A method and system for monitoring data optimization of a marine fuel supply system

Abstract

The application relates to the field of ship data transmission, and discloses a monitoring data optimization method and system for a ship fuel supply system. The method comprises the following steps: sending a detection message through a satellite communication link; obtaining a current bandwidth value, a delay value and a packet loss rate based on the detection message; calculating a theoretical optimal data block size according to the current bandwidth value, the delay value and the packet loss rate; dividing fuel monitoring data into data blocks matched with a determined actual data block size to obtain a block data sequence; and sending the block data sequence with an additional serial number and check information through the satellite communication link. Through the technical scheme, the problem of poor adaptability of data block size to link state in the prior art is solved, and the fuel monitoring data transmission efficiency under the satellite link is improved, and the packet loss retransmission probability is reduced.

Description

Technical Field

[0001] This application relates to the field of ship data transmission, and in particular to a method and system for optimizing monitoring data of a ship fuel supply system. Background Technology

[0002] Currently, ship-to-shore transmission of ship fuel monitoring data mainly relies on satellite communication links. However, satellite links in the marine environment have significant dynamic fluctuation characteristics: affected by factors such as ocean currents, weather, near-shore electromagnetic interference, and satellite coverage blind spots, the link bandwidth may drop sharply in a short period of time, the latency value fluctuates widely, and the packet loss rate often increases significantly due to signal attenuation.

[0003] The mainstream monitoring data transmission methods in the existing technology mostly adopt a block transmission strategy with fixed-size data blocks, that is, a uniform data block size is pre-set, and basic verification information is added before transmission via satellite link.

[0004] Existing fixed-block transmission strategies have insurmountable technical flaws. When the link is in an ideal state of high bandwidth, low latency, and low packet loss, using too small a fixed data block will result in an excessively high proportion of protocol overhead in the data block header, significantly reducing the efficiency of effective data transmission. Conversely, when the link enters a severe state of low bandwidth, high latency, and high packet loss due to weather interference, using too large a fixed data block will cause the transmission time of a single data block to far exceed the link's stability window, easily leading to data block loss or transmission timeouts, and thus triggering frequent retransmissions.

[0005] Solving this technical problem is a technical challenge that needs to be overcome by those skilled in the art. Summary of the Invention

[0006] This application provides a method for optimizing monitoring data of a ship fuel supply system to at least partially solve the above-mentioned technical problems.

[0007] To achieve the above objectives, according to a first aspect of this application, a method for optimizing monitoring data of a ship fuel supply system is provided, comprising:

[0008] Send probe messages via satellite communication link; obtain current bandwidth value, latency value and packet loss rate based on the probe messages;

[0009] The theoretically optimal data block size is calculated based on the current bandwidth, latency, and packet loss rate.

[0010] Determine whether the theoretically optimal data block size is within a preset range: if the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size; if the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size.

[0011] Fuel monitoring data is divided into data blocks that match the determined actual data block size to obtain a block data sequence;

[0012] The segmented data sequence is transmitted via a satellite communication link, with additional sequence numbers and verification information.

[0013] According to a second aspect of this application, a monitoring data optimization system for a ship fuel supply system is provided, comprising:

[0014] The first processing module is used to: send probe messages through a satellite communication link; and obtain the current bandwidth value, latency value, and packet loss rate based on the probe messages;

[0015] The second processing module is used to calculate the theoretically optimal data block size based on the current bandwidth value, latency value, and packet loss rate.

[0016] The third processing module is used to: determine whether the theoretically optimal data block size is within a preset range; if the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size; if the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size.

[0017] The fourth processing module is used to: divide the fuel monitoring data into data blocks that match the determined actual data block size to obtain a block data sequence;

[0018] The fifth processing module is used to: transmit the block data sequence with additional sequence numbers and verification information via a satellite communication link.

[0019] In summary, the embodiments of this application, through the above technical solutions, solve the problem of poor adaptability between data block size and link status in the prior art, improve the data transmission efficiency of fuel monitoring under satellite links, and reduce the probability of packet loss and retransmission.

[0020] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a flowchart illustrating the steps of a method for optimizing monitoring data of a ship fuel supply system provided in an exemplary embodiment of this application.

[0023] Figure 2 This is a schematic diagram of a monitoring data optimization system for a ship fuel supply system provided in an exemplary embodiment of this application;

[0024] Explanation of reference numerals in the attached drawings: 01, First processing module; 202, Second processing module; 203, Third processing module; 204, Fourth processing module; 205, Fifth processing module. Detailed Implementation

[0025] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0026] This application provides a method for optimizing monitoring data of a ship's fuel supply system. Please refer to [link / reference]. Figure 1 The method for optimizing monitoring data of a ship fuel supply system provided in this application includes the following steps:

[0027] Step 101: Send a probe message through the satellite communication link; obtain the current bandwidth value, latency value and packet loss rate based on the probe message.

[0028] Specifically, in the satellite communication link between the ship terminal and the shore end, the ship terminal first sends probe messages to the shore end. The shore end then calculates and obtains the current link status parameters, namely bandwidth, latency, and packet loss rate, based on feedback information. The probe message is a specific format data frame used to probe the transmission characteristics of the satellite communication link. It typically includes a timestamp recording the time of transmission, a unique sequence number to distinguish different probe messages, and a data integrity checksum to verify whether the message is corrupted. This allows the receiving end to provide feedback on the transmission status based on the message content. The bandwidth value mentioned above refers to the maximum amount of data that a satellite communication link can transmit per unit time, reflecting the transmission capacity of the link. The higher the bandwidth value, the greater the amount of data that the link can transmit per unit time. The latency value refers to the total time difference between when a probe message is sent from the ship terminal, transmitted to the shore via the satellite link, and then when the shore sends an acknowledgment message back to the ship terminal. It reflects the real-time performance of the link. The lower the latency value, the smaller the time lag in data transmission. The packet loss rate refers to the ratio of the number of probe messages lost within a set time to the total number of probe messages sent. It reflects the transmission reliability of the link. The lower the packet loss rate, the lower the probability of messages being lost during transmission.

[0029] Step 102: Calculate the theoretically optimal data block size based on the current bandwidth, latency, and packet loss rate.

[0030] Specifically, based on the obtained bandwidth, latency, and packet loss rate, the theoretically optimal data block size is calculated. This theoretically optimal data block size refers to the optimal data block size that simultaneously balances data transmission efficiency and reliability under the current link conditions of bandwidth, latency, and packet loss rate. The unit is usually bytes. The core logic is as follows: when link bandwidth is high, latency is low, and packet loss rate is low, the theoretically optimal data block size can be appropriately increased to reduce data block header protocol overhead and improve effective data transmission efficiency; when link bandwidth is low, latency is high, and packet loss rate is high, the theoretically optimal data block size needs to be appropriately decreased to reduce the risk of single data block transmission failure, reduce the number of retransmissions, and avoid transmission timeouts or data loss due to excessively large data blocks.

[0031] Step 103: Determine whether the theoretically optimal data block size is within a preset range: If the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size; if the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size.

[0032] Specifically, to avoid the theoretically optimal data block size exceeding the hardware constraints of the ship's terminal or failing to meet the actual transmission capacity of the satellite link, it is necessary to compare it with a preset range to ultimately determine the actual data block size used. The aforementioned preset range refers to a reasonable range of data block sizes pre-set based on the ship's terminal's hardware storage capacity, data processing speed, and typical transmission characteristics of the satellite communication link. It includes a minimum threshold and a maximum threshold: the minimum threshold is typically set based on the minimum efficiency of a single data processing operation by the ship's terminal to avoid excessively small data blocks causing frequent processing of fragmented data and increasing CPU load; the maximum threshold is typically set based on the maximum stable data volume transmitted per satellite link operation to avoid excessively large data blocks exceeding the link's buffer capacity and triggering transmission interruptions.

[0033] For example, if the preset range is [512Byte, 8192Byte], when the theoretically optimal data block size is calculated to be 400Byte, which is less than the minimum threshold of 512Byte, the actual data block size is 512Byte; when the theoretically optimal data block size is calculated to be 9000Byte, which is greater than the maximum threshold of 8192Byte, the actual data block size is 8192Byte; when the theoretically optimal data block size is calculated to be 4096Byte, which is within the preset range, the actual data block size is directly taken as 4096Byte.

[0034] Step 104: Divide the fuel monitoring data into data blocks that match the determined actual data block size to obtain a block data sequence.

[0035] Specifically, fuel monitoring data refers to various monitoring parameters collected during the operation of a ship's fuel supply system, including data such as fuel level, fuel temperature, fuel pressure, fuel flow rate, engine speed, and sensor operating status. A block data sequence refers to an ordered set of data blocks formed by dividing continuous fuel monitoring data according to the actual data block size; each data block contains a segment of continuous monitoring data.

[0036] Step 105: Transmit the block data sequence with additional sequence number and verification information via satellite communication link.

[0037] Specifically, to ensure the receiving end can identify the transmission order of data blocks, determine whether data is lost, and verify data integrity, a sequence number and checksum information are appended to each data block in the segmented data sequence before being transmitted to the shore via a satellite communication link. The sequence number refers to a unique identifier assigned to each data block in the segmented data sequence, typically an incrementing integer, such as starting from 1. The receiving end can use the sequence number to determine whether the data blocks were transmitted in order and whether any are missing. For example, if data blocks numbered 1 and 3 are found, it can be determined that data block number 2 is missing. The checksum information refers to a check value generated based on the data block content using an algorithm such as Cyclic Redundancy Check (CRC32). After receiving the data block, the receiving end recalculates the checksum using the same algorithm. If it matches the appended checksum information, it indicates that the data block was not damaged during transmission; if they do not match, it indicates that the data block is damaged and a retransmission is required.

[0038] In existing technologies, when ship fuel monitoring data is transmitted via satellite links, a fixed data block size strategy can lead to excessive protocol overhead and reduced effective transmission efficiency when link bandwidth is high and latency is low. Conversely, a fixed large data block significantly increases the probability of packet loss and retransmission when link bandwidth is low and latency is high. This invention avoids reliance on fixed link assumptions by acquiring bandwidth, latency, and packet loss rate parameters reflecting the dynamic state of the link; it calculates the theoretically optimal data block size based on link parameters to achieve adaptation between data block size and link state; and it determines the actual data block size based on a preset range. This solves the problem of poor adaptation between data block size and link state in existing technologies, improving the data transmission efficiency of fuel monitoring data via satellite links and reducing the probability of packet loss and retransmission.

[0039] In some embodiments, obtaining the current bandwidth value, latency value, and packet loss rate based on the probe message includes:

[0040] Set initial link monitoring parameters; the initial link monitoring parameters include the initial detection frequency;

[0041] Based on the initial detection frequency, a detection message is sent via a satellite communication link; the detection message includes a timestamp, a sequence number, and a check field.

[0042] After receiving the probe message, the receiving end returns an acknowledgment message containing the original sequence number and the receiving timestamp;

[0043] The delay value is calculated based on the sending time of the probe message and the receiving time of the acknowledgment message;

[0044] The current bandwidth value is calculated based on the number of probe messages and the amount of data successfully transmitted per unit time.

[0045] The packet loss rate is calculated based on the ratio of the number of lost probe packets to the total number of probe packets sent.

[0046] The latency, bandwidth, and packet loss rate are output as link status parameters.

[0047] In some embodiments, the theoretically optimal data block size is calculated based on the current bandwidth value, latency value, and packet loss rate, including:

[0048] The current bandwidth value, latency value, and packet loss rate are used to construct a three-dimensional input feature vector;

[0049] The theoretically optimal data block size is obtained by inputting the three-dimensional input feature vector into a lightweight multilayer perceptron neural network model pre-deployed on the ship terminal;

[0050] The multilayer perceptron neural network model consists of multiple layers of neurons; neurons in adjacent layers transmit information through connections, and each connection has a corresponding weight parameter. By adjusting the weight parameter, a nonlinear mapping of the input features is achieved.

[0051] The specific structure of the multilayer perceptron neural network model includes:

[0052] Input layer; the input layer consists of three neuron nodes, which respectively receive the current bandwidth value, latency value, and packet loss rate;

[0053] Hidden layer; the hidden layer consists of multiple neuron nodes, uses the ReLU activation function, and extracts a high-order representation of the input features through nonlinear transformation;

[0054] Output layer: Consists of a single neuron node, which outputs an estimate of the theoretically optimal data block size;

[0055] After training, the multilayer perceptron neural network model undergoes model compression optimization, which specifically includes: analyzing the contribution of the connection weights between neuron nodes to the model output, and removing connections whose contribution is lower than a set threshold and their corresponding weight parameters.

[0056] Specifically, before inputting the multilayer perceptron neural network model, the acquired link state parameters are first converted into a format, integrating the bandwidth value, latency value, and packet loss rate into a three-dimensional input feature vector. The three-dimensional input feature vector refers to a vector structure in the form of [bandwidth value, latency value, packet loss rate], where bandwidth value, latency value, and packet loss rate are used as the three-dimensional feature components respectively. For example, if the current bandwidth value is 100kbps, the latency value is 500ms, and the packet loss rate is 5%, then the corresponding three-dimensional input feature vector is [100, 500, 5%]. Before inputting it into the model, the feature components in the vector also need to be standardized to map the values ​​to the [0,1] interval to avoid model inference bias caused by differences in parameter magnitudes.

[0057] The standardized 3D input feature vector is input into the local model of the ship terminal, and the model calculates and outputs the theoretically optimal data block size. The lightweight multilayer perceptron neural network model is an optimized model adapted to resource-constrained devices by simplifying the network structure based on the traditional multilayer perceptron, such as reducing the number of hidden layers, controlling the size of neurons, and optimizing it. This model needs to be trained offline and pass compatibility testing before deployment on the ship terminal.

[0058] A multilayer perceptron neural network model consists of multiple layers of neurons. Information is transmitted between neurons in adjacent layers through connections, each connection having a corresponding weight parameter. Adjusting these weight parameters enables a non-linear mapping of input features. To ensure the model accurately fits the complex relationship between link parameters and data block size, the core logic of the model's structure must be clearly defined: the basic unit of the model is the neuron node, with multiple neurons divided into layers to form a multilayer structure; neurons in adjacent layers are connected, each connection corresponding to a weight parameter; this weight parameter is a value learned during model training and is used to control the strength of information transmission from the previous layer to the next. The larger the absolute value of the weight parameter, the stronger the influence of the information transmission of the corresponding connection; through the coordinated adjustment of the weight parameters of multiple layers of neurons and the non-linear processing of the activation function, the model can achieve a non-linear mapping of input features.

[0059] The specific structure of the multilayer perceptron neural network model includes an input layer, a hidden layer, and an output layer. The input layer is the entry point for the link parameters to enter the model, and its structure needs to match the dimension of the input feature vector. In this application, the input layer is set with three neuron nodes. Each node only receives parameters of one dimension of the three-dimensional input feature vector. The first neuron node receives the bandwidth value, the second receives the delay value, and the third receives the packet loss rate.

[0060] Hidden layers are the core layers in a model that fit complex relationships. They are usually set to 1-2 layers, and the number of neurons in each layer is determined according to the complexity of the training data. The neurons in the hidden layers use the ReLU activation function. The core function of the hidden layers is to extract higher-order representations of the input features. That is, through multi-layer weight calculation and activation function processing, the input bandwidth, latency, and packet loss rate are transformed into abstract features that can reflect the complex relationship between the three. For example, high bandwidth, low latency, and low packet loss correspond to the feature logic that requires larger data blocks to improve efficiency.

[0061] The output layer is the output of the computation results. It has only one neuron node, and its output value is the estimated value of the theoretically optimal data block size. After receiving the high-order features transmitted from the hidden layer, the output layer neuron performs a linear transformation to map the abstract features into specific byte values ​​to obtain the final result. The output layer will round the result to ensure that the data block size is an integer.

[0062] In some embodiments, the method further includes:

[0063] A data buffer is established to store the link status parameters and corresponding actual transmission performance feedback obtained within the most recent N time windows. The link status parameters include bandwidth, latency, and packet loss rate. The actual transmission performance feedback includes the number of successfully transmitted data blocks, the number of retransmissions, and the effective throughput.

[0064] The model fine-tuning process is triggered when any of the following conditions are met: the amount of data stored in the buffer reaches the preset update threshold, or the distribution offset of the statistical distribution of the current link state parameters relative to the original training dataset exceeds the set tolerance threshold.

[0065] Model fine-tuning specifically includes:

[0066] The cached link state parameters are used as input features to the multilayer perceptron neural network model to obtain the predicted theoretical optimal data block size;

[0067] Obtain the actual transmission performance feedback corresponding to the input features as a supervision signal; calculate the performance evaluation index based on the supervision signal;

[0068] A loss function is constructed and a loss value is obtained, wherein the loss value is the difference between the expected performance score corresponding to the predicted theoretical optimal data block size and the performance evaluation score corresponding to the actual supervision signal;

[0069] The gradient descent algorithm is used to update the output layer weights of the multilayer perceptron neural network model step by step; during the update process, the weight parameters of the hidden layers are kept unchanged.

[0070] After updating the model weights, evaluate the performance of the updated model using the reserved validation dataset; calculate the prediction error of the model before and after the update on the validation dataset; if the prediction error of the updated model is...

[0071] If the reduction exceeds the preset improvement threshold, the updated model is accepted.

[0072] The updated model is used to replace the original model for calculating the theoretically optimal data block size.

[0073] Specifically, the data buffer refers to a dedicated storage area within the ship's terminal used for temporary storage of specific data. Its capacity is set according to the value of N and the size of a single data entry, and it adopts a first-in, first-out (FIFO) storage rule. Actual transmission performance feedback refers to the performance indicators generated during actual transmission after the data block size is determined based on the link status parameters within the window. Among these, the number of successfully transmitted data blocks refers to the total number of data blocks within the window that are not lost, not damaged, and confirmed by the receiving end; the number of retransmissions refers to the total number of times all data blocks within the window are retransmitted due to loss or damage; and the effective throughput refers to the amount of effective data successfully transmitted per unit time within the window.

[0074] The system sets dual trigger conditions to ensure timely fine-tuning of the model when there is sufficient data or significant changes in the link, avoiding excessively frequent or delayed fine-tuning. The preset update threshold refers to the minimum amount of cached data required to trigger fine-tuning. The statistical distribution offset refers to the difference between the probability distribution of the link state parameters in the current M consecutive time windows and the probability distribution of the original training dataset used during model training, usually calculated using KL divergence. The set tolerance threshold refers to the maximum range of distribution offset allowed. When the offset exceeds this threshold, it indicates that the current link state has exceeded the coverage of the original training data, and the original parameters of the model can no longer be adapted, requiring fine-tuning to be triggered.

[0075] The first step in fine-tuning is to use the model's prediction results generated from the cached data as the basis for subsequent loss calculations. Specifically, all link state parameters are extracted from the data cache, constructed into a three-dimensional input feature vector, and standardized. These vectors are then input one by one into the currently used multilayer perceptron neural network model. The model output corresponding to each input vector is recorded, which represents the theoretically optimal data block size for prediction in that link state. This process must ensure that the input order is consistent with the temporal order of the cached data to avoid data misalignment that could lead to incorrect matching of subsequent supervision signals.

[0076] To determine the quality of the model's predictions, the predicted values ​​need to be correlated with the actual transmission performance to generate a supervision signal and quantify it into a performance evaluation metric. The supervision signal refers to the actual transmission performance feedback within the same time window as the input features. Its purpose is to provide the model with a basis for judging the accuracy of its predictions. If the data block size predicted by the model under a certain link state corresponds to a high effective throughput and low retransmission count in the actual transmission performance feedback, then the prediction is accurate; otherwise, it is inaccurate. The performance evaluation metric quantifies the supervision signal into a calculable score. The calculation logic can be configured as needed; a higher score indicates a better actual transmission performance and more accurate model predictions.

[0077] If a static model, trained offline and then permanently deployed, is used, it cannot adapt to the dynamic changes in satellite links, and the parameters of the static model based on the original training data cannot match the new link state. This leads to a larger deviation in the predicted optimal data block size, a decrease in transmission efficiency, and a surge in retransmissions. If a fine-tuning method with full-layer weight updates is used, it will consume a large amount of computing power from the ship's terminal. This invention solves the technical problems of static models being unable to adapt to link changes and the high resource consumption and unstable performance of full-layer updates.

[0078] In some embodiments, the method further includes, prior to segmenting the fuel monitoring data into data blocks:

[0079] The status field in the fuel monitoring data is analyzed to identify whether there are any abnormal indicators in the fuel monitoring data; the abnormal indicators include fuel pressure change indicator, flow abnormal indicator, sensor failure indicator, liquid level abnormal indicator, and temperature abnormal indicator;

[0080] If the parsing result shows the presence of any anomaly flag, the corresponding data segment containing that anomaly flag will be marked as high priority; if the parsing result shows the absence of any anomaly flag, the corresponding data segment will be marked as normal priority.

[0081] For data segments marked as high priority, independent blocks are formed using a first preset upper limit data block size. The method for determining the first preset upper limit includes: obtaining the maximum acceptable transmission delay; statistically analyzing the transmission delay distribution corresponding to different data block sizes based on historical performance data of the satellite communication link; determining the maximum data block size that can meet real-time requirements based on the maximum acceptable transmission delay and the transmission delay distribution; and setting the maximum data block size as the first preset upper limit.

[0082] For data segments marked as ordinary priority, fuel monitoring data are divided into data blocks that match the determined actual data block size.

[0083] Specifically, before data segmentation, the status information of the fuel monitoring data is parsed to filter out abnormal data that needs to be transmitted first. The aforementioned status field is a specific data segment in the fuel monitoring data frame used to identify the system's operating status. Its content is generated in real time by the sensors or control units of the ship's fuel supply system and is used to quickly mark whether the data is associated with anomalies. Anomaly flags are identifier bits in the status field used to characterize specific anomaly types. Fuel pressure mutation flag: When the change in fuel pressure per unit time exceeds a preset safety threshold, the corresponding bit in the status field is set to 1, used to warn of pipeline blockage and leakage risks. Flow anomaly flag: When the fuel flow rate is lower than the minimum operating threshold or higher than the maximum safety threshold, the corresponding bit in the status field is set to 1, used to reflect abnormal fuel supply rates. Sensor fault flag: When the output signal of sensors such as fuel level and temperature exceeds the normal range, the corresponding bit in the status field is set to 1. The status field contains several identifiers: a flag indicating unreliable monitoring data; a liquid level anomaly flag (set to 1 when fuel level is below the minimum warning threshold or above the maximum safety threshold), and a temperature anomaly flag (set to 1 when fuel temperature is below the solidification threshold or above the high temperature warning threshold), indicating abnormal fuel physical condition. During analysis, reading each identifier in the status field allows for quick determination of any of these anomalies in the monitoring data.

[0084] Fuel monitoring data is prioritized based on the presence or absence of anomaly indicators. High-priority data requires priority transmission and real-time assurance; this type of data is related to the safety of the ship's fuel system and must be transmitted to shore as quickly as possible for fault diagnosis. Normal-priority data is used only for routine monitoring and has lower real-time requirements, such as routine data when fuel levels are stable; this data can be transmitted based on efficiency, while ensuring the transmission of high-priority data. Priority markers are appended to the header of the corresponding data segment.

[0085] To ensure the real-time transmission of high-priority data, a smaller-than-usual block size strategy must be adopted, with a clearly defined upper limit for block size. The first preset upper limit refers to the maximum allowed size of high-priority data blocks, and this value must be less than the minimum threshold within a preset range, ensuring smaller high-priority data block sizes and faster transmission. Ordinary priority data, without urgent security requirements, can be divided using the determined actual data block size, balancing transmission efficiency and resource consumption.

[0086] In some embodiments, after transmitting the segmented data sequence with additional sequence numbers and check information via a satellite communication link, the method further includes:

[0087] After receiving the segmented data sequence, the receiving end checks the continuity of the data block sequence based on the sequence number information of the received data blocks;

[0088] The same Cyclic Redundancy Check (CRC) algorithm used at the sending end is used to verify the check information of each data block;

[0089] Identify missing or corrupted data blocks based on continuous inspection and verification results;

[0090] If the missing data block belongs to the normal priority category, the missing data block is skipped directly without a retransmission request; based on the received normal priority data blocks, a time series interpolation algorithm is used to reconstruct the fuel state trend; the interpolation algorithm includes linear interpolation, polynomial interpolation, or Kalman filtering algorithm;

[0091] If the missing data block is of high priority, a high-priority retransmission request is immediately sent to the sender. The retransmission request includes the sequence number, priority, and timestamp information of the missing data block. A retransmission timeout timer is set, and if the retransmission data is not received within the timeout period, the retransmission request is resent.

[0092] Specifically, when the receiving end continuously receives the block data sequence, it uses the sequence number information attached to the data block header to determine whether the data transmission is disordered or missing; the continuity check refers to the receiving end sorting through the received data blocks in ascending order of sequence number to identify the situation of sequence number gaps.

[0093] While checking continuity, the receiving end needs to verify the integrity of each received data block to determine whether data corruption has occurred due to link interference. The sending end calculates a checksum based on the data block content and appends it to the end of the data block. The receiving end recalculates the checksum for the received data blocks using the same algorithm. The checksum information is the CRC checksum appended by the sending end. The receiving end compares the recalculated checksum with the received checksum information: if they match, it means the data block was not corrupted during transmission; if they do not match, it means the data block has bit errors or is missing, and must be determined as a corrupted data block.

[0094] For abnormal data blocks of ordinary priority, a skip and reconstruction strategy is adopted to avoid consuming satellite link bandwidth due to retransmission of ordinary data; the aforementioned direct skipping of missing data blocks means not initiating a retransmission request to the sender, reducing link interaction overhead; the time series interpolation algorithm refers to using received, temporally adjacent ordinary priority data blocks, such as missing data through linear interpolation. For abnormal data blocks of high priority, an immediate retransmission strategy is adopted to ensure that critical data associated with security is not lost.

[0095] In some embodiments, the contribution of connection weights between neuron nodes to the model output is analyzed, and connections with a contribution below a set threshold and their corresponding weight parameters are removed, including:

[0096] For a trained multilayer perceptron neural network model, iterate through all connection weights between neurons in adjacent layers. Calculate the sensitivity index for each connection weight. ;in Represents the index of the previous layer of neurons. Represents the index of the next layer of neurons; the sensitivity index Used to quantify the impact of connection weights on the accuracy of model output. Where L is the loss function value of the model on the preset validation set. This represents the gradient of the connection weights with respect to the loss function.

[0097] An extreme link scenario test set is constructed, which covers typical harsh link conditions in ship satellite communication, including high packet loss rate conditions, low bandwidth conditions, high latency conditions, and combinations of the above conditions; each condition in the test set includes corresponding link status parameters and a matching optimal data block size label.

[0098] All connection weights are assigned according to sensitivity metrics Sort the connections by weight from smallest to largest and select the least sensitive connections for pruning verification. Then, reset the least sensitive connection weights to zero while keeping other weight parameters unchanged. Input the extreme link scenario test set into the model and calculate the model's prediction error for data block size in each extreme scenario. If the prediction error of the model in all extreme scenarios does not exceed the preset tolerance after resetting to zero, it is determined that the connection has minimal impact on the model's accuracy and the connection is removed.

[0099] If the prediction error in any extreme scenario exceeds the preset tolerance, the connection will be retained.

[0100] Specifically, before pruning, it is necessary to identify connection weights that have little impact on model accuracy. The core of this is to quantify the degree of contribution through sensitivity metrics. In particular, this involves traversing the connection weights of all adjacent layers in the model. Calculate the sensitivity index S for each connection. ij ;| | represents the absolute value of the connection weight. The larger the absolute value of the weight, the stronger its impact on information transmission. Let L be the gradient of the weight with respect to the loss function L, reflecting the magnitude of the change in the loss function when the weight changes slightly. The larger the absolute value of the gradient, the more significant the impact of the weight change on the loss value of the model output. The product of L and L is S. ij The overall weighting reflects the degree of contribution; S ijThe smaller the value, the less impact the weight has on the model's output accuracy, and the more suitable it is to be pruned; conversely, the larger the value, the greater its contribution, and it should be retained. During the calculation, L is calculated using a preset validation set used during offline model training to ensure that the gradient reflects the contribution of the weight in actual application, rather than just fitting the training set.

[0101] Pruning needs to avoid the problem of acceptable accuracy in normal scenarios but a sharp drop in accuracy in extreme scenarios. Therefore, it is necessary to build a special extreme scenario test set to verify the pruning effect.

[0102] Reference Figure 2 The second embodiment of the present invention provides a monitoring data optimization system for a ship fuel supply system, comprising:

[0103] The first processing module 201 is configured to: send probe messages via a satellite communication link; and obtain current bandwidth, latency, and packet loss rate based on the probe messages.

[0104] The second processing module 202 is used to: calculate the theoretically optimal data block size based on the current bandwidth value, latency value, and packet loss rate;

[0105] The third processing module 203 is used to: determine whether the theoretically optimal data block size is within a preset range; if the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size; if the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size.

[0106] The fourth processing module 204 is used to: divide the fuel monitoring data into data blocks that match the determined actual data block size to obtain a block data sequence;

[0107] The fifth processing module 205 is used to: transmit the block data sequence with additional sequence number and verification information via a satellite communication link.

[0108] It should be noted that the monitoring data optimization system for a ship fuel supply system provided in this embodiment of the invention is used to execute all the process steps of the monitoring data optimization method for a ship fuel supply system in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0109] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0110] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0111] The embodiments, implementation methods, and related technical features of this application can be combined and substituted for each other without conflict.

[0112] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for optimizing monitoring data of a ship fuel supply system, characterized in that, include: Send probe messages via satellite communication link; The current bandwidth value, latency value, and packet loss rate are obtained based on the probe messages; The theoretically optimal data block size is calculated based on the current bandwidth, latency, and packet loss rate. Determine whether the theoretically optimal data block size is within a preset range: if the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size; if the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size. Fuel monitoring data is divided into data blocks that match the determined actual data block size to obtain a block data sequence; The segmented data sequence, with additional sequence numbers and verification information, is transmitted via a satellite communication link; The theoretically optimal data block size is calculated based on the current bandwidth, latency, and packet loss rate, including: The current bandwidth value, latency value, and packet loss rate are used to construct a three-dimensional input feature vector; The theoretically optimal data block size is obtained by inputting the three-dimensional input feature vector into a lightweight multilayer perceptron neural network model pre-deployed on the ship terminal; The multilayer perceptron neural network model consists of multiple layers of neurons; neurons in adjacent layers transmit information through connections, and each connection has a corresponding weight parameter. By adjusting the weight parameter, a nonlinear mapping of the input features is achieved. The specific structure of the multilayer perceptron neural network model includes: Input layer; the input layer consists of three neuron nodes, which respectively receive the current bandwidth value, latency value, and packet loss rate; Hidden layer; the hidden layer consists of multiple neuron nodes, uses the ReLU activation function, and extracts a high-order representation of the input features through nonlinear transformation; Output layer: Consists of a single neuron node, which outputs an estimate of the theoretically optimal data block size; The multilayer perceptron neural network model is compressed and optimized after training, specifically including: analyzing the contribution of the connection weights between neuron nodes to the model output, and removing connections whose contribution is lower than a set threshold and their corresponding weight parameters. The method further includes: A data buffer is established to store the link status parameters and corresponding actual transmission performance feedback obtained within the most recent N time windows. The link status parameters include bandwidth, latency, and packet loss rate. The actual transmission performance feedback includes the number of successfully transmitted data blocks, the number of retransmissions, and the effective throughput. The model fine-tuning process is triggered when any of the following conditions are met: the amount of data stored in the buffer reaches the preset update threshold, or the distribution offset of the statistical distribution of the current link state parameters relative to the original training dataset exceeds the set tolerance threshold. Model fine-tuning specifically includes: The cached link state parameters are used as input features to the multilayer perceptron neural network model to obtain the predicted theoretical optimal data block size; Obtain the actual transmission performance feedback corresponding to the input features as a supervision signal; calculate the performance evaluation index based on the supervision signal; A loss function is constructed and a loss value is obtained, wherein the loss value is the difference between the expected performance score corresponding to the predicted theoretical optimal data block size and the performance evaluation score corresponding to the actual supervision signal; The gradient descent algorithm is used to update the output layer weights of the multilayer perceptron neural network model step by step; during the update process, the weight parameters of the hidden layers are kept unchanged. After updating the model weights, evaluate the performance of the updated model using the reserved validation dataset; calculate the prediction error of the model before and after the update on the validation dataset; if the prediction error of the updated model is... If the reduction exceeds the preset improvement threshold, the updated model is accepted. The updated model is used to replace the original model for calculating the theoretically optimal data block size.

2. The method according to claim 1, characterized in that, Based on the probe messages, the current bandwidth value, latency value, and packet loss rate are obtained, including: Set initial link monitoring parameters; the initial link monitoring parameters include the initial detection frequency; Based on the initial detection frequency, a detection message is sent via a satellite communication link; the detection message includes a timestamp, a sequence number, and a check field. After receiving the probe message, the receiving end returns an acknowledgment message containing the original sequence number and the receiving timestamp; The delay value is calculated based on the sending time of the probe message and the receiving time of the acknowledgment message; The current bandwidth value is calculated based on the number of probe messages and the amount of data successfully transmitted per unit time. The packet loss rate is calculated based on the ratio of the number of lost probe packets to the total number of probe packets sent. The latency, bandwidth, and packet loss rate are output as link status parameters.

3. The method according to claim 2, characterized in that, Before segmenting the fuel monitoring data into data blocks, the method further includes: The status field in the fuel monitoring data is analyzed to identify whether there are any abnormal indicators in the fuel monitoring data; the abnormal indicators include fuel pressure change indicator, flow abnormal indicator, sensor failure indicator, liquid level abnormal indicator, and temperature abnormal indicator; If the parsing result shows the presence of any anomaly flag, the corresponding data segment containing that anomaly flag will be marked as high priority; if the parsing result shows the absence of any anomaly flag, the corresponding data segment will be marked as normal priority. For data segments marked as high priority, independent blocks are formed using a first preset upper limit data block size. The method for determining the first preset upper limit includes: obtaining the maximum acceptable transmission delay; statistically analyzing the transmission delay distribution corresponding to different data block sizes based on historical performance data of the satellite communication link; determining the maximum data block size that can meet real-time requirements based on the maximum acceptable transmission delay and the transmission delay distribution; and setting the maximum data block size as the first preset upper limit. For data segments marked as ordinary priority, fuel monitoring data are divided into data blocks that match the determined actual data block size.

4. The method according to claim 3, characterized in that, After transmitting the segmented data sequence with additional sequence numbers and check information via a satellite communication link, the method further includes: After receiving the segmented data sequence, the receiving end checks the continuity of the data block sequence based on the sequence number information of the received data blocks; The same Cyclic Redundancy Check (CRC) algorithm used at the sending end is used to verify the check information of each data block; Identify missing or corrupted data blocks based on continuous inspection and verification results; If the missing data block belongs to the normal priority category, the missing data block is skipped directly without a retransmission request; based on the received normal priority data blocks, a time series interpolation algorithm is used to reconstruct the fuel state trend; the interpolation algorithm includes linear interpolation, polynomial interpolation, or Kalman filtering algorithm; If the missing data block is of high priority, a high-priority retransmission request is immediately sent to the sender. The retransmission request includes the sequence number, priority, and timestamp information of the missing data block. A retransmission timeout timer is set, and if the retransmission data is not received within the timeout period, the retransmission request is resent.

5. The method according to claim 4, characterized in that, Analyze the contribution of connection weights between neuron nodes to the model output, and remove connections and their corresponding weights that contribute less than a set threshold, including: For a trained multilayer perceptron neural network model, iterate through all connection weights between neurons in adjacent layers. Calculate the sensitivity index for each connection weight. ;in Represents the index of the previous layer of neurons. Represents the index of the next layer of neurons; the sensitivity index Used to quantify the impact of connection weights on the accuracy of model output. Where L is the loss function value of the model on the preset validation set. This represents the gradient of the connection weights with respect to the loss function. An extreme link scenario test set is constructed, which covers typical harsh link conditions in ship satellite communication, including high packet loss rate conditions, low bandwidth conditions, high latency conditions, and combinations of the above conditions; each condition in the test set includes corresponding link status parameters and a matching optimal data block size label. All connection weights are sorted by sensitivity index Sort the connections by weight from smallest to largest and select the least sensitive connections for pruning verification. Then, reset the least sensitive connection weights to zero while keeping other weight parameters unchanged. Input the extreme link scenario test set into the model and calculate the model's prediction error for data block size in each extreme scenario. If the prediction error of the model in all extreme scenarios does not exceed the preset tolerance after resetting to zero, it is determined that the connection has minimal impact on the model's accuracy and the connection is removed. If the prediction error in any extreme scenario exceeds the preset tolerance, the connection will be retained.

6. A monitoring data optimization system for a ship fuel supply system, characterized in that, include: The first processing module is used to: send probe messages via a satellite communication link; The current bandwidth value, latency value, and packet loss rate are obtained based on the probe messages; The second processing module is used to calculate the theoretically optimal data block size based on the current bandwidth value, latency value, and packet loss rate. The third processing module is used to: determine whether the theoretically optimal data block size is within a preset range; if the theoretically optimal data block size is less than the minimum threshold of the preset range, then the minimum threshold is used as the actual data block size. If the theoretically optimal data block size is greater than the maximum threshold of the preset range, then the maximum threshold is used as the actual data block size; if the theoretically optimal data block size is within the preset range, then the theoretically optimal data block size is used as the actual data block size. The fourth processing module is used to: divide the fuel monitoring data into data blocks that match the determined actual data block size to obtain a block data sequence; The fifth processing module is used to: transmit the block data sequence with additional sequence numbers and verification information via a satellite communication link; The theoretically optimal data block size is calculated based on the current bandwidth, latency, and packet loss rate, including: The current bandwidth value, latency value, and packet loss rate are used to construct a three-dimensional input feature vector; The theoretically optimal data block size is obtained by inputting the three-dimensional input feature vector into a lightweight multilayer perceptron neural network model pre-deployed on the ship terminal; The multilayer perceptron neural network model consists of multiple layers of neurons; neurons in adjacent layers transmit information through connections, and each connection has a corresponding weight parameter. By adjusting the weight parameter, a nonlinear mapping of the input features is achieved. The specific structure of the multilayer perceptron neural network model includes: Input layer; the input layer consists of three neuron nodes, which respectively receive the current bandwidth value, latency value, and packet loss rate; Hidden layer; the hidden layer consists of multiple neuron nodes, uses the ReLU activation function, and extracts a high-order representation of the input features through nonlinear transformation; Output layer: Consists of a single neuron node, which outputs an estimate of the theoretically optimal data block size; The multilayer perceptron neural network model is compressed and optimized after training, specifically including: analyzing the contribution of the connection weights between neuron nodes to the model output, and removing connections whose contribution is lower than a set threshold and their corresponding weight parameters. Also includes: A data buffer is established to store the link status parameters and corresponding actual transmission performance feedback obtained within the most recent N time windows. The link status parameters include bandwidth, latency, and packet loss rate. The actual transmission performance feedback includes the number of successfully transmitted data blocks, the number of retransmissions, and the effective throughput. The model fine-tuning process is triggered when any of the following conditions are met: the amount of data stored in the buffer reaches the preset update threshold, or the distribution offset of the statistical distribution of the current link state parameters relative to the original training dataset exceeds the set tolerance threshold. Model fine-tuning specifically includes: The cached link state parameters are used as input features to the multilayer perceptron neural network model to obtain the predicted theoretical optimal data block size; Obtain the actual transmission performance feedback corresponding to the input features as a supervision signal; calculate the performance evaluation index based on the supervision signal; A loss function is constructed and a loss value is obtained, wherein the loss value is the difference between the expected performance score corresponding to the predicted theoretical optimal data block size and the performance evaluation score corresponding to the actual supervision signal; The gradient descent algorithm is used to update the output layer weights of the multilayer perceptron neural network model step by step; during the update process, the weight parameters of the hidden layers are kept unchanged. After updating the model weights, evaluate the performance of the updated model using the reserved validation dataset; calculate the prediction error of the model before and after the update on the validation dataset; if the prediction error of the updated model is... If the reduction exceeds the preset improvement threshold, the updated model is accepted. The updated model is used to replace the original model for calculating the theoretically optimal data block size.

Images