Ship ballast water pressure loading and unloading control system based on large model

Abstract

The invention discloses a ship ballast water pressure loading and unloading control system based on a large model. The ship ballast water pressure loading and unloading control system is used for improving intelligence and efficiency of ship ballast water operation. According to the system, firstly, data such as ship navigation states, ballast water tank water levels and ship body postures are collected in real time through a multi-source sensor; analyzing and reasoning the collected data in the trained large neural network model; the large model outputs an optimal control strategy for ballast water loading and unloading, and the decision control module generates a specific execution instruction to drive a valve or a pump or other devices to complete ballast unloading actions. The system is also provided with a man-machine cooperation interface and a feedback mechanism, and the model can be finely adjusted and optimized according to a sailor instruction or real-time feedback information. Compared with a traditional control method, the method has the advantages that ship stability, energy efficiency and environmental protection requirements are fused through a large model, a ballast scheme can be automatically adjusted, rapid and efficient ballast replacement is achieved, and ship navigation safety and economical efficiency are remarkably improved.

Description

Technical Field

[0001] This patent relates to the field of ship control technology, and specifically to a ship ballast water ballast unloading control system based on a large model. Background Art

[0002] In the shipping industry, whether manned or unmanned, managing ballast water during port loading and unloading is a complex and crucial task. Ballast water primarily regulates the ship's stability, draft, trim, and roll, ensuring safe navigation and performance. When loading and unloading cargo, changes in cargo weight cause changes in the ship's center of gravity and buoyancy, necessitating appropriate ballast water loading and unloading operations.

[0003] Currently, traditional ballast water management methods rely primarily on operator experience and manual control. Operators must constantly assess the ship's status in real time based on the progress of loading and unloading cargo, and manually control the rate at which ballast water is added or unloaded.

[0004] However, this approach has many problems. First, human judgment is subjective and limited. Operators may make mistakes due to fatigue, lack of experience or other factors, which may affect the stability and safety of the ship. Second, manual control is inefficient. During the cargo loading and unloading process, operators are required to pay continuous attention to the loading and unloading progress and perform operations frequently. This not only increases the labor intensity of operators, but also easily leads to untimely operations, prolonging the ship's stay in the port and reducing shipping efficiency. In addition, some existing automated control technologies are often based on simple preset rules and algorithms, which cannot fully take into account various complex actual situations. They lack the ability to perceive and analyze the overall status of the ship in real time and comprehensively, making it difficult to achieve efficient and accurate ballast water control. Summary of the Invention

[0005] The present invention aims to solve the above-mentioned problems.

[0006] To this end, the technical solution adopted by the present invention is as follows: a ship ballast water ballast unloading control system based on a large model, the system comprising the following steps:

[0007] The data acquisition module is used to install various sensors on the ship to collect real-time ship status data, including the liquid level and pressure of the ship's ballast tanks, the weight of the cargo, and the ship's inclination angle, as well as port environment data, and transmit the collected data to the data processing module;

[0008] The data processing module is used to pre-process the collected raw data, including cleaning, normalization, and feature extraction, and input the pre-processed data into the large model;

[0009] The large model training and optimization module is used to train the large model using a large amount of historical data and optimize the large model using a reinforcement learning algorithm, so that the large model can predict the optimal ballast water control instructions based on the current ship status and port environment data;

[0010] The decision-making module is used by the large model to output ballast water loading or unloading instructions based on the input data. The decision-making module also considers the operator's instructions to achieve human-machine collaborative control. When the operator issues an instruction, the large model will evaluate and adjust the instruction based on the actual flow rate and liquid level changes of the ballast water;

[0011] The execution module is used for the ballast water equipment to receive instructions from the decision module and automatically adjust the ballasting or unloading process of the ballast water. The execution module has a feedback mechanism that can monitor the actual flow and liquid level changes of the ballast water in real time and transmit the feedback information back to the decision module so that the large model can make real-time adjustments.

[0012] Furthermore, in the data acquisition module, liquid level sensors and pressure sensors are installed in each ballast water tank of the ship, weight sensors are installed in the cargo loading and unloading area, and tilt sensors are installed around the ship. Port environmental data is obtained through a weather station and a port data system. All sensors and data interfaces transmit data to the data processing module through wired or wireless communication.

[0013] The port environment data includes but is not limited to wind speed, wind direction and tide.

[0014] Furthermore, in the data processing module, a sliding average filtering algorithm is used to smooth the liquid level and pressure data;

[0015] For cargo weight and ship inclination angle, a statistical method is used to detect and remove outliers;

[0016] After data cleaning, all data were normalized, and principal component analysis (PCA) was used to reduce the dimensionality of the normalized data.

[0017] Furthermore, in the large model training and optimization module, sensor data and corresponding ballast water control instructions for different ship types and different cargo loading and unloading scenarios are collected as historical data, the historical data are labeled and classified, a large model is constructed using a deep learning framework, and a reinforcement learning algorithm is used to train and optimize the large model.

[0018] Furthermore, in the decision-making module, the large model is constructed based on a multi-layer neural network, and the relevant parameters of the ballast water control instructions are output. The decision-making module calculates the evaluation value of the operator's instructions through an evaluation function. When the evaluation value exceeds the set threshold, a prompt is issued to the operator and a recommended instruction is provided. The decision-making module uses an error correction mechanism to dynamically adjust the instructions in real time according to the feedback information of the execution module.

[0019] Furthermore, in the execution module, the ballast water equipment includes but is not limited to water pumps and valves, and receives instructions issued by the decision module through the control system, and the instructions include at least ballast water flow instructions;

[0020] The control system uses a fuzzy adaptive PID control algorithm to adjust the speed of the water pump, calculates the target valve opening according to the ballast water flow command and the pressure difference before and after the valve, and controls the valve opening through an electric actuator. The actual flow and liquid level changes of the ballast water are monitored in real time through flow sensors and liquid level sensors installed on the ballast water pipeline, and the feedback information is transmitted back to the decision module.

[0021] Compared with the prior art, the advantages of the present invention are:

[0022] (1) The present invention uses a large model trained on a large amount of historical data and optimized through reinforcement learning to perform precise analysis and decision-making, accurately adjusting the ballast water according to the real-time status of the ship and environmental information, effectively avoiding ship stability problems caused by human subjectivity and limitations (such as fatigue, lack of experience, etc.) and misjudgment, thereby greatly improving the safety of ships during loading and unloading.

[0023] (2) The present invention realizes the automation and intelligence of ballast water management. The data acquisition module automatically collects data in real time, the data processing module automatically cleans, pre-processes and extracts features, the large model automatically analyzes and makes decisions, and the execution module automatically adjusts the ballast water, reducing the manual operation of operators, reducing labor intensity, shortening the time ships stay in ports, and thus improving shipping efficiency;

[0024] (3) The present invention uses a large model to learn complex real-world situations such as different ship types, different cargo loading and unloading scenarios, and different port environments. It utilizes a multi-layer neural network structure built with a deep learning framework and reinforcement learning optimization to adapt to various changes, demonstrating strong adaptability and flexibility.

[0025] (4) The present invention utilizes the decision-making module to fully utilize the intelligent decision-making capabilities of the large model while retaining the operator's control. The evaluation function assesses the rationality and safety of the operator's instructions and provides recommended instructions when the instructions are unreasonable, thus achieving human-machine collaborative operation and improving system reliability and stability. At the same time, the feedback mechanism of the execution module allows the decision-making module to dynamically adjust instructions in real time based on feedback, ensuring accurate and stable ballast water regulation. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0027] Figure 1 It is a schematic diagram of the overall process flow of the present invention;

[0028] Figure 2 This is a diagram of the large model training and optimization module architecture of the present invention;

[0029] Figure 3 This is a flow chart of the decision module of the present invention. DETAILED DESCRIPTION

[0030] To achieve the above objectives, the present invention is implemented through the following technical solutions.

[0031] Reference Figure 1 As shown, the present invention is a ship ballast water ballast unloading control system based on a large model. The system consists of main parts such as a data acquisition module, a data processing module, a large model training and optimization, a decision module, and an execution module.

[0032] Ultrasonic level sensors are installed in each ballast tank of the ship. The measurement principle is based on the propagation time of ultrasonic waves in liquids. The calculation formula for the liquid level h is: The propagation speed v of ultrasound in liquid is affected by factors such as the density ρ and temperature T of the liquid, and can be calculated using the empirical formula v = v0 + a(T-T0) + bρ, where v0 is the speed of sound at the reference temperature T0, and a and b are constants.

[0033] Installed at the bottom of the ballast tank, it measures the ballast water pressure (P) according to the liquid pressure formula P = ρgh, where ρ is the density of the ballast water, g is the acceleration due to gravity, and h is the liquid level. The pressure sensor's measured value, combined with the known ballast water density, can verify the accuracy of the level sensor's measurements.

[0034] Install a strain gauge weight sensor in the cargo loading and unloading area. Its working principle is based on the fact that the resistance change of the strain gauge is proportional to the stress, that is, Assume that the initial resistance of the strain gauge is R0, the resistance change after being subjected to force is ΔR, the strain coefficient is K, and the stress is σ. By measuring the resistance change, the weight W of the cargo can be calculated.

[0035] Accelerometer-based tilt sensors are installed at the bow, stern, and midship key locations to measure the ship's tilt angles θ in the x and y directions. x and θ y The acceleration component a output by the accelerometer x and a y The relationship with the tilt angle is

[0036] Obtain the wind speed V at the port through the weather station w , wind direction α w The wind speed sensor uses a three-cup anemometer, and its output pulse frequency f is proportional to the wind speed V w In a linear relationship, that is, V w =kf+c, where k and c are calibration constants. The wind direction sensor uses a wind vane, and the angle signal it outputs corresponds to the wind direction α w .

[0037] Obtain tide data through the port data system, tide height H t As time t changes, the tidal change curve H can be fitted by historical tidal data. t =f(t) in order to predict the tide height in real time.

[0038] All sensors and data interfaces transmit data to the data processing module via wired or wireless communication.

[0039] Adopt industrial Ethernet protocols, such as Modbus TCP, to ensure the stability and reliability of data transmission.

[0040] Low-power wide area network protocols such as LoRaWAN or ZigBee are used, which are suitable for distributed sensor nodes on ships. Assuming the amount of data sent by the sensor node is D and the communication rate is r, the data transmission time is To ensure real-time data transmission, the data transmission cycle must be reasonably arranged according to the sensor data update frequency and communication rate. At the same time, the AES encryption algorithm is used to encrypt the transmitted data to prevent data from being stolen or tampered with during transmission.

[0041] The sliding average filter algorithm is used for smoothing. Assume that the original data sequence collected is x1, x2, ..., x n , the sliding window size is m, then the data y after sliding average filtering i for Where i=1,2,…,n, when j<1 or j>n, x j The value can be determined based on the boundary processing strategy (such as mirror filling).

[0042] Use statistical methods to detect and remove outliers. Let the mean of the data be μ and the standard deviation be σ. For a given threshold k, if a data point x satisfies |x - μ| > kσ, it is considered an outlier and removed.

[0043] Use the minimum-maximum normalization method to map the data to the [0,1] interval. Let the original data be x and the normalized data be x', then where x min and x max The same method is used to process port environmental data (such as wind speed, wind direction, tide, etc.).

[0044] The principal component analysis (PCA) method is used. Assume that the original data matrix is Where n is the number of samples and p is the number of features.

[0045] First, the data is centralized. Where 1 is an n-dimensional all-one vector and μ is the mean vector of the data.

[0046] Then calculate the covariance matrix Perform eigenvalue decomposition on the covariance matrix S and obtain the eigenvalues ​​λ1≥λ2≥…≥λ p and the corresponding eigenvectors u1,u2,…,u p .

[0047] Select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the projection matrix U=[u1,u2,…,u k ]. The centralized data Projected onto U, we get the reduced dimension feature matrix

[0048] like Figure 2 As shown in Figure 2, the large model training and optimization module collects sensor data and corresponding ballast water control instructions for different ship types and different cargo loading and unloading scenarios as historical data. Let the historical data set be D = {d1, d2, ..., d n}, each sample d i =(s i ,c i ), contains the sensor data vector s i and ballast water control command vector c i .

[0049] The sliding average filter algorithm is used to remove noise. Assume that the original data sequence is x1, x2,…, x m , the sliding window size is k, then the data after sliding average filtering Where j ≥ k.

[0050] Unify the data of different ranges to the [0,1] scale. For data x, the normalized data where x min and x max are the minimum and maximum values ​​of the data, respectively.

[0051] A multi-layer neural network is constructed using a deep learning framework such as TensorFlow as a large model. Assume that the number of nodes in the input layer is m, which corresponds to the number of features of the processed sensor data. The number of hidden layers is l, and the number of nodes in the hth hidden layer is n. h , the number of nodes in the output layer is p, which corresponds to the number of parameters of the ballast water control instructions.

[0052] The weight matrix from the input layer to the first hidden layer is The bias vector is The weight matrix from the hth hidden layer to the h+1th hidden layer is The bias vector is The weight matrix of the output layer is The bias vector is

[0053] For an input vector s, the output of the first hidden layer is h*1 = σ(W*1s+b*1), where σ is an activation function, such as the ReLU function, and σ(z) = max(0,z). The output of the hth hidden layer is h*h = σ(W*hh*h-1+b*h), and the output of the output layer is c = W*l+1h*l+b*l+1.

[0054] Deep Q-Network (DQN) is used to train and optimize large models.

[0055] Define the state space Ship status and environment information, action space is the ballast or unloading instruction of ballast water. Let Q(s,a) be the state Take action Q value.

[0056] The goal of DQN is to learn an optimal Q function Q * (s,a), so that the cumulative discounted reward is maximized. Let the reward function r(s,a,s') represent the reward obtained when taking action a in state s to transfer to state s', and the discount factor is γ∈[0,1].

[0057] Use the experience replay mechanism to store the agent's experience (s, a, r, s') in the experience replay buffer During training, A batch of experience data is randomly sampled for learning.

[0058] Loss Function where y=r+γmax*a'Q(s',a';θ - ), θ is the parameter of the current network, θ - The parameters of the target network are periodically copied from the current network. The network parameters θ are updated using the gradient descent method, i.e. θ←θ-αabla θ L(θ), where α is the learning rate.

[0059] The dataset is divided into a training set and a test set. The model is trained on the training set and evaluated on the test set. The mean square error (MSE) is used as the evaluation metric. The number of samples in the test set is N, and the predicted value is The true value is c*i, then By continuously optimizing the model structure and parameters, the MSE is reduced and the prediction accuracy and generalization ability of the model are improved.

[0060] like Figure 3 As shown, the decision module receives the data x=[x1, x2, ..., x n ], feature transformation and inference are performed through multi-layer neural networks. Let the input of the lth layer be h (l-1) , the output is h (l) , then h (l) =σ(W (l) h (l-1) +b (l) ), where W (l) is the weight matrix of layer l, b (l) is the bias vector of the lth layer, σ is the activation function, such as the ReLU function σ(z)=max(0,z). After multiple layers of transformation, the output layer outputs the relevant parameter vector y=[y1,y2,…,y m ].

[0061] The operator can input the command u=[u1,u2,…,u m The decision module introduces the evaluation function E(u,y * ) Evaluate the rationality and safety of the instruction, where y * It is the optimal instruction obtained by the large model based on the current data. Evaluation function E(u,y * )=α∥uy * ∥*2 2 +β∑*i=1 m γ i f i (u i ), ∥·∥*2 is the Euclidean norm, α and β are weight coefficients, γ i is the safety constraint weight of the i-th parameter, f i (ui ) is the safety constraint function of the i-th parameter. For example, for the ballast water flow parameter u1, the safety constraint function where u 1min and u 1max are the lower and upper safety limits of ballast water flow respectively. * ) exceeds a certain threshold θ, the decision module determines that the instruction is unreasonable, issues a prompt to the operator, and provides a recommended instruction y based on the results of the large model suggest =λy * +(1-λ)u, where λ is a weight factor, ranging from [0,1], and is adjusted according to actual conditions.

[0062] The decision module continuously receives feedback information from the execution module r=[r1,r2,…,r k ]. Using the error correction mechanism, let the error vector be e=yr, and the updated instruction y new =y + ηe, where η is the learning rate, which controls the adjustment step size. By adjusting the instructions based on feedback information, the ballast water regulation process is ensured to be accurate and stable, adapting to real-time changes in ship status and port environment.

[0063] The ballast water execution module is composed of water pumps, valves and other equipment, and communicates with the decision module through an industrial-grade programmable logic controller (PLC). The PLC receives the output of the decision module including the ballast water flow rate Q through industrial Ethernet or field bus (such as Profibus). d and valve opening u and other key parameters to ensure that the instructions are accurately transmitted to the execution equipment.

[0064] The pump speed n is proportional to the ballast water flow Q, that is, The water pump speed is regulated by the fuzzy adaptive PID control algorithm. The traditional PID controller control law is: Where u(t) is the controller output, K p , K i and K d are proportional, integral and differential coefficients respectively, e(t)=Q d -Q is the error between the set flow rate and the actual flow rate. The fuzzy adaptive PID control algorithm is based on the error e(t) and the error change rate. Adjust the three parameters of PID controller in real time. The input of fuzzy controller is error e and error change rate The output is ΔK p , ΔK i and ΔK d , used to correct K p , K i and K d , the corrected parameter is Kp =K p0 +ΔK p , K i =K i0 +ΔK i , K d =K d0 +ΔK d , where K p0 , K i0 and K d0 are the initial PID parameters.

[0065] The valve opening u affects the ballast water flow resistance and flow rate, and the flow coefficient C of the linear flow characteristic valve v It is proportional to the valve opening u, that is, C v =C vmax u, where C vmax is the flow coefficient when the valve is fully open. According to the Bernoulli equation and the continuity equation, the ballast water flow Where ΔP is the pressure difference before and after the valve, ρ is the density of ballast water. The control system is based on the flow command Q d Calculate the required flow coefficient C based on the current pressure difference ΔP before and after the valve vreq , and then get the valve target opening u req The valve is driven by an electric actuator, and the PLC controls the action of the electric actuator by outputting an analog signal (such as 4-20mA) to make the valve opening reach the target value.

[0066] High-precision flow sensors and liquid level sensors are installed on the ballast water pipeline to monitor the actual ballast water flow Q and liquid level h in real time, and transmit the signal to the PLC through the analog input module. The PLC compares the actual flow Q with the set flow Q d Compare the calculated error e(t), and at the same time according to the liquid level change rate Flow control is corrected. When the error e(t) exceeds the set threshold, the decision module adjusts the command in real time based on the feedback information. For example, if the actual flow rate is less than the set flow rate, the pump speed is increased or the valve opening is widened; otherwise, the valve opening is reduced. This feedback mechanism ensures that the ballast water regulation process is accurate and stable, meeting the real-time ballast water needs of ships loading and unloading cargo.

[0067] In summary, this embodiment combines advanced large model technology, multi-sensor data fusion and intelligent control algorithm to form a complete set of ship ballast water intelligent control solutions. Figure 1 The system flow shown in this paper effectively enhances the automation and intelligence of ballast operations, enabling rapid, economical, and compliant ballast water management. The decision-making capabilities based on large models also lay the foundation for integrating more intelligent ship functions in the future.

[0068] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A ship ballast water unloading control system based on a large model, characterized in that: The system comprises the following steps, The data acquisition module is used to install various sensors on the ship to collect real-time ship status data, including the liquid level and pressure of the ship's ballast tanks, the weight of the cargo, and the ship's inclination angle, as well as port environment data, and transmit the collected data to the data processing module; The data processing module is used to pre-process the collected raw data, including cleaning, normalization, and feature extraction, and input the pre-processed data into the large model; The large model training and optimization module is used to train the large model using historical data and optimize it using a reinforcement learning algorithm, enabling the large model to predict the optimal ballast water control instructions based on the current ship status and port environment data; The decision-making module is used by the large model to output ballast water control instructions based on the input data. The decision-making module also considers the operator's instructions to achieve human-machine collaborative control. When the operator issues an instruction, the large model will evaluate and adjust the instruction based on the actual flow and level changes of the ballast water; The execution module is used for the ballast water equipment to receive instructions from the decision module and automatically adjust the ballasting or unloading process of the ballast water. The execution module has a feedback mechanism that can monitor the actual flow and liquid level changes of the ballast water in real time and transmit the feedback information back to the decision module so that the large model can make real-time adjustments.

2. The system according to claim 1, wherein: In the data acquisition module, liquid level sensors and pressure sensors are installed in each ballast water tank of the ship, weight sensors are installed in the cargo loading and unloading area, and tilt sensors are installed at the bow, stern and midship positions of the ship. Port environmental data is obtained through the weather station and port data system. All sensors and data interfaces transmit data to the data processing module through wired or wireless communication; The port environment data includes but is not limited to wind speed, wind direction and tide.

3. The system according to claim 1, wherein: In the data processing module, a sliding average filtering algorithm is used to smooth the liquid level and pressure data; For cargo weight and ship inclination angle, a statistical method is used to detect and remove outliers; After data cleaning, all data were normalized, and principal component analysis (PCA) was used to extract features from the normalized data.

4. The system according to claim 1, wherein: In the large model training and optimization module, sensor data and corresponding ballast water control instructions under different ship types and different cargo loading and unloading scenarios are collected as historical data, the historical data are labeled and classified, a large model is constructed using a deep learning framework, and a reinforcement learning algorithm is used to train and optimize the large model.

5. The system according to claim 1, wherein: In the decision-making module, the large model is constructed based on a multi-layer neural network and outputs the relevant parameters of the ballast water control instructions. The decision-making module calculates the evaluation value of the operator's instructions through an evaluation function. When the evaluation value exceeds the set threshold, a prompt is issued to the operator and a recommended instruction is provided. The decision-making module uses an error correction mechanism to dynamically adjust the instructions in real time based on the feedback information from the execution module.

6. The system according to claim 1, wherein: In the execution module, the ballast water equipment includes but is not limited to water pumps and valves, and receives instructions from the decision module through the control system, and the instructions include at least ballast water flow instructions; The system uses a fuzzy adaptive PID control algorithm to adjust the speed of the water pump, calculates the target valve opening according to the ballast water flow command and the pressure difference before and after the valve, and controls the valve opening through an electric actuator. The actual flow and liquid level changes of the ballast water are monitored in real time through flow sensors and liquid level sensors installed on the ballast water pipeline, and the feedback information is transmitted back to the decision-making module.

Images