An intelligent ship communication network integration monitoring method and device

Abstract

The application provides a kind of intelligent ship communication network integration monitoring method and device, it is related to ship network communication technical field.The intelligent ship communication network integration monitoring method includes data acquisition, constructs cloud computing platform, data mining processing, feature extraction, model establishment, adjustment channel, task allocation and alarm and control steps.The application is combined by depth reinforcement learning model and alliance block chain network, realizes the dynamic adjustment of channel and task, when there is new data mining task, each node uploads the state information of itself to block chain according to the rules in smart contract, after the information is verified and confirmed by the consensus mechanism in block chain network, the size and distribution of data subset are dynamically adjusted according to the real-time energy consumption of node, ensure the efficient use of network resources, reduce time delay, improve the timeliness of data analysis, make intelligent ship can quickly respond to network changes, guarantee the safety of navigation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ship network communication, and in particular to an intelligent ship communication network integrated monitoring method and device. BACKGROUND

[0002] An intelligent ship refers to a ship that utilizes sensor, communication, Internet of Things, Internet, and other technical means to automatically perceive and obtain information and data of the ship itself, marine environment, logistics, port, and the like, and realizes intelligent operation in ship navigation, management, maintenance, and cargo transportation based on computer technology, automatic control technology, and big data processing and analysis technology.

[0003] In the use process of the intelligent ship, communication network monitoring is crucial. The intelligent ship needs to obtain data of marine environment, weather conditions, channel information, and the like in real time to make correct navigation decisions. These data are usually transmitted from a shore-based system or other ships to the intelligent ship through the communication network. If the communication network fails, the intelligent ship will not be able to obtain the information in time, which may face navigation risks. Communication network monitoring can timely discover network failures or abnormalities to ensure stable transmission of data, thereby ensuring the navigation safety of the intelligent ship. With the continuous development of intelligent ship technology, more and more intelligent functions will depend on the communication network.

[0004] The existing ship is affected by wind, waves, weather, distance, and other factors when sailing on the sea, resulting in unstable or interrupted signals, which in turn affects real-time transmission of data. With the expansion of the network scale, time delay data analysis efficiency is low, time delay is large, the credibility of the network signal channel is reduced, the state of each key device cannot be grasped in real time, and safety events cannot be predicted in advance to prevent accidents.

[0005] Therefore, it is necessary to provide a new intelligent ship communication network integrated monitoring method and device to solve the above technical problems. SUMMARY

[0006] In view of the technical problems in the prior art that the intelligent ship communication network has low time delay data analysis efficiency, large time delay, reduced credibility of the network signal channel, cannot grasp the state of each key device in real time, and cannot predict safety events in advance to prevent accidents, the present application provides an intelligent ship communication network integrated monitoring method and device.

[0007] To achieve the above purposes, the technical solution adopted by the present application is as follows:

[0008] An intelligent ship communication network integrated monitoring method, the method comprising the following steps:

[0009] S1, data acquisition: using sensors and instruments, obtaining ship image network data, running parameters and external related data; wherein, the running parameters include temperature, pressure, water level and smoke concentration; the external related data includes marine weather data, port surrounding traffic data;

[0010] S2, building a cloud computing platform: according to the image network data and running parameters, building a ship image network data transmission platform based on the cloud computing platform, optimizing the transmission process of the ship image network data transmission platform, obtaining a data transmission channel with strong anti-interference ability, the cloud computing platform pre-processes the received ship image network data, and transmits the pre-processed ship image network data to the monitoring center through the data transmission channel; wherein, the step of optimizing the transmission process of the ship image network data transmission platform is as follows:

[0011] Step 1, first, the correction value of the energy attenuation degree of the intelligent ship communication network channel is calculated, and the correction value is obtained to reflect the channel energy attenuation;

[0012] Step 2, according to the channel energy attenuation, the intelligent ship image network data is divided into several subsets, and the creation of the intelligent ship image network data transmission channel balanced signal attenuation model is realized;

[0013] Step 3, synchronously create intelligent ship communication network channel credibility set: the intelligent ship communication network synchronously sends multiple signals, the signals are transmitted according to the set rules, for the intelligent ship communication network channel credibility set, wherein, represents the credibility of the i-th channel, and i=1, 2, 3,..., j, j represents the number of channels in the network;

[0014] Step 4, by collecting the multi-path signals in the intelligent ship communication network credibility set, avoiding the mutual interference between nodes and channels when the network runs, optimizing the data transmission channel;

[0015] Step 5, according to the optimization result of the data transmission channel, realizing the transmission of the ship image network data, the total running period gap expression in the transmission process is as follows:

[0016]

[0017] wherein, is the running period of the K-th network node; is the running period gap between network node o and adjacent relay node r, unit is second, The runtime period of the network node o, the runtime period gap represents the time interval between two continuous data transmission tasks, and the length of the runtime period gap depends on the processing capacity of the network node and the current network load condition;

[0018] S3, data mining processing: the preprocessed ship image network data is segmented in the monitoring center, the Map Reduce framework is configured for parallel processing, and the parallel K-means clustering algorithm is adopted, so that the ship image network data is classified and mined;

[0019] S4, feature extraction: the ship image network data is fused with external related data, the convolutional neural network is used for feature extraction of the fused image data, and the recurrent neural network and its variants are used for feature extraction of the ship running parameter data, so that the image features and running parameter features of the ship are obtained;

[0020] S5, model establishment: the image features and running parameter features are fused through the multi-modal feature fusion technology to form a comprehensive feature vector, a deep reinforcement learning model is constructed by using a deep Q network, the comprehensive feature vector is input, the learning optimal channel and task allocation strategy are output through the interaction with the environment; wherein, the comprehensive feature vector includes visual information and running state information, the visual information includes the surrounding environment and obstacles of the ship, and the running state information includes the speed, direction and acceleration of the ship;

[0021] The deep reinforcement learning model constructed by using the deep Q network includes the following operation steps:

[0022] 1) Network architecture design: a deep Q network DQN containing 3 fully connected layers is constructed, the input layer dimension is the comprehensive feature vector dimension, and the output layer dimension is the action space size, corresponding to the combination of different channel, power and task acceptance decision;

[0023] Among them, the action space includes action composition and action vector representation, the action composition: the action of the agent includes selecting different channels for data transmission; adjusting the data transmission power, including high, medium and low three levels; accepting or rejecting new tasks, including accepting or rejecting two choices; the action vector representation: combining the action into an action vector: a= (c, p, t), wherein c is channel selection, p is transmission power, and t is task acceptance decision;

[0024] 2) Experience replay mechanism: an experience replay buffer is established for storing experience samples of the agent interacting with the environment;

[0025] 3) Target network update: initialize the main network and the target network, the parameters are the same, and every certain training step, copy the parameters of the main network to the target network;

[0026] 4) Training: design the reward function according to the task completion, energy consumption and channel utilization efficiency;

[0027] The training process is as follows: first, taking the intelligent ship communication network as the environment, regarding the network node as an agent, defining the state space and action space of the agent, the agent selects an action a according to the current state s using the e-greedy strategy, the e-greedy strategy randomly selects an action with a probability of , and selects the action with the maximum Q value with a probability of 1- , the agent performs the action in the environment to obtain the immediate reward r, which represents the immediate reward obtained after performing the action a in the state s, and moves to the next state s', stores (s, a, r, s') in the experience replay buffer, randomly extracts a certain number of samples from the experience replay buffer for training the main network, for each sample, calculates the target Q value , wherein is a discount factor, Q1 is the Q value function of the target network, is a discount factor, , which represents the Q value of the action a' selected in the next state s' so that the Q value is maximum; using the mean square error loss function to update the parameters of the main network, wherein Q2 is the Q value function of the main network, , which represents the Q value estimate given by the main network Q when taking action ai in state si, repeating the above training steps until the deep reinforcement learning model converges;

[0028] S6, channel and task allocation adjustment: building a consortium blockchain in the intelligent ship communication network, recording node energy, task execution and channel usage information, designing a smart contract to clearly define the channel and task allocation rule process, dynamically adjusting the data subset allocation according to the real-time state of the node, updating the node credit points, and reasonably optimizing the management of the channel and the task;

[0029] S7, alarm and control: continuously monitoring the ship state according to the updated node credit points, image signals and running parameters, and setting up an alarm mechanism according to the ship state.

[0030] Further, in the S1, the sensors include temperature sensors, pressure sensors, water level sensors and smoke sensors; the instruments include speedometers and tachometers.

[0031] Further, in the Step1, the correction value of the energy attenuation degree of the intelligent ship communication network channel is calculated, and the specific method is as follows:

[0032] In the calculation of the correction value of the channel energy attenuation degree of the intelligent ship communication network, first, by measuring the signal strength under different distances and conditions, a signal attenuation model is established;

[0033] The operation formula of the correction value is:

[0034]

[0035] Among them, The correction value of the channel energy attenuation, The actual measured signal energy, The theoretical signal energy, The environmental attenuation coefficient, d is the signal propagation distance; By comparing the actual measurement and the theoretical calculation value, the correction value is obtained to reflect the channel energy attenuation;

[0036] In the Step2, the signal attenuation model is created as follows:

[0037]

[0038] Among them, p is the total amount of scheduling association attributes of the intelligent ship communication network data terminal; d and c are the output control parameters and the number of pulse bandwidth of the channel respectively; E(y) is the joint function of time and scale; y is the clustering unit set of the channel; a and b represent the weight coefficients of signal strength and energy efficiency;

[0039] In the calculation of the weight coefficients of signal strength and energy efficiency, the formula used is:

[0040]

[0041] In the formula, Metric is a comprehensive evaluation index, which is used to quantify the advantages and disadvantages of channel selection; SignalStrength is the signal strength, and EnergyEfficiency is the energy efficiency;

[0042] In the Step3, the process of creating the intelligent ship communication network channel credibility set is as follows:

[0043] Signal synchronization transmission: each node in the intelligent ship communication network synchronously sends multiple signals, and the multiple signals are transmitted according to the preset rules;

[0044] Multi-path signal collection: the receiving node collects signals from different paths;

[0045] Credibility evaluation: according to the number, strength and signal quality of the collected multi-path signals, the credibility of each channel is evaluated;

[0046] Credibility set creation: Group the credibility assessment results of each channel into a set, called the channel credibility set ;

[0047] In Step 4, the data transmission channel is optimized based on the number of collected multipath signals, and the optimization control formula is:

[0048]

[0049] In Step 5, the implementation of ship image network data transmission includes the following steps:

[0050] Data blocking and scheduling: The ship image network data to be transmitted is divided into multiple data blocks, and scheduling is performed according to the load condition of the network node and the channel quality;

[0051] Channel selection and allocation: According to the current network state and channel credibility set, the optimal channel for data transmission is selected; at the same time, the size and allocation of the data subset are dynamically adjusted according to the real-time energy consumption of the node;

[0052] Data transmission and verification: Transmit the data blocks through the selected channel, and perform verification at the receiving end to ensure the integrity and accuracy of the data. During transmission, forward error correction coding technology is used to improve the reliability of data transmission;

[0053] Transmission efficiency evaluation: According to the actual situation of data transmission, evaluate the transmission efficiency.

[0054] Further, the operation of the parallel K-means clustering algorithm in Step S3 includes the following steps:

[0055] Step 1: According to the obtained image signal and running parameters, randomly select a ship image network data as target data, and set the target data as the center value of a single cluster and the original average value;

[0056] Step 2: Based on the average distance between the target data and each cluster, use dynamic weighted Euclidean distance to assign such data to the cluster with the smallest distance;

[0057] Among them, the use of dynamic weighted Euclidean distance to assign such data to the cluster with the smallest distance, specifically includes:

[0058] Feature vector normalization: For ship image network data other than the target data, extract its comprehensive feature vector X j =[x j1 ,x j2 ,...,x jn ]n is the feature dimension, and it is mapped to the interval [0,1] through linear normalization:

[0059]

[0060] wherein, is the set of all sample values of the kth feature; denotes the original value of the jth sample on the kth feature; denotes the value of the jth sample on the kth feature after linear normalization, mapped to the interval [0, 1];

[0061] Dynamic weighted distance calculation: define the cluster center vector as:

[0062] , calculate the weighted Euclidean distance between sample and cluster :

[0063]

[0064] wherein, the weight coefficient is dynamically adjusted according to the contribution of the feature to the clustering result, satisfying: equal to 1;

[0065] Optimal cluster assignment: assign sample to the cluster with the smallest distance:

[0066]

[0067] wherein, denotes the cluster to which sample is assigned, is the jth sample in the data set; arcmin refers to finding the cluster with the smallest distance under the weighted distance measurement; denotes the weighted distance; denotes the center of the ith cluster;

[0068] If there are multiple clusters with the same distance, the sample is preferentially assigned to the cluster with fewer samples to maintain balanced cluster size;

[0069] Step three, calculate the average value of each cluster, repeat the above steps until the criterion function of the algorithm converges.

[0070] Further, in the S4, a convolutional neural network CNN is used to extract features from the fused image data, including:

[0071] Normalization: scale the image pixel value to the interval [0, 1] by the formula , wherein x is the original pixel value and X1 is the normalized value;

[0072] After normalization, the image size is unified, and the size is adjusted using the bilinear interpolation algorithm;

[0073] The rotation angle range is set, and random rotation is performed;

[0074] A CNN model containing 5 convolutional layers, 3 pooling layers and 2 fully connected layers is constructed, the convolutional layer uses a 3x3 convolutional kernel, the step is 1, and the padding method is same, the pooling layer uses a 2x2 maximum pooling, the step is 2, the convolutional layer uses a ReLU activation function, the formula is f(x)=max(0,x); The Softmax activation function is used after the fully connected layer for multi-classification output, the fused image dataset is divided into training set, validation set and test set, the cross-entropy loss function is used, the formula is , wherein is the true label, is the predicted probability, the Adam optimization algorithm is used, the training set data is input into the model, the loss is calculated by forward propagation, the model parameters are updated by back propagation, the model performance is evaluated using the validation set every epoch, when the validation set loss no longer decreases, the preprocessed image data is input into the trained convolutional neural network CNN model, and the output of the last fully connected layer is taken as the image feature vector.

[0075] Further, the recurrent neural network is LSTM, and the feature extraction of the ship running parameter data using LSTM includes the following steps:

[0076] 1) Data preprocessing: remove outliers in the running parameter data, scale the running parameter data to the [-1, 1] interval, and use the min-max normalization formula , wherein and are the minimum and maximum values of the running parameter data, respectively, and the running parameter data is arranged in time sequence to form a time series dataset;

[0077] 2) Model construction: construct an LSTM model containing 2 layers of LSTM layers and 1 fully connected layer, set a certain number of hidden units for each LSTM layer, use a tanh activation function in the LSTM layer, the formula is: , and the fully connected layer uses a ReLU activation function;

[0078] 3) Model training: divide the time series dataset into training set, validation set and test set, use mean square error (MSE) loss function, adopt RMSprop optimization algorithm, input the training set data into the model, calculate the loss by forward propagation, update the model parameters by back propagation, evaluate the model performance using the validation set after training an epoch, stop training when the validation set loss no longer decreases, at this time, input the preprocessed driving parameter time series dataset into the trained LSTM model, and take the hidden state of the last time step as the driving parameter feature vector.

[0079] Further, in the S6, the alliance blockchain network is built in the ship communication network, each network node joins as a node of the blockchain, the distributed ledger feature of the blockchain is used to record the energy consumption, task execution and channel usage of each node, the smart contract is designed to define the rules and processes of channel and task allocation, when there is a new data mining task, each node uploads its energy state, channel quality and task execution capability to the blockchain according to the rules in the smart contract, the consensus mechanism in the blockchain network verifies and confirms the information uploaded by each node, only the information that has passed the consensus will be recorded in the blockchain, according to the real-time energy consumption of the node, the size and allocation of the data subset are dynamically adjusted, when a node has low energy, the size of the data subset allocated to the node is reduced, and part of the task is transferred to a node with sufficient energy, after the node executes the task, the task execution result and energy consumption are uploaded to the blockchain again, the smart contract verifies the task execution result, and updates the credit score of the node according to the verification result.

[0080] On the other hand, the application also provides a device suitable for the intelligent ship communication network integrated monitoring method described above, the device comprises:

[0081] A data acquisition unit for acquiring image network data, driving parameters and external related data of the ship;

[0082] A cloud computing platform construction unit for constructing a ship image network data transmission platform;

[0083] A data mining processing unit for segmenting and classifying and mining the preprocessed ship image network data;

[0084] A feature extraction unit for fusing the ship image network data and external related data, using convolutional neural network to extract features from the fused image data, and using recurrent neural network and its variants to extract features from the driving parameter data of the ship, to obtain image features and driving parameter features of the ship;

[0085] The model establishing unit fuses image features and driving parameter features through a multi-modal feature fusion technology to form a comprehensive feature vector, constructs a deep reinforcement learning model using a deep Q network, takes the comprehensive feature vector as input, trains through interaction with the environment, and outputs a learning optimal channel and task allocation strategy.

[0086] The channel and task allocation adjusting unit is used for building a consortium blockchain in the intelligent ship communication network, recording node energy, task execution and channel usage information, designing a smart contract to clearly define channel and task allocation rule processes, dynamically adjusting data subset allocation according to real-time node states, updating node credit points, and reasonably optimizing channel and task management.

[0087] The alarm and control unit continuously monitors the state of the ship and detects abnormalities or potential risks, triggers an alarm mechanism and takes control measures if there are abnormalities or potential risks.

[0088] Further, the device further comprises:

[0089] The data storage and management unit is responsible for storing the collected ship image signals, driving parameters and the results after data mining processing on the cloud or local server.

[0090] The network communication unit is responsible for establishing and maintaining a connection with the intelligent ship communication network node.

[0091] The user interface and interaction unit is used to provide an intuitive user interface to display the real-time state of the ship and the data mining results.

[0092] The system monitoring and maintenance unit monitors the running state of each unit of the system.

[0093] Compared with the prior art, the beneficial effects of the present application are:

[0094] 1. The present application uses the distributed ledger characteristics of the blockchain to record the energy consumption, task execution and channel usage state of each node, only the information that has passed the consensus will be recorded in the blockchain, ensuring the data in the transmission and storage process is tamper-proof and complete, avoiding data being stolen, tampered or forged in the transmission process, and further improving the reliability of data transmission.

[0095] 2、The application realizes dynamic adjustment of channels and tasks by combining a deep reinforcement learning model and a consortium blockchain network, when there is a new data mining task, each node uploads its state information to the blockchain according to the rules in the smart contract, the consensus mechanism in the blockchain network verifies and confirms the information, and then dynamically adjusts the size and allocation of the data subset according to the real-time energy consumption of the node, ensures efficient use of network resources, reduces latency, improves the timeliness of data analysis, enables the intelligent ship to quickly respond to network changes, and ensures navigation safety;

[0096] 3、The application adopts a data mining and model prediction mode, optimizes the navigation route, improves the operation efficiency and reduces the operation cost, and through continuous monitoring of the ship state and setting up an alarm mechanism, potential safety hazards can be found and handled in time, and the safety and reliability of the ship are improved;

[0097] 4、In the channel and task allocation adjustment process, the application designs a smart contract to define the rules and processes of channel and task allocation, the smart contract automatically executes the task allocation algorithm, dynamically adjusts the task allocation according to the node energy, channel quality, task priority and other factors, avoids human intervention and subjective factors, ensures the fairness and rationality of channel and task allocation, and at the same time, the smart contract also verifies the task execution result, updates the credit points of the node according to the verification result, the credit points are used as a reference basis for subsequent task allocation and resource allocation, encourages the node to actively complete the task and maintain a good network state, further enhances the credibility of the network channel, dynamically adjusts the size and allocation of the data subset, reduces the size of the data subset allocated to the node when the energy of the node is low, and transfers part of the task to the node with sufficient energy, realizes reasonable allocation and optimal utilization of network resources, improves the overall performance of the network, reduces latency, reduces network congestion, and ensures stable operation of the intelligent ship communication network.

[0098] Other features and advantages of the embodiments of the application will be described in detail in the subsequent specific embodiments part. BRIEF DESCRIPTION OF DRAWINGS

[0099] Figure 1 is a flow chart of the integrated monitoring method of the intelligent ship communication network according to the application;

[0100] Figure 2 is a structure block diagram of the ship image network data transmission platform architecture according to the application. DETAILED DESCRIPTION

[0101] In order to make the person skilled in the art better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application, so that the purpose, characteristics and advantages of the present application can be better understood. It should be understood that the embodiments shown in the drawings are not a limitation on the scope of the present application, but only to illustrate the essential spirit of the technical solutions of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should belong to the scope of protection of the present application.

[0102] Unless the context clearly requires otherwise, throughout the description and the claims, the words "comprise", "comprising", and the like are to be construed in an open, inclusive sense, as opposed to a closed or exclusive sense, that is as "including, but not limited to".

[0103] Reference throughout this specification to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Thus, the appearances of the phrase "in one embodiment" or "in an embodiment" in various places throughout this specification are not necessarily all referring to the same embodiment. Furthermore, the particular features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0104] As used in this specification and the appended claims, the singular forms "a", "an", and "the" include plural referents unless the content clearly dictates otherwise. It should be noted that the term "comprising" or "comprises" as used in this specification and the appended claims is not meant to exclude the presence of elements other than those listed in the claims.

[0105] In the following description, for the purpose of clarity, directional terms such as "front", "back", "left", "right", "upper", "lower", "inner", "outer", "over", "under", and "on" are used with reference to the orientation of the illustration of the figures as placed in the drawing, although it is understood that the present application can assume different orientations when placed in other use or environmental conditions.

[0106] The implementation details of the embodiments of the present application will be specifically described below with reference to the drawings. The following details are provided for the convenience of understanding only, and are not essential for implementing the present solution.

[0107] In some embodiments, the intelligent ship communication network integrated monitoring method of the present application has the process as shown in Figure 1 The operation steps include the following steps:

[0108] S1, data acquisition: using sensors and instruments to obtain image network data, running parameters and external related data of the ship;

[0109] The sensors include a temperature sensor, a pressure sensor, a water level sensor, and a smoke sensor.

[0110] The instruments include a speedometer and a tachometer.

[0111] The external related data includes marine weather data and port surrounding traffic data.

[0112] The running parameters include temperature, pressure, water level, and smoke concentration.

[0113] S2, constructing a cloud computing platform: according to the image network data and the running parameters, a ship image network data transmission platform is constructed based on the cloud computing platform, the transmission process of the ship image network data transmission platform is optimized, a data transmission channel with strong anti-interference ability is obtained, the cloud computing platform pre-processes the received ship image network data, and transmits the pre-processed ship image network data to the monitoring center through the data transmission channel.

[0114] The pre-processing of the received ship image network data includes denoising and compression of the data.

[0115] Further, the step of optimizing the transmission process of the ship image network data transmission platform is as follows:

[0116] Step 1, first, the correction value of the energy attenuation degree of the intelligent ship communication network channel is calculated, and the correction value is obtained to reflect the channel energy attenuation;

[0117] Step 2, according to the channel energy attenuation, the intelligent ship image network data is divided into several subsets, and the creation of the intelligent ship image network data transmission channel balanced signal attenuation model is realized;

[0118] Step 3, synchronously create an intelligent ship communication network channel credibility set: the intelligent ship communication network synchronously sends multiple signals, the signals are transmitted according to the set rules, for the intelligent ship communication network channel credibility set, wherein, represents the credibility of the i-th channel, and i=1, 2, 3,..., j, j represents the number of channels in the network.

[0119] Step 4, by collecting the multi-path signals in the intelligent ship communication network credibility set, avoiding mutual interference between nodes and channels during the operation of the network, the data transmission channel is optimized;

[0120] Step 5, according to the optimization result of the data transmission channel, the transmission of the ship image network data is realized, and the total running time gap expression in the transmission process is as follows:

[0121]

[0122] wherein, is the runtime period of the Kth network node; is the runtime period gap of the network node o and the adjacent relay node r, in seconds, is the runtime period of the network node o, the runtime period gap represents the time interval between two consecutive data transmission tasks, and the length of the runtime period gap depends on the processing capacity of the network node and the current network load condition;

[0123] In the Step 1, the correction value of the energy attenuation degree of the intelligent ship communication network channel is calculated, and the specific method is as follows:

[0124] In calculating the correction value of the energy attenuation degree of the intelligent ship communication network channel, first, a signal attenuation model is established by measuring the signal strength under different distances and conditions;

[0125] The calculation formula of the correction value is:

[0126]

[0127] wherein, is the correction value of the channel energy attenuation, is the actually measured signal energy, is the theoretical signal energy, is the environmental attenuation coefficient, and d is the signal propagation distance; by comparing the actual measurement value and the theoretical calculation value, the correction value is obtained to reflect the channel energy attenuation condition;

[0128] In the Step 2, according to the channel energy attenuation condition, the intelligent ship communication network image data is divided into several subsets to realize the creation of the intelligent ship data transmission channel equalization signal attenuation model, and the created signal attenuation model is:

[0129]

[0130] wherein, p is the total amount of the scheduling correlation attribute set of the intelligent ship communication network data terminal; d and c are respectively the output control parameter and the number of pulse bandwidth of the channel; E(y) is the joint function of time and scale; y is the clustering unit set of the channel; a and b represent the weight coefficients of the signal strength and the energy efficiency;

[0131] In calculating the weight coefficients of the signal strength and the energy efficiency, the formula used is:

[0132]

[0133] Metric is a comprehensive evaluation index for quantifying the pros and cons of channel selection, SignalStrength is signal strength, and EnergyEfficiency is energy efficiency;

[0134] The creation steps of the weight coefficients a and b of the signal strength and the energy efficiency are as follows:

[0135] 1. Analyzing requirements: first, analyze the factors that are valued in the channel selection process, i.e., signal quality or energy efficiency;

[0136] For example, in the case of needing to prioritize communication quality, the value of a can be increased; in the case of needing to save energy, the value of b can be increased.

[0137] 2. Setting initial values: based on the analysis results, set initial values for a and b, for example: a = 0.8, b = 0.2;

[0138] It should be noted that the initial values of a and b are preferably based on experience, experimental data, or simulation results.

[0139] 3. Adjustment and optimization: in actual application, according to the effects of channel selection, such as communication quality or energy consumption, the values of a and b are adjusted and optimized;

[0140] It should be noted that the optimization method can be realized by online learning, reinforcement learning, or genetic algorithm optimization algorithm;

[0141] For example: using reinforcement learning algorithms such as Q-learning or SARSA to learn the optimal channel selection strategy, adjust and optimize the values of a and b, and apply the learned optimal strategy to actual channel selection;

[0142] 4. Verification and confirmation: after adjustment and optimization, verify whether the new values of a and b meet the requirements of channel selection, and confirm in actual application.

[0143] In Step 3, the process of creating the intelligent ship communication network channel credibility set is as follows:

[0144] Signal synchronization transmission: each node in the intelligent ship communication network synchronously sends out multiple signals, and the multiple signals are transmitted according to a predetermined rule;

[0145] Multi-path signal collection: the receiving node collects signals from different paths;

[0146] Credibility evaluation: according to the number, strength, and signal quality of the collected multi-path signals, the credibility of each channel is evaluated;

[0147] Credibility set creation: Group the credibility assessment results of each channel into a set, called the channel credibility set ;

[0148] In Step 4, the number of collected multipath signals is used to optimize the control data transmission channel. The optimization control formula is

[0149]

[0150] In Step 5, the implementation of ship image network data transmission includes the following steps:

[0151] Data blocking and scheduling: The ship image network data to be transmitted is divided into multiple data blocks, and scheduling is performed according to the load condition of the network node and the channel quality;

[0152] Channel selection and allocation: According to the current network state and channel credibility set, the optimal channel for data transmission is selected; at the same time, according to the real-time energy consumption condition of the node, the size and allocation of the data subset are dynamically adjusted;

[0153] Data transmission and verification: Transmit the data blocks through the selected channel and perform verification at the receiving end to ensure the integrity and accuracy of the data. During transmission, forward error correction coding technology is used to improve the reliability of data transmission;

[0154] Transmission efficiency evaluation: According to the actual situation of data transmission, the transmission efficiency is evaluated.

[0155] Transmission efficiency can be calculated by the following formula:

[0156]

[0157] Where SuccessfulTransmittedBits represents the number of successfully transmitted bits, and TotalBitsToTransmit represents the total number of bits to be transmitted.

[0158] S3, data mining processing: In the monitoring center, the preprocessed ship image network data is segmented, the Map Reduce framework is configured for parallel processing, and the parallel K-means clustering algorithm is used to classify and mine the ship image network data to discover potential abnormal behavior, safety hazards or optimization space;

[0159] In some embodiments, referring to Figure 1 and Figure 2 , the operation of the parallel K-means clustering algorithm includes the following steps:

[0160] Step one, according to the obtained image signal and running parameter, randomly select a ship image network data as target data, set the target data as the center value of single cluster and original average value;

[0161] Step two, based on the average distance between the target data and each cluster, the ship image network data outside the target data is distributed to the cluster with the minimum distance by using dynamic weighted Euclidean distance;

[0162] Among them, the dynamic weighted Euclidean distance is used to distribute the data to the cluster with the minimum distance, specifically including:

[0163] Feature vector normalization: for the ship image network data outside the target data, extract its comprehensive feature vector X j =[x j1 ,x j2 ,...,x jn ], n is the feature dimension, and it is mapped to the interval [0, 1] by linear normalization:

[0164]

[0165] Among them, is the set of all sample values of the kth feature; x represents the original value of the jth sample on the kth feature, that is, the specific value of feature k in sample j,

[0166] Dynamic weighted distance calculation: define the center vector of cluster as:

[0167] , calculate the weighted Euclidean distance between sample and cluster :

[0168]

[0169] Among them, the weight coefficient is dynamically adjusted according to the contribution of the feature to the clustering result, which satisfies: Equal to 1;

[0170] Optimal cluster distribution: distribute sample to the cluster with the minimum distance:

[0171]

[0172] Among them, represents the cluster to which sample is distributed, is the jth sample in the dataset; arcmin refers to finding the cluster that minimizes the distance under the weighted distance metric; represents the weighted distance; represents the center of the ith cluster;

[0173] If there are multiple clusters with the same distance, the cluster with fewer samples is assigned preferentially to maintain the balance of cluster size;

[0174] Step three, calculate the average of each cluster, repeat the above steps until the criterion function of the algorithm converges.

[0175] The configuration of the Map Reduce framework includes setting the Map task and assigning the segmented data block to different computing nodes for processing.

[0176] S4, feature extraction: fuse the ship image network data with external related data, use convolutional neural network CNN to extract features from the fused image data, and use recurrent neural network and its variants such as LSTM or GRU to extract features from the ship running parameter data, obtain the image features and running parameter features of the ship, and comprehensively describe the state and operation of the ship through the features of multiple data sources; wherein GRU (Gated Recurrent Unit) is a simplified recurrent neural network (RNN) architecture, which is a variant of LSTM.

[0177] Further, the convolutional neural network CNN is used to extract features from the fused image data, including:

[0178] Normalization: scale the image pixel value to the interval [0, 1], assuming that the original image pixel value range is [0, 255], normalize by the formula , wherein x is the original pixel value and X1 is the normalized value;

[0179] After normalization, the image size is unified to 224x224 pixels, and the size is adjusted using the bilinear interpolation algorithm;

[0180] Random rotation, rotation angle range-15° to 15°, horizontal flip, vertical flip, etc. operations are performed to increase data diversity, and a CNN model including 5 convolutional layers, 3 pooling layers and 2 fully connected layers is constructed at this time, the convolutional layer uses a 3x3 convolutional kernel, the step is 1, the padding mode is same, the pooling layer uses 2x2 maximum pooling, the step is 2, the ReLU activation function is used after the convolutional layer, the formula is f(x) = max(0, x); the Softmax activation function is used after the fully connected layer for multi-classification output, the fused image dataset is divided into a training set 70%, a validation set 15% and a test set 15%, a cross-entropy loss function is used, the formula is , wherein is the true label, is the predicted probability, the Adam optimization algorithm is used, the learning rate is set to 0.001, the batch size is set to 32, the training set data is input into the model, the loss is calculated by forward propagation, the model parameters are updated by back propagation, the validation set is used to evaluate the model performance every epoch, and the training is stopped when the validation set loss no longer decreases. At this time, the preprocessed image data is input into the trained convolutional neural network CNN model, and the output of the last fully connected layer is taken as the image feature vector, assuming that the feature dimension is 1024.

[0181] In addition, the LSTM is also used to extract the running parameter data features of the ship in this step, wherein the LSTM (Long Short-Term Memory, Long Short-Term Memory Network) is a special recurrent neural network (RNN) architecture designed to solve the gradient vanishing / explosion problem faced by traditional RNN when processing long sequence data, and its process is as follows:

[0182] 1. Data preprocessing: remove outliers in the running parameter data, such as data points with speed values exceeding 2 times the maximum design speed of the ship, scale the running parameter data to the [-1, 1] interval, and use the min-max normalization formula , wherein and are the minimum and maximum values of the running parameter data, respectively, the running parameter data is arranged in chronological order to form time series data, assuming that the time step is 1 second and the sequence length is 100;

[0183] 2. Model construction: construct an LSTM model including 2 layers of LSTM layers and 1 fully connected layer, set the number of hidden units of each LSTM layer to 128, use the tanh activation function in the LSTM layer, the formula is: , and the ReLU activation function is used in the fully connected layer;

[0184] 3. Model training: The time series driving parameter dataset is divided into a training set 70%, a validation set 15%, and a test set 15%. The mean square error (MSE) loss function is used, with the formula: The RMSprop optimization algorithm is used, with a learning rate of 0.001 and a batch size of 32. The training set data is input into the model, and the loss is calculated by forward propagation. The model parameters are updated by backpropagation. The model performance is evaluated using the validation set after each epoch. When the validation set loss no longer decreases, the preprocessed driving parameter time series data is input into the trained LSTM model, and the hidden state at the last time step is taken as the driving parameter feature vector.

[0185] S5. Model establishment: The image features and driving parameter features are fused by multi-modal feature fusion technology to form a comprehensive feature vector. A deep Q network is used to construct a deep reinforcement learning model. The comprehensive feature vector is input into the model, which is trained by interacting with the environment. The learning optimal channel and task allocation strategy are output, providing intelligent decision-making basis for channel and task allocation adjustment.

[0186] The comprehensive feature vector includes visual information and driving state information. The visual information includes the surrounding environment and obstacles of the ship. The driving state information includes the speed, direction, and acceleration of the ship.

[0187] The specific operation steps of the comprehensive feature vector formed by fusing the image features and driving parameter features by multi-modal feature fusion technology are as follows:

[0188] 1. Feature alignment and preprocessing operation: First, the image features extracted from the convolutional neural network (CNN) and the driving parameter features extracted from the recurrent neural network (LSTM) are aligned in time or space dimensions. For example, if the image features are extracted based on the surrounding environment of the ship at a certain time, the driving parameter features should also be the speed, direction, and acceleration of the ship at the same time or nearby time. The image features and driving parameter features are normalized to have the same scale, which facilitates subsequent fusion. For example, the image features and driving parameter features are scaled to the [0, 1] interval.

[0189] 2. Feature extraction and enhancement operation: A pre-trained CNN model, such as VGG16, is used to extract features from the fused ship image data, resulting in a high-dimensional image feature vector. For example, a 512-dimensional image feature vector is extracted from the fully connected layer or pooling layer of the CNN An LSTM recurrent neural network is used to extract features from the ship's driving parameter data, capturing temporal dependencies, resulting in a time series feature vector. For example, a 128-dimensional time series feature vector is extracted from the last time step of the LSTM .

[0190] 3. Feature interaction design operation: design an interaction layer, such as using a fully connected layer or attention mechanism, to realize information interaction between image features and driving parameter features, and perform interaction operations on image features and driving parameter features in the interaction layer. For example, the concatenated and can be input into a fully connected layer, which realizes information exchange between features by learning a weight matrix, and outputs a fused feature representation;

[0191] For example: design a fully connected layer as an interaction layer, the input is the concatenated , , i.e. a 640-dimensional vector, and the output is a 640-dimensional interaction feature vector ;

[0192] 4. Feature transformation and fusion operation: perform nonlinear transformation on the interaction features, such as through a ReLU activation function and a fully connected layer, to capture complex relationships between features, and then fuse the transformed features to form a comprehensive feature vector;

[0193] For example: after the interaction layer, design a fusion layer containing a fully connected layer and a ReLU activation function, input into the fusion layer, and output a 640-dimensional comprehensive feature vector through nonlinear transformation;

[0194] If the dimension of the comprehensive feature vector is too high, dimension reduction techniques such as PCA can be used to reduce the dimension of the comprehensive feature vector;

[0195] PCA (Principal Component Analysis) is a commonly used unsupervised learning dimension reduction technique. In the intelligent ship communication network integration monitoring method, if the dimension of the comprehensive feature vector is too high, PCA is used to reduce its dimension;

[0196] PCA basic principle: PCA projects the original high-dimensional data into a low-dimensional space through linear transformation, while preserving the main variation direction in the data, i.e. the principal component. The principal component is a linear combination of the original features, sorted by the amount of variance they explain. The first principal component explains the largest variance in the data, the second principal component explains the largest part of the remaining variance, and so on;

[0197] 5. Output comprehensive feature vector: after feature interaction, transformation, and fusion, output the comprehensive feature vector , if not reduced in dimension, then , if reduced in dimension, then .

[0198] An agent is defined as: each communication node in the intelligent ship communication network, such as a router, a sensor node, is regarded as an agent;

[0199] The state space is defined as:

[0200] 1. State composition: the state of the agent includes the residual energy of the node, expressed in percentage form, ranging from 0 to 100%; the current channel quality, expressed in signal-to-noise ratio SNR, ranging from 0 to 50 dB; the task queue length, which is an integer, indicating the number of tasks waiting to be processed; and the data transmission rate Mbps;

[0201] 2. State vector representation: the above state information is combined into a state vector: s= (e, q, l, r), where e is the residual energy, q is the channel quality, l is the task queue length, and r is the data transmission rate;

[0202] The action space is defined as:

[0203] 1. Action composition: the action of the agent includes selecting different channels for data transmission, assuming there are 5 selectable channels numbered 1-5; adjusting the data transmission power, including high, medium and low three levels; accepting or rejecting new tasks, including accepting or rejecting two choices;

[0204] 2. Action vector representation: the action is combined into an action vector: a= (c, p, t), where c is the channel selection, p is the transmission power, and t is the task acceptance decision.

[0205] In some embodiments, the deep Q network in this step builds a deep reinforcement learning model, including the following operation steps:

[0206] 1. Network architecture design: a deep Q network DQN containing 3 fully connected layers is built, the input layer dimension is the dimension of the comprehensive feature vector, assuming it is 4-dimensional, the number of hidden units in the middle layer is 64 and 32 respectively, and the output layer dimension is the size of the action space, assuming it is 5x3x2=30-dimensional, corresponding to the combination of different channels, power and task acceptance decisions;

[0207] 2. Experience replay mechanism: an experience replay buffer with a capacity of 10000 is established at this time, which is used to store experience samples of the agent interacting with the environment;

[0208] 3. Target network update: initialize the main network and the target network, the parameters are the same, copy the parameters of the main network to the target network every 100 training steps;

[0209] 4. Training: design a reward function according to the task completion, energy consumption and channel utilization efficiency; for example, give a positive reward of 10 for successfully completing a task, give a negative reward of -0.1 for consuming 1% of energy, and give a positive reward of 2 for increasing the channel utilization rate by 10%;

[0210] wherein the training process is as follows: first, taking the intelligent ship communication network as the environment, regarding the network nodes as agents, defining the state space and action space of the agents, the agents selecting actions a according to the current state s using the e-greedy strategy, the e-greedy strategy randomly selecting actions with a probability of and selecting the action with the maximum current Q value with a probability of 1- The agents perform the actions in the environment to obtain an immediate reward r, indicating the immediate reward obtained after performing the action a in the state s, and move to the next state s', and store (s, a, r, s') in the experience replay buffer. A certain number of samples are randomly extracted from the experience replay buffer for training the main network. For each sample, the target Q value is calculated, wherein is a discount factor, Q1 is the Q value function of the target network, is a discount factor, indicates the Q value of the action a' selected in the next state s' so that the Q value is maximum; the mean square error loss function is used to update the parameters of the main network, wherein Q2 is the Q value function of the main network, indicates the Q value estimate given by the main network Q when taking action ai in state si, and the above training steps are repeated until the deep reinforcement learning model converges.

[0211] S6, Channel and task allocation adjustment: after the deep reinforcement learning model is constructed, a consortium blockchain network is built in the intelligent ship communication network, each network node joins as a node of the blockchain, and the distributed ledger feature of the blockchain is used to record the energy consumption, task execution and channel usage state of each node. At this time, a smart contract is designed to define the rules and processes of channel and task allocation. When there is a new data mining task, each node uploads its energy state, channel quality and task execution capability to the blockchain according to the rules in the smart contract. The consensus mechanism in the blockchain network verifies and confirms the information uploaded by each node. Only the information that has passed the consensus will be recorded in the blockchain. According to the real-time energy consumption of the nodes, the size and allocation of the data subsets are dynamically adjusted. When a node has low energy, the size of the data subset allocated to the node is reduced, and part of the task is transferred to a node with sufficient energy. After the node executes the task, the task execution result and energy consumption are uploaded to the blockchain again. The smart contract verifies the task execution result, and updates the credit points of the nodes according to the verification result.

[0212] It should be noted that the energy consumption, task execution and channel usage status of each node are recorded by using the distributed ledger characteristics of the blockchain, and only the information that has passed the consensus will be recorded in the blockchain. This mechanism ensures the integrity and non-tamperability of data during transmission and storage, avoids data theft, tampering or forgery during transmission, and further improves the reliability of data transmission.

[0213] Further, for the construction of the alliance blockchain network in this step: select the Hyperledger Fabric platform, deploy Fabric nodes on each node in the intelligent ship communication network, including ordering nodes, peer nodes and organization management nodes;

[0214] Network configuration: configure the channels, chaincode, i.e. smart contract and member management service of the blockchain network, to ensure that the nodes can communicate and interact with data safely and reliably;

[0215] Among them, each node regularly collects its own energy consumption data, obtains real-time power, task execution status, including task completion time, success rate, etc. and channel usage status, including channel occupancy time, channel quality indicators, etc. through the energy monitoring module;

[0216] The collected data is encapsulated according to the transaction format of the blockchain to generate transaction data. For example: The transaction data is submitted to the blockchain network through the SDK of Fabric, and recorded on the blockchain through the consensus mechanism;

[0217] In addition, the smart contract design in this step includes the following design rules:

[0218] Rule making: define channel and task allocation rules, such as nodes with sufficient energy, e.g. remaining energy > 50%, are given priority in task allocation;

[0219] Nodes with good channel quality, such as SNR > 30 dB, are given priority in using high-quality channels;

[0220] According to the task urgency, set the emergency level 1-5, 5 being the most urgent, and sort the task priority.

[0221] Contract writing: first write the smart contract code to implement the logic of the above rules. For example, define a function allocateTask to make task allocation decisions based on the information uploaded by the nodes and the rules;

[0222] Contract deployment: deploy the written smart contract to the Fabric blockchain network, and through chaincode installation and instantiation operations, all nodes can call the contract.

[0223] Task allocation process is:

[0224] 1. Information upload: when there is a new data mining task, each node uploads its energy state, channel quality, and task execution capability to the blockchain according to the rules in the smart contract. For example, node1 uploads information:

[0225]

[0226] 2. Consensus verification: the consensus mechanism in the blockchain network, such as the PBFT consensus algorithm, verifies and confirms the information uploaded by each node, and each node verifies the signature of the received transaction. When more than two-thirds of the nodes reach a consensus, the transaction is recorded in the blockchain;

[0227] 3. Dynamic adjustment: according to the real-time energy consumption of the node, dynamically adjust the size and allocation of the data subset. For example, when the energy of node2 is lower than 30%, reduce the size of the data subset allocated to the node, such as from the original 100 data to 50 data, and transfer part of the task to the node with sufficient energy node3.

[0228] 4. Task execution and verification: after dynamic adjustment, the node executes the allocated task, such as data mining calculation, and the node completes the task, and encapsulates the task execution result, such as data mining accuracy 95%, processing time 10 seconds and energy consumption, for example: this task consumes 5% of the electricity, and uploads it to the blockchain network, and the smart contract verifies the task execution result, checks whether the accuracy reaches the preset threshold, such as 90%. According to the verification result, update the credit score of the node, for example, the task is successfully completed and the accuracy meets the standard, the credit score increases by 10 points; if it does not meet the standard, the credit score decreases by 5 points. Credit score is used for reference in subsequent task allocation and resource allocation.

[0229] S7, alarm and control: according to the updated credit score, image signal and driving parameter, continuously monitor the ship state, and set up an alarm mechanism according to the ship state.

[0230] It should be noted that through the combination of deep reinforcement learning model in S5 and alliance blockchain network in S6, dynamic adjustment of channel and task is realized. When there is a new data mining task, each node uploads its state information to the blockchain according to the rules in the smart contract, and the consensus mechanism in the blockchain network verifies and confirms the information, and dynamically adjusts the size and allocation of the data subset according to the real-time energy consumption of the node, ensuring efficient use of network resources, reducing latency, improving the timeliness of data analysis, enabling intelligent ships to quickly respond to network changes and ensure navigation safety;

[0231] It needs to be further explained that by using data mining and model prediction, through the fusion of multi-source data, the model can more accurately perceive environmental changes such as wind and waves, traffic density, and provide the basis for path planning. In the environment of intelligent ship communication network, define node state, such as define remaining energy and action, define remaining energy, such as define channel quality and task queue, define action: such as define channel selection, power adjustment and task acceptance, through interaction with the environment, such as: select the optimal channel to transmit data, the model learns the optimal action strategy under different environments, when CNN detects obstacles in front, the model adjusts the action selection to high signal-to-noise ratio channel transmission obstacle avoidance instruction, while LSTM predicts the trend of ship movement, dynamically plans the detour path, optimizes the navigation route, improves the operation efficiency and reduces the operation cost, and through continuous monitoring of the ship state and setting up alarm mechanism, it can timely find and handle potential safety hazards, improve the safety and reliability of the ship;

[0232] It needs to be further explained that in the process of adjusting channel and task allocation, the intelligent contract is designed to define the rules and processes of channel and task allocation, and the intelligent contract automatically executes the task allocation algorithm, dynamically adjusts the task allocation according to the node energy, channel quality, task priority and other factors, avoids human intervention and subjective factors, ensures the fairness and rationality of channel and task allocation, at the same time, the intelligent contract also verifies the task execution result, updates the credit points of the node according to the verification result, the credit points are used as the reference basis for subsequent task allocation and resource allocation, encourages the node to actively complete the task and maintains good network state, further enhances the credibility of network channel, solves the problem that the existing ship communication network is affected by wind and waves, weather, distance and other factors when sailing at sea, resulting in unstable or interrupted signal, which affects the real-time transmission of data, and with the expansion of network size, the data analysis efficiency is low, the time delay is large, the credibility of network signal channel is reduced, it is impossible to grasp the state of each key equipment in real time, so as to prevent accidents and other problems in advance, by dynamically adjusting the size and allocation of data subset, when a node has low energy, reduce the size of the data subset allocated to the node, and transfer part of the task to the node with sufficient energy, realize the reasonable allocation and optimal utilization of network resources, improve the overall performance of the network, reduce the time delay, reduce the occurrence of network congestion, and ensure the stable operation of intelligent ship communication network.

[0233] In some embodiments, a device suitable for an integrated monitoring method of an intelligent ship communication network comprises:

[0234] The data acquisition unit is used to acquire image signals, running parameters and external related data of the ship, such as surrounding environment pictures captured by a camera and running parameters such as speed, direction, heading angle and position information, wherein, in the data acquisition unit, a GPRS DTU wireless data terminal is adopted to receive real-time ship network data;

[0235] The cloud computing platform construction unit is used to construct a ship image network data transmission platform, i.e. a cloud, for transmitting, storing and processing ship image network data. The cloud platform provides powerful computing capacity and storage space, supports fast processing and analysis of large-scale data, and can efficiently process and analyze collected ship data to timely discover potential problems and provide protection for safe operation of the ship.

[0236] The data mining processing unit is used to segment the preprocessed data and classify and mine the ship image network data. A parallelized K-means clustering algorithm is used to deeply mine the ship image network data to extract valuable information and patterns.

[0237] The feature extraction unit fuses the ship image network data and external related data, uses a convolutional neural network to extract features from the fused image data, and uses a recurrent neural network and its variants to extract features from the running parameter data of the ship to obtain image features and running parameter features of the ship.

[0238] The model establishment unit fuses the image features and running parameter features by a multi-modal feature fusion technology to form a comprehensive feature vector, uses a deep Q network to construct a deep reinforcement learning model, takes the comprehensive feature vector as input, trains through interaction with the environment, and outputs a learning optimal channel and task allocation strategy.

[0239] The channel and task allocation adjustment unit is used to build a blockchain in the ship communication network to record node energy, task execution and channel usage information, design an intelligent contract to clearly define channel and task allocation rule processes, dynamically adjust data subset allocation according to real-time node states, update node credit points, and reasonably optimize management of channels and tasks.

[0240] The alarm and control unit continuously monitors the state of the ship and detects abnormalities or potential risks, triggers an alarm and takes corresponding control measures when abnormalities or potential risks are detected. The measures include emergency stopping of the ship, starting the fire fighting system, adjusting the navigation plan and sending alarm information.

[0241] It should be noted that the data acquisition unit uses a GPRS DTU wireless data terminal to acquire data in real time, ensuring the timeliness and integrity of the data. The cloud computing platform construction unit provides powerful computing capacity and storage space, supports fast processing and analysis of large-scale data, can efficiently process and analyze collected ship data, and timely discovers potential problems to provide protection for safe operation of the ship.

[0242] It should be noted that the data mining processing unit uses the parallel K-means clustering algorithm to deeply mine the ship image network data, which can extract valuable information and patterns, such as the regularity of ship navigation trajectory and the trend of surrounding environment changes, to provide scientific basis for ship operation decision-making.

[0243] It should be further noted that the model establishment unit uses a deep Q network to build a deep reinforcement learning model, which is trained through interaction with the environment to learn the optimal channel and task allocation strategy, improving the efficiency and performance of the ship communication network and reducing resource waste.

[0244] It should be further noted that the channel and task allocation adjustment unit builds a consortium blockchain in the ship communication network, records node energy, task execution and channel usage information, designs an intelligent contract to clarify the rule process, dynamically adjusts the data subset allocation according to the real-time state of the node, updates the node credit points, and reasonably optimizes the management of channels and tasks. The distributed ledger feature of the blockchain ensures the information's non-tamperability and traceability, and the intelligent contract realizes the automated execution of the allocation rules, reducing human intervention and improving the fairness and transparency of resource allocation.

[0245] In some embodiments, the intelligent ship communication network integrated monitoring device can further include:

[0246] A data storage and management unit is responsible for storing the collected ship image signals, running parameters and the results of data mining processing on the cloud or local server;

[0247] A network communication unit is responsible for establishing and maintaining connections with ship communication network nodes to ensure real-time and reliable transmission of data, including image signals, running parameters, alarm information and control instructions;

[0248] A user interface and interaction unit is used to provide an intuitive user interface to display real-time state of the ship, data mining results and other information, and users can perform monitoring, querying, setting and other operations through the interface to realize effective interaction with the system;

[0249] A system monitoring and maintenance unit monitors the running state of each unit of the system, including data acquisition, cloud computing platform, data mining processing, model establishment, channel and task allocation adjustment, alarm and control.

[0250] The present application integrates data acquisition, cloud computing, data mining and model prediction technologies to realize comprehensive monitoring and real-time analysis of the state of the ship communication network, improve monitoring efficiency, and optimize network resource allocation by building an energy consumption model and dynamically adjusting channel and task allocation, thereby reducing energy consumption and operating costs.

[0251] The foregoing merely illustrates the principles of the application and application of its more particular aspects and embodiments. This description and the examples are not intended to limit the scope of the application to these precise forms. Many variations and modifications of the embodiments described above will present themselves, to persons of ordinary skill in the art, based on the concept disclosed. For example, the scope of the application is not limited to the specific details of the above-described exemplary embodiments, but is defined by the claims appended hereto. Any figure reference in the claims is to be construed as an example only, and not a limitation of the scope of the claims.

[0252] Furthermore, it should be understood that although the description above has been set forth in the context of implementations, the present application is not limited to these implementations alone. Rather, numerous modifications are possible in light of the teachings below. Some particular implementations merely reveal one independent technical solution among others. Any such modifications to the implementations would still be encompassed by the present application, which is defined by the scope of the claims appended hereto.

Claims

1. An integrated monitoring method for intelligent ship communication networks, characterized in that, The method includes the following steps: S1. Data Acquisition: Using sensors and instruments, acquire image network data, driving parameters, and external relevant data of the ship; wherein, the driving parameters include temperature, pressure, water level, and smoke concentration; the external relevant data includes marine meteorological data and port surrounding traffic data; S2. Constructing a Cloud Computing Platform: Based on the image network data and driving parameters, a ship image network data transmission platform is constructed using a cloud computing platform. The transmission process of the ship image network data transmission platform is optimized to obtain a data transmission channel with strong anti-interference capabilities. The cloud computing platform preprocesses the received ship image network data and transmits the preprocessed ship image network data to the monitoring center through the data transmission channel. The steps for optimizing the transmission process of the ship image network data transmission platform are as follows: Step 1: First, calculate the correction value for the channel energy attenuation of the intelligent ship communication network to obtain the correction value to reflect the channel energy attenuation. Step 2: Based on the channel energy attenuation, the intelligent ship image network data is divided into several subsets to create an equal signal attenuation model for the intelligent ship image network data transmission channel; Step 3: Synchronously create a channel reliability set for the intelligent ship communication network: The intelligent ship communication network simultaneously transmits multiple signals, which are transmitted according to a set rule. This is a set of channel reliability for intelligent ship communication networks, where... Let represent the credibility of the i-th channel, where i = 1, 2, 3, ..., j, and j represents the number of channels in the network; Step 4: By collecting multipath signals within the credibility set of the intelligent ship communication network, interference between nodes and channels during network operation is avoided, and the data transmission channel is optimized. Step 5: Based on the optimization results of the data transmission channel, realize the transmission of ship image network data. The total runtime gap expression during the transmission process is as follows: in, This refers to the runtime segment of the Kth network node; This is the runtime interval between network node o and its neighboring relay node r, in seconds. The runtime segment of network node o is defined as the time interval between two consecutive data transmission tasks. The length of the runtime segment interval depends on the processing capacity of the network node and the current network load. S3. Data Mining and Processing: In the monitoring center, the preprocessed ship image network data is segmented, the MapReduce framework is configured for parallel processing, and the parallelized K-means clustering algorithm is used to classify and mine the ship image network data. S4. Feature Extraction: The ship image network data is fused with external related data, and a convolutional neural network is used to extract features from the fused image data. At the same time, a recurrent neural network and its variants are used to extract features from the ship's driving parameter data to obtain the ship's image features and driving parameter features. S5. Model Establishment: Image features and driving parameter features are fused using multimodal feature fusion technology to form a comprehensive feature vector. A deep reinforcement learning model is constructed using a deep Q-network. The comprehensive feature vector is used as input, and the model is trained through interaction with the environment to output the optimal channel and task allocation strategy. The comprehensive feature vector includes visual information and driving state information. The visual information includes the ship's surrounding environment and obstacles, and the driving state information includes the ship's speed, direction, and acceleration. The construction of a deep reinforcement learning model using a deep Q-network includes the following steps: 1) Network architecture design: Construct a deep Q-network (DQN) with 3 fully connected layers. The input layer dimension is the dimension of the comprehensive feature vector, and the output layer dimension is the size of the action space, corresponding to different combinations of channels, power and task acceptance decisions; The action space includes action components and action vector representation. The action components are: the agent's actions include selecting different channels for data transmission; adjusting the data transmission power, including high, medium, and low levels; and accepting or rejecting new tasks, including two choices: accept or reject. The action vector representation combines actions into an action vector: a = (c, p, t), where c is the channel selection, p is the transmission power, and t is the task acceptance decision. 2) Experience replay mechanism: Establish an experience replay buffer to store experience samples of the agent's interaction with the environment; 3) Target network update: Initialize the main network and the target network with the same parameters. At regular training steps, copy the parameters of the main network to the target network. 4) Conduct training: Design a reward function based on task completion, energy consumption, and channel utilization efficiency; The training process is as follows: First, using an intelligent ship communication network as the environment, network nodes are considered as agents. The state space and action space of the agents are defined. Based on the current state s, the agent selects action a using an e-greedy policy. The probability of randomly selecting an action is 1- The agent selects the action with the highest current Q-value based on probability. The agent performs the action in the environment and receives an immediate reward *r*, representing the immediate reward obtained after performing action *a* in state *s*. It then transitions to the next state *s'*. The sequence (*s*, *a*, *r*, *s'*) is stored in the experience replay buffer. A certain number of samples are randomly drawn from the experience replay buffer for training the main network. For each sample, the target Q-value is calculated. ,in It is the discount factor, and Q1 is the Q-value function of the target network. This represents the Q-value of the action chosen to maximize the Q-value in the next state s'; the mean squared error loss function is used. Update the parameters of the main network, where Q2 is the Q-value function of the main network. Indicates that in state s i Take action a i At that time, the Q-value estimate given by the main network Q is used to repeat the above training steps until the deep reinforcement learning model converges; S6. Channel and Task Allocation Adjustment: Build a consortium blockchain in the intelligent ship communication network to record node energy, task execution and channel usage information, design smart contracts to clarify channel and task allocation rules and processes, dynamically adjust data subset allocation according to the real-time status of nodes, update node credit scores, and reasonably optimize and manage channels and tasks. S7. Alarm and Control: Based on the updated node credit score, image signal and driving parameters, continuously monitor the ship's status and set up an alarm mechanism according to the ship's status.

2. The method according to claim 1, characterized in that, In S1, the sensors include a temperature sensor, a pressure sensor, a water level sensor, and a smoke sensor; the instruments include a speedometer and a tachometer.

3. The method according to claim 1, characterized in that, In Step 1, the correction value for the energy attenuation of the intelligent ship communication network channel is calculated, and the specific method is as follows: When calculating the correction value for the energy attenuation of the channel in the intelligent ship communication network, a signal attenuation model is first established by measuring the signal strength under different distances and conditions. The formula for calculating the correction value is: in, This is a correction value for channel energy attenuation. The signal energy obtained from actual measurement. Theoretically, signal energy Here, d represents the environmental attenuation coefficient, and d represents the signal propagation distance. By comparing actual measurements with theoretical calculations, a correction value is derived to reflect the channel energy attenuation. In Step 2, the signal attenuation model is created as follows: Where p is the total set of scheduling-related attributes of the data terminal of the intelligent ship communication network; d and c are the output control parameters and pulse bandwidth of the channel, respectively; E(y) is the joint function of time and scale; y is the clustering unit set of the channel; a and b represent the weighting coefficients of signal strength and energy efficiency. The formula used to calculate the weighting coefficients for signal strength and energy efficiency is as follows: In the formula, Metric is a comprehensive evaluation index used to quantify the quality of channel selection, SignalStrength is the signal strength, and EnergyEfficiency is the energy efficiency. In Step 3, the process of creating the intelligent ship communication network channel credibility set is as follows: Synchronous signal transmission: Each node in the intelligent ship communication network simultaneously sends out multiple signals, and the multiple signals are transmitted according to preset rules; Multipath signal collection: The receiving node collects signals from different paths; Credibility assessment: The credibility of each channel is assessed based on the quantity, strength, and quality of the collected multipath signals. Creation of credibility sets: The credibility evaluation results of each channel are combined into a set, called the channel credibility set. ; In Step 4, the data transmission channel is optimized based on the number of collected multipath signals. The optimization control formula is as follows: In Step 5, the transmission of ship image network data includes the following steps: Data segmentation and scheduling: The ship image network data to be transmitted is segmented into multiple data blocks and scheduled according to the load of network nodes and channel quality; Channel selection and allocation: Select the optimal channel for data transmission based on the current network status and channel reliability set; at the same time, dynamically adjust the size and allocation of data subsets based on the real-time energy consumption of nodes; Data transmission and verification: Data blocks are transmitted through the selected channel and verified at the receiving end to ensure the integrity and accuracy of the data. During transmission, forward error correction coding technology is used to improve the reliability of data transmission. Transmission efficiency assessment: Evaluate transmission efficiency based on the actual data transmission situation.

4. The method according to claim 1, characterized in that, Step S3, which uses a parallelized K-means clustering algorithm, includes the following steps: Step 1: Based on the acquired image signals and driving parameters, randomly select a ship image network data as the target data. Let the target data be the center value and the original average value of a single cluster. Step 2: For ship image network data other than the target data, based on the average distance between the data and each cluster, use dynamic weighted Euclidean distance to allocate this data to the cluster with the smallest distance. Specifically, the allocation of such data to the cluster with the smallest distance using dynamically weighted Euclidean distance includes: Feature vector normalization: Extracting the comprehensive feature vector X from ship image network data other than the target data. j =[x j1 ,x j2 ,...,x jn ], where n is the feature dimension, and it is mapped to the interval [0,1] through linear normalization: in, Let k be the set of all sample values ​​of the k-th feature. This represents the original value of the j-th sample in the k-th dimension feature; This represents the value of the j-th sample after linear normalization on the k-th feature, mapped to the interval [0,1]. Dynamic weighted distance calculation: defining clusters The center vector is: Calculate samples with cluster Weighted Euclidean distance: Among them, the weighting coefficient The clustering results are dynamically adjusted based on the contribution of features to the clustering results, satisfying the following: It equals 1; Optimal cluster allocation: assigning samples Assign to the cluster with the smallest distance: in, Indicates sample The clusters that were assigned It is the j-th sample in the dataset; arcmin refers to finding the cluster that minimizes the distance under a weighted distance metric. Indicates the weighted distance; Indicates the center of the i-th cluster; If multiple clusters are at the same distance, the sample will be assigned to the cluster with the smaller number of samples to maintain a balanced cluster size. Step 3: Calculate the average value for each cluster. Repeat the above steps until the algorithm's criterion function converges.

5. The method according to claim 1, characterized in that, In step S4, a convolutional neural network (CNN) is used to extract features from the fused image data, including: Normalization: Scaling image pixel values ​​to the [0,1] range using the formula Normalization is performed, where x is the original pixel value and X1 is the normalized value; After normalization, the image size is unified, and the size is adjusted using a bilinear interpolation algorithm; Set the rotation angle range and perform random rotation; Construct a CNN model containing 5 convolutional layers, 3 pooling layers, and 2 fully connected layers. The convolutional layers use 3×3 kernels with a stride of 1 and same padding. The pooling layers use 2×2 max pooling with a stride of 2. The convolutional layers are followed by the ReLU activation function, f(x) = max(0,x). The fully connected layers are followed by the Softmax activation function for multi-class output. The fused image dataset is divided into training, validation, and test sets, and the cross-entropy loss function is used. ,in For real labels, To predict probabilities, the Adam optimization algorithm is used. The training set data is input into the model, the loss is calculated by forward propagation, and the model parameters are updated by backpropagation. The model performance is evaluated using a validation set after each training epoch. Training stops when the validation set loss no longer decreases. At this point, the preprocessed image data is input into the trained convolutional neural network (CNN) model, and the output of the last fully connected layer is taken as the image feature vector.

6. The method according to claim 5, characterized in that, The recurrent neural network is an LSTM. The steps for extracting features from the ship's navigation parameter data using LSTM include: 1) Data preprocessing: Remove outliers from the driving parameter data, scale the driving parameter data to the [-1,1] interval, and use the min-max normalization formula. ,in and These are the minimum and maximum values ​​of the driving parameter data. The driving parameter data is then arranged in chronological order to form a time series dataset. 2) Model Construction: Construct an LSTM model consisting of two LSTM layers and one fully connected layer. Each LSTM layer has a certain number of hidden units. The tanh activation function is used in the LSTM layers, with the following formula: Fully connected layers use the ReLU activation function; 3) Model Training: The time series dataset is divided into training, validation, and test sets. The mean squared error (MSE) loss function is used, and the RMSprop optimization algorithm is employed. The training set data is input into the model, and forward propagation is performed to calculate the loss. Backward propagation is used to update the model parameters. The model performance is evaluated using the validation set after each training epoch. Training is stopped when the validation set loss no longer decreases. At this point, the preprocessed driving parameter time series dataset is input into the trained LSTM model, and the hidden state of the last time step is taken as the driving parameter feature vector.

7. The method according to claim 1, characterized in that, In step S6, a consortium blockchain network is established within the ship communication network. Each network node joins as a blockchain node. Utilizing the distributed ledger characteristics of the blockchain, the energy consumption, task execution status, and channel usage of each node are recorded. Smart contracts are designed to define the rules and processes for channel and task allocation. When a new data mining task arrives, each node uploads its energy status, channel quality, and task execution capabilities to the blockchain according to the rules in the smart contract. The consensus mechanism in the blockchain network verifies and confirms the information uploaded by each node. Only information that has passed consensus is recorded in the blockchain. Based on the real-time energy consumption of the nodes, the size and allocation of data subsets are dynamically adjusted. When a node has low energy, the size of the data subset allocated to that node is reduced, and some tasks are transferred to nodes with sufficient energy. After a node executes a task, it uploads the task execution result and energy consumption status back to the blockchain. The smart contract verifies the task execution result and updates the node's credit score based on the verification result.

8. An apparatus applicable to the integrated monitoring method for intelligent ship communication networks according to any one of claims 1-7, characterized in that, The device includes: The data acquisition unit is used to acquire the ship's image network data, driving parameters, and other relevant external data. The cloud computing platform building unit is used to build a ship image network data transmission platform; The data mining processing unit is used to segment, classify, and mine the preprocessed ship image network data. The feature extraction unit fuses the ship image network data with external related data, uses a convolutional neural network to extract features from the fused image data, and uses a recurrent neural network and its variants to extract features from the ship's driving parameter data, thereby obtaining the ship's image features and driving parameter features. The model building unit fuses image features and driving parameter features through multimodal feature fusion technology to form a comprehensive feature vector. It uses a deep Q-network to build a deep reinforcement learning model. With the comprehensive feature vector as input, it is trained through interaction with the environment and outputs the optimal channel and task allocation strategy. The channel and task allocation adjustment unit is used to build a consortium blockchain in the intelligent ship communication network, record node energy, task execution and channel usage information, design smart contracts to clarify channel and task allocation rules and processes, dynamically adjust data subset allocation according to the real-time status of nodes, update node credit scores, and reasonably optimize and manage channels and tasks. Alarm and control unit: continuously monitors the ship's status and detects anomalies or potential risks. If an anomaly or potential risk is found, it triggers an alarm mechanism and takes control measures.

9. The apparatus according to claim 8, characterized in that, The device further includes: The data storage and management unit is responsible for storing the collected ship image signals, driving parameters, and data mining results on the cloud or local server. The network communication unit is responsible for establishing and maintaining connections with intelligent ship communication network nodes; User interface and interaction unit, used to provide an intuitive user interface to display the real-time status of the ship and the results of data mining; The system monitoring and maintenance unit monitors the operational status of each unit in the system.

Images