels. The fixed time period can be set to 20-40 minutes, and the data sampling frequency is 0.5-2 seconds per sampling.

[0024] According to an embodiment of the present invention, S1 includes: The preprocessing specifically includes wavelet threshold denoising, missing value repair, outlier removal, and data normalization. Wavelet threshold denoising uses the db3 wavelet basis function, with a decomposition level of 3 layers and a threshold value of 0.02-0.05 to denoise the collected data for high-frequency noise. The missing value repair adopts piecewise linear interpolation to filter and fill short-term missing data with a duration of less than 2 seconds within a single period. The outlier removal adopts the 3σ criterion to remove abnormal data points that exceed the normal operating condition parameter range. The normalization process adopts the maximum-minimum normalization algorithm to uniformly map all parameter data to a preset standard range.

[0025] It can be understood here that wavelet threshold denoising is used to filter out high-frequency noise generated by electromagnetic interference from ship navigation and equipment vibration; the preset standard interval can be set to [0,1] to eliminate the difference in the dimensions of different parameters.

[0026] According to an embodiment of the present invention, S2 includes: In each monitoring cycle, N time points are set; At each time point, multi-dimensional power generation parameters and multi-dimensional temperature control parameters are obtained from the collected data. The parameter change rate of adjacent nodes is weighted and calculated for the power generation status to obtain the comprehensive power change value. The combined temperature control change value is obtained by weighting the rate of change of parameters of adjacent nodes for the temperature control status.

[0027] It can be understood here that N>=10. Multi-dimensional power generation parameters and multi-dimensional temperature control parameters are user-defined parameters; for example, power generation capacity and working fluid flow rate parameters can be selected.

[0028] The comprehensive change value is calculated as follows: ; Where P is the rate of change of a certain parameter, and Q1 is the weight value. These represent the values ​​of a parameter at the current time point, the previous time point, and the next time point. For multi-dimensional parameters, the average rate of change of the corresponding multiple parameters is used as the comprehensive change value.

[0029] The working fluid flow rate is the mass of the circulating working fluid (such as R245fa, pentane, etc.) flowing through the evaporator per unit time in an Organic Rankine Cycle (ORC) or similar waste heat recovery system, expressed in kg / h or t / h.

[0030] According to an embodiment of the present invention, step S3 specifically includes: At a given time point, a two-dimensional feature vector is generated based on the changes in comprehensive power and comprehensive heat transfer. Multiple feature vectors are calculated for each monitoring cycle, and a feature vector set is formed. In a feature vector set, the feature vector set of a single monitoring period is used as an independent clustering sample. K initial cluster centers are set, and the number of cluster centers, the iteration convergence threshold and the maximum number of iterations are preset. The distance between vectors is calculated using cosine distance. Multiple optimal vector center points of the feature vector set of this period are selected through iterative clustering. The mean value of multiple vector center points obtained by clustering for a single monitoring period is calculated to obtain the mean vector center point that uniquely represents the overall fluctuation characteristics of that period.

[0031] This can be understood as follows: the feature vector is used to comprehensively characterize the combined waste heat characteristics of the power generation and heat exchange states assessed by the ORC system at a certain time point. Based on this, a feature vector set is formed for each monitoring period. Furthermore, the similarity and consistency between the vector sets are analyzed to determine the trend of waste heat state characteristics in the monitoring period. Monitoring periods with consistent characteristics are identified as periods with long-term fluctuation trends, and corresponding data are extracted for prediction. Each monitoring period corresponds to a feature vector set.

[0032] A multi-dimensional numerical average is calculated for multiple optimal vector centers, and the resulting mean vector is used as the mean vector center. The preset number of cluster centers is set to a small number, such as 2-4 classes, the iteration convergence threshold is set to 0.001-0.005, and the maximum number of iterations is 80-120.

[0033] According to an embodiment of the present invention, S3 further includes: The feature similarity between the center points of the mean vectors of different monitoring periods is calculated using cosine similarity. Feature similarity intervals are set based on the maximum distance, the shortest distance and the preset similarity threshold. If two consecutive monitoring periods are selected, and the distance between the center points of the corresponding two mean vectors is within a similar interval, then the two consecutive monitoring periods are marked, and all monitoring periods are identified to obtain the long-term fluctuation period. The remaining unmarked periods are taken as short-term fluctuation periods.

[0034] Here, based on the maximum distance, the shortest distance, and a preset similarity threshold, a feature similarity interval is set. Specifically, the distance D between the center points of vectors from two monitoring cycles is calculated, and the maximum and minimum values, D_MAX and D_MIN, are selected. The interval [0, |D_MAX-D_MIN|×P1] is set as the similarity discrimination interval. When the distance D belongs to this interval, it is determined that the two corresponding monitoring cycles have consistent residual heat characteristics, and the two consecutive monitoring cycles are marked as long-term cycles, that is, the two sets of monitoring cycle characteristics are highly consistent, and there is a long-term stable residual heat fluctuation trend. The remaining monitoring cycles are determined to be cycles with discrete characteristics and no stable temporal pattern, which are significantly affected by short-term random disturbances, and these cycles are marked as short-term fluctuation cycles. P1 is the preset similarity threshold, which is an adjustable percentage value.

[0035] Figure 2 A schematic diagram of the training of the LSTM prediction model of the present invention is shown; According to an embodiment of the present invention, step S4 includes: For long-term and short-term fluctuation cycles, the corresponding power generation parameters and temperature control parameters are obtained from the collected data, resulting in the first and second collected data, respectively. Construct an LSTM time series prediction model, specifically a two-layer stacked deep neural network model. The prediction model includes an input layer, a two-layer LSTM hidden layer, a fully connected mapping layer, and an output layer. The datasets whose parameter fluctuation range is within a preset threshold compared to the first and second collected data are selected from the historical collected data and labeled as the first training set and the second training set. The first and second training sets are imported into the prediction model to perform long-term and short-term fluctuation prediction training of power generation status, respectively, until the preset number of iterations is reached. The first and second collected data are serialized based on time nodes to obtain the power generation parameter sequence and the temperature control parameter sequence. The sequences are then imported into the prediction model. Long-term and short-term prediction periods are set to predict the power generation status and temperature control status. Long-term and short-term prediction features are generated based on the prediction data.

[0036] Here, long-term series prediction is performed, with the prediction period set to 40-80 minutes, to uncover the long-term gradual change pattern of waste heat and output the long-term prediction characteristics of the system's power generation long-term trend, the steady-state change of heat exchange temperature, and the long-term deviation of the working fluid flow rate. For short-term fluctuation periods, short-term series prediction is performed, with the prediction period set to 5-15 minutes, to capture the instantaneous change pattern of waste heat and output the short-term prediction characteristics of the system's operating conditions short-term fluctuation amplitude and the instantaneous deviation of parameters.

[0037] In the prediction model, the input layer dimension is matched with multiple core acquisition parameters after preprocessing. The number of neurons in the two-layer LSTM hidden layer is set to 100-150 and 50-80 respectively. The number of model training iterations is set to 150-250 times. The initial learning rate is set to 0.0008-0.0012. The dropout regularization coefficient is set to 0.2-0.3, which effectively prevents the model from overfitting during training and improves the prediction generalization ability.

[0038] In serialized prediction, the power generation state parameters can be selected as power generation capacity and working fluid flow rate, and the temperature control parameters can be selected as heat exchange temperature, heat exchange flow rate, etc.

[0039] In the historical data collection, select the data set whose parameter fluctuation range is within the preset threshold of the first and second data collection. Specifically, the preset threshold can be set to 10%. Compare the historical parameters with the first and second data collection respectively. If the parameter deviation rate is within 10%, the corresponding historical data is extracted as feature similar data and marked as the training set.

[0040] According to an embodiment of the present invention, step S5 specifically includes: Based on the long-term prediction characteristics, the long-term increasing or decreasing trend of waste heat is analyzed, and the operating speed of the working fluid pump, the total amount of working fluid circulation, the heat exchange temperature of the evaporator chamber, and the operating pressure of the condenser chamber of the ORC system are adjusted to generate the first adjustment scheme. Based on the short-term prediction characteristics, the instantaneous change disturbance of waste heat is analyzed, and the instantaneous circulation rate of the working fluid, the opening degree of the temperature control valve, and the flow rate of the heat exchange medium are monitored and adjusted to generate a second adjustment scheme.

[0041] It can be understood that the first adjustment scheme is a long-term steady-state optimization scheme, which adapts to the trend of long-term predicted characteristics. In response to the long-term increasing or decreasing trend of waste heat, it slowly, continuously and linearly adjusts the operating speed of the working fluid pump, the total amount of working fluid circulation, the heat exchange temperature of the evaporator chamber, and the operating pressure of the condenser chamber in an active, linear manner to match the long-term trend of waste heat and ensure the long-term efficient operation of the system. The second adjustment scheme is a short-term anti-disturbance optimization scheme, which adapts to the random fluctuation of short-term predicted characteristics. In response to the instantaneous sudden change of waste heat, it quickly fine-tunes the instantaneous circulation of the working fluid, the opening of the temperature control valve, and the flow rate of the heat exchange medium to quickly offset the short-term operating disturbance and ensure the instantaneous stable operation of the system.

[0042] In a preferred embodiment, the adjustment scheme is specifically configured as follows: when the long-term prediction characteristics show an upward trend in waste heat, the working fluid pump speed is increased synchronously and gradually, the total amount of working fluid circulation is increased, and the heat exchange temperature of the evaporation condition is raised; when the long-term prediction characteristics show a downward trend in waste heat, the working fluid pump speed is gradually decreased, the amount of working fluid circulation is reduced, and the heat dissipation efficiency of the condenser is optimized; when the short-term prediction characteristics show a sudden surge in waste heat, the amount of working fluid circulation is slightly reduced, and the intensity of condensation heat dissipation is enhanced; when the short-term prediction characteristics show a sudden drop in waste heat, the amount of working fluid circulation is slightly increased, and the evaporation heat exchange condition is stabilized, thereby achieving graded, precise, and efficient recovery of low-temperature waste heat.

[0043] According to an embodiment of the present invention, it further includes: Short-term prediction sequences for power generation status and short-term prediction sequences for temperature control status are obtained based on the prediction data. The short-time prediction sequences for power generation and temperature control were linearly fitted using the least squares method, resulting in two fitting curves. If the fitting loss in the fitted curve is less than the preset loss and the fitted curve is linearly increasing, then the current ORC system is determined to be in a non-steady-state period. Power characteristics, temperature control characteristics, and corresponding prediction data are collected in real time to generate waste heat steady-state early warning information and send the early warning information to the preset terminal.

[0044] It can be understood here that the linear fitting uses univariate regression linear fitting, with time nodes as independent variables and corresponding state parameters as dependent variables for fitting, and the fitting coefficients are determined by least squares search. A preset loss can be set to 10%. When the deviation rate between the fitted value and the actual predicted value is within 10%, it is determined that the fitting loss is less than the preset loss. In this embodiment, linear fitting prediction of waste heat characteristics is performed based on short-term predicted fluctuation data. If linear growth exists, it indicates that the waste heat system is in an unstable state, with some over- or under-adjustment, which can easily lead to low low-temperature waste heat recovery efficiency and poor system operational stability, requiring corresponding early warning settings.

[0045] Figure 3 A block diagram of an LSTM-based marine low-speed engine waste heat ORC system operation optimization system is shown.

[0046] A second aspect of the present invention also provides an LSTM-based operational optimization system for a marine low-speed engine waste heat ORC system. The system includes a memory, a processor, and an interface. The memory includes an LSTM-based operational optimization program for the marine low-speed engine waste heat ORC system. When executed by the processor, the LSTM-based operational optimization program for the marine low-speed engine waste heat ORC system performs the following steps: S1. During multiple monitoring cycles, power generation status data and temperature control status data are collected through the ORC system of the marine low-speed engine, and the collected data is preprocessed. S2. In each monitoring cycle, multiple time nodes are set, and power characteristics and temperature control characteristics are analyzed based on power generation status data and temperature control status data, and comprehensive power change value and comprehensive heat exchange change value are generated for each time node. S3. Construct a feature vector set based on a monitoring cycle according to the comprehensive power change value and the comprehensive heat exchange change value. Select the vector center point of the feature vector set by clustering. Based on the vector center point, identify the cycle with feature consistency in multiple monitoring cycles and mark the long-term fluctuation cycle and the short-term fluctuation cycle. S4. Based on the collected data, an LSTM prediction model is introduced to perform serialized prediction of the corresponding power generation and temperature control states in the long-term and short-term fluctuation cycles, and to obtain long-term and short-term prediction features based on the waste heat state. S5. Generate the first and second adjustment schemes of the ORC system based on the long-term and short-term prediction characteristics, respectively.

[0047] When the system is running, it can perform one or more steps of the above-described LSTM-based method for optimizing the operation of marine low-speed engine waste heat ORC systems.

[0048] A third aspect of the present invention also provides a computer-readable storage medium comprising an LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system. When the LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system is executed by a processor, it implements the steps of the LSTM-based operation optimization method for a marine low-speed engine waste heat ORC system as described in any of the preceding claims.

[0049] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0050] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0051] In addition, in the various embodiments of the present invention, each functional unit can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0052] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] Alternatively, if the integrated units of this invention are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this invention, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0054] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. An LSTM-based method for optimizing the operation of a marine low-speed engine waste heat ORC system, characterized in that, include: S1. During multiple monitoring cycles, power generation status data and temperature control status data are collected through the ORC system of the marine low-speed engine, and the collected data is preprocessed. S2. In each monitoring cycle, multiple time nodes are set, and power characteristics and temperature control characteristics are analyzed based on power generation status data and temperature control status data, and comprehensive power change value and comprehensive heat exchange change value are generated for each time node. S3. Construct a feature vector set based on a monitoring cycle according to the comprehensive power change value and the comprehensive heat exchange change value. Select the vector center point of the feature vector set by clustering. Based on the vector center point, identify the cycle with feature consistency in multiple monitoring cycles and mark the long-term fluctuation cycle and the short-term fluctuation cycle. S4. Based on the collected data, an LSTM prediction model is introduced to perform serialized prediction of the corresponding power generation and temperature control states in the long-term and short-term fluctuation cycles, and to obtain long-term and short-term prediction features based on the waste heat state. S5. Generate the first and second adjustment schemes of the ORC system based on the long-term and short-term prediction characteristics, respectively.

2. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 1, characterized in that, S1 specifically includes: The power generation status data includes real-time power generation, multi-level power generation condition labels, real-time working fluid flow rate, and working fluid inlet and outlet pressure parameters. The temperature control status data includes waste heat exchange temperature difference, heat exchange temperature, heat exchanger inlet and outlet temperatures, temperature control adjustment range, and heat exchange flow rate parameters. During the data collection process, multiple monitoring cycles are continuous and uninterrupted time-series monitoring cycles, while the duration of a single monitoring cycle is a fixed time period.

3. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 2, characterized in that, S1 includes: The preprocessing specifically includes wavelet threshold denoising, missing value repair, outlier removal, and data normalization. Wavelet threshold denoising uses the db3 wavelet basis function, with a decomposition level of 3 layers and a threshold value of 0.02-0.05 to denoise the collected data for high-frequency noise. The missing value repair adopts piecewise linear interpolation to filter and fill short-term missing data with a duration of less than 2 seconds within a single period. The outlier removal adopts the 3σ criterion to remove abnormal data points that exceed the normal operating condition parameter range. The normalization process adopts the maximum-minimum normalization algorithm to uniformly map all parameter data to a preset standard range.

4. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 3, characterized in that, The S2 includes: In each monitoring cycle, N time points are set; At each time point, multi-dimensional power generation parameters and multi-dimensional temperature control parameters are obtained from the collected data. The parameter change rate of adjacent nodes is weighted and calculated for the power generation status to obtain the comprehensive power change value. The combined temperature control change value is obtained by weighting the rate of change of parameters of adjacent nodes for the temperature control status.

5. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 4, characterized in that, Specifically, S3 is: At a given time point, a two-dimensional feature vector is generated based on the changes in comprehensive power and comprehensive heat transfer. Multiple feature vectors are calculated for each monitoring cycle, and a feature vector set is formed. In a feature vector set, the feature vector set of a single monitoring period is used as an independent clustering sample. K initial cluster centers are set, and the number of cluster centers, the iteration convergence threshold and the maximum number of iterations are preset. The distance between vectors is calculated using cosine distance. Multiple optimal vector center points of the feature vector set of this period are selected through iterative clustering. The mean value of multiple vector center points obtained by clustering for a single monitoring period is calculated to obtain the mean vector center point that uniquely represents the overall fluctuation characteristics of that period.

6. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 5, characterized in that, S3 further includes: The feature similarity between the center points of the mean vectors of different monitoring periods is calculated using cosine similarity. Feature similarity intervals are set based on the maximum distance, the shortest distance and the preset similarity threshold. If two consecutive monitoring periods are selected, and the distance between the center points of the corresponding two mean vectors is within a similar interval, then the two consecutive monitoring periods are marked, and all monitoring periods are identified to obtain the long-term fluctuation period. The remaining unmarked periods are taken as short-term fluctuation periods.

7. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 6, characterized in that, The S4 includes: For long-term and short-term fluctuation cycles, the corresponding power generation parameters and temperature control parameters are obtained from the collected data, resulting in the first and second collected data, respectively. Construct an LSTM time series prediction model, specifically a two-layer stacked deep neural network model. The prediction model includes an input layer, a two-layer LSTM hidden layer, a fully connected mapping layer, and an output layer. The datasets whose parameter fluctuation range is within a preset threshold compared to the first and second collected data are selected from the historical collected data and labeled as the first training set and the second training set. The first and second training sets are imported into the prediction model to perform long-term and short-term fluctuation prediction training of power generation status, respectively, until the preset number of iterations is reached. The first and second collected data are serialized based on time nodes to obtain the power generation parameter sequence and the temperature control parameter sequence. The sequences are then imported into the prediction model. Long-term and short-term prediction periods are set to predict the power generation status and temperature control status. Long-term and short-term prediction features are generated based on the prediction data.

8. The method for optimizing the operation of a marine low-speed engine waste heat ORC system based on LSTM according to claim 7, characterized in that, Specifically, S5 is: Based on the long-term prediction characteristics, the long-term increasing or decreasing trend of waste heat is analyzed, and the operating speed of the working fluid pump, the total amount of working fluid circulation, the heat exchange temperature of the evaporator chamber, and the operating pressure of the condenser chamber of the ORC system are adjusted to generate the first adjustment scheme. Based on the short-term prediction characteristics, the instantaneous change disturbance of waste heat is analyzed, and the instantaneous circulation rate of the working fluid, the opening degree of the temperature control valve, and the flow rate of the heat exchange medium are monitored and adjusted to generate a second adjustment scheme.

9. An LSTM-based operational optimization system for waste heat recovery (ORC) systems of marine low-speed engines, characterized in that, The system includes: a memory, a processor, and an interaction interface. The memory includes an LSTM-based optimization program for the operation of a marine low-speed engine waste heat ORC system. When the processor executes the LSTM-based optimization program for the marine low-speed engine waste heat ORC system, it implements the steps of the LSTM-based optimization method for the operation of a marine low-speed engine waste heat ORC system as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes an LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system. When the LSTM-based operation optimization program for a marine low-speed engine waste heat ORC system is executed by a processor, it implements the steps of the LSTM-based operation optimization method for a marine low-speed engine waste heat ORC system as described in any one of claims 1-8.

Images